Citation
Application of computer vision and electronic nose technologies for quality assessment of color and odor of shrimp and salmon

Material Information

Title:
Application of computer vision and electronic nose technologies for quality assessment of color and odor of shrimp and salmon
Creator:
Luzuriaga, Diego Andres, 1970-
Publication Date:
Language:
English
Physical Description:
xx, 268 leaves : ill. ; 29 cm.

Subjects

Subjects / Keywords:
Ammonia ( jstor )
Colors ( jstor )
Computer vision ( jstor )
Electronics ( jstor )
Nose ( jstor )
Odors ( jstor )
Sensors ( jstor )
Shrimp ( jstor )
Storage time ( jstor )
Sulfites ( jstor )
Dissertations, Academic -- Food Science and Human Nutrition -- UF ( lcsh )
Food Science and Human Nutrition thesis, Ph.D ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis (Ph.D.)--University of Florida, 1999.
Bibliography:
Includes bibliographical references (leaves 253-267).
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Diego Andres Luzuriaga.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. §107) for non-profit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
Resource Identifier:
030477897 ( ALEPH )
81792845 ( OCLC )

Downloads

This item has the following downloads:


Full Text
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ESNOZGLRW_DMWVAG INGEST_TIME 2015-03-09T20:31:06Z PACKAGE AA00028860_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES



PAGE 1

APPLICATION OF COMPUTER VISION AND ELECTRONIC NOSE TECHNOLOGIES FOR QUALITY ASSESSMENT OF COLOR AND ODOR OF SHRIMP AND SALMON By DIEGO ANDRES LUZURIAGA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1999

PAGE 2

To my wife, Elena, for all her love, encouragement, and support. To my daughter, Maria Victoria, for precious moments during the preparation of this work. To my parents, Oscar and Cecilia, for the invaluable opportunity they gave me throughout my life to obtain the best education possible.

PAGE 3

ACKNOWLEDGMENTS I would like to express my sincere thanks and appreciation to my advisor, Dr. Murat O. Balabaa Throughout my studies at the University of Florida, Dr. Balaban gave me an invaluable amount of advice, guidance, and friendship. Without his help this project could not be achieved. Moreover, Dr. Balaban helped me in all aspects of my career and encouraged me to participate in other activities, learn other disciplines, network with people around the world, and develop my teaching skills. He is the main reason for most of my achievements and successes in my academic career. Appreciation is also extended to Drs. O'Keefe, Portier, Sims, and Teixeira for their guidance and recommendations for the success of this study. I want to thank Sea Grant for partial funding of this project (Project No. R/LR-Q17). Gratitude is extended to Walter Staruszkiewicz and Jim Barnett from the Food and Drug Administration for allowing me to participate in the FDA Decomposition Workshops, and for providing shrimp samples for this study. Finally, my sincere grateful esteem to the students and friends in the laboratory: Ferruh Erdogdu, Figen Korel, and Asli Odabasi, for their help in running some of the experiments. Also my appreciation to the panelists that helped in the sensory studies. iii

PAGE 4

TABLE OF CONTENTS page ACKNOWLEDGMENTS iii LIST OF TABLES vii LIST OF FIGURES xiii ABSTRACT xix CHAPTERS 1 INTRODUCTION 1 Importance of Shrimp and Salmon in the United States 1 Current Quality Evaluation of Seafood Products 2 Importance of Rapid Sensory Methods for Quality Evaluation of Seafood 3 Objectives 4 Visual Analysis 4 Odor Analysis 5 2 LITERATURE REVIEW 6 Quality Evaluation of Seafood Products 6 Visual Attributes 6 Odor Attributes 14 Computer Vision 22 Image Processing 23 Applications of computer vision 24 Color Systems 26 RGB Color System 31 Electronic Nose 33 Electronic Nose Technology 35 Sensor Technology 38 Applications to Food Products 40 Pattern Recognition Techniques 42 iv

PAGE 5

Current Shortcomings and Future Improvements of the Electronic Nose 46 3 COLOR ANALYSIS: SOFTWARE DEVELOPMENT 48 Introduction 48 Hardware Used in the Color Computer Vision System 50 Software Development 51 Image Acquisition 52 Image Processing 56 Data Display 69 Color Data Analysis 77 4 COLOR AND MELANOSIS EVALUATION ON DIFFERENT SPECIES OF SHRIMP STORED AT DIFFERENT TEMPERATURES USING COMPUTER VISION 80 Introduction 80 Materials and Methods 82 Shrimp samples 82 Color Analysis 83 Data Analysis 84 Results and Discussion 85 Conclusions 108 5 AUTOMATION AND CORRELATION OF COLOR AND ODOR EVALUATION OF SALMON 110 Introduction 110 Materials and Methods 113 Salmon Samples and Storage Conditions 113 Moisture Content and Water Activity Measurements 114 Sensory Evaluation 114 Electronic Nose Measurements 115 Color Analysis 117 Data Analysis 118 Results and Discussion 121 Conclusion 146 6 EVALUATION OF DECOMPOSITION ODOR IN RAW AND COOKED SHRIMP: CORRELATION BETWEEN ELECTRONIC NOSE READINGS, ODOR SENSORY EVALUATION, AND AMMONIA LEVELS 148 v

PAGE 6

Introduction 148 Materials and Methods 151 Shrimp Samples 151 Sensory Evaluation 153 Electronic Nose Measurements 154 Ammonia Analysis 155 Data Analysis 155 Results and Discussion 156 Conclusion 180 7 USE OF THE ELECTRONIC NOSE TO DETECT CHEMICALS USED IN SHRIMP 188 Introduction 188 Materials and Methods 191 Shrimp Samples 191 Chemicals Used and Sample Treatments 1 92 Electronic Nose Measurements 193 Sensory Evaluation 194 Ammonia Analysis 195 Moisture Content and Water Activity Measurements 195 pH Measurements 196 Microbial Analysis 196 Data Analysis 197 Results and Discussion 197 Conclusions 214 8 CONCLUSIONS 229 APPENDICES A COLOR ANALYSIS SOFTWARE 235 B COLOR DATA FOR SALMON 237 C DATA FOR SHRIMP TREATED WITH DIFFERENT CHEMICALS 239 REFERENCES 253 BIOGRAPHICAL SKETCH 268 vi

PAGE 7

LIST OF TABLES Table page 2-1. Scale used to describe and rate occurrence of melanosis (blackspot) on shrimp . . 8 2-2. Shrimp sensory standards used by USDC inspectors 15 2-3. Salmon sensory standards used by USDC inspectors 16 2-4. Odor thresholds of important nitrogenous compounds 19 2-5. Classes of shrimp decomposition defined by the FDA 22 2-6. The most common color systems used in color technology 28 27. Common pattern recognition techniques used with electronic nose data 43 31 . Parameters calculated for each blob by the color analysis program 61 32. Comparison of known L*a*b* values of color standards with L*a*b* measurements from a colorimeter and the color machine vision system developed in this study 68 41 . Average color values (% of total view area of shrimp) for different species of shrimp. Color was measured immediately after thawing the shrimp (Day 0). (n=75) 88 4-2. Average L*a*b* values for different species of shrimp, as measured by the color machine vision system immediately after thawing the shrimp (day 0). RGB values and color names are given, (n = 75, ± = standard deviation) .... 88 4-3. Color changes in brown shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n = 24) 91 vii

PAGE 8

4-4. Color changes in pink shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n = 25) 93 4-5. Color changes in tiger shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n = 25) 95 4-6. Color changes in white shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n = 25) 97 4-7. Average L, a and b values for brown and pink shrimp during storage at different temperatures (n = 25) 100 4-8. Average L, a and b values for tiger and white shrimp during storage at different temperatures (n = 25) 101 4-9. Color blocks identified as melanosis in different species of shrimp 103 410. Amount of melanosis (average % of the total view area of shrimp) in different species of shrimp during storage at different temperatures (n = 25) 105 51 . Moisture content (% wet basis) and water activity (% relative humidity) of salmon fillets during storage at different temperatures (Experiment 1, Summer). (± = std. deviation, n = 3) 121 5-2. Moisture content (% wet basis) and water activity (% relative humidity) of salmon fillets during storage at different temperatures (Experiment 2, Fall). (± = std. deviation, n = 3) 122 5-3. Sensory scores for odor, color and overall evaluation of fresh salmon fillets stored at different temperatures. Sensory score of 1 = good, 10 = bad .... 123 5-4. Color blocks whose levels were above 2% of the total area of the samples analyzed. Color blocks with a / represent colors included in the combined data set (electronic nose and color) or when color alone was used to obtain the DF A models 126 5-5. Correctly classified cases obtained from the classification matrix for the DFA of electronic nose readings, color data and the combination of both when correlated with odor, color and overall sensory grades, respectively. (Values are % of correctly classified samples) 128 viii

PAGE 9

5-6. Correctly classified cases obtained from the classification matrix for the DFA of electronic nose readings, color data, and combination of both when correlated with storage time. (Values are % of correctly classified samples) . 135 5-7. Statistical significance (p-values) from DFA of the electronic nose sensors contribution to the prediction of group membership 138 5-8. Parameters for equation 5-1, fitted to color data of salmon stored at different temperatures (Experiment 1 , Summer) 139 59. Parameters used for the linear fit to calculate energy of activation and kg for color changes in salmon stored at different temperatures (Experiment 1, Summer) 140 61 . DFA coefficients for sensory evaluation scores correlated to electronic nose sensor readings. Shrimp samples collected from decomposition workshops (intact samples) 158 6-2. DFA coefficients for sensory evaluation scores correlated to electronic nose sensor readings. Shrimp samples collected from decomposition workshops (chopped samples) 159 6-3. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with sensory grades of shrimp samples collected from decomposition workshops 160 6-4. DFA coefficients for sensory evaluation scores correlated to electronic nose sensor readings. Different species of shrimp stored at 2°C for 14 days 171 6-5. DFA coefficients for storage time correlated to electronic nose sensor readings. Different species of shrimp stored at 2°C for 14 days 172 6-6. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with sensory grades for different species of shrimp stored at 2°C for 14 days 173 6-7. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with storage time for different species of shrimp stored at2°Cfor 14 days 179 6-8. Ammonia levels (ppm) and sensory scores of different species of raw and cooked shrimp stored at 2°C for 14 days 186 ix

PAGE 10

7-1 . Moisture content of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite (average of three samples) 200 7-2. Water activity (% relative humidity) of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite 201 7-3. Microbial load of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite (average of two samples) 202 7-4. pH of shell-on pink shrimp treated with different levels of bleach, phosphate, and sulfite (average of two samples) 204 7-5. Ammonia levels of shell-on pink shrimp treated with different levels of phosphate and sulfite (average of two samples) 205 7-6. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of bleach. Superscripts denote statistical significance among the means at the p-level of 0.05 206 7-7. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of phosphate. Superscripts denote statistical significance among the means at the p-level of 0.05 206 7-8. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of sulfite. Superscripts denote statistical significance among the means at the p-level of 0.05 207 7-9. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by storage time (both replicates together). DFA models were obtained for each chemical. (Results for Figure 7-1) 210 7-10. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of bleach used to treat shrimp. DFA models were obtained for each measurement time 212 7-11. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of phosphate used to treat shrimp. DFA models were obtained for each measurement time 212 7-12. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of sulfite used to treat shrimp. DFA models were obtained for each measurement time 213 x

PAGE 11

7-13. Eigenvalues for the first two factors obtained by PC A for the combined data set of replicates for each chemical at each storage time 213 7-14. DFA coefficients for bleach concentration correlated to electronic nose sensor readings 224 7-15. DFA coefficients for phosphate concentration correlated to electronic nose sensor readings 225 7-16. DFA coefficients for sulfite concentration correlated to electronic nose sensor readings 226 7-17. Statistical significance (pvalues) from DFA of the electronic nose sensors contribution to the prediction of group membership 227 A1 . Description of the 64-color block scheme used in the Color Analysis software 235 B-l . Color data of samples of salmon fillets stored at different temperatures (first experiment: Summer) 237 B-2. Color data of samples of salmon fillets stored at different temperatures (second experiment: Fall) 238 C-l. Moisture content of shell-on shrimp treated with different levels of bleach . . . 239 C-2. Moisture content of shell-on shrimp treated with different levels of phosphate 240 C-3. Moisture content of shell-on shrimp treated with different levels of sulfite ... 241 C-4. Microbial load of shell-on shrimp treated with different levels of bleach 242 C-5 . Microbial load of shell-on shrimp treated with different levels of phosphate . . 243 C-6. Microbial load of shell-on shrimp treated with different levels of sulfite 244 C-7. pH of shell-on shrimp treated with different levels of bleach 245 C-8. pH of shell-on shrimp treated with different levels of phosphate 246 C-9. pH of shell-on shrimp treated with different levels of sulfite 247 xi

PAGE 12

C-10. Ammonia levels of shell-on shrimp treated with different levels of phosphate .248 C-l 1 . Ammonia levels of shell-on shrimp treated with different levels of sulfite .... 249 C-12. Sensory data for shrimp treated with different levels of bleach solutions 250 C-l 3. Sensory data for shrimp treated with different levels of phosphate solutions . . 25 1 C-l 4. Sensory data for shrimp treated with different levels of sulfite solutions 252 xii

PAGE 13

LIST OF FIGURES Figure page 2-1. CIE 1931 chromaticity diagram, Yxy color space. Source: Minolta Corp 29 2-2. RGB color system 31 23. Schematic of sensor head and sample vessel of an electronic nose with static sampling 36 31 . Main steps of the color analysis program to obtain information from a color image 51 3-2. Main screen of the color analysis software 53 3-3. Options for color data analysis screen 54 3-4. Screen for video camera adjustments 54 3-5. Screen of an image opened from a file or captured from a video camera 55 3-6. Screen for selection of the binary threshold 58 3-7. Screen of a binary image 58 3-8. Options for blob analysis screen 60 3-9. Screen of an image after blob analysis and object identification 60 3-10. Screen of a blob analysis report 62 3-11. Screen with regions of interest selected by the user 63 3-12. Grouping of color blocks in the RGB color space. Source: Precetti (1995) ... 64 xiii

PAGE 14

3-13. Options for color scheme selection screen 64 3-14. Options for color calibration screen 66 3-15. Screen with color data report 72 3-16. Screen with L*a*b* color information results 72 3-17. Plot color data selection screen 73 3-18. Screen with a histogram of color data 73 3-19. Screen with the description of the 64-color block scheme 74 3-20. Screen with histograms of red, green and blue color intensities 75 3-21. Color conversion utility screen .' 76 322. Color conversion utility screen interacting with a color image 77 41 . Discrimination of shrimp species based on the 64-color block data, n = 75 shrimp per cluster 85 4-2. Color profiles of different species of shrimp immediately after thawing. Bars = average of 75 shrimp, whiskers = ± 1 standard deviation 87 4-3. Color changes of brown shrimp during storage at different temperatures, points represent the average color of 24 shrimp 90 4-4. Color changes of pink shrimp during storage at different temperatures, points represent the average color of 25 shrimp 92 4-5. Color changes of tiger shrimp during storage at different temperatures, points represent the average color of 25 shrimp 94 4-6. Color changes of white shrimp during storage at different temperatures, points represent the average color of 25 shrimp 96 4-7. Average L*a*b* values of different species of shrimp stored at different temperatures, measured by the color machine vision system, n= 25 102 4-8. Development of melanosis in brown and pink shrimp during storage at different temperatures. Line represents a linear fit 106 xiv

PAGE 15

49. Development of melanosis in tiger and white shrimp during storage at different temperatures. Line represents a linear fit 107 51 . Portion of salmon fillet used in the experiments. Black square is region of interest to extract color information 117 5-2. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades correlated with electronic nose and color data (Experiment 1, Summer) 124 5-3. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades correlated with electronic nose and color data (Experiment 2, Fall) 125 5-4. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades with electronic nose alone, color data alone and combination of both. (Experiment 1, Summer) 129 5-5. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades with electronic nose alone, color data alone and combination of both. (Experiment 2, Fall) 130 5-6. DFA model of salmon fillets overall quality correlated with electronic nose and color data combined, with the validation data set from variable storage temperature studies 1 34 5-7. DFA of electronic nose and color data vs storage time of salmon fillets (Experiment 1, Summer) 136 5-8. DFA of electronic nose and color data vs storage time of salmon fillets (Experiment 2, Fall) 137 5-9. Color changes vs time for salmon stored at different temperatures 141 5-10. Measured and predicted colors of salmon fillets stored under variable storage temperature conditions 142 5-11. Correlation of sensory scores with color from the flesh of salmon fillets 143 5-12. L*a*b* values from flesh of salmon fillets during storage at different temperatures (Experiment 1, Slimmer) 144 xv

PAGE 16

513. L*a*b* values from flesh of salmon fillets during storage at different temperatures (Experiment 2, Fall) 145 61 . DFA of raw Mexican brown shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 161 6-2 DFA of raw Thai pink shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 162 6-3. DFA of raw Thai tiger shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 163 6-4. DFA of Ecuadorian white shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 164 6-5. DFA of raw Mexican white shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 165 6-6. DFA of raw Ecuadorian white shrimp odor based on sensory grades and electronic nose readings (including shrimp contaminated with diesel). Samples obtained from FDA workshops 167 6-7. DFA of all shrimp species odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops 168 6-8. Sensory grades versus ammonia levels of raw shrimp samples (All species pooled together, samples obtained from FDA workshops) 169 6-9. DFA of raw and cooked pink shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days 174 6-10. DFA of raw and cooked tiger shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days 175 6-11. DFA of raw and cooked white shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days 176 6-12. DFA of different species of shrimp (white, pink, tiger) odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days ... 177 6-13. DFA of fresh raw and cooked shrimp samples based on electronic nose readings. Day 0 of storage study 178 xvi

PAGE 17

6-14. DFA of raw and cooked pink shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days 181 6-15. DFA of raw and cooked tiger shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days 182 61 6. DFA of raw and cooked white shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days 183 6-17. Ammonia levels of different species of raw and cooked shrimp during storage at 2°C. Lines represent exponential fits 1 84 618. Sensory grades versus ammonia levels of raw and cooked shrimp samples stored at 2°C for 14 days 185 71 . DFA results of the correlation of electronic nose readings and storage time at 2°C of shrimp treated with different chemicals 209 7-2. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 0 hrs of treatment 215 7-3. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 24 hrs of treatment 216 7-4. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 48 hrs of treatment 217 7-5. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 0 hrs of treatment 218 7-6. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 24 hrs of treatment 219 7-7. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 48 hrs of treatment 220 xvii

PAGE 18

7-8. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 0 hrs of treatment 221 7-9. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 24 hrs of treatment 222 71 0. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 48 hrs of treatment 223 xviii

PAGE 19

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy APPLICATION OF COMPUTER VISION AND ELECTRONIC NOSE TECHNOLOGIES FOR QUALITY ASSESSMENT OF COLOR AND ODOR OF SHRIMP AND SALMON By Diego Andres Luzuriaga August 1999 Chairperson: Murat O. Balaban Major Department: Food Science and Human Nutrition Current techniques for shrimp and salmon quality evaluations rely on sensory methods. These procedures are subjective, prone to error, and difficult to quantify. Chemical analyses are seldom used by the seafood industry due to the complexity and length of time these methods require. Automated evaluation of color and odor is desirable to reduce subjectivity and discrepancies and assist with the creation of standards for inspectors worldwide. The objectives of this study were to develop color machine vision techniques for visual evaluation and to test electronic nose sensors for odor assessment of raw and cooked shrimp and fresh raw salmon. A color machine vision system was developed to analyze the color of seafood samples. Hardware consisted of a light box, a video camera, and a frame grabber. Software developed for the Windows environment was able to determine the color of food xix

PAGE 20

samples by discretizing the RGB (red, green, blue) color system into 64, 512, or 4096 colors, and by giving color information in different color scales (RGB, L*a*b*, XYZ, or Munsell). The system was able to analyze the color of samples with non-uniform color surfaces, to predict the amount of melanosis (black spots) in shrimp, and to measure color changes of shrimp and salmon during storage. An electronic nose with twelve conducting polymer sensors was used to measure odors of shrimp and salmon stored at different temperatures, with different levels of spoilage, and treated with different chemicals. Discriminant function analysis was used as the pattern recognition technique to differentiate samples based on odors. Results showed that the electronic nose could discriminate differences in odor due to storage time and spoilage levels for shrimp and salmon, and species and food additives in shrimp. Results also showed good correlation of sensor readings with sensory scores and chemical concentrations. Overall, the electronic nose showed good sensitivity and accuracy. Results from this work could lead to methodologies that will assist in the objective and repeatable quality evaluation of shrimp and salmon. These methods have potential in industrial and regulatory applications where rapid response, no sample preparation, and no need for chemicals are required. Furthermore, expertise in sensory evaluation may be captured and used by instruments. xx

PAGE 21

CHAPTER 1 INTRODUCTION Importance of Shrimp and Salmon in the United States The shrimp industry in the United States (U.S.) is very important, especially in the Gulf and South Atlantic states. In 1997, 71% of the total U.S. landings of shrimp were produced in this region. Shrimp were the most important species in value of the U.S. landings in 1997 and ninth in quantity. During the same year, the United States imported fishery products valued at $7.0 billion, including $2.9 billion worth of shrimp, accounting for 3 7% of the value of total edible imports. Domestic shrimp represented only 1 8% of the total shrimp processed in the U.S. during that year. The U.S. annual per capita consumption of shrimp (all preparations) has increased in the last 20 years from 0.73 to 1.23 Kg (U.S. Department of Commerce, 1998). High-quality fresh salmon is a valuable commodity worldwide. In 1992 the world commercial catch of salmon was 1,467,000 metric tons, which increased to 2,102,000 metric tons in 1996. In 1997 the U.S. commercial landings of salmon represented 12% of the world production, with 257,000 metric tons valued at $270 million, being the fourth fishery product of importance, both in quantity and value. The same year, the U.S. imported $344 million of salmon, the third largest imported fishery product by value. In 1

PAGE 22

2 1997 salmon was the second most important fishery product of export for the U.S., generating $308 million (U.S. Department of Commerce, 1998). Current Quality Evaluation of Seafood Products The increasing consumption and importance of shrimp and salmon in the U.S., makes quality an important factor that should be monitored for a safe and high quality commercial product. Current quality evaluation of most seafood products relies on subjective methods that use the senses of vision, smell and touch. The inspector examines the product for visual signs of quality, such as color, shape, size, etc.; smells to detect any perceptible sign of spoilage, off-odors or presence of chemicals; and touches to feel the texture or structure of the product. This evaluation is subjective, time consuming, susceptible to error, hard to repeat and to compare with standards worldwide. Moreover, chemical analyses are seldom used by the shrimp and salmon industry because of their complexity and length of time required by these methods. This may create discrepancies between the evaluation results of suppliers, buyers and inspection agencies. Sensory tests depend on the observer's ability to accurately estimate quality parameters related with the appearance, odor or texture of the product. The inspector relies on his/her experience and interpretation, and external factors could influence the final decision. Inspectors are subject to fatigue, sickness and loss of sensitivity with time. Furthermore, most procedures for evaluation by the producers, the industry or the government are not standardized. Shrimp and salmon are highly perishable and expensive foods, consequently, their quality should be more carefully analyzed using methods that

PAGE 23

3 ideally would be quantitative, reproducible, fast and non-destructive. Automation of quality evaluation using computers and sensors is a potential alternative for the seafood industry. Importance of Rapid Sensory Methods for Quality Evaluation of Seafood Fresh seafood products have a very short shelf life, and should be marketed as fast as possible. Part of the commercialization process is the grading, evaluation and/or inspection of the product, especially for imported fishery products. This evaluation needs to be accomplished rapidly if the product is to retain its quality and safety, without waiting too long for the results. Most of the seafood industry and regulatory agencies rely on sensory evaluation as the main tool to rapidly evaluate product quality. Therefore, there is a need for objective and rapid methods to evaluate the quality of raw fresh salmon and raw shrimp that can assist in the development of common standards between the industry, regulatory agencies and international markets. Also, there is a need to understand the correlation between sensory evaluation and machine measurements, which is vital for the reliability of rapid predictive assessment of quality attributes. Emerging technologies can provide alternative means and/or aid in the process of evaluating seafood products in a fast manner. Computers and sensors can mimic the inspectors' evaluations, providing more objective approaches, methodologies and results. These systems will not be affected by fatigue and could be used in different settings, locations and with different products. Visual inspection of seafood products can be done by using video cameras and image processing software to grade or evaluate the product

PAGE 24

based on color, appearance, size or shape. Arrays of conducting polymer, metal oxide or surface acoustic wave sensors can be used to detect and quantify the presence of volatiles in a sample. The responses from these sensors can then be correlated to the quality of the product, or could be used as an estimator of spoilage, microbial count, storage time, or degree of contamination by different chemical substances. These technologies, known as 'color machine vision systems' and 'electronic noses,' can provide objective results in minutes or even seconds. The seafood industry, government agencies and ultimately the consumer will benefit from these methodologies because more samples can be analyzed in the same amount of time, with repeatable and reliable results. These systems could be used to standardize some of the methodologies used for shrimp and salmon quality evaluation. Objectives The overall objective of this project was to develop and test new technologies for automated quality evaluation of shrimp and fresh raw salmon, by using color machine vision techniques for visual evaluation, and electronic nose sensors for odor assessment. The specific objectives of the project were: Visual analysis 1 . To develop a color analysis software that could be used to assess the color of foods in the RGB, L*a*b*, XYZ, or Munsell color systems;

PAGE 25

5 2. To measure color changes of different species of shrimp during storage at different temperatures, generate reference color histograms, and quantify the amount of melanosis (black spots); 3. To measure color changes of raw fresh salmon during storage at different temperatures, and develop models that could be used to predict the quality of salmon fillets; Odor analysis 1 . To measure electronic nose sensor response to odor changes of fresh raw salmon fillets stored at different temperatures, and obtain predictive models to estimate their quality; 2. To combine electronic nose sensor data and color data to obtain predictive models for the overall quality of salmon fillets stored at different temperatures. 3. To measure odor changes of raw and cooked shrimp during storage by using an electronic nose; 4. To evaluate the odors of decomposition of raw shrimp using an electronic nose, ammonia analysis, different sample preparations and sensory panels; 5. To measure the response of the electronic nose to food additives of importance in shrimp (sodium hypochlorite, sodium tripolyphosphate and sodium metabisulfite).

PAGE 26

CHAPTER 2 LITERATURE REVIEW Quality Evaluation of Seafood Products Visual Attributes Food products are usually evaluated with the senses of sight, smell, taste and touch. These senses indirectly examine the microbiological, chemical and physical changes that have occurred in the product. The first sense used by consumers when they are exposed to a food product is sight. When consumers look at a food, they will have an impression of the overall quality of the product. Therefore, visual evaluation of foods is very important. Several quality parameters, such as color, shape, size, appearance, etc., can be measured by just looking at the product. In seafood products, color is a good visual indicator of quality. Color of shrimp Shrimp species are classified by their color. Shrimp color depends mainly on that of the shell, and is influenced by age, size, harvest season and location, diet, etc. (Dore and Frimodt, 1987). Despite this variation of shell color, the meat is generally white. White shrimp comes in varying shades of greyishwhite and aqua, with tints of green, blue and 6

PAGE 27

sometimes red. Brown shrimp can be reddish to grey-brown with occasionally blue-purple hues. Pink shrimp vary from light to rose pink and can darken to resemble brown shrimp. Tiger shrimp has a mixture of dark grey colors with tints of green and blue, black stripes found parallel to each segment, and sometimes some yellow and red colors are present in the legs. There is a vast amount of variation in color between species and sometimes among species. For each individual species, the original color can change over time due to storage conditions, storage temperatures, packaging, etc. This color change could be obvious as in the formation of black spots or melanosis, or it could be a gradual change to a different level of grey or pink. The red-orange color after cooking occurs as a carotenoid-protein complex closely associated with oil and is located primarily at the interface between the meat and the shell (Collins and Kelley, 1969). Melanosis, which is only a cosmetic change, is the most important color attribute used in quality evaluation of a shrimp sample. Melanosis in shrimp Melanosis is a type of enzymatic browning that occurs postmortem in some Crustacea. Shrimp melanosis, commonly termed as "black spot", is a surface discoloration caused by enzymatic formation of precursor compounds which can polymerize spontaneously and/or react with cellular constituents to form insoluble pigments (Savagon and Sreenivasan, 1978). This is a cosmetic change which can reduce commercial value and consumer acceptance of shrimp. The endogenous shrimp enzyme, polyphenol oxidase, which catalyzes the initial step in blackspot formation, remains active throughout

PAGE 28

8 post-harvest processing unless the shrimp are frozen or cooked. The polyphenol oxidase activity can resume in raw shrimp upon thawing (McEvily et al., 1991). Evaluation of shrimp melanosis is done by trained experts who examine the shrimp visually. Otwell and Marshall (1986) described a scale from 0 to 10 to rate the occurrence of melanosis on shrimp (Table 2-1). Table 21 . Scale used to describe and rate occurrence of melanosis (blackspot) on shrimp MELANOSIS SCORE DESCRIPTION 0 Absent 2 Slight, noticeable on some shrimp 4 Slight, noticeable on most shrimp 6 Moderate, noticeable on most shrimp 8 Heavy, noticeable on most shrimp 10 Heavy, totally unacceptable Source: Otwell and Marshall, 1986. This rating system is related to the recommendations of the National Marine Fisheries Service for rating of raw shrimp (CFR, 1998). A score of 4 or below corresponds to Grade A, whereas a score between 4 and 8 corresponds to grade B, and represents a measurable defect that results in a devalued product. A score of 8 or above represents an unacceptable product. Melanosis usually begins as a blackening of the membrane which connects the ends of overlapping segments (Alford and Fieger, 1952). It forms in the presence of oxygen and can appear in a few hours after catching the shrimp, or it can be delayed up 2 or 3

PAGE 29

9 days depending on handling procedures. Melanosis in shrimp is characterized by the darkening of the shell, tail, fins, legs, antennae and head (Otwell and Marshall, 1986) and its formation can be inhibited by the use of sulfite compounds (McEvihy et aL, 1991). Visual defects in shrimp Some of the most common visual defects found in shrimp are: • Yellowing: excessive amounts of sulfite often show an unusual yellowing of the underside of shrimp (legs, tail, etc.) as well as a bleached appearance (Ravelo, 1990). • Meat translucency and/or glossiness: the use of polyphosphates, which aids in moisture retention in the product, often shows a glossy appearance of the meat and an increase in the translucency of the shrimp tail. • Improper deheading (excessive throat meat): throats are those portions of flesh and/or extraneous material from the head (cephalothorax) that remain attached to the first segment after deheading (CFR, 1998). Excessive throats are considered a defect. • Pieces of shrimp, broken or damaged shrimp: any portion of a shrimp that contains less than 85% whole segments is considered a piece. As a general rule, the amount of damaged and broken shrimp in a package should be less than 5% by weight, except for salad shrimp (over 70 count) for which 15% is acceptable (Ravelo, 1990). Presence of legs (pleopods), antennae, loose shell: these factors are related to handling of shrimp during packaging. Excessive presence of legs (for peeled

PAGE 30

10 shrimp), antennae and loose shell (for shell-on shrimp) are considered defects. Extraneous material, such as shrimp head, loose shell, attached legs and other nonharmful debris should not exceed one instance per pound of shrimp (Ravelo, 1990). Improperly deveined: the digestive tract of shrimp is the black vein on its back side. This vein is usually removed in peeled shrimp. A portion of vein in the tail section, or any vein less than two segments long is not considered a defect (Ravelo, 1990). Color of salmon Salmon's flesh color is one of the most important attributes used during its quality evaluation. Color depends on the species, genetic differences, the spawning cycle, feed and other environmental factors. Compared to shrimp, flesh color of salmon is not an absolute guide for species differentiation. Sockeye salmon has the reddest flesh, while pink salmon has the palest. Moreover, color of individual fish vary greatly from one another (Dore, 1990). The flesh color is also a measure of the freshness of the fish. Surveys show that consumers equate freshness with the vibrancy of the flesh color (Beaudoin, 1997). Fresh salmon has an attractive vivid pink-orange color. This color will change into dull pink, or even beige during storage. In farm-raised salmon, flesh color could depend on the feed given to the animal. Many salmon farmers use carotenoid compounds in the feed to give the salmon the proper color that will attract the consumer's attention The fish farming industry is working to develop strains of fish which have redder flesh and which absorb

PAGE 31

11 more easily the pigments from their feed. It is expected that salmon farmers will be able to supply quite precise colors as the market requires (Dore, 1990). Markets differ in their color requirements, in Japan they prefer the deep color of sockeye, while in France they prefer paler colors, such as those of Atlantic or coho salmon. Visual defects in salmon The most commonly found visual defects found in salmon are: • Gaping: disruption of the tissue or muscle due to breakdown of the connective tissue. • Presence of parasites: parasites usually cause visible damage in the flesh and can be easily distinguished in the pink flesh by using techniques like candling. • Browning of belly flap tissue: the fillet portion closest to the head of the fish has a white belly flap which is mainly composed of lipids. These lipids will undergo oxidation and will change color to a pale yellow or light brown, which is related to the storage time of the fillet. • Color of gills and eyes: fresh salmon has very bright red gills which turn into a pale brown when the fish has been stored for a certain period of time. The eyes of the fish are shiny and turgid when fresh, but they become pale and depressed when they start to loose freshness. Appearance of the skin: the skin of fresh salmon is bright and shiny with all the scales attached properly. As time goes by, the skin loses its shhiness and becomes dull, while the scales start to come out very easily.

PAGE 32

12 Instrumental techniques to measure the color of shrimp and salmon Color of food products can be assessed by various analytical methods: sensory analysis using trained panelists for descriptive or comparative tests, comparison of fish samples with standardized colors, or by instrumental analysis based on light reflectance from flesh samples. When results from the various analytical methods are to be compared, difficulties may arise due to lack of references in sensory analysis, lack of standardized colors for comparison tests, or instrument design and sample preparation used for instrumental analysis (Skrede et al., 1989). Spectral characteristics have frequently been used to evaluate the pigmentation of salmonids, however, very little research has focused on shrimp spectral properties. Salmon flesh has a uniform color, making it easier to be analyzed with current instrumental techniques. Salmon color has been described by the parameters L, a and b where L represents lightness, a redness, and b yellowness of the sample (Skrede and Storebakken, 1986). These color parameters are commonly obtained from commercially available trisitimulus colorimeters. A colorimeter is composed of a light source, a set of three filters which duplicate the responses of the three types of receptors in the human eye, a photocell and the object to be measured (Francis and Clydesdale, 1975). Colorimeters measure the color of a sample by shining a beam of light at a certain angle on the surface of the object. The light reflected from the sample is sensed by a photocell which converts the light information into its tristimulus values. These values are usually in the CIE XYZ ( International Commission on Illumination) color system, which can be converted into other color

PAGE 33

13 systems such as L*a*b*, Munsell, etc. Colorimeters measure the overall reflectance from the surface of the sample. In the case of foods with different colors or non-uniform surfaces, a single tristimulus value is not a good representation of the actual colors present in the sample. Very few researchers have studied the color of shrimp. Shrimp have different colors, small sizes and round surfaces, making it difficult to use a colorimeter to measure the surface spectral properties. To overcome this problem, Huang et al. (1996) blended the shrimp into a paste and used a colorimeter to measure the color changes of marinated shrimp during storage. However, this procedure is not valid when one is interested in the surface color of the shrimp. Computer vision can be an alternative to measure color of irregular surfaces. Color of salmon flesh depends on the content of certain carotenoids. The color of the flesh can be changed with the diet. Colorimeters have been widely used to assess these color changes, and to find correlations between carotenoid levels and the amount of redness in the flesh (Saito, 1969; Schmidt and Cuthbert, 1969; Skrede and Storebakken, 1986; Skrede et al., 1989). Colorimeters have also been used for color grading of canned salmon and to predict the color of processed salmon from the color of raw fish (Francis and Clydesdale, 1975). Skrede et al. (1990) developed a color card for raw flesh of astaxanthin-fed salmon, based on instrumentally obtained L*a*b* values. These procedures focused only on the red portion of the flesh, however an overall color evaluation of salmon fillets involves the flesh, the belly flap and the connective tissue surrounding the muscle.

PAGE 34

14 Odor Attributes Smell is often the second sense that is used to judge the quality of food products. The odor of seafood products has been widely used as one of the main indicators of quality since ancient times (Botta, 1995). The odor alone can be used as a decision making tool to accept or reject a batch of shrimp, salmon or any other seafood product. Generally, when seafood has a fishy, spoiled or putrid odor it is rejected. The term "fishy" odor is not applied exclusively to fish, since there is not a single fishy odor, but a number of fishy odors. These odors arise mainly during the bacterial spoilage of seafood, and are closely related to the odors of certain chemicals such as trimethylamine (TMA), dimethylamine (DMA), ammonia or hydrogen sulfide (Stansby, 1962). Evaluation of off-odors in seafood Perkins (1992) defined fresh seafood as fish that exhibits a clean and natural odor, with physical characteristics representative of the species in good condition. The odor of freshly caught shrimp or salmon is mild. Shrimp odors have been described as typical of the sea and seaweeds. If shrimp are held on ice from the time they are caught, they will retain their high quality for a period of about one week. During this time, there will be little or no objectionable fishy odor (Gorga and Ronsivalli, 1988; Campbell and Williams, 1952; Matches, 1982; Shamshad et al., 1990). Fresh salmon has no detectable odors, however some researchers define the fresh odor as that of sliced cucumbers (Dore, 1990). Inspectors usually smell the gut cavity to detect any off-odors. Salmon stored at 3°C has a shelf life of 6 to 8 days with low off-odor formation (Haard and Lee, 1982).

PAGE 35

15 The odor of shrimp and salmon should be judged only when the product is in the unfrozen state, preferably at room temperature. In the frozen state, the odor is not completely detectable. According to the Code of Federal Regulations (CFR, 1998) fresh shrimp flavor and odor are best described as mild and pleasant. Spoiled shrimp begin to emit an ammonia smell. The United States Department of Commerce (USDC) inspectors base their judgement on their experience and grade shrimp and salmon according to Table 2-2 and Table 2-3, respectively. Compared to other muscle foods, seafood products are significantly higher in low molecular weight non-protein nitrogen compounds (Finne, 1982). This results in their unique, delicate and different flavors. These compounds are also responsible for the rapid deterioration of fresh seafood by serving as substrates for typical spoilage organisms, which convert them into obnoxious smelling volatile bases (Finne, 1992). Deterioration of the odor of freshly caught seafood is caused by facultative aerobic bacteria, (Gorga and Ronsivalli, 1988), bacterial enzymes, and naturally occurring enzymes (Finne, 1982). Table 2-2. Shrimp sensory standards used by USDC inspectors GRADE 3 CRITERIA A Pleasant flavor and odor characteristic(s) of freshly caught shrimp that is free from off-flavors and odor of any kind. B Lacking of good flavor and odor characteristics of freshly caught shrimp, but free from objectionable off-flavors or off-odors of any kind. Substandard Meets either grade A or B criteria in sensory tests, but not in nonsensory test, such as weight, defects, etc. a : Products that have objectionable flavors and odors are not eligible for grading. Source: CFR, 1998.

PAGE 36

16 Table 2-3. Salmon sensory standards used by USDC inspectors GRADE CRITERIA A Good flavor and odor. The fish flesh has the good flavor an odor characteristic of the indicated species of salmon, and is free from rancidity and from off-flavors and off-odors. B Reasonably good flavor and odor. The fish flesh may be somewhat lacking in the good flavor an odor characteristic of the indicated species of salmon, is reasonably free from objectionable off-flavors and off-odors. Substandard Substandard flavor and odor. The flavor and odor fail to meet the requirements of reasonably good flavor and odor. Source: Dore, 1990. In order to express the quality of shrimp in terms of chemical parameters, a large number of reports have suggested the determination of a variety of chemical compounds. Changes in pH (Bethea and Ambrose, 1962), trimethylamine (Ruiter and Weseman, 1976; Shamshad et al., 1990; Fatima et al., 1988), dimethylamine (Ruiter and Weseman, 1976), total volatile nitrogen, (Ruiter and Weseman, 1976; Cobb and Vanderzant, 1975; Malle and Poumeyrol, 1989), indole (Chang et al., 1983; Niola and Valletrisco, 1986), inosine monophosphate and hypoxanthine (Fatima et al., 1981), and total volatile nitrogen/amino nitrogen ratio (Cobb and Vanderzant, 1975) have been either used or suggested as quality indices for shrimp. In the case of salmon, researchers have tried to find a chemical analysis that can be correlated to that clean, natural odor and that could be used as an index of quality. Some of the methods considered were trimethylamine oxide (Hebard et al., 1982), rapid estimation of volatile amines (Storey et al., 1984), diagnostic strip for trimethylamine (Wong et al., 1988), determination of ammonia and total volatile nitrogen (Halland and

PAGE 37

17 Njaa, 1988), development of biosensors to measure hypoxanthine (Mulchandani et al., 1989), measurement of nucleotide degradation in fish muscle (Kennish and Kramer, 1987), levels of ethanol concentration (Hollingworth and Throm, 1982; Kelleher and Zall, 1983) and other chemical analyses that could be used to correlate salmon quality or freshness with sensory evaluations. The chemical composition of seafood is one of the important factors affecting the flavor of the final product. According to Whitfield (1988), other flavors, mainly offflavors can result from: (a) the microbial and auto-oxidative spoilage of fresh material through incorrect handling and storage, (b) the adsorption of chemicals from an industrially polluted environment, and (c) the bioaccumulation of components naturally present in either the organism's diet or its environment. In shrimp, spoilage odors are primarily caused by bacterial action, generally Gram negative bacteria of the genera Pseudomonas , Moraxella and Acinetobacter (Nickelson, 1992). Metabolites formed by these bacteria include TMA, DMA, ammonia, and formaldehyde (FA). Some of the most important substrates for bacterial spoilage are free amino acids and trimethylamine oxide (TMAO). Chemicals responsible for off-odors There are several compounds that are responsible for the off-flavors or off-odors in shrimp and salmon, and in seafood in general. Free amino acids . The free amino acid pool serves as an immediate substrate to typical spoilage organisms. The most common mechanism by which microorganisms utilize free amino acids is deamination. Bacterial deamination can proceed in several

PAGE 38

18 different pathways depending on the enzymatic constitution of the organism and the environmental conditions (Finne, 1992). However, independent of the prevailing mechanism, deamination of free amino acids will result in the formation of ammonia which is the primary component of the decomposed flavor of seafood (Finne, 1992; Nickelson, 1992). Ammonia has been suggested as an objective quality index for fresh seafood, and is the major component of the total volatile nitrogen (TVN) which is often used as a quality indicator for fresh fish (Cobb et al., 1973b). Decarboxylation is another mechanism by which microorganisms utilize free amino acids in seafood. Decarboxylation of amino acids produces an amine and carbon dioxide. Many of these amines are unique in structure and have been identified as objective quality indices in shrimp (Finne, 1992). The amines that have attracted most attention are putrescine, cadaverine and indole which come from decarboxylation of arginine, lysine and tryptophan, respectively (Nickelson, 1992). Trimethvlamine oxide (TMAO) . TMAO and its breakdown products are among the most studied compounds in the decomposition of marine foods. TMAO is only found in saltwater fish and shellfish, and it varies with species, season, size, age and environmental conditions (Huss, 1988). Two different chemical reactions are responsible for the breakdown of TMAO in postmortem muscle, both being responsible for the deterioration of quality. During refrigerated or iced storage, TMAO is enzymatically reduced to TMA by TMAO-reductase positive microorganisms. The final product of this decomposition is ammonia, whereas TMA, DMA or FA are intermediates. The most common Gram-negative spoilage organisms have the capability of reducing TMAO to

PAGE 39

19 TMA which is a volatile amine with odor and flavor characteristics similar to ammonia (Hebard et al, 1982). Because of the direct relationship to bacterial growth and decomposition, TMA has been suggested as a spoilage indicator by a number of researchers (Ruiter and Weseman, 1976; Shamshad et al., 1990; Fatima et al., 1988). For example, the limit for acceptable shrimp in some sectors of the Australian and Japanese markets is 5 mg TMA nitrogen per lOOg (Montgomery et al., 1970). The second breakdown reaction of TMAO in post-mortem muscle is the formation of equimolar amounts of DMA and FA. This reaction can take place both enzymatically and non-enzymatically. When fresh shrimp are held on ice, the enzymatic breakdown of TMAO by TMAO-reductase dominates, forming high levels of TMA. However, in frozen shrimp, where bacterial activity is reduced or stopped, DMA and FA are formed by the breakdown of TMAO (Finne, 1992). Thresholds in the detection of ammonia, TMA and DMA are listed in Table 2-4. Table 2-4. Odor thresholds of important nitrogenous compounds SUBSTANCE CONCENTRATION IN PPM Ammonia 110 DMA 30 TMA 0.6 Source: Regenstein et al., 1982. Trimethvlamine . TMA has been studied extensively in fish; however, Chang et al. (1983) concluded that TMA was not a good quality indicator for shrimp. TMA was not detected in the ice-stored shrimp before 8 days of storage, at which time shrimp with a I

PAGE 40

20 total plate count of 10 8 cfu/g was considered to be of poor quality. Montgomery et al. (1970) reported that acceptable shrimp in Australia and Japan have limits of 5 mg TMAN/lOOg and 30 mg TVN-N/lOOg. Ohashi et al. (1991) measured TMA, DMA, ammonia and other gases that evolved from several different fish muscles. They observed that in general the concentration of ammonia evolved from fish muscle increased rapidly with the progress of putrefaction, while that of DMA was relatively low and remained nearly constant. In contrast, the concentration of TMA was known to increase even before apparent putrefaction. Most of the methods used to measure TMA and total volatile bases are time consuming, require sophisticated equipment and are difficult to apply for an online inspection. These methods are based on titration, distillations, chromatographic separations, etc. Ammonia . Ammonia has been suggested as an index of quality for crab meat (Steinbrecher, 1973). LeBlanc and Gill (1984) used ammonia as an objective quality index in squid which correlated well with TVN. They observed that the ammonia production in squid appeared to be linear with postmortem time of storage, and is sensitive to temperature of storage, physical abuse and method of handling. Temperature had a positive effect on the generation of ammonia in oysters during storage, at 0°C and 8°C it was slow, with a rapid increase at 25°C (Hamaoka and Sasaki, 1992). In shrimp, most of the total volatile nitrogen is ammonia (Cobb et al., 1973a; Ruiter and Weseman, 1976, Finne, 1982). Ammonia concentration was shown to increase during storage (Cheuk and Finne, 1984) and there was a good correlation between the concentration of ammonia and traditional spoilage indicators (Ward et al., 1979; Cheuk and Finne, 1984; Finne, 1982).

PAGE 41

21 Ward et al. (1979), using an ammonia-specific electrode, demonstrated the relationship between storage time, total microbial numbers and ammonia concentration during refrigerated storage of fresh shrimp. Cheuk and Finne (1984) observed that during storage of fresh shrimp and crab meat at 3.5°C, ammonia, urea, TVN and TMA increased as the microbial population increased. As they expected, there was a close relationship between ammonia and total volatile nitrogen. The shrimp used for this study were spoiled after 7 days at 3.5°C; at this time the total plate count was 1 .69xl0 7 organisms/g, the TMA concentration was 4.52 mg of TMA-N/100 g and the ammonia content was 23.98 mg of NHj/lOO g. Ammonia has been suggested as an objective index of fresh seafood quality (Ward et al., 1979). They concluded that most of the other chemical parameters are seldom used by the shrimp industry due to the complexity, time consumption and inconsistency of the methods. Therefore, Luzuriaga et al. (1997a) developed a rapid method to measure ammonia in the headspace of a shrimp sample by using a commercially available ammonia selective electrode and a custom-built container. The procedure was simple, no sample preparation was required, and analysis time was less than three minutes. Indole . Indole had been proposed as an indicator of decomposition for shrimp and oysters since the 1940's. Indole is used by the U.S. Food and Drug Administration (FDA) to validate the sensory evaluation of shrimp decomposition. According to FDA, shrimp decomposition could be classified in three groups or classes (Federal Register, 1981) (Table 2-5).

PAGE 42

22 Table 2-5. Classes of shrimp decomposition defined by the FDA GRADE DESCRIPTION INDOLE LEVELS Class 1 Fresh aroma, no odor identifiable as decomposition <25 ug/lOOg Class 2 Slight odor of decomposition >25 ug/lOOg Class 3 Strong odor of decomposition >50 ug/lOOg Source: Federal Register, 1981. Chang et al. (1983) concluded that high indole levels indicate decomposition, but shrimp of poor or unacceptable quality may not necessarily contain indole. Indole alone is not a suitable indicator of the quality of fresh or frozen shrimp, but when used in conjunction with other quality evaluation methods it can be of value in assessing prior high temperature abuse. Computer Vision Computer vision uses a computer, software and a video camera to obtain an image of an object. The image is digitized by using image processing techniques, and divided in small elements or regions, called pixels. Each pixel has the information of grey levels for a black-and-white image, or the levels of the three primary colors (red, blue and green, RGB) for a color image. Computer vision can use different optical sensing schemes, but video cameras are most commonly applied (Elster and Goodrum, 1991). Two terms commonly mentioned along with computer vision are image processing and pattern recognitioa Image processing involves a series of steps that can enhance an image and extract some information (Paulsen and McClure, 1986). Pattern recognition involves

PAGE 43

23 procedures that allows the computer to identify and classify an object based on certain characteristics (Gonzalez and Wintz, 1977). Computer vision as a tool has not been used to a full extent in agriculture, mainly due to the complexity and variability of images in agricultural products. But in the last 8 to 10 years there has been a steep increase in applications, mainly for quality evaluation and inspection (Krutz and Precetti, 1991; Okamura et al., 1993; Liao et al., 1994; Luzuriaga, 1995; Luzuriaga et al., 1997b; Miller and Drouillard, 1997; Cardarelli et al., 1998), grading (Miller and Delwiche, 1988; Diehl et al., 1990; Guedalia, 1997; Heinemann et al., 1997), and automatic fruit and vegetable harvesting (Sites and Delwiche, 1988; Miller, 1987). Image Processing There are many software techniques available to manipulate digital images. A primary function of image processing is to create a new image by altering the data in such a way that the features of interest are enhanced and the noise is reduced or eliminated. The processing system contains the electronics and software programs to perform image grabbing, image enhancement, feature extraction and output formatting (Galbiati, 1990). The data processing could be done by hardware or software. When the data are handled by hardware, the system runs faster, but is less flexible. Usually a frame grabber will perform several of the operations already mentioned. These operations will reduce the raw image data into a set of relevant information. Based on the extent of data reduction, computer vision techniques can be categorized into three levels (Sarkar, 1991):

PAGE 44

24 • Low Level Operations. These operations are used on every pixel of the image, and perform functions such as reduction by converting a color or grey scale image into a binary image, image enhancement, morphological operations (erosion, dilation, thinning, skeletonization), and logical or arithmetic operations. Both the input and output of these operations are digital images. • Intermediate Level Operations. They obtain a feature set or statistics from an image, resulting in multiple orders of magnitude reduction in data. Some operations include image segmentation, encoding, data compression, texture analysis, histograms, image transforms (Hough, Fourier, etc.) and others. • High Level Operations. At this level the data from previous levels are analyzed and decisions are made depending on the outcome of the analysis. In computer vision applications there are several paths that the user can take to obtain the same results. Some of them will be faster, which is desired for on-line inspection where a decision has to be taken immediately after the image is acquired. In these cases most of the low and intermediate level operations are done in hardware, whereas the decision or high level operations are implemented in software. A pplications of computer vision Computer vision is being used in many areas of engineering research. The range of possible applications is growing steadily, especially in the area of automation. In the food and agricultural area, research has focused on grading and sorting by means of computer vision (Affeldt, 1991). For example, Ghate et al. (1993) developed an automated peanut

PAGE 45

25 maturity detection system. Gunasekaran et al. (1988) detected soybean seed coat and cotyledon cracks by image processing. Okamuraetal. (1993) developed a computer vision system to grade raisins. Rehkugler and Throop (1989) developed image processing algorithms for apple defect detection. Rigneyetal. (1992) graded asparagus with computer vision. Sarkar and Wolfe (1984) developed a computer vision system capable of separating fresh market tomatoes based on visual quality. Miller and Drouillard ( 1 997) developed an on-line system to analyze the color, shape and blemishes of citrus. Cardarelli et al. (1998) developed a high resolution machine vision system for non-destructive internal inspection and classification of rough rice. Marine resources have also benefitted from this research. Hatano et al. (1989) used digital image processing to find an objective criterion to predict the flesh redness from the spawning coloration of fall chum salmon. Diehl et al. (1990) described geometric and physical properties of oyster meat by computer vision. Li (1990) presented a method to detect oyster hinge line by computer vision. Pau and Olafsson (1991) published a book on fish quality control by computer vision, including parasite detection, length measurements, fish inspection and sorting, etc. Strachan et al. (1990) recognized fish species by shape analysis of images. Strachan (1993) developed a machine to measure the length of fish by computer vision. Ling et al. (1988) developed an adaptive tresholding technique for shrimp images. Ling and Searcy (1989) developed a feature extraction system for a vision based shrimp deheader. Kassler et al. ( 1 993) developed a prototype for automatically grading and packing prawns into single-layer consumer packs, in which each prawn is approximately straight and has the same orientation; this system

PAGE 46

26 combines computer vision and robotics. Balaban et al. (1994) evaluated the count and uniformity ratio of tiger and white shrimp. Count was estimated from the correlation of weight to view area of individual shrimp. They concluded that the weight of individual shrimp can be accurately determined by computer vision. Luzuriaga et al. (1997b) developed a system to grade melanosis and to measure color of white shrimp. Newman (1998) used a computer vision system to evaluate the color of raw tuna stored at different temperatures. Color Systems Color can be measured and described in physical terms. However, the actual color that we perceive is the result of a complex series of processes in the human visual system (Rossotti, 1983). Color has different meanings, the most common are perceived color and psychophysical color. Perceived color is defined as the aspect of a visual perceptual phenomenon, distinct from form, shape, size, position, gloss or texture, that enables a person to distinguish between elements of the visual field and to characterize the elements by color names such as white, black, yellow, red, green, gray brown, orange, pink purple and so on (Robertson, 1992). Psychophysical color was defined by the CIE (Commission Internationale de l'Eclairage) as a characteristic of visible radiation by which an observer may distinguish such differences between fields of view of the same shape, size, position and structure as may be caused by differences in the spectral composition of the radiation (Robertson, 1992). Psychophysical color is usually specified in quantitative terms such as the tristimulus values.

PAGE 47

27 Researchers in the area of color have been trying to develop methodologies for the scaling of the attributes of color perception and to quantify color discrimination. Different types of color scales or systems have been defined throughout the years (Table 2-6). Wyszecki and Stiles (1982) defined a color scale as a series of ordered numbers which represent observable gradations of a given attribute, or gradations of a combination of attributes of color perception. The full physical specification of a color stimulus is multidimensional. Each color has a specific radiance intensity at each wavelength in the visible spectrum (Giorgianni and Madden, 1998). This color representation is not an intuitive way to define the color of an object. Most color systems define their colors in terms of the two-dimensional graph known as the 193 1 CIE chromaticity diagram. The CIE tried to standardize the definition of colors, and defined colors in terms of the sensor response curves. This resulted in the tristimulus values. The tristimulus values are the basis of the XYZ color system and the CIE diagram The diagram was derived from the X, Y, Z imaginary primaries spectral functions, also known as color matching functions. The three-dimensional space was reduced to two dimensions by removing the color intensity, and created the xyY color system (Figure 2-1). The history of color description revolves around the CIE diagram (Former and Meyer, 1997). This diagram has been modified and transformed to develop other color systems. Munsell, XYZ and L*a*b* color systems are among the most important color systems used in the food industry. The Munsell system has been used for color matching. However, for quantitative measurements, XYZ and L*a*b* color scales are used, and are

PAGE 48

28 Table 2-6. The most common color systems used in color technology COLOR MODEL VARIABLES COMMENTS light spectra continuos (wavelength) Always exact, impractical RGB red, green, blue Color space used by computers AIL yV, I , Zj iriMUIlulUa VdiUCb (~*r\lr\r cnci(*p TArmQ li^^rl rwr TTh i^oiur s>pacc luriiutuzcu uy v_*m, yvY xy i v \/ — r* ri tvi in 'it i ( m t v v = liiminonpp A. v LIU UlllallCliy, 1 lUIIllllailCC Similar tn YY7 Hunter Lab Li — llglllllcss, a, o — ciiruiiiaiiciiy coordinates oysicin ucinitu irom a aiiu a reference white point L*a*b* L = lightness, a, b = chromaticity coordinates Similar to Hunter Lab, also know as CIELAB Munsell hue, value, chroma used for color matching CMY cyan, magenta, yellow subtractive analog to RGB CMYK cyan, magenta, yellow, black used for printing presses R YR icu. yciiuw, uiuc uscu lor pdiniing YTO i n t * Ml ci t v tv/l-int/^ncitA/ r\liif»_intf*«cif \r uiiciioiiy, icu-iiiiciioiiy, uiuc-micxiMiy ubc iur DioaULabi television HSI hue, saturation, intensity close to perception, used for r»r\ lor HpQprirvHnn ^/UlUl UvoCl 1LH14J11 HSV hue, saturation, value similar to HSI, used for computer calculations HSB hue, saturation, brightness identical to HSV HSL hue, saturation, lightness similar to HSI, used for computer calculations Adapted from: Former and Meyer, 1997. commonly encountered in colorimeters. Using mathematical transformations, these color scales can be transformed into the RGB color system, which is the system used in computer vision applications (Shin, 1995).

PAGE 49

29 Yxy Color System (CIE 1931) 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 X Figure 2-1. CIE 1931 chromaticity diagram, Yxy color space. Source: Minolta Corp.

PAGE 50

30 In the Munsell system, colors are specified by hue, value and chroma. Hue is designated by one of several ways, all equivalent and all based on a 1 00-point hue scale. Either a pure numeric (ranging from 0 to 1 00), a color name, a color name abbreviation, or a mixed alphanumeric symbol may be used (Kuehn and Luxenberg, 1 968). The alphanumeric notation is most commonly used. Value is a measure of lightness expressed on a scale from 0 to 10. Black has the value 0 and white has the value 10. Equal value steps represent subjectively equal lightness differences (Billmeyer and Saltzman, 1981). Chroma is a measure of saturation, expressed on a scale from 0 to 14 or further. Zero is neutral (achromatic or gray) color, while larger numbers represent more saturation. Equal chroma steps represent subjectively equal saturation steps. The Munsell color system was developed to aid researchers in naming their colors by comparing the color of the object with that in the "The Munsell Book of Color", a book that contains painted representations of more than 1500 colors. The Lab color system was developed by Hunter in 1942. It became an industry standard because instruments were available to provide color measurements. The Lab color scale is one of the uniform color spaces. It is uniform because it does not require nonlinear transformations of the 1931 CIE XYZ system (Billmeyer and Saltzman, 1981). The L value represents the lightness of the color. The values a and b are the chromaticity coordinates and express the actual color. Redness or greeness can be expressed as a single number, usually called a. The a value is positive if the color is red and negative if the color is green. Similarly, yellowness or blueness is expressed by the coordinate b, which is positive for yellow and negative for blue. The Hunter Lab values were adopted

PAGE 51

31 by the CIE in 1976 with some mathematical transformations of the L value. This created the CIELAB or CIE L*a*b* notation. RGB Color System The most accurate way to represent a single color would be to store the values of the entire emitted or reflected spectra of that color from 400 nra to 700 ran, or by referring that single color to a known color standard (Former and Meyer, 1997). However, in an image that contains thousands of pixels is not practical to have a full spectrum, or a color standard number for each pixel. Therefore, the most common way for a computer to represent a color is through the RGB nomenclature. RGB stands for red, green and blue and refers to the representation of a particular color by three numbers, a red, a green, and a blue value. Trichromatic theory states that any color can be represented by the combination of these three basic colors. RGB is the most frequently used system for image processing (Precetti, 1995). The RGB system is a three dimensional space represented as a color cube (Hearn and Baker, 1994). The cube axes represent the primary colors (Figure 2-2). Color intensity increases along each coordinate as the point moves away from the origin. The origin of the cube represents black (Searcy and Reid, 1989). The primary colors and their complements with maximum intensities are placed at the vertices of the cube. The diagonal connecting white and black represents grays that have equal intensities in the three primary colors (Burger and Gillies, 1989; Hearn and Baker, 1994).

PAGE 52

32 Figure 2-2. RGB color system The maximum intensity for the three primary colors can be normalized to the range from 0 to 1 (Burger and Gillies, 1989), but due to the internal arrangement of the computer, the red, green and blue components have integer values ranging from 0 to 255 (represented by 1 byte=8 bits). These values will be related to the intensity value for each one of three colors. A triplet of 255, 255, 255 represents white, 0,0,0 represents black, 255,0,0 represents red, and so on (Burger and Gillies, 1989). This method of storing color information is known as the 24-bit color (8 bits per color). The combination of these 24 bit colors in an organized manner will create a 24-bit color image. All possible combinations of intensity values for the red, green and blue components will create more than 16.7 million colors (256 3 ), that a computer will be able to display. The human eye

PAGE 53

33 can not distinguish that many colors at one time, therefore the reduction in the number of colors is done by regrouping colors in blocks (Precetti, 1995). All colors within a block are represented by the block's center color. The 24-bit RGB format is a device dependent color system. Color displays, digital cameras, printers or scanners often will show different colors for the same RGB triplet (Former and Meyer, 1997). Standardization can be achieved by defining the three sets of primaries used by each device and storing those in the image to apply a gamma correction factor (Hetzroni and Miles, 1994). No set of three RGB primaries can generate all possible colors. Therefore, there are colors that cannot be represented by 24-bit RGB. In any case, regardless of the problems, 24-bit RGB color is the universal choice for digitized color (Giorgianni and Madden, 1998). Visual attributes constitute an important part of the overall quality attributes of foods, as well as odor, flavor and texture. Advances in computers and electronics provide opportunities to automate many processes by doing rapid, more objective, and standardized quality evaluations by computer vision. Electronic Nose There are three sensory systems in humans that are responsible for the sensation of flavor. These are gustation (sense of taste), olfaction (sense of smell) and the trigeminal sense (responsive to irritant chemical species) (Gardner and Bartlett, 1994). The sense of smell arises from the stimulation of the human olfactory system by volatile compounds. These compounds go into the nasal cavity and across the olfactory area or epithelium.

PAGE 54

34 The chemosensory receptors located at the surface of the olfactory hairs or cilia have receptor binding proteins that will bind with the volatile compounds (Pearce, 1997a). They will amplify the signal and send it to the brain by means of the axons. The brain will process the signal and compare it with previous knowledge and try to define the odor. Researchers are trying to model this process, but precise details are still unknown. The main unknown is the method of interaction between an odor molecule and a receptor site. The human nose has approximately 100 million olfactory cells. There are a small number of receptor proteins (-1000), therefore, the receptor cells have partially overlapping sensitivities (Bartlett et al., 1997). These cells are believed to have a high sensitivity (in the ppb range), low specificty and only live on average for about 22 days (Gardner and Bartlett, 1994). These cells sense odorant molecules which are typically hydrophobic and polar, with molecular masses up to 300 Daltons. A single molecule can have a distinct odor. However, more natural smells or flavors are a complex mixture of chemical species and contain hundreds of constituents (Dodd et al., 1992). The human nose is the main instrument used in many industries to evaluate the smell or flavor of food products. However, it is subjective, prone to error, difficult to correlate among different people and is subject to fatigue. The industry also uses techniques such as gas chromatography (GC), gas chromatography-mass spectrometry (GC-MS), or gas chromatography-olfactometry (GCO) to analyze odors. These techniques are time consuming, expensive and sometimes inadequate in identifying the key odor compounds. Consequently there is a demand for an electronic instrument that can mimic the human sense of smell and provide low-cost and rapid sensory information.

PAGE 55

35 Electronic Nose Technology The development of an instrument to specifically detect odors started with Moncrieff (1961). Later in 1964-65 several researchers started studying the redox reactions of odorants at an electrode, and modulation of electrical conductivity and of contact potential by odorants. Twenty years later, Persaud and Dodd (1982) presented for the first time the concept of an electronic nose as a chemical sensor array system for odor classification. In the late 1980's the term electronic nose was commonly used in the area of research for the design of artificial olfactory systems. Nowadays, the electronic nose has various synonyms such as artificial nose, mechanical nose, odor-sensing system, sensor array system Gardner and Bartlett (1994) defined an electronic nose as an instrument, which comprises an array of electronic chemical sensors with partial specificity, and an appropriate pattern recognition system, capable of recognizing simple or complex odors. The main components of an electronic nose are: sample handling mechanisms, • an array of chemical sensors, signal preprocessing and conditioning, and • pattern recognition techniques. The electronic nose technology tries to simulate the olfactory process with fewer sensors and with software designed to analyze the responses from the sensors. Each chemical sensor represents a group of olfactory receptors and produces a time-dependent electrical signal in response to an odor. Any noise or sensor drift may be reduced using

PAGE 56

36 Sensor head Purge oul Purge in Seal Sample vessel Figure 2-3 . Schematic of sensor head and sample vessel of an electronic nose with static sampling signal preprocessing techniques. Finally, classification and memorization of odors in the brain is equivalent to the use of pattern recognition techniques. Sample handling procedures are responsible for obtaining the vapor above a sample and transporting the vapor to the sensor array. Static and dynamic sampling are the two methods that are currently being used. A static system is designed to measure the headspace above a liquid or solid (Figure 2-3). The system consists of a sensor head (compartment with the sensor array) and a sample vessel. The sensor array and the sample vapor remain in separate sealed compartments. Purging of the sample headspace and sensor compartment is carried out before the analysis to eliminate any foreign odor. This is done by passing compressed air or any other inert gas, such as nitrogen, for a certain period of time. Once the sample has reached equilibrium the door between the two compartments is opened and the test starts. In the case of conducting polymer sensors, an integral DC power supply maintains a constant current through the sensors. When the

PAGE 57

37 sensors are in contact with the headspace of the sample, the sensor resistance changes. The corresponding voltage change across the sensor is measured, and the resulting analog signal is digitized and sent to the computer. Acquisition of the change in conductivity of the sensors is carried for a given period of time. Once the test is completed, the door is closed and the sensor head and sample vessel are purged for further samples. In the dynamic sampling procedure the electronic nose works similarly to a GC. A sample is placed in a closed container. The sample headspace will equilibrate with the vapors coming off from the sample. Once headspace equilibration is achieved, a sample of the headspace is obtained, and injected into the sampling port of the electronic nose. The sample will be carried by an inert gas to the sensor array. The sensor will change its electrical properties and send a signal to the computer. Similar to the static system, the sensor array must be cleaned with an inert gas before analyzing the next sample. The preprocessing and conditioning of the analog response of the sensor are responsible for maximizing the information provided by the sensor signal. This is done by using signal conditioning circuits, potential dividers, constant voltage sources, and an analog-to-digital converter (Corcoran, 1993). The measurement of electrical parameters (change in current or voltage) uses techniques that optimize system sensitivity. These reduce the effect of noise by modulating the sensor signal and amplify it to a suitable level. System noise can and will be affected by variations in the electronic circuitry as well as connections between the sensors and the circuit (Hodgins, 1997). The signal conditioning digitizes the response of the sensors and generates an output that is then analyzed with pattern recognition techniques to define the odors or volatiles present in the sample.

PAGE 58

38 Sensor Technology There are different types of materials and technologies that are being used to manufacture sensors that are useful for odor detection. The types of sensors that are being used commercially in electronic noses are semiconductor metal oxides, conducting polymers and surface acoustic wave sensors (Bartlett et al., 1997; Hodgins, 1997). Other types of sensors that have potential or have been used are biosensors, enzyme sensors, electrolytic sensors, platinum hot wire detectors and Schottky devices (Shurmer, 1990). Sensor technology is changing very rapidly and more stable, sensitive and faster response sensors are being developed. Semiconductor metal oxides Metal oxide sensors have been available for many years. Two main types have been developed: thick film metal oxides, also known as Taguchi sensors, and thin film, which are commonly used in commercial electronic noses. These sensors are made with a platinum heater coil coated with alumina. As current passes through the coil the metal oxide heats up. The reaction between the vapor and the metal oxide causes a change in resistance at a fixed temperature. This change in resistance can be measured and related to the odor being monitored. In metal oxides, chemisorbed oxygen [0] reacts irreversibly with the odorant [R] to produce combined molecules [RO] and liberated conducting electrons [e ]. Electron mobility increases and therefore, overall electrical conductivity of the material changes (Tan et al., 1995). These sensors typically operate between 400600°C to avoid interference from water and to aid rapid response and recovery times

PAGE 59

39 (Bartlett et al., 1997). They are sensitive to combustible materials, such as alcohols, but are less sensitive at detecting sulfuror nitrogen-based odors. Conducting polymers Conducting polymers have unique electrical properties that make them suitable for gas detection. They have the advantage that a wide variety of suitable materials exist for their manufacture. The main types are poly-pyrrole and poly-anilines. The electrical conductivity of the polymer changes in the presence of volatile compounds. This change occurs rapidly and reversibly. The adsorbed odor molecules are believed to cause a swelling of the polymers and to interfere with charge transfer within the polymer (Corcoran, 1993). Conducting polymer sensors are nonspecific, which means that a number of different compounds will interact with the polymer material. These sensors operate at room temperature and have quite a good sensitivity, typically between 0. 1 and 100 ppm (Bartlett et al., 1997). Conducting polymers respond to moisture, therefore, caution should be taken when analyzing samples with different moisture contents. Surface acoustic wave devices This type of sensor has been in research and development for 5 to 10 years (Hodgins, 1997). The principle of operation is that a surface wave is generated in a material that absorbs the compounds of interest. The surface wave is normally generated using a quartz resonator. The frequency of operation is usually 100 Mhz to 1 Ghz, and depends on the sensitivity required by the system When the sensor is not exposed to a vapor, it will have a certain resonant frequency. Once the sensor is in contact with the volatiles, there will be a change of mass in the sensor material, and therefore, a change in

PAGE 60

40 the resonant frequency (Ballantine and Wohltjen, 1989). This change in frequency is the response or output from the sensor to the volatiles present in the sample analyzed. These sensors have higher sensitivity than conducting polymers (Hodgins, 1997). However, they are more selective, which means that a larger number of these sensors are needed to cover all vapors that are likely to occur in food products. A pplications to Food Products Several review articles have mentioned the benefits and impact of electronic noses, their theoretical and practical uses in the food industry (Springett, 1991; Tan et al., 1995; Shiers and Farnell, 1995; Bartlett et al., 1997). Quality control is the most important area where this technology could be beneficial, especially since HACCP and ISO 9000 are enforced by regulatory agencies. Using an electronic nose would allow measurement of the odor of a food product from raw material receiving, all the way to the final product. Many different stages can be monitored to ensure that the right conditions are maintained. Even though an on-line electronic nose does not exist yet, continuous monitoring would ensure early detection of malodors and ultimately be cost effective. In the area of grains and beans substantial work has been done to discriminate among coffee cultivars, coffee from different origins, and coffee aromas (Aishima, 1991; Tan et al., 1995; Delaure et al., 1996). Borjesson et al. (1996) used an electronic nose to classify grains and therefore reduce the inspector's exposure to grains that can be contaminated with aflatoxins.

PAGE 61

Non-alcoholic and alcoholic drinks are one of the areas with great potential for electronic noses. Several studies monitored the flavor and aroma of beer and its raw materials (Pearce et al., 1993; Tomlinson, 1995; Tomlinson et al, 1995; Zimmermann and Leclercq, 1995). Lucas and Castan (1995) and Weber and Poling (1996) studied the aroma of pure hop and blends used in beer making. Viaux et al. (1996) used an electronic nose to help in the determination of the technical specifications of some additives and technological aids used in the sparkling wine process. In the fruits and vegetables area, volatiles of fresh squeezed orange juice (Bazemore et al., 1996) and citrus juice aroma volatiles (Hodgins, 1995; Hodgins and Simmonds, 1995) have been studied. Bazemore et al. (1997) used metal oxide sensors to discriminate between grapefruit juices of different cultivars. Noh and Ko (1997) used an electronic nose with conducting polymers to discriminate the country of origin for ginseng, garlic and carrots. Maul et al. (1997) assessed the ability of a sensor array to nondestructively identify and classify tomato fruit exposed to different harvesting and postharvest handling treatments. Werlein and Watkinson (1997) compared the sensory quality of "sous vide" and conventionally processed carrots, green beans and potatoes using metal oxide sensors and sensory panels. In the dairy area, sensor arrays have been used for differentiation of enzyme modified cheese slurries (Jin and Harper, 1996), and to determine the role of fatty acids in the aroma profiles of Swiss cheese (Harper et al., 1996). The meat industry has also benefitted from the electronic nose. Winquist et al. (1993) estimated the quality of ground beef, while Turhan et al. ( 1 998) detected adulteration of ground beef with pork. Metal

PAGE 62

42 oxide sensors were used to differentiate between six different brands of sausages (Tan et aL, 1995). In the seafood area, applications of the electronic nose have been done in differentiation of odors in shrimp stored on ice (Balaban and Luzuriaga, 1996), storage of fresh tuna at different temperatures (Newman, 1998), monitoring of haddock and cod freshness (Olafsson et aL, 1992), and recognition of fish storage time (di Natale et aL, 1996). Although the electronic nose is rapid and objective in quantifying odors, little work has been published on the correlation of electronic nose sensor data with sensory evaluation in seafood products. Pattern Recognition Techniques Pattern recognition techniques are being widely used today in different fields of research. There are a large number of pattern recognition techniques (multivariate statistics) currently available. The decision of what method to use depends on the objectives of the experiment, or what that the researcher is looking for. Some examples of multivariate analysis are: principal components analysis (PC A), discriminant function analysis (DFA), cluster analysis (CA), factor analysis (FA), multiple regression (MR). There are other methods, which are not purely statistical, but they could be used with multivariate data. These methods are artificial neural networks (ANN) and fuzzy logic. A summary of the most common techniques is listed in Table 2-7. Multivariate analysis deals with measurements which have several variables of interest (sensor outputs in the case of the electronic nose) that could be used to predict,

PAGE 63

Table 2-7. Common pattern recognition techniques used with electronic nose data METHOD PARAMETRIC SUPERVISED LINEAR Principal component analysis no yes yes Discriminant function analysis yes yes yes Multiple regression yes yes yes Cluster analysis no no yes Artificial neural networks no yes no Fuzzy logic no no no Adapted from: Gardner and Hines, 1997. discriminate or classify groups present in the data. Multivariate techniques attempt to extract information describing the simultaneous relationships between the variables (Corcoran, 1993). Principal component analysis Principal component analysis is a linear non-parametric technique used to extract relationship among possible groups present in the electronic nose sensor data. PCA is a linear technique and assumes that the response vectors are well described in Euclidean space (Bartlett et al., 1997). Principal component analysis puts together those sensor responses that are very similar, and relegates to a different component those sensor responses that are different from the first. The goal of PCA is to achieve an optimum description of a given data set in a dimension which is smaller than the number of variables (number of sensors). The reduction is usually done to a 2 or 3-dimensional space, so visualization of similarities and differences could be apparent. This method is mainly used

PAGE 64

44 when a-priori information on how the data are classified is not known, or when similarities among groups of samples that are being analyzed with the sensor arrays are desired (Powers, 1989). Discriminant function analysis Discriminant function analysis is a parametric procedure to identify relationships between qualitative criterion variables (i.e. sensory descriptors, odor classes, etc.) and quantitative variables (sensor response). It attempts to classify samples into known groups by constructing linear relationships for the sensor data and the criterion set of variables. These relationships will aid in the separation of odor classes or sensory descriptors. This method generates functions that are used to classify samples. Each discriminant function is calculated for which the F-ratio of the analysis of the variance is maximized (Gardner and Hines, 1997). Discriminant function analysis can also be applied using stepwise procedures. The benefit of stepwise DFA is that the model will find which variables (sensors) discriminate better among the different descriptors (odor, sensory, etc.). It will select the significant variables and construct the functions. Multiple regression Multiple regression is similar to DFA except that it is used when the predicting variables and the variable to be predicted are continuous. In general, multiple regression allows determination of which variables are the best predictors for the dependent variable or property that is being measured. Multiple regression can be used to correlate sensor data with physical or chemical measurements. Customarily, the degree to which two or

PAGE 65

more sensor readings (independent or X variables) are related to the dependent (Y) variable is expressed in the correlation coefficient R. Artificial neural networks Artificial neural networks have been widely used to analyze sensor data (Corcoran, 1993; Bartlett et al., 1997; Gardner and Hines, 1997). This procedure has several advantages over conventional techniques including adaptability, noise and fault tolerance, and fast operation once the network has been trained (Shurmer, 1990). A major disadvantage is the obscure nature of the classifier. Statistical techniques (DFA, PCA, etc.) can provide measures of confidence in the classification and also give additional information to help interpret which sensors are the most useful for discrimination. In neural networks, the knowledge is retained in the weights of the hidden layer (Corcoran, 1993). A typical ANN is composed of three layers. The input layer is made of input neurons, usually equal to the number of inputs or sensors present in the electronic nose. The hidden layer does the processing and classifies the data, and can have multiple configurations. The output layer gives the classification results. It usually contains the same amount of neurons as classes to be identified. Fuzzy logic Fuzzy logic has also been investigated for the analysis of sensor data (Berrie, 1 997). It mirrors human thinking and makes it easier for human experience to be integrated into decision-making or control algorithms. It uses membership functions to classify data. It tolerates uncertainties in input data and is capable of producing an output from noisy or incomplete data. Fuzzy logic complements classical methods by offering

PAGE 66

46 solutions where mathematical models reach their limits. In the case of sensor data, fuzzy logic applications are very limited but Berrie (1997) expects that fuzzy evaluation methods will establish themselves as reliable alternatives to conventional sensor signal processing. The application of pattern recognition techniques for odor identification is still at an early stage. More work must be done to properly classify odors based on sensor data by using the techniques mentioned earlier, or by developing new methodologies to analyze the vast amount of data generated from sensor arrays from electronic noses. Current Shortcomings and Future Improvements of the Electronic Nose The electronic nose systems rely on sensors made from different materials. These materials have a useful life of one year in the case of conducting polymers and two years for metal oxides. After this period of time sensors will have a reduced response to volatiles decreasing the signal to noise ratio (Bartlett, 1994). This loss in ability to respond to volatiles occurs gradually and is known as sensor drift. Sensor technology is getting better and new materials with less sensor drift are being used. Moreover, electronic nose manufacturers are correcting sensor drift by calibration procedures and data analysis. The electronic nose has great potential in quality control applications in markets worldwide. The system can be used to standardize odor detection in different factories and at different locations. However, transferability of data from one electronic nose to another (same manufacturer) has not been fully tested. Even though there are no data available in the literature regarding this issue, personal communications with electronic

PAGE 67

47 nose users have mentioned that data obtained with one instrument is different from that of another instrument with the same sensors. Data analysis methodologies must be developed to transfer data from one system to another for the electronic nose to become the standard machine for olfaction of food samples. Gardner and Bartlett (1996) have started exploring the performance definition and standardization of electronic noses. However, there are studies that have discussed the portability of data from one machine to another. Different types of sensors do not have similar discrimination abilities for volatiles coming from the same product. Robie (1997) compared two sensor technologies (conducting polymers and metal oxides) for their discrimination ability with tobacco products and flavors. In the case of cut tobacco, the conducting polymer sensors had discriminated better than metal oxides. However, in flavors from tobacco, such as menthol, metal oxides performed better. Therefore, research is focusing in finding which sensors respond and discriminate better in different food products. This will create a bank of sensors that customers can choose based on their product (Bartlet et al., 1997). Also, the concept of hybrid electronic noses are being introduced, where combination of sensor technologies can be used to optimize odor detection. Very soon the electronic nose market will have portable electronic noses that can be taken to the field, processing plants, fishing boats, etc. with rapid sensor responses.

PAGE 68

CHAPTER 3 COLOR ANALYSIS: SOFTWARE DEVELOPMENT Introduction Color and appearance are usually one of the primary criteria for acceptance or rejection of a food. Yet, important as this may be, there is a surprisingly wide variation in color which fits within preconceived notions of what the acceptable color of a food should be (Francis and Clydesdale, 1975). Moreover, color influences our sense of taste of many foods and our decision making process to select, purchase and/or eat different food products, especially produce and meats. Color can be measured and described in physical terms. However, the actual color that we perceive is the result of a complex series of processes in the human visual system (Rossotti, 1983). The full physical specification of a color stimulus is multidimensional. Each color has a specific radiance intensity at each wavelength in the visible spectrum (Giorgianni and Madden, 1998). This color representation is not an intuitive way to define the color of an object. Therefore, different types of color scales or systems have been defined over time to interpret color perception and to quantify color discrimination (Former and Meyer, 1997). Most color systems define their colors based on the tristimulus values X, Y and Z adopted by the Committee Internationale d'Eclairage (CIE, 48

PAGE 69

49 International Commission on Illumination), and represented in the two-dimensional graph known as the 1931 CIE chromaticity diagram. The important color systems used in food colorimetry are XYZ, xyY, Hunter Lab, or CIELab, and Munsell (Hutchings, 1994). Color in foods has been measured by visual inspection, spectrophotometry and tristimulus colorimetry. Most color measurements of foods are based on the reflectance of light from the surface of the food. Visual evaluation is very simple and fast, and results in an immediate perception of the color and appearance of the food. However, human perception of color differs from person to person, it is difficult to quantify the amount or intensity of the color, and to describe or define the color present in a food. Color measurements with spectrophotometers and tristimulus colorimeters are widely used in food samples with uniform surfaces and colors. The color measured by a colorimeter is described as a single value in the XYZ or L*a*b* color scales. In the case of foods with different colors or non-uniform surfaces, a single value is not a good representation of the actual colors present in the sample. Researchers had chopped or blended their samples to obtain a uniform paste and then be able to measure the "average"color. However, in many cases this procedure is inappropriate because the original color and appearance of the food is lost. Computer vision can be an alternative for food color measurement. In a computer vision system, an image of the food sample is digitized. The image is divided into small elements or regions, called pixels. Each pixel has the information of the levels of the three primary colors (red, green and blue = RGB color system). By using image processing and pattern recognition techniques the computer vision system can identify and classify an

PAGE 70

50 object based on color and appearance (Gonzalez and Wintz, 1977). This methodology can provide a complete description of all the colors present in the sample and the amounts of each color. With this procedure, one can look at each individual pixel and obtain the color information, therefore, samples with different colors throughout the surface can be analyzed very easily. Computer vision can handle samples of different shapes, sizes, surface texture, and colors. The only restriction is the size of the light box where the samples are placed. Therefore, the objectives of this study were to assemble the hardware necessary for a computer vision system, and to develop a computer program that can assess and quantify the color of foods with non-uniform colors and surfaces in the RGB, L*a*b*, XYZ or Munsell color systems. Hardware Used in the Color Computer Vision System The hardware for the color machine vision system consisted of a light box, color video camera, frame grabber and computer. The light box was built of 100% clear acrylic safety glazing sheets as described by Luzuriaga et al. (1997b). The box had top and bottom lighting with 2 fluorescent lights each (45.7 cm, 15 watt Chroma 50, F15T8-C50, General Electric, Cleveland, OH). The dimensions of the chamber were 42.5 cm (w) x 61 cm (1) x 68.6 cm (h). The inside walls of the chamber were painted white (flat white No. 1502, Krylon, Sherwin Williams Co., Solon, OH) to reflect light in all directions and to niinimize shadow formation. The chamber side of the boxes had a white translucent Polycast acrylic (No. 2447, Polycast Technology Corp., Stanford, CT) 6.35 mm thick with

PAGE 71

51 5 1 % light transmission. This was near the recommendation of Arnarson et al. ( 1 988) to provide diffused light. The CCD (charge coupled device) color video camera (Sony SSC-S20, Sony, Japan) with 24 bit color, 525 lines, and a horizontal resolution of > 460 lines was placed at the top lighting box between the 2 fluorescent lights. A lens (focal length 6-12 mm, model 12VM612, Tamron Industries Inc, New York, NY) was positioned 49.5 cm above the surface of the bottom lighting box. The S-video output of the camera was connected to a color frame grabber (Matrox Meteor, Matrox, Canada) placed inside a Pentium 133Mhz PC computer. The frame grabber was capable of acquiring 24-bit color images with 640 x 480 pixels digitization. Software Development The color analysis software was developed in Microsoft Visual Basic Professional (Version 6.00, Microsoft Inc., Redmond, WA). The program ran in the Microsoft Windows 95 and Figure 31 . Main steps of the color analysis program to obtain information from a color image Color Analysis Software Initialize program Grab image Open image file Obtain bii lary image Find objects (blobs) Extract blob information Extract RGB color data Extract L*a*b* color data Color calibration Display blob and color data Analyze color data

PAGE 72

52 Windows 98 operating systems. It required dynamic link libraries (DLL) from Matrox Imaging Library (Ver. 5.0, Matrox Image Processing Group, Quebec, Canada), Color Science Library (Ver. 2.0, Computer Graphics Systems Development Corporation, Mountain View, CA), Graphics Server (Ver. 5.10, Bits Per Second Ltd, Brighton, England) and Formula One (Ver. 4.1, Visual Components, Lenexa, KS). Figure 3-1 shows a schematic of the main steps used in the program to extract color information from a color image. The program (Figure 3-2) consisted of three main sections: image acquisition, image processing and data analysis display. Image Acquisition The program was designed to grab 24-bit color images from a video camera with a Matrox Meteor frame grabber installed in the computer. If such a device was not present, then the program could analyze images previously saved in a file. The program allowed the user to select the option to grab images from a video camera, a video player/recorder (VCR) with a S-VHS connector, or a VCR with a composite connector (Figure 3-3). When the program started, it checked for the presence of the Matrox Meteor frame grabber. If the program could not find the frame grabber, then the option to grab images was disabled. Image acquisition started by initializing the video camera. Software communicated with the frame grabber and with the video camera. During initialization the program asked for the maximum image size (pixels in the X and Y direction) and number of color bands (color images contain 3 color bands: red, green and blue). Based on the size of the

PAGE 73

53 Color Analysis ma File Edit Process Image Color analysis tools Qptions Window Help ft 241 G: 255 B: 233 x-134 y-2 Borders Alignment Pattern A1 [Filename" A B C D E F G H 1 iFilename .ego-2\PROGRAMS\Temp\shnmp.tif 2 Date 4/2/99 5 3 Time 7:50:34 AM 4 Blob Num. 0 1 2 3 4 Average 5 Number 0 1 2 3 4 6 Area (pixe 2079 1713 1614 2434 2220 2012 7 Area (user 0 0 0 0 0 0 8 Perimeter 363 306 270 429 390 351 6 9 MaxX in * i r 90 . ni 166 CO. ~ 249 99 220 a 3Q TV |\ Sheet! / Figure 3-2. Main screen of the color analysis software

PAGE 74

54 \ Options for Analysis , Binary threshold .'. ^ Color analysis Analysis options \ Blob analysis Color calibration — Show binary image at ^ Display color data as the end of analysis % of total area ic* i ui nni iSave colors with % T Select muttjpleROIs r hjgh er than: P Multiple data in same file / reset for new file Input Signal: P" Video camera r vca s-vhs <~ VCR. composite Apply Reset Close Figure 3-3. Options for color data analysis screen image and the color bands, space in memory was allocated for further processing. The CCD camera used in this system was limited to a maximum image size of 640x480 pixels. Once initialization was carried out, the program showed live video (continuous grabbing) from the video camera. The frame grabber was able to show 30 frames per second, however, the number of frames displayed in the computer monitor depended on the memory in the video card. During continuous grabbing, the program allowed the user to change brightness, contrast, hue or saturation of the image (Figure 34). If needed, these camera djust Digitizer Parameters JLl 4 J J J Save _] |148 Brightness _] |l 60 Contrast _] [o Hue |l 60 Saturation Restore Default OK Figure 3-4. Screen for video camera adjustments

PAGE 75

55 settings could be changed to approximate the color of the captured image to that of the real color of the sample. These four parameters ranged from 0 to 255. The program allowed the user to save or load previously saved camera settings. The default filename extension was *.cfg. The default values for the camera settings were: brightness 148, contrast = 160, hue = 0, and saturation = 160. When the user decided to capture an image, a command was sent to the frame grabber to freeze one frame and to display it in a window (Figure 3-5). This image was | Ft 146 G:120 8:81 x-384 y431 Figure 3-5. Screen of an image opened from a file or captured from a video camera

PAGE 76

56 then used for color analysis, or it was saved in a file. Images could be saved in four different formats: bitmap (consists of a rectangular matrix of numbers, where each number represents the appearance of that pixel. In the case of color images, the number is three bytes long, each byte consisting of red, green and blue values), TIFF (Tag Image File Format, Version 6.0), MIM (Matrox image format, a regular TIFF 6.0 file format with extra information included in the comment field) and raw format (the contents are dumped directly (byte stream) into the file and no header is added. Color bands are dumped one after the other). The software was able to read previously saved image files with the same formats described earlier. The software opened the image, obtained the size of the image from the header section of the file, and displayed it in a window. When the opened or grabbed image was displayed, the contents were placed in memory, in a three-dimensional array of byte data type: ColorArray (i, j, n) = red, green or blue color intensity from 0 to 255 where: i, j was the location of the pixel in the X, Y position in the image, and n had values of 0, 1 or 2, which represented each of the three primary colors: red, green and blue respectively. This array was used during image processing to extract the color information. Image Processing Several image processing techniques are used to extract size, shape and color information from the image. The first step was to find the objects present in the image,

PAGE 77

57 also known as blob analysis. The program was also able to look at regions of interest selected by the user, rather than blobs identified by the program. Once the blobs were found or regions of interest were chosen, color information was extracted. Blob analysis Blob analysis extracted connected regions of pixels (blobs or objects) within an image. This can be done when the object color is different from the background color. The software was capable of differentiating a clear background (white) from a colored foreground (objects), based on a given color threshold set by the user (Figure 3-6). This color threshold is an RGB value, which represents the minimum red, green and blue intensity values of the pixels present in the background. If the background were completely white, then the color threshold value could be set to RGB 255, 255, 255. However, the background of most images are not completely white, therefore the default value used in the program was RGB 220, 220, 220. This meant that for the default color threshold, a background pixel was defined as a pixel whose RGB values met the following conditions: red > 220 and green > 220 and blue > 220. Based on the color threshold, the program obtained a binary image from the 24-bit color image (Figure 3-7). Background pixels were painted black and objects were painted white. The binary image was saved in a two-dimensional array (Binarylmage (i, j), where i, j are the X, Y positions of the pixel in the image) of boolean data type (0 = black and 1 = white). This twodimensional array was used to extract color information from the foreground pixels. From the binary image, the program identified connected regions of pixels surrounded by background pixels (black). Each region was given a unique number known

PAGE 78

^Options for Analysis Binary threshold Color analysis Analysis options Blob analysis Color calibration jlJ J jJ |220 Red J _d |220 Green J jJ |220 Blue Threshold color: Apply Beset Close Figure 3-6. Screen for selection of the binary threshold Q go 2\PnOGRAMS\Temp\Shrimp1.tif | ft 250 G:254 B 249 x-237 y-9 Figure 3-7. Screen of a binary image

PAGE 79

59 as the blob number. This number identified each single blob for further analysis. Based on user input, the program deleted or excluded blobs that did not satisfy certain criteria (Figure 3-8). Small blobs could be eliminated from the final result. The program did not consider blobs with areas smaller than those input by the user as in Figure 3-8. Usually in an image of 640 x 480 pixels, small blobs (less than 100 pixels) are noise, unless objects in which the user is interested are very small. Also, if blobs were touching the edges of the image, the program allowed the user to exclude them from the analysis. When dimensions (area, perimeter, etc.) of the blob are important, then blobs touching the edges should not be included for further analysis. However, for color information only, blobs touching the edges can be considered during the analysis. A color image of the blobs that met the criteria set by the program were displayed in a window with a black rectangle surrounding each object. The rectangle boundaries are based on the maximum and minimum X and Y coordinates of the most extreme pixels in the object (Figure 3-9). This rectangle was used to verify that the software found the objects present in the image. A label was written in the upper right corner of each rectangle surrounding the blob. The label represented the blob number and the area in pixels of the blob. The label allowed the user to identify each blob, and refer to it in the color analysis results. The blob number, area and coordinates of the surrounding box for each blob were saved in memory for the color analysis part. The software analyzed each blob to extract several features and obtain different measurements (Figure 3-10). Most of these measurements were obtained by using

PAGE 80

60 \ Options for Analysis Binary threshold , Color analysis , . .. > Analysis options Blub analysis Color calibration r Connectivity C 4-neighbors P 8-neighbors Convert blob area to mm*, in 2 , etc. After analysis display: r Outline <• Boxes r None Remove blob touching. .. 17 Ed 9 es Min. area to remove blob: 100 App^ Reset CJose Figure 3-8. Options for blob analysis screen 'Q go ?\PROGRAMS\Temp\Shrimpl til | R Z50 G: 254 B: 250 x-619 y-14 Figure 3-9. Screen of an image after blob analysis and object identification

PAGE 81

61 Table 3-1. Parame 1 1 A t f* 11111.1 1 1 * ters calculated for each blob by the color analysis program PARAMETER DESCRIPTION Area (pixels) The number of foreground pixels in a blob (holes are not counted) Area (user units) Conversion from pixels to mm 2 , in 2 , etc. by using the image of an object with known dimensions Perimeter This is the total length of edges in a blob with an allowance made for the staircase effect that is produced when diagonal edges are digitized (inside corners are counted as 1.414, rather than 2.0). A single pixel blob (area =1) has a perimeter of 4.0. Maximum X, Maximum Y, Minimum X, Minimum Y These are the coordinates of the extreme left, top, right, and bottom pixels, respectively, of a blob. This coordinates were used to draw the rectangle that surrounds each blob. Center of gravity X, center of gravity Y This is the X and Y positions of the center of gravity of a blob. Convex perimeter This is an approximation of the perimeter of the convex hull of a blob. It is derived from several Feret diameters; so, a larger number of Ferets gives a more accurate result (Matrox, 1997) Compactness This value is a minimum for a circle (1 .0) and is derived from the perimeter (p) and area (A). The more convoluted the shape, the greater the value. It is equal to: p 2 /47tA Roughness This is a measure of how rough a blob is and is equal to perimeter / convex perimeter. A smooth convex object will have the minimum roughness of 1 .0. Length This is a measure of the true length of an object, although it can only be applied to certain object types because it is derived from the perimeter (P) and area (A) assuming that P = 2(length + breadth) and A = length* breadth. Breadth This is a measure of the true breadth of an object, with the same advantages and disadvantages as length. Elongation This value is equal to: length / breadth.

PAGE 82

62 (sataia asiaa nana a ee Arial Fonts di ia zJ BEE Add sheet I Delete sheet Borders 90% 3| III BBiLDBl * jo fl Pattern J. Sort Num format A1 |Filename ft a D c B r ft ** u n J * J Ulob Num. n U 4 1 2 3 A 1 6 7 f Number o 1 2 3 4 — ! 6 7 — | 6 Area (pixel 9687 10109 9B81 9681 8456 1235 9677 9511 98 7 Area (user 0 0 0 0 0 0 0 9 Perimeter 687 731 717 693 563 198 629 606 7 0 MaxX 591 153 310 448 416 364 266 585 5 10 MaxY 244 276 240 224 301 352 424 384 4 11 MinX 449 59 157 321 206 310 50 431 3 12 MinY 39 59 80 52 221 305 320 266 3 13 CogX 535 96 223 394 322 336 136 524 4 14 CogY 119 175 135 111 256 330 381 317 4 15 Conv. perir 537 230 480 368 413 138 399 338 5 16 Compactn 3.879 4.216 4.145 3.958 2.99 2.54 3.262 3.075 4 6 17 Roughnes 1 278 3.178 1.494 1.883 1.384 1.433 1.575 1.791 14 18 Length 312 335 328 316 247 84 280 267 3. 19 Breadth 30 30 30 30 34 14 34 35 20 Elongation 10.087 11.154 10.932 10 338 7.254 5 808 8.125 7.526 21 i\>\\ Sheetl / ILil, J Figure 3-10. Screen of a blob analysis report functions from the Matrox Imaging Library. Table 3-1 lists the parameters that were calculated for each blob. Most measurements were based on pixels. However, units of area or length are easier to understand and are better descriptors of the real dimensions of the object. Therefore, the program was capable of converting pixels into units of area (mm 2 , in 2 , etc.), by selecting in Figure 3-8 the check box labeled: "Convert blob area to mm 2 , in 2 , etc.". This conversion of units was done by grabbing an image with an object of known area. The area was entered by the user in the program, and the software calculated a conversion factor automatically.

PAGE 83

63 Analysis of regions of interest The program also allowed the user to select regions of interest to be analyzed, rather than analyzing an entire blob or all blobs present in the image. In this case, the user has the flexibility to select a rectangle portion of the image and obtain the color information of that area only. This was done by dragging the mouse on top of the image and selecting the region of interest (Figure 3-11). Binarization was done on the region ofinterest to eliminate any background pixels present in that region. Except for area, blob parameters (Table 3-1) were not calculated for the regions ofinterest. .go-2\PROGRAMS\Tcmp\Shrimp1 tit Add ROI Analyze color | gear all ROI R:251 Q.2SA B: 250 x161 y-97 Figure 3-11. Screen with regions of interest selected by the user

PAGE 84

64 Extraction of color information The color machine vision system was able to look at more than sixteen million colors (256 3 , the combination of the three color bands' intensities), but due to the complexity of displaying the frequency of sixteen million colors, the colors were regrouped in 64, 512 or 4096 colors (Figure 3-12). Each color band was divided in four, eight or sixteen sections respectively. Therefore, the three dimensional color array for the 64 colors was 4x4x4 (Red x Green x Blue), for the 5 12 colors was 8x8x8, and for 4096 colors was 16x16x16. Each one of 16 million colors belonged to a color block. Therefore, any color in the block was assigned to the block's center color. Color blocks were identified by numbers. In the case of 64 colors blocks, numbers ranged from 0 to 63, color block 0 was black (RGB 32, 32, 32), while 63 was white (RGB 224, 224, 224), similar color numbering schemes were used for the 5 12 and 4096 color blocks. Color data was extracted from each blob. The program looked at every pixel in the blob and read the RGB values, which were previously stored in an array (ColorArray (i, j, n)). The RGB values were converted to the 64, 512 or 4096 color blocks (Eq. 3-1, 3-2, 3-3), depending on user selection (Figure 3-13). Figure 3-12. Grouping of color blocks in the RGB color space. Source: Precetti(1995).

PAGE 85

65 Options lor Analysis m Binary threshold Color analysis Analysis options Blob analysis / V Color calibration Color scheme for color histogram ^ 4 colors per axis (64 r 8 colors per axis color blocks) (51 2 color blocks) P 16 colors per axis (4096 color blocks) f~ Data as color spectrum Apply Reset Close Figure 3-13. Options for color scheme selection screen 64 color blocks = (B\64) + (G\64)(2 2 ) + (R\64)(2 4 ) (3-1) 512 color blocks = (B\32) + (G\ 32)(2 3 ) + (R\32)(2 6 ) (3-2) 4096 color blocks = (B\ 16) + (G\ 16)(2 4 ) + (R\ 16)(2 8 ) (3-3) where: R = red (0 to 255), G = green (0 to 255) and B = blue (0 to 255). "\" = division of two numbers that used the integer portion of the result only. Color information was saved in memory as the amount of pixels belonging to each one of the color blocks in the blob. A user-defined data type variable was used to store the color information for each blob: BlobColorHisto(m).ColorHisto(n) = number of pixels where: m = the blob identification number (0 to total number of blobs in the image) and n = the color block number (0 to 63, 51 1 or 4095 depending on user selection). This variable saved the amount of pixels present in blob m that belonged to the color block n.

PAGE 86

66 Color was also measured in the L*a*b* color system. The program read the RGB values for every pixel in the blob. Each value was converted to the XYZ color system, and then this number was converted to the L*a*b* color system. Conversions were done using functions from the Color Science Library, which were based on equations described by Wyszecki and Stiles (1 982). The L*a*b* values for every pixel in the blob were averaged using equation 3-4. '» Jr. (3-4) where: n = the blob number in the image, ij = X,Y coordinates of the pixel of blob n, A = area of blob n in pixels, and L = L value from the L*a*b* color space. Similar equations were used to calculate the average a and b color parameters. Standard deviations for the L, a and b values were also calculated. Color calibration The RGB color system in a machine vision system is device dependent (Former and Meyer, 1997). Temporal changes of the light source, reflection from adjacent objects and the camera's sensitivity will cause objects to appear slightly different. Therefore, color calibration is needed to correct the colors of the acquired image. Color calibration is done by comparing the values of the observed color with those of a color standard. Calibration procedures for color images have been developed by Hetzroni and Miles (1994) and Chang (1994). They involved extensive computations based on the spectral

PAGE 87

67 power distribution of the light source, the spectral reflectance of the object and the spectral response of the video camera's detector. Implementation of procedures developed in those studies was beyond the scope of this project, however they will be necessary if this program will be used to compare color data from images taken at different times, in different sites and under different hardware conditions. In this study, a one-point calibration procedure was used. The program acquired the image of a color standard of known XYZ or L*a*b* values. The known values of the standard were saved in memory and compared with the measured L*a*b* values from the image of the color standard captured by the system (Figure 3-14). The difference of the L*a*b* values was used as the correction factor. Calibration was implemented in the images by reading every pixel in the image, converting the RGB value of each pixel into the L*a*b* system, applying the correction factor and then converting it back to RGB. Table 3-2 shows the L*a*b* values for commercial color standards (Gardner Laboratory Inc., Bethesda, Maryland and Hunter Associates Laboratory, Inc., Fairfax, Virginia),and the measured Options for Analysis El Binary threshold / s Color analysis / \ Analysis options / v. Blob analysis 17 Apply color calibration to image C Known XYZ shift values x |61.6 <* Calculate XYZ shift values Y |63.8 |Light yellow X-61 . 6 Y=63.8ZM4.3 wj Z |443| Reset Color calibration Select a color tile or enter the XYZ values for the color standard CJose Figure 3-14. Options for color calibration screen

PAGE 88

68 values using a colorimeter (Colorgard 2000, BYK-Gardner, Bethesda, Maryland) and the color machine vision system developed in this study. Table 3-2. Comparison of known L*a*b* values of color standards with L*a*b* measurements from a colorimeter and the color machine vision system developed in this study Color standard" Color standard values Colorimeter measurements b Machine vision measurements 0 L a b L a b L a b white 92.7 -1.2 0.2 91.3 -1.1 0.3 98.7 -5.2 6.6 dark gray 21.8 -0.8 -0.3 22.3 -0.7 -0.3 26.0 -7.7 5.5 light gray 44.7 0.7 1.2 45.6 0.4 1.1 50.9 -8.8 11.5 light yellow 78.3 -1.8 22.3 79.9 -2.1 23.0 87.2 -13.0 40.4 dark yellow 74.3 6.4 47.8 75.5 5.1 48.6 82.5 -3.3 83.1 olive 31.3 -2.9 18.0 31.2 -3.4 18.0 34.1 -12.3 33.2 orange 43.9 19.7 24.0 44.3 18.7 25.0 46.0 14.3 44.4 pale red 24.2 19.7 5.4 24.5 22.6 5.0 24.6 9.2 6.9 a : descriptive color name of the color standard b : average of three measurements c : average of 4270 pixels The known values of the color standards and colorimeter measurements were done with an illuminant C (represents average daylight with a color temperature of 6774°K (Wyszecki and Stiles, 1982)). However, the machine vision system used a 5000°K fluorescent light (illuminant D 50 , which simulates color qualities of noonday summer sun (General Electric, 1993)). Hardware conditions were different; therefore, the measured L*a*b* values with the color machine vision system varied from the color standard values. On the other hand, the values measured with the colorimeter were fairly close to the color

PAGE 89

69 standards, since both were measured under the same lighting conditions. This is important because measurements with a colorimeter and the machine vision system cannot be compared directly. Color data should be transformed from one light source to the other using equations (Wyszecki and Stiles, 1982), or an approximation can be obtained by implementing the color calibration procedure to the image. Once calibration was applied to the image, the software allowed the user to see two images on screen, the original image and the calibrated image. Color data were then obtained from the calibrated image. Color calibration should be done with a color standard similar to the color of the sample to be analyzed. When reporting color data, the color standard should be mentioned to fully describe the conditions under which the color of a sample was measured, as suggested by Chervin et al. (1996). Data Display Data generated by the color analysis software were presented graphically and numerically. Data were divided in two types: blob analysis data and color data. Blob analysis reports Blob analysis data refers to physical measurements (Table 3-2) of the blobs present in the image. As explained earlier, blob analysis data were displayed in a spreadsheet, which was included within the color analysis program (Figure 3-10). The spreadsheet was developed using subroutines from the Formula One software. The spreadsheet had all the basic functions of any commercial spreadsheet. Both blob analysis and color data obtained by the program were displayed in the spreadsheet. Data from the spreadsheet were saved

PAGE 90

70 in a Formula One file format (*.vts), text format (*.txt) or in Microsoft Excel format (*.xls). Color data reports The color analysis program generated a matrix of n x m color data, where n = number of color blocks (64, 512 or 4096), and m = number of objects in the image. Depending on what the user selected in the program (Figure 3-3), color data were reported in pixels (frequency of pixels of any given color block in the blob) or percentage (percent of the total area of the blob covered by a given color block). Color data were displayed in the spreadsheet, where columns represented blobs in the image and rows represented color blocks (Figure 3-15). The last column displayed in the spreadsheet was the average for the blobs present in the image. Color data also included the average and standard deviations of the L*, a* and b* values for each blob. Figure 3-16 shows an example of the L*a*b* data, and the area of each blob used to calculate the average and standard deviation. Graphical representation of the color data were displayed in the form of histograms. Figure 3-17 shows the window where the user selected the blobs and colors of interest to be plotted. The program allowed selection of all color blocks to display a full color profile of any blob, or selection of color blocks whose values were above a certain number of pixels or percentage of the total area of the blob. This helped simplify color data and look only at the most important colors present in a sample. Color data histograms showed the color block numbers in the X axis and the area of the blob or amount of pixels belonging to that color block in the Y axis (Figure 3-18). The painted

PAGE 91

71 colors in X axis in Figure 3-18 represented the actual color at the center of the color block. This gave a visual appreciation of the colors present in the sample. The graphs were plotted in the Windows Metafile format and could be saved to a graph file or copied and pasted in another application. Data used to generate each graph could be exported to the spreadsheet, allowing the user to plot the data in another graphics software. The program displayed a descriptive list of the 64, 512 or 4096 color blocks (Figure 3-19). The list was displayed in a spreadsheet. It contained the RGB values at the center of each color block, the corresponding L*a*b* value, the NBS color name, and a painted rectangle with the color of the center of the block. Color data were also visualized as histograms of red, green and blue pixel intensity values (Figure 3-20). Histograms could be generated for each object in the image or for all of them The X axis represented the red, green or blue intensity values from 0 to 255, and the Y axis was the number of pixels for each intensity value. These histograms did not provide specific information of the colors present in the image, but showed the distribution of the intensities of each color band. The general brightness and contrast characteristics of the image could be determined from these histograms. If pixels were clustered in the low intensity values, it was expected to have a dark image. Conversely, higher values were related to bright images. Color conversion utility The color conversion utility converted color data from one color scale to another (Figure 3-21). The color scales used were: RGB, Color blocks (64, 512 and 4096), L*a*b*, XYZ, and Munsell. Equations used to convert from one color space to another

PAGE 92

72 Color Analysis 5le Edit IaE Erocess Image 2°'°' annlysis tools Options ifVjndow Help ~l m a| &\ x| i t Sheet9 bibb sbub ma ai aaai 100% Atial 1TB Fonts | Add sheet | Delete sheet | Borders | Alignment | Pattern Sort | Num format | 3 A1 IRIename B 539 J____ 540 LabL 541 StdDev L H 542 543 544 Lab a StdDev a Lab b 545 StdDev b 48.5 13.68 4.07 6.36 26.08 7.38 52.28 13.83 9 16 5.94 32.31 9.63 58.14 14.38 8.28 6.11 29 85 9.69 59.36 12.4 5.55 5.84 29.72 7.63 63.98 ,10.53 •*6.14 5.29 35.24 876 43.7 16.35 5 95 4.06 11.9 763 56.03 11.54 7.98 6.01 33.35 8.73 I 5 1 546 NBS Namelight olive bstrong yellclight yellow light yellow light yellow grayish brcstrong yellcmoder_ 547 1 1 »KSheet1 f Figure 3-15. Screen with color data report \mxmm 1SI m m OH = EEJLU i h°°* Zl I f| Bt Ml HID! Arial jjpo ~j [Bj[£j[aJ HO I L*J® T](2£] tdS SS I *l Fonts | Add sheet | Delete sheet | Borders | Alignment | Pattern Sort | Num format I A540 |Numb er A c O E F G H I * 540 Number 1 2 3 4 5 6 541 542 Lab L StdDev L 60.62 13.85 521 49.47 14.97 53.43 59.43 65.02 55.21 45.15 5< 15.19 15.65 7.85 11.79 5.76 14.81 17.87 1: 543 Lab a 3.97 8.85 7.78 5.79 644 StdDev a 6 03 6.36 6.15 6.45 5.6 6.17 4.11 645 Lab b 29.28 25.78 31.8 29.38 34.65 29.43 11.71 3: 646 647 648
PAGE 93

73 Plot color data SELECT DATA TO PLOT Graph title Color data Data sets (spreadsheet 0 B.\Shrimp1 01 1 C : \Shrimp1 tit 2 2 D (Shiimpl tH 3 3 E \Shrimpl tit 4 A . F:\Shrimp1.tif5 5: G:\Shrimp1.trf6 6: H:\Shrimp1.tif7 7: l:\Shrimp1.tif8 8: J:\Shrirtip1.W9 9: K:\Shrimp1.tifavg Select data set(s) to be graphed 'YaxistKle | Color (% of total area) Colorthreshold (x > threshold) |i c All data sets P Selected data sets Update color list <* Group data by colors Group data by data set Show data labels r Show legend ^ Show colors in X axis Export data Select ell data sets Figure 3-17. Plot color data selection screen Color Analysis Graph 3 |-|q|xl| Color data 209 218 281 282 290 346 354 355 418 419 427 «" 20-S s Shrimp 4 Figure 3-18. Screen with a histogram of color data

PAGE 94

74 y 64 color blocks Block* R G B L a b NBS Name NBS# Color 16 96 32 32 16.93 30.30 18.26 deep reddish brown 41 17 96 32 96 19.54 38.49 -24.20 very deep reddish purple 239 18 96 32 160 25.26 53.50 -58.82 vivid violet 205 19 96 32 224 32.85 70.25 -87.53 vivid purplish blue 194 20 96 96 32 34.21 -8.88 36.24 moderate olive 107 21 96 96 96 35.39 0.00 0.00 dark gray 266 22 96 96 160 38.41 18.55 -37 54 strong purplish blue 196 23 96 96 224 43.20 40.99 -7050 vivid purplish blue 194 24 96 160 32 56.28 -42.32 55.91 strong yellow green 117 25 96 160 96 56.86 -3622 29.34 strong yellowish green 131 26 96 160 160 58.42 -21.57 -6.74 light bluish green 163 27 96 160 224 61.11 -051 -41.79 brilliant blue 177 28 96 224 32 78.23 -67.47 74.34 vivid yellowish green 129 29 96 224 96 78.58 -63.42 55.26 vivid yellowish green 129 30 96 224 160 79.51 -53.05 23.35 brilliant green 140 31 96 224 224 81.18 -36.52 -11.04 brilliant bluish green 159 32 160 32 32 31.97 51.92 38.20 vivid red 11 33 160 32 96 33.26 55.63 -1.97 deep purplish red 256 34 160 32 160 36.52 64.37 -4003 vivid purple 216 35 160 32 224 41.61 76.71 -72.79 vivid violet 205 36 160 96 32 42.60 23.91 46.81 strong brown 55 37 160 9R 9fi 43 48 ?R31 1210 nrns/ish rp.ri 19 Figure 3-19. Screen with the description of the 64-color block scheme

PAGE 95

75 Green Color Histogram for Blob # 2 250t Figure 3-20. Screen with histograms of red, green and blue color intensities

PAGE 96

76 / \ Color Conversion r| 0.0435 |~~73 G | 0.6334 [212 B| 0.0735 I 90 Color blocks 64 [29 Average color 64 ^2 spectra 512 pTT 4096 1237 L*| 74.03 a1 -64 45 b*| 52.82 X |2577 Y I 46.76 Z " 14.62 Munsell System H| 0 50 |G _tJ V| 7 25 C| 13.60 0.50G 7.25/13.60 \ Recalculate Show Color Dictionary NBS System name jvivid yellowish green (129) Delay (ms) T ~3 J Figure 3-21. Color conversion utility screen were obtained from the Color Science Library, which based its calculations on equations reported by Wyszecki and Stiles (1982), and Billmeyer and Saltzman (1981). Additionally, it gave the name of the color based on the ISCC-NBS (InterSociety Color CouncilNational Bureau of Standards) Color Dictionary (ISCC-NBS, 1955). The NBS color dictionary provides precise description of 267 color names in terms of portions of the Munsell color system. The naming system is based on 28 hue names plus five neutrals, and a set of modifiers. Each color is also described by a tristimulus value of the centroid of each color. The program interacted with the color conversion utility to give color information of any given pixel in the image. This was done by moving the cursor over the image and pointing at any pixel. The color conversion utility described that pixel color in the color scales mentioned above, it gave a visual representation of the color, and gave the NBS

PAGE 97

77 color name (Figure 3-22). This allowed the user to obtain a quick overview of the color or colors present in the image. Color Data Analysis Color data analysis is important in extracting meaningful information out of the large data sets generated from the software. Several techniques could be used to analyze £2 Colur Analys File Edit Process Image Color analysis tools Options Window Help \3 S If Color Conversion ...ego-2\PROGRAMS\Temp\Sal... n x r| 0.6527 |215 G| 0.1195 [7o9~ B I 0 0226 \~5B Color blocks 64 [52 64 G~ spectra 512 (ibT 4096 3427 Show Color Dictionary Average color L" a* b* 54 64 41 88 50 99 31.60 22.59 483 Munsell System "otT [yr~T| 530 H V, cf 12 07 0 73YR 5.30/1 2.07 Recalculate Delay (ms) NBS System name | strong reddish orange (35) Figure 3-22. Color conversion utility screen interacting with a color image

PAGE 98

78 these multivariate data sets, where each color block can be considered as a variable. For a given sample, a 64, 512 or 4096-color profile was obtained. This color profile would be a unique description or fingerprint of the sample. This fingerprint can be used for identification and classification of samples based on color. Depending on the application and the information needed, several approaches were used in this study to analyze color data. In samples where color changes over time were being studied, a plot of color trends were extracted from the data. The most important colors were selected and plotted over time. The quantity of some colors in the sample diminished and even disappeared, while others increased and in some cases new colors appeared. This approach helped visualize gradual changes of colors, and estimated rates of color loss or color formation. Equations were fitted to these color trends to predict color changes, or to predict other attributes correlated with the color of the samples. In samples with a uniform color, average L*a*b* values obtained by the software were used to describe the color of the sample. This is similar to those obtained with a colorimeter. The L*a*b* measurements were mainly used when samples had a uniform surface color, otherwise results were meaningless. Average L*a*b* values of two or more colors will give another color which is not related to the original ones. Multivariate statistics, such as discriminant function analysis and multiple regression were used to classify samples based on their color profiles. Classification functions were obtained to help classify samples or predict a certain attribute of the sample based on its color profile. These techniques were useful when dealing with multiple

PAGE 99

79 variables, and when correlating color data with other measured attributes such as sensory analysis or other quality attributes. Furthermore, multivariate statistics could be used to correlate quality attributes of a sample with a combined data set of color data and electronic nose sensor data. This will generate overall predictive or classification functions based on visual and odor properties of the sample.

PAGE 100

CHAPTER 4 COLOR AND MELANOSIS EVALUATION ON DIFFERENT SPECIES OF SHRIMP STORED AT DIFFERENT TEMPERATURES USING COMPUTER VISION Introduction Shrimp quality is usually evaluated based on visual, odor and texture properties. Current evaluation and inspection of shrimp relies mainly on sensory tests, and to some extent on instrumental analysis. Visual quality is measured by inspectors who evaluate the color of shrimp and look for visible defects such as melanosis (black spots formed by enzymatic activity). Luzuriaga et al. (1997b) reported that color of white shrimp changed during storage. However, there are no data regarding other shrimp species. Color changes are caused by chemical and microbiological changes in the tissue. Amines are being produced due to microbial growth, which increases the pH of the tissue. When there is a high level of amines, usually the shrimp turns pink, similar to when shrimp is cooked. This change in color is due to denaturation of the proteins. However, when these color changes appear, the shrimp most probably has a strong fishy odor. On the other hand, melanosis or black spots can be formed immediately after harvest. If shrimp is not properly handled and cooled, the enzyme polyphenoloxidase will start forming dark colored spots on the shell of the shrimp. This color is not appealing to the consumer and 80

PAGE 101

81 therefore, shrimp will be considered of poor quality, price will be reduced, or the end use of the shrimp will be different (peeled, breaded, etc.). There is no standardized instrumental technique for color and/or melanosis evaluation of shrimp. Color of shrimp is difficult to measure with colorimeters. Colorimeters measure the reflected light (color) from a sample by shining a beam of light on its surface. The result from the measurement is a single tristimulus color value in the L*a*b* or XYZ color scale. In the case of shrimp, which have different colors, different sizes and non uniform surfaces, a single tristimulus value is not a good representation of the actual colors present in the sample. Therefore, new procedures are needed to quantify the colors present in shrimp. Computer vision has been widely used to automate quality evaluation. In the agricultural and food sector, this has included non-destructive inspection and classification of rough rice (Cardarelli and Berhardt, 1998), detecting cracks in eggs (Elster and Goodrum, 1991), peanut maturity from images of surface texture (Ghate et al., 1993), and grading of apples (Rehkugler and Throop, 1989), peaches (Miller and Delwiche, 1988), asparagus (Rigney et al., 1992), tomatoes (Sarkar and Wolfe, 1984), and citrus (Miller and Drouillard, 1997) based on color, ripeness and/or blemishes. Fish and prawns have been automatically sorted according to shape, length and orientation in a processing line (Kassler et al., 1993; Strachan, 1993; Strachan et al., 1990). Computer vision was used to calculate the weight, uniformity ratio and count of shrimp (Balaban et al., 1994), and to measure color and melanosis of white shrimp (Luzuriaga et al., 1997b). Computer vision and image processing techniques could be alternatives to quantify the color of shrimp

PAGE 102

82 samples, which can help determine the visual quality of this valuable commodity with a standardized, fast and simple procedure. The overall objective of this study was to use a color machine vision system to evaluate visual attributes of different species of shrimp (brown, pink, tiger and white). The specific objectives were: 1) to differentiate, measure and quantify the color of different species of shrimp during storage at three different temperatures; and 2) to objectively measure melanosis levels in different species of shrimp stored at three different temperatures. Materials and Methods Shrimp samples Frozen, headless, 35-44 count/Kg shell-on shrimp (brown Penaeus aztecus . pink P. duorarum , tiger P. monodon and white P. vannamei ) were purchased from Lombardi*s Seafood (Orlando, FL). Shrimp were not treated with sulfites. Frozen blocks of shrimp were thawed under running tap water. Each shrimp specie was divided into 3 groups, each group containing 25 individual shrimp, except for brown shrimp which had 24. Each group was stored at a different temperature: 1.5, 7.0 and 13.0°C, for 1 1, 7, and 4 days, respectively. Shrimp were stored in 1 gal freezer Ziploc bags, which were not closed during storage, and which had holes cut throughout the bag, allowing air circulation for melanosis development.

PAGE 103

83 Color Analysis Every day of the study, 24-bit color images of shrimp were obtained using the computer vision system described in chapter three. Eight to nine shrimp were placed in the light box with front and bottom lighting, the door of the light box was closed, and the image was captured. A total of three images were captured every day for each specie at each storage temperature to accommodate the 24 or 25 shrimp. The settings for the video camera were: hue = 0, saturation = 170, contrast = 180 and brightness =168 (settings ranged from 0 to 255). After the images were captured, the shrimp were placed back in the plastic bag and stored in the pre-established cooler temperature. Shrimp remained at room temperature for no more than 1 5 minutes during image capturing. Images of the same side of the shrimp were obtained throughout the study. Images were saved as Tiff (version 6.0) format for further color and melanosis analysis. The color analysis program described in chapter three was used to extract the color information from the images. During image acquisition a standardized color tile was placed next to the shrimp sample in the light box. For brown, pink and white shrimp an orange color tile (L = 43.9, a = 19.7, b = 24.0) was used, while for tiger shrimp a dark green color tile (L = 3 1 .3, a = -2.9, b = 1 8.0) was used. Images were calibrated in the color analysis software using the values of the color standards. The image calibration was done prior to extracting color information. Color of the calibrated images was reported in the RGB and L*a*b* color systems. In the RGB color system, it was reported in the 64color blocks scheme (Appendix A). A 64-color block histogram was obtained for every

PAGE 104

84 shrimp throughout storage. The value for each color block was the percent of the area of an individual shrimp covered by a given color block. The summation of values for all color blocks was 1 00%, which was equal to the total surface area of the shrimp. Shrimp melanosis was also reported as the percentage of the area of the shrimp covered by dark spots. Melanosis is not only represented by black areas, but also areas that were dark or brown (Luzuriaga et al., 1997b). The melanotic areas for each shrimp specie were further analyzed using the Color Analysis software to identify the color blocks of the areas considered as melanosis. Raw data can be obtained from Dr. Murat Balaban, Food Science and Human Nutrition Deptartment at the University of Florida, filename: WthesisUext files\chap 4\shrimp species color data.txt. Data Analysis Color data from day 0 (immediately after thawing) were analyzed in Statistica for Windows ('98 edition, StatSoft Inc., Tulsa, OK) using discriminant function analysis (DFA). DFA was used to construct predictive functions to classify shrimp species based on the differences in color of the shrimp samples. Seventy five shrimp per specie were included for analysis. Color blocks with color values above 2% were selected for analysis, and were reduced to 2 discriminant functions to calculate the correct classification rate. Average color profiles for the 75 shrimp were plotted to observe the differences in the color patterns among the four shrimp species. The color profiles included the 1 0 most important color blocks present in the 4 different shrimp species. Also the average L, a

PAGE 105

85 and b values for the 75 shrimp were obtained using the color machine vision system and the Color Analysis software. Color data were also analyzed by plotting selected colors over time and looking at trends. The five most representative color blocks for each shrimp specie were selected. Changes in the average L, a and b values for the different shrimp species were obtained and plotted over time. Melanosis data (percent of the view area of the shrimp covered with melanosis) were plotted over time to observe the development of black spots in the different species of shrimp. Results and Discussion Shrimp species can be easily differentiated by looking at the color of the raw shrimp. However, color machine vision requires pattern recognition techniques to differentiate them Figure 4-1 shows the discrimination of shrimp species based on color data from the 64-color block scheme, using DFA as the pattern recognition technique. The overall classification rate for the DFA model was 91 .9%. This model was obtained by combining the color data from day 0 of the three storage temperatures for each specie, resulting in 75 shrimp per specie. From the 64-color block data, only 19 color blocks were selected. These color blocks were present in the different species in levels higher than 2% of total area of the shrimp view area. Based on the DFA model Tiger shrimp were 100% correctly classified, meaning that the color profiles for tiger shrimp were different from the other species tested in this study. However, brown, pink and white shrimp were not perfectly separated. Brown shrimp had similar color characteristics to

PAGE 106

86 Shrimp Species o Brown Pink o Tiger -5 1 • . . . . . . , i -6^-2 02468 10 A Whlte Function 1 Figure 41 . Discrimination of shrimp species based on the 64-color block data, n = 75 shrimp per cluster white and pink. As seen in Figure 41 , brown shrimp were in between the white and pink shrimp clusters. The individual classification rates for brown, pink and white shrimp were 86.1%, 88.0% and 93.3% respectively. All the missclassified samples of pink and white shrimp were classified as brown. However, none of the pink shrimp samples were classified as white, and vice versa. This means that pink shrimp can be differentiated from white shrimp based on the 64-color block scheme. Even though brown shrimp formed a cluster, some brown shrimp have color profiles similar to those of pink or white shrimp. Figure 4-2 shows the color profiles for all four shrimp species. From the 19 color blocks that had color levels higher than 2%, only the 10 most important color blocks were selected to plot the color profiles. It can be seen that tiger shrimp has a completely different color profile than the other three species. However, brown, pink and white had some similarities, corroborating the results form the DFA. The most important color for

PAGE 107

87 all shrimp species was color 41 (RGB 160, 160, 96), which is responsible for the pale grayish olive color. On average, 46% of the area of brown and white shrimp was covered by this color, while in pink and tiger it was 29%. Two other important colors were color block 57 (RGB 224, 160, 96), which is the moderate orange found in pink shrimp, and color block 21 (RGB 96, 96, 96), which is the dark gray present in tiger shrimp. Both of these colors had average color values of 25%. Most all other colors were close to 10% of the area or below. This demonstrated that shrimp is a multicolor object, and therefore a single tristimulus value (XYZ or L*a*b*) will not be sufficient to quantify the colors present in a shrimp sample. In fact, the Color Analysis software measured the color of the shrimp in the L*a*b* color scale. Average L, a and b values and their respective NBS 70 60 50 Color Blocks Brown Pink Tiger White Shrimp Species 58 42 41 25 20 62 36 37 21 57 n = 75 Figure 4-2. Color profiles of different species of shrimp immediately after thawing. Bars = average of 75 shrimp, whiskers = ± 1 standard deviation

PAGE 108

88 Table 4-1 . Average color values (% of total view area of shrimp) for different species of shrimp. Color was measured immediately after thawing the shrimp (Day 0). (n = 75) Color RGB values Color Shrimp Specie Block No. representation Diuwn PintI 11 IK Tiger White 41 160, 160, 96 Ave. 46.7 29.1 28.7 45.5 St.dev. 14.64 14.66 12.98 9.69 57 224, 160, 96 Ave. 6.6 25.5 0.1 1.8 St.dev. 10.22 18.36 0.02 2.19 21 96, 96, 96 Ave. 4.3 1.1 24.6 9.3 St.dev. 4.78 1.14 10.83 8.86 37 160, 96, 96 Ave. 13.4 11.0 0.1 6.8 St.dev. 8.31 10.13 0.15 3.49 36 160, 96, 32 Ave. ? 4.5 8.5 0.1 1.2 St rlev JL.UCY. 4. 12 7.54 0.01 0.91 62 224, 224, 160 Avg. 5.4 9.2 4.9 8.7 St.dev. 1.95 3.69 1.63 5.08 42 160, 10, 160 Avg. 5.0 2.0 10.3 9.8 St.dev. 2.66 0.99 5.79 5.19 20 96, 96, 32 Avg. 5.7 2.1 7.0 5.8 St.dev. 5.71 3.10 5.70 2.79 25 96, 160, 96 Avg. 0.1 0.1 13.3 0.7 St.dev. 0.06 0.03 5.76 1.66 58 224, 160, 160 Avg. 3.1 4.0 0.1 3.7 St.dev. 2.70 2.33 0.04 3.64 TOTAL (%) 94.8 92.6 89.2 93.3 Table 4-2. Average L*a*b* values for different species of shrimp, as measured by the color machine vision system immediately after thawing the shrimp (day 0). RGB values and color names are given, (n = 75, ± = standard deviation) Shrimp Specie Brown Pink Tiger White LabL* 57.15 ±5.18 61.91 ±6.04 53.43 ± 4.62 58.32 ±5.33 Lab a* 2.30 ± 3.03 8.01 ±3.67 -10.71 ±0.99 -2.62 ± 1.59 Labb* 27.95 ± 3.60 34.02 ± 3.40 18.58 ±2.35 23.47 ± 2.79 NBS Color name dark greyish yellow light yellowish brown moderate yellow green dark greyish yellow R 166 188 132 159 G 144 152 142 150 B 101 102 107 111 Color Representation

PAGE 109

(National Bureau of Standards) color name for the 75 shrimp from each specie are shown in Table 4-2. Both brown and white shrimp L*a*b* values were classified as a dark greyish yellow. Average L*a*b* values were similar for both species. Pink shrimp was considered a light yellowish brown, which is far from the characteristic pink or orange color of this specie. Tiger shrimp has the most non-uniform color distribution, however, the average L*a*b* values gave a moderate yellow green color. Tiger shrimp is mainly greyish blue with green, yellow, black and some blue lines. Therefore, a unique color value will not give the true representation of the color of the shrimp. It is important to observe the amount of variation in the color data (Table 4-1). Even though there were 75 observations for each specie, the standard deviations in some cases were larger than the average color value (Table 4-1). This showed that shrimp color is not an easy parameter to measure, considerable number of replicates are needed, and good pattern recognition techniques are required to evaluate the visual quality attributes of shrimp based on color. In this experiment the shrimp for each specie came from the same location, season, harvest and processing conditions. These variables must be added to develop a comprehensive model to differentiate shrimp species based on color. Color of shrimp changed during storage (Figures 4-3, 4-4, 4-5 and 4-6). The figures show the average color values (n = 25) of the 5 most important color blocks for each specie. Data for these figures are listed in Tables 4-3, 4-4, 4-5 and 4-6, with their respective standard deviations. Colors changed in different patterns, some increased during storage, while others decreased. Some colors remained fairly constant, while others increased slowly and then they decreased. The trends for the different storage

PAGE 110

90 Brown Shrimp 1.5°C 50 | 40 I 30 1 o 20 | 10 u ' 3 4 5 6 7 Storage time (days) Brown Shrimp 7.0°C 3 4 Storage time (days) Brown Shrimp 13.0°C 10 Color Block No -— 41 — 57 — 20 — 37 -— 21 Color Block No — 41 — 57 — 20 — 37 — 21 Color Block No — 41 — 57 — 20 — 37 — 21 0 12 3 4 Storage time (days) Figure 4-3. Color changes of brown shrimp during storage at different temperatures, points represent the average color of 24 shrimp

PAGE 111

91 Table 4-3. Color changes in brown shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n = 24) Storage temp. o i or age time (days) Color 41 Color 57 Color 20 Color 37 Color 21 Avg. St. dev. Av £St. dev. Avg. St. dev. Avfi. St. dev. Avg. St. dev. 1.5°C A U 43.0 16.90 7.7 12.99 6.9 6.70 14.1 8.53 4.3 5.03 1 1 40.4 12.98 6.8 11.32 8.7 6.32 16.8 8.73 5.2 4.34 I 37.1 16.53 4.2 8.42 13.5 11.44 16.7 9.65 ' 6.4 4.62 1 J 34.2 14.65 1.3 2.41 13.8 9.01 16.8 8.63 11.9 9.19 A 30.5 15.02 1.2 2.40 16.0 7.22 18.4 9.73 12.7 8.04 5 29.7 15.21 1.9 5.20 16.7 9.21 16.7 7.45 13.0 7.62 6 27.9 13.90 0.5 0.72 16.7 8.16 19.2 7.61 13.5 7.61 7 23.1 15.42 1.1 2.31 18.3 8.76 18.9 9.35 15.2 7.95 8 25.8 14.78 3.0 6.11 16.7 9.05 17.2 8.14 13.5 7.24 9 23.3 13.40 2.0 3.60 19.7 8.02 20.6 8.92 10.2 7.14 10 21.2 12.17 3.2 6.44 20.5 9.99 21.4 7.76 7.7 4.83 11 19.5 11.31 2.5 5.17 19.2 7.57 24.7 8.07 10.6 6.09 7.0°C 0 45.2 13.88 7.0 9.27 6.4 6.16 13.6 8.38 4.2 4.65 1 40.1 14.43 2.6 3.49 11.1 6.98 17.1 7.43 6.9 6.06 2 33.2 14.16 1.0 1.48 14.1 5.82 16.4 8.60 13.8 10.29 3 30.3 16.14 0.7 0.80 15.3 7.70 16.6 9.09 14.4 10.17 4 27.4 16.19 0.6 0.94 18.7 8.96 17.4 9.28 14.1 9.22 5 27.0 12.47 0.5 1.01 18.4 7.38 19.2 7.15 13.9 7.05 6 24.0 12.59 0.9 1.44 19.3 7.83 21.4 9.22 12.0 6.54 7 21.3 12.59 1.1 1.52 21.2 10.26 21.9 9.63 10.2 5.22 13.0°C 0 51.9 11.74 5.1 8.02 3.7 3.39 12.4 8.26 4.3 4.86 1 41.1 14.01 2.0 2.89 9.3 6.28 16.8 6.72 9.4 8.43 2 32.8 16.55 1.6 3.21 14.5 8.22 17.2 7.42 13.1 9.38 3 28.6 14.51 2.7 4.78 15.2 7.86 22.3 10.53 10.3 9.08 4 25-5 14.41 . 4,37 '7-8 10.45 2U -7J2.. , 8.0 6,41

PAGE 112

92 Pink Shrimp 1.5°C 50 40 30 20 10 0 e 5 Q 1 ' 1 1 1 ' 'J Color Block No — 57 — 41 — 36 1 -— 37 4 5 6 Storage time (days) Pink Shrimp 7.0°C 10 11 20 3 4 Storage rime (days) Pink Shrimp I3.0°C Color Block No — 57 -— 41 — 36 — 37 — 20 Color Block No — 57 — — 41 — 36 -— 37 -°20 Storage time (days) Figure 4-4. Color changes of pink shrimp during storage at different temperatures, points represent the average color of 25 shrimp

PAGE 113

93 Table 4-4. Color changes in pink shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n 25) Storage temp. Storage time (days) Color 57 Color 41 Color 36 Color 37 | Color 20 Ave. St. dev. Ave. St. dev. Ave St. dev. Avj? St. dev. St Hev 1.5°C 0 23.6 19.29 29.5 14.87 8.6 9.54 11.0 10.85 1 9 — . 1 — 1 22.7 16.89 30.8 13.86 7.2 5.40 12.1 8.72 3 2 5 94 2 16.7 15.73 35.5 13.56 7.6 5.70 13.8 8.44 4.3 3.46 3 18.2 15.78 31.8 14.18 8.2 7.53 14.1 7.12 5.6 6.15 4 20.3 16.60 27.4 11.62 10.9 6.27 14.3 9.79 6.9 6.55 5 20.9 17.41 24.9 11.93 10.8 5.82 17.0 10.47 6.2 5.13 6 24.8 19.42 22.3 11.84 11.2 7.82 15.6 9.57 6.4 5.19 7 26.1 19.13 22.4 10.56 12.6 7.31 15.1 9.13 5.3 4.34 8 29.5 20.06 18.7 9.84 13.7 7.33 15.2 11.01 4.0 2.52 9 32.5 21.68 14.7 7.17 14.0 9.89 15.5 9.86 4.1 3.35 10 33.7 19.23 15.8 9.40 14.0 7.82 15.0 9.32 3.5 2.51 11 31.9 18.18 15.7 7.34 12.1 7.27 17.1 9.23 4.2 3.38 7.0°C 0 25.9 17.44 28.2 14.30 9.9 7.41 11.2 9.25 2.2 4.23 1 17.2 14.71 33.5 12.98 9.0 6.78 15.0 8.49 3.9 6.49 2 16.9 15.04 29.3 12.95 12.8 9.67 14.6 8.99 6.0 5.33 3 22.6 15.45 26.7 11.11 10.6 7.09 15.5 10.24 5.4 4.59 4 26.2 16.22 21.2 10.33 15.8 9.69 11.8 6.87 6.2 6.76 5 27.5 18.56 18.0 8.72 15.4 9.61 16.8 8.53 4.1 3.41 6 28.7 19.84 16.3 8.68 14.2 8.58 18.7 10.26 3.8 2.56 7 29.9 18.31 14.1 7.89 16.2 7.41 17.5 8.96 3.4 2.47 13.0°C 0 27.1 18.88 29.5 15.37 7.2 5.07 10.7 10.64 2.2 2.67 1 21.0 15.68 30.3 11.53 6.4 3.64 16.7 10.97 4.4 4.25 2 27.3 19.87 22.9 12.17 12.5 8.55 14.7 9.37 4.4 3.39 3 35.1 20.55 16.8 11.20 12.0 6.73 15.0 8.68 2.7 2.37 4 35.2 22.29 11.2 4.99 19.4 12.01 12.7 8.16 3.1 3.18

PAGE 114

Tiger Shrimp 1.5°C ! — o S 3 U 4 5 6 Storage time (days) Tiger Shrimp 7.0°C 10 1 c * o 0 U 3 4 Storage time (days) Tiger Shrimp 13.0°C 50 40 30 1 20 I 10 Color Block No — 41 — 21 — 0 — 20 -o42 0 12 3 4 Storage time (days) Figure 4-5. Color changes of tiger shrimp during storage at different temperatures, points represent the average color of 25 shrimp

PAGE 115

95 Table 4-5. Color changes in tiger shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n 25) Storage temp. Storage time Color 41 Color 21 Color 0 Color 20 Color 42 (days) — 7 Avg. St. dev. Avg. St. dev. Ave* St. dev. Aver St Hpv A vet St. dev. 1.5°C 0 27.4 13.47 26.8 1 1.53 1.6 1.48 7.8 5 60 *J . \J\J 4.24 1 34.3 14.25 24.6 9.37 2.1 1.90 8.6 7 24 8 4 3.03 2 25.8 15.15 29.8 12.62 3.3 3.23 8.6 6.26 9.9 6.42 3 25.1 14.99 32.1 11.48 3.9 3.21 9.8 5.98 8.3 3.71 4 23.8 13.90 34.9 10.97 4.6 2.94 11.0 4.34 7.6 2.42 5 22.7 14.03 34.7 10.81 6.1 4.17 12.3 5.51 7.7 2.92 6 22.9 13.46 33.7 9.84 6.9 6.84 13.7 6.50 6.8 2.27 7 21.8 11.41 35.4 8.02 6.8 4.02 15.1 5.99 6.1 2.24 8 23.6 12.16 33.1 9.47 . 7.7 6.97 14.5 6.76 5.6 2.06 9 22.8 13.26 30.5 7.84 8.8 7.45 18.6 7.82 4.6 1.51 10 21.9 13.48 32.0 8.99 8.5 6.60 19.7 7.43 4.5 1.31 11 24.5 14.62 32.0 10.16 7.2 5.55 17.0 7.44 5.1 1.59 7.0°C 0 29.0 13.15 25.0 11.24 1.6 1.38 7.3 5.40 9.2 4.50 1 29.7 16.55 25.5 11.93 3.2 3.94 9.1 6.88 8.8 4.99 2 26.6 15.95 28.7 12.21 4.0 4.48 9.6 7.03 8.0 2.73 3 25.0 14.20 31.8 9.07 5.1 4.83 10.8 7.14 6.6 1.75 4 23.2 14.89 32.1 9.95 6.7 5.40 15.1 7.35 5.9 2.30 5 24.3 16.69 31.9 11.39 7.7 8.94 14.1 7.22 5.7 1.92 6 25.5 15.30 29.9 9.90 8.7 9.55 14.9 7.06 4.9 1.49 7 25.4 14.37 29.2 8.33 8.4 9.64 17.6 8.43 4.6 1.70 13.0°C 0 29.8 12.72 21.9 9.49 1.1 1.87 5.9 6.14 12.7 7.49 1 29.2 15.52 26.1 11.80 2.2 3.23 8.4 6.64 9.2 4.50 2 27.1 16.23 28.4 11.96 3.8 4.57 10.5 7.86 7.5 3.12 3 28.5 17.30 28.5 11.50 4.9 4.74 12.7 7.17 6.0 2.67 4 25.7 13.85 29.7 10.14 4.9 4.60 18.6 8.51 4.3 1.27

PAGE 116

96 White Shrimp 1.5°C I 9 5 5 0 50 40 30 20 10 0 o 3 0 4 5 6 Storage time (days) White Shrimp 7.0°C 10 Color Block No — 41 — 21 — 20 -•62 42 0 12 3 4 Storage time (days) White Shrimp 13.0°C 5 6 7 Color Block No — 41 — 21 — 20 -— 62 — 42 Color Block No — 41 21 20 62 -o_ 42 1 2 3 Storage time (days) Figure 4-6. Color changes of white shrimp during storage at different temperatures, points represent the average color of 25 shrimp

PAGE 117

97 Table 4-6. Color changes in white shrimp during storage at different temperatures. Data are averages (% of total view area of shrimp) of the most significant colors (n 25) Storage temp. Storage time Color 41 Color 21 Color 20 Color 62 Color 42 (days) Ave. St. dev. Avg. St. dev. Ave. St. dev. Ave. St. dev. Ave. St. dev. 1.5°C 0 44.2 10.75 10.8 9.03 6.2 1.98 9.4 7.1 1 8.7 4.23 1 43.2 9.70 11.5 9.14 7.5 2.22 7.1 2.81 9.2 6.23 2 39.9 11.91 15.7 12.31 7.6 2.88 6.3 3.26 9.7 6.90 3 39.1 10.55 14.8 10.86 8.0 3.52 5.8 2.67 9.8 4.86 4 37.8 12.17 13.7 10.55 11.2 5.97 6.0 3.96 5.8 4.64 5 35.6 11.03 16.2 11.43 10.1 3.66 4.6 2.47 8.9 6.28 6 31.6 12.18 17.0 10.76 12.0 5.44 4.8 3.53 7.3 4.92 7 32.5 13.05 19.4 11.00 12.1 4.48 3.8 2.59 7.5 3.71 8 32.1 13.69 18.5 10.97 14.8 6.25 3.0 1.81 6.1 2.81 9 32.1 11.99 15.7 9.35 15.3 5.53 2.8 1.51 4.3 2.04 10 32.6 14.96 13.4 9.24 18.0 8.11 2.3 1.41 4.3 1.93 11 30.7 14.75 15.1 8.83 17.6 6.00 1.7 0.71 4.0 1.53 7.0°C 0 47.8 8.44 8.7 8.49 6.0 3.31 7.5 3.08 9.1 4.64 1 42.0 10.77 12.2 10.42 7.3 3.38 6.5 2.78 10.0 6.77 2 41.7 11.53 13.8 11.31 8.0 2.52 5.1 1.57 9.0 7.49 3 37.7 10.40 16.3 8.99 8.9 3.30 4.5 1.72 9.0 5.45 4 35.6 13.08 18.0 11.47 12.5 3.69 3.4 1.29 6.6 2.85 5 34.3 12.74 16.5 8.65 12.7 4.70 3.3 1.44 6.4 3.67 6 31.9 9.74 14.9 7.02 15.2 3.83 2.7 0.91 4.2 1.48 7 30.6 15.82 15.4 8.13 18.0 7.39 2.2 0.84 4.3 1.88 13.0°C 0 44.4 9.69 8.4 9.19 5.3 2.95 9.1 4.19 11.7 6.16 1 44.8 9.59 9.7 8.85 7.7 2.86 7.1 2.65 6.0 2.47 2 39.7 15.96 13.9 12.62 10.4 5.05 5.6 2.02 6.3 3.07 3 38.9 14.37 12.2 8.98 12.0 6.37 4.6 1.56 5.0 1.60 4 34.6 13.95 8.8 6.84 15.3 5.64 3.6 1.12 3.4 1.03

PAGE 118

98 temperatures were very similar. The main difference was that shrimp at 13.0°C changed color faster compared to the lower temperatures. However, in most cases the color after 4 days of storage at 13.0°C looked somewhat similar to that at 7 days at 7.0°C, and to that of 1 1 days at 1.5°C. In general for all shrimp species, color 41 (RGB 160, 160, 96) decreased during storage, while other colors started to form. In brown shrimp, color 20 (RGB 96, 96, 32) (Figure 4-3 and Table 4-3), which is a dark olive, increased during storage. Color block 20 was considered a melanotic color (Table 4-9). In general, brown shrimp did not change its color too much, except for the formation of melanosis. The other colors in brown shrimp changed only slightly. Pink shrimp changed color to a bright orange during storage. This was seen by the increase in color 57 (RGB 224, 160, 96) (Figure 4-4 and Table 4-4). It was also noted that pink shrimp had a stronger ammonia odor compared to the other species. Formation of ammonia at the surface increases the pH to an alkaline level. When shrimp is exposed to a high pH, the color changes to a cooked appearance. Similar observations were also reported by Luzuriaga et al. (1997b) in white shrimp. Color block 36 (160, 96, 32) increased slightly in pink shrimp. This color was responsible for a brown appearance, which was identified as melanosis. The other colors remained fairly constant. Tiger shrimp changed color the least compared to the other species. This was observed visually and also by the color block data analysis (Figure 4-5 and Table 4-5). Throughout storage, tiger shrimp remained with the original colors, except for an increase in the melanotic colors. Color 41 (RGB 160, 10, 96) and 42 (RGB 160, 160, 160)

PAGE 119

99 decreased slightly, losing the greyish appearance. However colors 0 (RGB 32, 32, 32) and 20 (RGB 96, 96, 32) increased over time, denoting an increase in the melanosis level. White shrimp showed development of melanosis by an increase in color block 20 (RGB 96, 96, 32), while color 62 (RGB 224, 224, 160) and 42 (RGB 160, 160, 160) decreased over time (Figure 4-6 and Table 4-6). Color 62 is a pale greenish yellow, while 42 is a medium grey, which were probably converted into melanotic colors. Similar to tiger, white shrimp did not have major changes in color except for the black spots formed during storage. L*a*b* values from the Color Analysis software were obtained during storage. Figure 4-7 and Table 4-7 and 4-8 show the average trends for these parameters. All shrimp species at the different storage temperatures had a slight decrease in the L value. L is related to the lightness, which denotes a loss of bright clear colors, which could be related to the formation of dark colors such as melanosis. Except for pink shrimp, all other species had a decrease of about 12 to 15 units in the L value. In pink shrimp L decreased at the beginning, but then it leveled off. The a value showed an increasing trend in all species. Theoretically, a higher a value denotes a change into the red colors; however, it is difficult to visualize where the red color is being formed in shrimp. The a values are close to 0, therefore the colors are close to gray or pale colors. The b value shows different trends in the different species. In pink, tiger and white shrimp it remained fairly constant. However, in brown shrimp the b value decreased for a few days and then it started to increase. In all species the b value of the shrimp stored at 13.0°C increased. An increase in the b value means a change into the yellow region. As seen from the data,

PAGE 120

100 Table 4-7. Average L, a and b values for brown and pink shrimp during storage at Storage temp. Storage time (days) Shrimp Specie Brown Pink L a b L a b Avg. Stdev. Avg. St.dev. Avg. St.dev. Avg. St.dev. Avg. St.dev. Ave. St.dev. 1 5°C 0 56.5 5.84 3.0 3.13 28.7 4.03 62.4 6.90 6.9 3.42 33.5 2.88 1 54.5 5.48 3.1 3.00 27.0 3.25 60.7 6.24 7.5 3.77 32.3 3.53 2 51.8 6.47 2.3 2.34 26.0 2.68 59.1 5.41 6.6 3.74 31.2 2.44 3 49.7 5.99 1.8 2.35 22.3 2.91 57.9 5.46 7.8 3.59 30.6 2.38 4 47.9 5.23 1.7 2.26 22.7 2.94 57.0 6.16 8.1 3.68 32.8 2.95 5 46.9 6.56 2.2 2.35 22.2 3.50 56.9 6.23 9.6 3.52 32.3 3.25 6 45.5 5.39 2.3 2.12 21.4 2.87 56.6 6.48 10.9 3.46 32.2 3.09 7 44.7 6.78 2.4 2.12 21.2 2.68 56.7 6.33 11.9 3.25 33.4 3.12 8 45.6 7.84 3.7 2.44 21.3 3.53 57.9 5.86 13.0 2.97 35.0 3.60 9 44.3 6.28 4.4 2.38 23.6 3.17 57.7 6.65 14.4 3.01 33.9 3.30 10 44.3 6.03 6.3 2.41 25.3 3.02 57.3 4.84 15.5 2.85 34.8 3.09 11 43.8 5.56 5.9 2.69 23.3 3.14 56.9 5.46 15.2 2.76 33.0 3.19 7.0°C 0 57.0 5.29 2.4 3.38 28.8 3.29 61.3 5.46 8.7 3.93 34.7 3.35 1 52.8 5.21 2.0 2.51 25.8 3.05 58.7 5.34 7.7 3.83 31.4 2.32 2 49.2 5.19 0.9 2.57 22.5 2.92 56.7 6.27 7.8 3.23 32.0 2.26 3 47.5 6.13 1.5 2.48 21.5 2.79 57.0 5.62 9.7 2.92 32.1 2.87 4 45.5 5.92 1.5 2.27 22.2 2.94 56.3 6.70 11.0 3.19 34.4 3.19 5 45.6 4.21 2.1 2.39 22.1 2.58 56.3 5.49 13.6 3.42 33.8 2.41 6 45.0 4.59 3.4 2.68 22.9 2.67 56.4 5.48 14.7 3.27 33.1 2.28 7 44.0 4.76 4.6 2.85 23.8 2.96 56.2 4.59 16.0 3.56 34.9 2.77 13.0°C 0 58.0 4.42 1.5 2.44 26.4 3.01 62.1 5.87 8.4 3.54 33.9 3.92 1 53.1 4.55 1.7 2.32 24.0 3.38 58.6 5.11 8.2 3.91 30.9 2.68 2 49.4 5.51 1.6 2.04 22.7 3.27 57.8 5.45 10.6 3.86 33.6 2.40 3 48.0 5.68 4.5 2.14 23.7 2.90 58.5 4.85 15.1 3.54 34.3 2.74 4 47.1 5.47 5.4 2.40 26.7 3.21 56.9 6.44 16.5 2.53 37.0 2.78

PAGE 121

101 Table 4-8. Average L, a and b values for tiger and white shrimp during storage at Storage temp. Storage time (days) Shrimp Specie Tiger White L a b L a b Ave, St.dev. Ave.. St.dev. Ave, St.dev. Avg. St.dev. Ave. St.dev. Ave. St.dev. 1.5°C 0 52.6 4.52 -10.5 0.90 18.6 1.67 57.7 5.44 -3.1 1.49 23.3 3.09 1 52.5 4.48 -8.5 0.96 19.0 1.93 55.7 4.64 -2.1 1.59 22.8 2.69 2 50.6 6.06 -8.8 0.77 17.3 2.13 53.6 5.37 -3.0 1.51 21.3 3.02 3 49.0 5.62 -8.2 0.79 17.1 1.78 52.8 5.62 -2.3 1.55 20.5 2.70 4 47.4 4.69 -7.7 0.57 17.6 2.18 51.9 6.89 -2.0 1.35 23.5 3.32 5 46.1 5.87 -7.2 0.69 17.3 2.33 50.4 5.43 -1.8 1.49 20.6 3.16 6 45.0 6.29 -7.2 0.90 17.3 1.94 49.4 6.47 -1.5 1.63 21.3 3.23 7 44.4 4.90 -6.9 0.73 17.9 1.84 48.0 6.36 -1.9 1.32 19.5 2.61 8 44.4 6.18 -6.4 0.85 17.8 2.19 47.0 5.85 -1.6 1.40 20.7 2.59 9 43.3 6.37 -7.2 0.75 19.6 2.07 46.8 5.75 -0.7 1.35 22.3 2.64 10 42.6 6.20 -5.7 0.85 19.3 2.35 46.0 6.13 -0.3 1.39 22.9 2.47 11 44.2 6.21 -5.8 0.76 19.1 2.47 44.8 5.44 0.2 1.27 22.1 2.33 7.0°C 0 52.9 4.41 -10.7 1.02 18.7 2.44 57.8 4.89 -2.2 1.87 23.6 2.94 1 51.6 6.59 -9.8 1.01 19.0 2.54 55.2 5.16 -1.5 1.75 21.7 2.89 2 49.6 5.83 -9.2 0.95 18.0 2.68 53.0 4.12 -1.7 1.77 21.7 2.73 3 47.6 5.58 -8.6 0.84 17.9 2.13 51.6 4.06 -1.6 1.70 20.1 1.90 4 45.2 6.18 -8.1 0.86 18.5 2.51 49.0 4.95 -1.5 1.47 20.8 2.52 5 45.0 7.36 -7.6 1.12 18.3 2.75 48.3 5.16 -0.8 1.51 20.8 1.75 6 44.4 7.50 -7.3 1.03 18.9 2.83 46.9 3.62 0.2 1.33 22.1 1.61 7 44.0 7.19 -6.5 1.11 19.6 2.52 45.8 5.24 0.2 1.52 22.2 1.62 13.0°C 0 54.7 4.84 -10.9 1.04 18.4 2.87 59.5 5.65 -2.5 1.28 23.5 2.41 1 51.9 5.72 -9.7 0.86 18.6 2.35 55.9 5.13 -1.1 1.17 24.0 2.68 2 49.4 5.78 -9.4 0.83 18.6 2.43 52.8 5.98 -1.0 1.17 23.2 3.02 3 47.8 6.45 -8.8 0.93 19.6 2.58 51.1 4.93 -0.0 1.16 23.0 2.32 4 46.0 5.64 -7.7 0.88 21.2 2.27 49.9 4.43 1.2 1.35 125.7 2.56

PAGE 122

102 Brown Shrimp 0123456789 10 11 Storage time (days) 0123456789 10 11 Storage time (days) 0 1 23456789 10 11 Storage time (days) WUtc Shrimp 0 123456789 10 11 Storage time (days) Figure 4-7. Average L*a*b* values of different species of shrimp stored at different temperatures, measured by the color machine vision system, n=25

PAGE 123

103 L*a*b* values followed similar patterns during storage, but they were difficult to correlate to the real changes of all the colors present in the shrimp, except for the L value that had some correlation with melanosis. Melanosis or black spots appeared in all shrimp species during storage. Melanosis is not only a black color, but other dark brown and dark green colors. Table 4-9 shows a list of the colors considered to be responsible for the melanotic areas in the different species of shrimp. All melanotic colors had the B (blue) component as 32, while R (red) and G (green) are 32, 96 or 160. All shrimp species had color blocks 0 (RGB 32, 32, 32), 16 RGB (96, 32, 32) and 20 (96, 96, 32) as melanotic colors. Color 36 (RGB 160, 96, 32) was also found in brown, pink and white shrimp. However, color 4 (RGB 32, 96, 32) was only found in tiger shrimp, and color 32 (RGB 160, 32, 32) in pink shrimp. Therefore, the amount of melanosis present in shrimp was calculated as the sum of the percentages of the color blocks listed in Table 4-9. The result was the amount of area of the shrimp covered by melanosis (Table 4-10). Table 4-9 Color blocks identified as melanosis in different species of shrimp Color Block No. RGB value at the center of the color block Color Shrimp Specie representation Brown Pink Tiger White 0 32, 32, 32 * 4 32, 96, 32 16 96, 32, 32 20 96, 96, 32 32 160, 32, 32 36 160, 96, 32

PAGE 124

104 Visual observations throughout the experiments correlated very well with the data obtained from the color analysis. It was observed that melanosis was not developing on every shrimp at the same time. Some shrimp had more melanosis compared to others. This observation was confirmed by the large standard deviations of the calculated area covered by melanosis (Table 4-10). Pink shrimp was the specie where this observation was more pronounced. On the other hand, brown shrimp had a more uniform formation and distribution of melanosis. Figures 4-8 and 4-9 show the development of melanosis during storage at the three different temperatures. It was clear that the rate of formation of melanosis was higher with an increase in storage temperature. However, the levels of melanosis at the end of storage of the samples at 7.0° and 13.0°C were lower than the levels formed at 1.5°C. In the higher temperatures, experiments were stopped when the odor of the samples was unacceptable. Therefore, at the lowest temperature, the melanosis levels were higher, but without an objectionable odor. In the samples used in this experiment the rate of off-odor formation was faster than the rate of melanosis development. For all shrimp species, except pink, the development of melanosis followed a linear trend. The R 2 values were above 0.93 and in some cases close to 1.0. The slope of the line represented the increase in melanosis per day of storage, while the intercept denoted the initial amount of melanosis at day 0. It was observed that at the end of storage, brown shrimp had the highest levels of melanosis, compared to the other species. This was corroborated with the value of the slope, which were the largest for all shrimp species. In contrast, pink shrimp did not develop as much melanosis, therefore having the lowest

PAGE 125

105 Table 4-10. Amount of melanosis (average % of the total view area of shrimp) in Storage temp. Storage time (days) Shrimp Specie Brown Pink Tiger White Avg. St. dev. Avg. St. dev. Avg. St. dev. Avg. St. dev. 1.5°C 0 14.0 9.67 11.1 11.66 11.3 7.26 9.5 3.00 1 15.9 9.12 11.4 10.25 12.2 9.46 12.0 3.83 2 21.5 14.78 12.9 8.71 14.1 10.32 13.7 5.00 3 20.8 11.98 15.4 12.15 16.0 9.98 15.0 6.08 4 24.3 9.66 19.6 12.05 18.1 7.52 20.2 8.68 5 26.6 13.78 18.8 9.72 21.1 10.06 19.2 6.09 6 27.4 12.38 19.7 12.50 23.4 13.28 22.4 8.85 7 29.8 14.28 20.3 12.77 24.6 9.97 22.3 8.36 8 28.5 16.35 19.3 9.44 25.2 14.06 26.1 9.51 9 35.1 15.57 19.9 13.71 30.5 15.70 27.3 9.19 10 38.3 16.14 19.2 9.99 31.0 14.45 30.9 1 1.54 11 34.5 14.81 18.1 10.60 26.8 13.18 31.1 10.66 7.0°C 0 12.6 8.70 12.7 11.61 11.0 7.12 9.5 4.90 1 18.1 8.94 14.1 12.19 14.6 11.41 11.6 5.25 2 21.0 8.14 20.6 14.81 16.3 11.73 14.6 4.60 3 23.8 11.65 17.8 11.68 18.9 12.17 15.6 5.70 4 29.1 12.87 24.2 17.50 25.1 12.77 20.7 7.18 5 28.3 10.20 21.4 12.88 24.9 15.22 22.5 8.93 6 31.6 10.85 19.9 11.16 26.8 16.66 26.0 7.18 7 36.4 13.43 21.4 10.13 28.9 17.30 29.1 11.57 13.0°C 0 6.9 5.07 10.0 8.22 8.5 8.59 7.9 4.08 1 14.0 9.09 11.9 7.28 12.7 10.36 12.6 4.03 2 21.1 10.49 18.3 11.52 17.1 13.22 17.3 7.89 3 24.6 11.42 16.0 9.44 20.7 12.86 19.9 8.70 4 32.5 13.33 24.8 16.72 26.2 13.07 25.4 7.41

PAGE 126

106 Figure 4-8. Development of melanosis in brown and pink shrimp during storage at different temperatures. Line represents a linear fit

PAGE 127

; ; 7.0°C 0123456789 10 11 " 130 ° c Storage time (days) White Shrimp 40 35 area) 30 view 25 1 20 <*• 5 15 10 I 5 0 R 2 i 5°c = 0.980 y = 1.98 x + 9.89 R 7.o°c 0.988 y 2.84 x + 8.77 R 2 i3o°c = 0.992 y = 4.23x + 8.17 4 5 6 7 Storage time (days) 10 11 Storage Temperature o 1.5°C 7.0°C 13.0°C Figure 4-9. Development of melanosis in tiger and white shrimp during storage at different temperatures. Line represents a linear fit

PAGE 128

108 values for the slope. Pink shrimp had a different behavior in the development of melanosis. At the beginning of storage melanosis increased, however, halfway during storage, it leveled off, and even diminished slightly. As mentioned earlier, pink shrimp had a stronger ammonia odor, with a probable increase of pH which could have affected the enzyme polyphenoloxidase. All shrimp species had an initial melanosis level of about 10% of the total view area of the shrimp. It would be expected that the level should be close to 0%. However, the areas responsible for this initial level of melanosis were the telson and uropods (tail) of the shrimp. In live shrimp these areas have a slightly darker color than the shell however after harvest, they are very prone for melanosis development. In almost every shrimp used in this experiment, the telson and uropods were already dark after thawing. Tiger shrimp developed melanosis similarly as the other species. When one visually inspects tiger shrimp for melanosis it is more difficult to detect it, because it blends with the original colors of the shrimp. From the color analysis, it was shown that one of the most representative colors was color block 21 (RGB 96, 96, 96), which is a dark gray. This color was not identified as melanosis, therefore, making it easier to evaluate tiger shrimp for melanosis. Conclusions The human eye can distinguish the colors of any object in a fraction of a second. However, it is not capable of quantifying the amount of each color present in the object, nor defining the colors in a standardized manner. The color analysis software used in this

PAGE 129

109 study was capable of measuring the colors present in the different species of shrimp. It showed that the color profiles were capable of differentiating the species of shrimp used in this experiment. The 64-color block scheme was used to quantify the colors present in shrimp during storage, showing that certain colors changed over time. Even though the Color Analysis software can measure a higher color block resolution (512 or 4096), it was not used because of the complexity in observing trends over time. However, from the discrimination point of view, pattern recognition techniques may benefit from a higher color resolution in identifying the different species of shrimp. Other pattern recognition techniques could be beneficial to analyze color data generated by the software. Due to its multivariate characteristic, simple procedures will not extract much information from the data. This study also showed that melanosis can be easily measured and quantified with the color machine vision system and the Color Analysis software. This will benefit the industry and government agencies when evaluating shrimp for melanosis. It will eliminate the subjectivity of the current sensory evaluation, making it more objective and easy to quantify. The procedure is very simple and it takes few seconds to get the results. The purpose of this study was not to develop models to predict color changes or melanosis development. However, a comprehensive database including different shrimp species, harvest locations, year seasons, and processing conditions should be acquired to develop models to predict and estimate possible color changes during storage conditions. These data combined with other quality attributes such as electronic nose sensor readings, and texture changes can provide an overall quality evaluation procedure.

PAGE 130

CHAPTER 5 AUTOMATION AND CORRELATION OF COLOR AND ODOR EVALUATION OF SALMON Introduction High quality fresh salmon is a valuable commodity. Its world commercial catch increased from 1.5 million tons in 1992 to 2.1 million tons in 1996. In 1997 the United States (U.S.) commercial landings of salmon represented 12% of the world production, being the fourth fishery product of importance, both in quantity and value. The same year, the U.S. imported $344 million worth of salmon, the 3 rd largest imported fishery product by value. In 1997 salmon was the second most important fishery product exported by the U.S., generating $308 million (U.S. Department of Commerce, 1998). Currently, inspectors evaluate salmon quality for visual, odor and texture attributes. These subjective inspections are susceptible to error, difficult to quantify and to compare with standards worldwide. Moreover, chemical analyses are seldom used by the salmon industry due to complexity and length of time required. Therefore, there is need for objective and rapid methods to evaluate the quality of raw fresh salmon. Such methods can assist in the development of common standards between the industry, regulatory agencies and international markets. Also, the correlation between sensory 110

PAGE 131

Ill evaluation and machine measurements is vital for the reliability of rapid predictive assessment of quality attributes. The odor of seafood products has been widely used as one of the main indicators of quality since ancient times (Botta, 1995). Perkins (1992) defined fresh seafood as having a clean, natural odor and physical characteristics representative of the species in good condition. Researchers have tried to find chemical analyses that can be correlated with that clean, natural odor and that could be used as an index of quality (Hebard et al., 1982; Halland and Njaa, 1988; Kennish and Kramer, 1987; Hollingworth and Throm, 1982; Kelleher and Zall, 1983). Most of the chemical analyses require sample preparation, use of hazardous chemicals and technical expertise to perform the tests. Recent developments in sensor technology and electronic noses made it possible to objectively evaluate the odor of food samples (Corcoran, 1993). The electronic nose is an array of non-specific sensors that allows detection of a wide range of volatile compounds, thus making this technology very versatile (Hodgins, 1997). Even though the electronic nose is a rapid and objective method that could be used to quantify odors, little work has been published which shows correlation of electronic nose sensor outputs and sensory results in seafood products (Balaban and Luzuriaga, 1996; Olafsson et al., 1992; di Natale et al., 1996). Salmon's flesh color is an important attribute used during its quality evaluation. Consumers equate freshness with the vibrancy of the flesh color (Beaudoin, 1997). Fresh salmon has an attractive vivid pink-orange color, which changes to a dull pink, or even beige during storage. In farm-raised salmon, flesh color depends on the feed

PAGE 132

112 (carotenoids). Colorimeters have been widely used to assess these color changes, mainly to find correlations between carotenoid levels and the amount of redness in the flesh (Saito, 1969; Schmidt and Cuthbert, 1969; Skrede and Storebakken, 1986; Skrede et al., 1989). Some of these studies blended the flesh to obtain a paste with a uniform color, which changed the overall visual appearance and colors present in the salmon. Color machine vision uses a video camera to grab an image of a sample. The image is digitized into small elements called pixels, which contain color information. This procedure allows researchers to obtain a profile of the colors present in the sample, rather than a single color value given by colorimeters, which is the average reflectance of the surface being analyzed. A color machine vision system has been used by Luzuriaga et al. (1997b) to evaluate the color of shrimp and quantify the amount of melanosis. This procedure gives an overall color evaluation of the sample by defining and quantifying the colors present in the sample. Quality evaluation of salmon does not depend on odor or color evaluation alone. Results from both quality attributes are used during inspection of the commercial product. An objective overall quality evaluation of salmon fillets can be performed instrumentally by combining data from the two emerging technologies of electronic nose and color machine vision. The combination of sensor data and color profiles should give a more detailed description of the quality of the sample, thus making it easier to quantify different levels of quality. Using discriminant function analysis (DFA) as the pattern recognition technique on the combined data set may give more accurate predictive functions of salmon quality, rather than using electronic nose data or color profiles alone.

PAGE 133

113 The objectives of this study were 1) to measure odor and color changes of salmon fillets stored at different temperatures with sensory panels, an electronic nose, and a color machine vision system; 2) to obtain predictive models for overall quality of salmon using DFA on the combined data set of electronic nose sensor and color data; 3) to compare the DFA models for the combined data set with that of electronic nose or color data alone; 4) to test the predictive models with salmon fillets stored in variable temperature environments, and 5) to develop a model to predict the color of salmon based on storage temperature and time using the Arrhenius approach. Materials and Methods Salmon Samples and Storage Conditions Atlantic salmon ( Salmo salar) fillets (1.5-2 kg each) from Chile were obtained fresh. The study was replicated with fish obtained during summer (May) and during fall (November). Fish arrived within 48 hrs of harvest in summer and 72 hrs in fall. During transportation, fillets were packaged individually in plastic wrap and were shipped in styrofoam boxes with cold packs. Fillets were cut into three pieces from head to tail. The portion closest to the head was used for this study. The portions were stored in cold rooms set at 3 different temperatures: 1.8°, 7° and 1 1.7°C, and kept for 10, 7 and 5 days, respectively, for the summer experiment, and for 8, 6 and 4 days, respectively, for the fall experiment. Six fillet portions (replicates) were used for each storage temperature and each fillet portion came from a different fish.

PAGE 134

114 Two variable temperature storage studies were also done to test the predictive models for odor and color changes. The first set of fillets was kept at 1 .8°C for the first day, then at 1 1 .7°C during the second and third days, and at 1 .8°C for the rest of the study until day 7. The second set was kept at 1 .8°C for the first two days, then at 1 1 .7°C during the third day and at 1 .8°C for the rest of the study until day 1 0 and day 7 in the summer and fall experiments, respectively. Moisture Content and Water Activity Measurements Moisture content of the fillet portions was measured in triplicate at days 1, 4, 7 and 10 during storage using the oven method (AO AC, #24.003, 1980). Water activity (aj was measured using a Rotronic Hygroscop DT (Rotronic, Huntington, NY). A 5 g piece of salmon was placed in a plastic cup provided by Rotronic and placed in the a„, meter. Readings were done in duplicate and taken after approximately 30 to 45 minutes when equilibration was achieved. The temperature at which a„, was measured was 24° ± 0.5°C for the summer experiment and 23° ± 0.5°C for the fall experiment. Sensory Evaluation The odor and color of the fillet portions were analyzed by a 6-member trained sensory panel consisting of professors and graduate students, 24 50 years of age, from the Food Science and Human Nutrition Dept. at the University of Florida. Panelists were trained before the study with salmon fillets obtained from the same supplier using a 10point scale (1 = good and 10 = bad). The odor scale used different descriptors: 1 = mild,

PAGE 135

115 seawater and typical fresh fish odor, 10 = putrid offensive odor. Values between 2 and 3 were described as mild fishy odors, between 4 and 5 as stronger less pleasant fishy smell, 6 and 7 as souring fishy smell, and above 8 were described as strong sour, rancid and putrid odors. The color scale was based on color changes, where 1 = vivid pink orange and 10 = beige-pink color. Numbers in between described the disappearance of the pink color and the formation of the beige color. Values between 3 and 4 represented dull pink orange, 5 and 6 were pale pink, and between 7 and 8 was a dull pink with tints of beige. Samples were evaluated for odor and color every day during the study. All 6 panel members smelled and looked at the fillets together and reached a common decision. The United States Food and Drug Administration (FDA) evaluates the odor of fishery products based on three categories : class I, class II and class III (CFR, 1997) . The 10-point sensory scale used in this study was condensed to the three FDA classes. Class I odor was samples with sensory grade from 1 to 4 (named grade 'A' in this study); Class II odor ranged from 5 to 7 (named grade 'B'); and Class III odor ranged from 8 to 10 (named grade 'C'). Electronic Nose Measurements The first study in summer used an electronic nose (e-NOSE model D, Neotronics Inc., Flowery Branch, GA) equipped with twelve conducting polymer sensors (sensor types: 298, 297, 283, 279, 278, 264, 263, 262, 261, 260, 259 and 258) to quantify the sensor responses to odor changes in salmon fillets during storage. In the second study, a e-NOSE 4000 ( Neotronics Inc., Flowery Branch, GA) was used, it was also equipped

PAGE 136

116 with twelve conducting polymers, however, sensor types were different (sensor types: 483, 478, 464, 463, 462, 461, 460, 459, 458, 401, 298 and 297). The day the experiment started, the model D e-Nose was calibrated with distilled water, following the manufacturer's recommendation. The e-Nose 4000 electronic nose was also calibrated the day the experiment started, however a solution of propylene glycol (75% v/v) was used as recommended by the manufacturer (propylene glycol 100% solution from Fisher Scientific, No. P-355-20, Fair Lawn, NJ). A 60-g piece of salmon fillet was taken out of the cooler one hour prior to analysis to let the sample equilibrate to room temperature (2 1 .0 to 23 .0°C). She replicates were analyzed on each day for each storage temperature, and 3 replicates were done for both variable temperature studies. In the fall experiment, 6 replicates were analyzed for the variable temperature studies. The piece of salmon was placed in a 140-ml beaker and placed in the sampling vessel of the electronic nose. Every day prior to the experiments the electronic nose was turned on and compressed air (CGA Grade D, Strate Welding Supply Inc, Jacksonville, FL) was passed through the sensors for at least 30 min. For each replicate, the vessel was purged with compressed air for 2 min to eliminate any foreign odor from the environment. Then, the sensor head was purged for 4 min with compressed air. During this time, the sample volatiles were equilibrating in the headspace of the vessel. Then, sensor response data was acquired for 3 min for the first experiment in summer, and for 4 min in the second experiment in fall. Total analysis time for each sample took 9 min and 10 min, respectively. Readings at 4 min exposure of the sensors to the samples were used for data analysis. At the end of the day, electronic nose sensors

PAGE 137

117 were cleaned with compressed air for at least 30 min. Electronic nose sensor data can be obtained from Dr. Murat Balaban. Food Science and Human Nutrition Department at the University of Florida, filenames: "\\thesis\text files\Chap5\salmonl sensor and color data.txt" and "\\thesis\text files\Chap5\salmon2 sensor and color data.txt". Color Analysis The color machine vision system described by Luzuriaga et al. (1997b) was used to grab images in the summer experiment. This system had a Computer Eyes RT (Digital Vision Inc., Dedham, MA) frame grabber. However, the second experiment was done with the system described in chapter 3, which had a different frame grabber (Matrox Meteor, Matrox, Canada). The remaining hardware was the same for both systems. In the first experiment video camera settings were: hue = 50, saturation = 50, contrast = 35 and brightness = 30 (settings ranged from 0 to 100). In the second experiment the settings were: hue = 255, saturation = 255, contrast = 200 and brightness =180 (settings ranged from 0 to 255). Individual pieces of salmon fillets were placed in the light box, which was illuminated with front and back lighting flourescent lamps. Pictures of the flesh of the salmon fillet portion were taken and saved in a computer file. Six replicates were analyzed on each day for each storage temperature, and 3 replicates were done for both of the variable temperature studies. The same replicates were used throughout the experiment. The color analysis program was used to evaluate the color of the fillets by using the 64color block scheme (Appendix A). Color analysis was done on a selected region of the

PAGE 138

118 Figure 5-1. Portion of salmon fillet used in the experiments. Black square is region of interest to extract color information flesh portion. This region was flesh without portions of the belly flap or thick connective tissue (Figure 5-1). Color was reported in the RGB and L*a*b* color systems. In the RGB color system it was reported in the 64-color blocks scheme. The value for each color block was the percent of the area of the fillet covered by a given color block. The summation of values for all color blocks was 100%, which was equal to the total surface area of the fillet. Average color block data can be found in Appendix B. Data Analysis Electronic nose sensor readings, color data and sensory scores were analyzed in Statistica for Windows ('98 edition, StatSoft Inc., Tulsa, OK) using discriminant function analysis (DFA) as reported by other researchers (Gardner and Hines, 1997; Gardner and Bartlett, 1992). Electronic nose data and 64-color block data were combined to obtain an

PAGE 139

119 overall quality estimation based on storage time and sensory scores. DFA was used to develop predictive models for classification of samples based on the three grades (grade A, B, or C), and storage time. The 12 sensor outputs and color block data were reduced to 2 discriminant functions. These functions were used to map the data in two dimensional plots and observe separation between groups. Correct classification rates and the coefficients for each function were obtained using Statistics The sensor data from the two variable temperature studies were not included in the data set to obtain the discriminant functions for the model. Instead, these were used for validation, i.e. to determine whether the discriminant functions provided reliable means of classifying these salmon fillets into one of the sensory grades (grade A, B, or C). DFA was also used to analyze 64-color block data alone. Classification functions of the correlation between color data and storage time, as well as sensory scores, were obtained. Data from the three storage temperatures were pooled together and correlated with sensory scores using DFA to obtain predictive functions. Similar to the sensor data and color data combined, validation of the predictive functions was performed with the variable temperature studies, to obtain the number of correctly classified cases. The same approach was used to analyze the electronic nose sensor readings alone. Color data from experiment 1 (Summer) were also analyzed by plotting selected colors over time and looking at trends. Equations were fitted to the data to predict storage time or sensory score based on the amount of a selected color present in the sample. Colors that changed monotonically were further evaluated assuming that color changed with time at constant temperature as a first order reaction:

PAGE 140

120 Color = Color o * e k ' (5-1) where: Color = color (% of the total area) present in salmon fillet at time t Color 0 = initial color (% of total area) k = rate constant (1/day) (+ = increasing color, = decreasing color) t = storage time (days) Using the Arrhenius approach, a plot of ln(k) versus the reciprocal of temperature (°K) was used to obtain the energy of activation and reaction rate. Therefore, the prediction of color was given by the following equation: Ea Color = Color x e K e * * ' ( 5-2 ) o where: k 0 = rate constant (1/day) E a = activation energy (J/mol) R = gas constant (8.3144 J/mol-°K) T = storage temperature (°K) Color data were also correlated with sensory score. Equations were fitted to plots of the amount of a given color vs. sensory score (scale from 1 to 10). The best correlation of color with sensory score was obtained using the following polynomial equation: Color = a + b (score) + c (score) 2 (5-3) where: grade = sensory score (1 = good, 10 = bad) a, b and c = estimated parameters Average L*a*b* values, as explained in chapter 3 (Equation 3-4), were plotted over time. L*a*b* values of all pixels representing the selected regions of flesh of the salmon fillets were used to calculate the averages. Plots depicted changes in the overall color of the samples.

PAGE 141

121 Results and Discussion Conducting polymer sensors can respond to changes in the headspace humidity of the sample being tested (Bartlett et al., 1997; Hodgins, 1997; Shurmer, 1990). This study monitored the humidity of the samples analyzed. Moisture content and water activity of the salmon fillets in experiment 1 and experiment 2 did not change with storage time for any of the three temperature storage conditions, and for the two variable temperature studies (Tables 5-1 and 5-2). However, the moisture content in experiment 1 was higher (72% ± 1.77 wet basis) than that of experiment 2 (66% ± 1.55 wet basis). Water activities in experiment 1 and 2 were 97.3% ±0.15 and 97.7% ±0.15, respectively. The difference in moisture content was due to an increase in the fat content of the fillets in experiment 2 (13.84% ± 0.65) compared to experiment 1 (9.82% ± 0.53) as measured by Soxhlet fat extraction. Table 5-1. Moisture content (% wet basis) and water activity (% relative humidity) of salmon fillets during storage at different temperatures (Experiment 1, Summer). (± = std. deviation, n = 3) Storage time (days) Storage Temperature 1.8°C 7°C 11.7°C Variable temperature 1 Variable temperature 2 1 % H 2 0 %RH 72.5 ± 1.85 97.4 ±0.14 68.9 ± 1.08 97.1 ±0.00 71.6 ± 1.94 97.3 ±0.21 73.7 ±0.30 97.2 ± 0.07 72.3 ± 2.38 97.2 ± 0.07 4 % H 2 0 %RH 72.7 ± 1.50 97.4 ± 1.21 70.8 ± 1.25 97.3 ±0.21 72.4 ± 1.07 97.2 ± 0.07 73.8 ±0.54 97.4 ±0.14 72.8 ±0.14 97.5 ± 0.07 7 % H 2 0 %RH 72.1 ±0.53 97.4 ±0.14 70.9 ± 0.89 97.3 ± 0.07 70.7 ± 1.36 97.4 ±0.21 69.4 ± 0.72 97.3 ± 0.07 10 % H 2 0 %RH 73.6 ± 0.26 97.4 ±0.14 73.1 ± 1.38 97.4 ± 0.07 (^ measured at 24° ± 0.5°C)

PAGE 142

122 Table 5-2. Moisture content (% wet basis) and water activity (% relative humidity) of salmon fillets during storage at different temperatures (Experiment 2, Fall). (± = std. deviation, n = 3) Storage time (days) Storage Temperature 1.8°C 7°C 11.7°C Variable temperature 1 Variable temperature 2 1 % H 2 0 % RH 65.28 ± 1.06 97.9 ± 0.07 66.05 ±0.12 97.7 ± 0.21 66.14 ±2.19 97.6 ± 0.07 65.20 ± 1.09 97.7 ±0.14 66.67 ± 2.52 97.8 ± 0.00 4 % H 2 0 % RH 65.01 ± 1.59 97.8 ± 0.07 65.91 ± 0.94 97.8 ± 0.07 65.84 ± 0.86 97.7 ± 0.21 66.76 ± 0.80 97.6 ± 0.28 67.13 ± 1.41 97.6 ± 0.07 7 % H 2 0 % RH 66.10 ± 1.27 97.7 ±0.21 66.59 ± 2.06 97.7 ±0.14 64.09 ± 2.64 97.5 ±0.14 (a w measured at 23° ± 0.5°C) Color and odor measurements in experiment 2 were carried out until day 8 of storage. This was due to the fact that samples arrived 72 hrs after harvest compared to 48 hrs for samples in experiment 1 . This was reflected in the quality of the fillets, both in odor and color, due to the loss of one day of shelf life. Panelists gave better sensory scores to samples in experiment 1 compared to those in experiment 2 (Table 5-3). In general it was observed that color did not change as fast as odor did. However, the overall sensory score given by panelists was highly correlated with the odor score, and in some cases it was the average of color and odor scores. Figures 5-2 and 5-3 show the DFA results correlating electronic nose sensor readings and color data to sensory scores at each storage temperature for experiment 1 and experiment 2, respectively. The twelve sensor outputs and color block data (Table 54) were reduced to two discriminant functions to locate points in the two-dimensional plots. For the three storage temperatures 1.8°, 7° and 1 1.7°C, DFA clearly separated the data into the three sensory grades A, B, or C. The correct classification rates for the

PAGE 143

123 Table 5-3. Sensory scores for odor, color and overall evaluation of fresh salmon fillets stored at different temperatures. Sensory score of 1 = good, 1 0 = bad Storage time (days) Storage Temperai ure Sensory attribute 1.8°C 7°C 11.7°C Vari tern able p. 1 Van tem able p. 2 # r # 2 b # l a # 2 b # l a # 2 b # r # 2 b # l a # 2 b odor 1 2 1 2 1 2 1 2 1 2 1 color 1 1 1 1 1 1 1 1 1 1 overall 1 2 1 2 1 2 1 2 1 2 odor 1 3 1 4 2 4 1 4 1 5 2 color 1 2 1 2 1 3 1 2 1 2 overall 1 2 1 3 2 3 1 3 1 4 odor 2 5 3 6 5 6 5 6 3 5 3 color 3 4 5 4 5 5 4 5 3 4 overall 2 4 4 5 5 6 4 5 3 5 odor 2 6 6 8 7 9 6 7 5 7 4 color 3 6 5 6 7 8 5 6 4 5 overall 3 6 5 7 7 9 5 6 4 6 odor 3 7 8 8 9 8 8 6 8 5 color 4 7 8 7 10 5 7 5 6 overall 3 7 8 8 9 7 7 5 7 odor 3 8 9 9 9 9 8 9 6 color 5 8 9 9 8 8 6 8 overall o o o y 8 9 7 9 odor 4 8 10 10 8 10 7 color 4 9 8 1 A 10 7 9 overall 4 9 9 10 7 10 odor 5 9 9 8 color overall 6 5 10 9 8 8 odor 6 9 9 color overall 6 6 8 9 odor 7 10 10 color overall 8 8 9 10 a : Experiment 1 , Summer b : Experiment 2, Fall

PAGE 144

Storage temperature 1.8°C 124 -8 .2 0 -2 0 2 Function 1 Storage temperature: 1°C \ -2 0 Function 1 Storage temperature: 1 1 .7°C -IX -12 -6 Function I Sensory Grade o A (1-4) d B(5-7) o C(8-10) 0 /© o \ / o o \ O CP 1 V 8 . Sensory Grade ° A (1-4) ° B(5-7) o C(7-10) 0 0 o \ % y >° o J _ o ^/ / o o \ t \ a \ V Sensory Grade ° A (1-4) a B (5-7) o C(7-10) Figure 5-2. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades correlated with electronic nose and color data. (Experiment 1, Summer)

PAGE 145

Storage temperature: 1.8°C 125 d Storage temperature: 7°C Storage temperature: 11.7°C Sensory Grade o A (1-4) D B (5-7) 0 C (8-10) Sensory Grade ° A (1-4) ° B (5-7) ° C (7-10) Sensory Grade A (1-4) B (5-7) C (7-10) Figure 5 -3 . DF A of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades correlated with electronic nose and color data (Experiment 2, Fall)

PAGE 146

126 Table 5-4. Color blocks whose levels were above 2% of the total area of the samples analyzed. Color blocks with a / represent colors included in the combined data set (electronic nose and color) or when color alone was used to obtain the DFA models Color Block No. (RGB) c u & 16 32 36 48 52 53 57 w (96,32, 32) (160,32,32) (160,96,32) (224,32,32) (224,96,32) (224,96,96) (224,160,96) # 1 / / / / / / / #2 / / / y / / discriminant functions were 100% except for the 1.8°C set in experiment 1, which was 98.3% (Table 5-5). Cluster separation into the three sensory grades was very clear (Figures 5-2 and 5-3). As expected, the distance between the cluster from grade A and grade C was greater when compared with grade B. This could be due to the fact that sensory analysis was carried out by panelists that may have mis-classified some samples, since change in smell is not as drastic at low temperatures. For the 7° and 1 1.7°C storage temperatures, separation into the three sensory grades was very clear. As expected, the distance between the cluster from grade A and grade C was greater when compared with grade B. Electronic nose data alone had classification rates between 90 and 95%, when sensor data were correlated with sensory score (Table 5-5). This demonstrates the odor as perceived by panelists at each storage temperature can be discriminated fairly easily with the electronic nose. In the case of color alone, classification rates ranged from 66% to 93% (Table 5-5). In experiment 1, correctly classified cases were higher than those in experiment 2. Color data are not as highly correlated with the grades given by the

PAGE 147

127 panelists. However, when electronic nose sensor data and color data sets were combined, the overall quality of the samples was perfectly discriminated. This points to the potential advantage of discretizing color data, to combine it with the electronic nose sensor data for better overall quality evaluatioa Data from the three storage temperatures were pooled together and analyzed with DFA (Figures 5-4 and 5-5). The correct classification rates were 92.1% and 93.5% for experiment 1 and experiment 2, respectively. As expected, the correct classification rates are lower since there is increased variability due to the combination of all the odors and colors from all temperatures. During sensory evaluation of the salmon fillets, panelists detected differences in odors between temperatures. The putrid odor at 1 .7°C which was given a score of 8 or 9 was different from the putrid odor at 1 1 .8°C with the same sensory score (as verbally described by panelists). At different storage temperatures there will be selective growth of different types of microflora, and the metabolites from them will be different. In addition, reaction rates at the three temperature storage conditions are different. At the highest temperature condition (1 1 .7°C), panelists detected a stronger rancid odor compared to those at lower temperatures. Therefore, the bad sensory scores for those samples were not only related to microbial spoilage, but also to rancidity. In the case of color, panelists did not detect differences in color based on storage temperature. The only common observation was the gradual change of color to a beige brown as time progressed. There was also a layer of slime that was being formed on the fillet. As expected, the higher the storage temperature the faster the formation of the slime.

PAGE 148

128 Table 5-5. Correctly classified cases obtained from the classification matrix for the DFA of electronic nose readings, color data and the combination of both when correlated with odor, color and overall sensory grades, respectively. (Values are % of correctly classified samples) Storage Temp. Data includec inDFA model Experiment 1 , Summer Experiment 2, Fall n Sensory score Overall classif. n Sensory score Overall classif. A (1-4) ti (5-7) C {Q 1 A\ L (5-1UJ A /I 1 \ A (1-4) B (5-7) L (8-10) 1.8°C e-nose ou 95.2 83.3 83.3 Ol O 48 100.0 100.0 83.3 93.8 color 88.1 75.0 66.7 65.5 48 58.3 55.6 88.9 68.8 en-c aa 01) 100.0 91.7 100.0 AO i 98.3 A O 48 100.0 100.0 100.0 100.0 7°C e-nose AD HZ 88.9 100.0 94.4 qo o JO 100.0 83.3 94.4 94.4 color AD HZ 66.7 100.0 93.3 1 l.o 50 58.3 100.0 61.1 66.7 en-c AO HZ 100.0 100.0 100.0 1 AA A 36 100.0 100.0 100.0 100.0 11.7°C e-nose JU 100.0 91.7 100.0 OA 7 yO. 1 OA Z4 100.0 100.0 100.0 100.0 color 91.7 91.7 100.0 yi.5 Z4 91.7 83.3 100.0 91.7 en-c 30 100.0 100.0 100.0 100.0 24 100.0 100.0 100.0 ion o All temp." e-nose 132 81.9 86.7 90.0 84.9 108 88.9 80.0 81.0 83.3 color 132 80.6 73.3 79.2 78.6 108 69.4 63.3 66.6 66.7 en-c 132 94.4 96.7 79.2 92.1 108 91.7 93.3 95.2 93.5 Variable temp. b e-nose 51 86.7 83.3 79.2 82.4 84 88.9 80.0 55.6 71.4 color 51 86.7 33.3 45.8 54.9 84 88.8 26.7 77.8 61.9 en-c 51 83.3 88.8 66.7 78.4 84 77.8 76.7 88.9 82.1 n number of electronic nose readings used to obtain the discriminant functions "data for 1.8°, 7° and 1 1.7°C pooled together b validation set: data from both variable temperature studies pooled together, DFA model obtained from all temperatures was used to obtained the classification rates

PAGE 149

Electronic nose data 129 c fcu -1 0 1 Function 1 Color data Function 1 Electronic nose and color data Sensory Grade ° A(l-4) D B (5-7) ° C(8-10) Sensory Grade A (1-4) B(5-7) C(8-10) Sensory Grade 0 A (1-4) n B (5-7) ° C(8-10) Figure 5-4. DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades with electronic nose alone, color data alone and combination of both. (Experiment 1, Summer)

PAGE 150

Electronic nose data 130 B O Figure 5-5. Function 1 Color data -1 0 1 Function 1 Electronic nose and color data Sensory Grade A (1-4) B (5-7) C (8-10) •a o\\ o ° 8 2&\ © t> < 8 o # -oo ""S Sensory Grade o A (1-4) a B (5-7) ° C (8-10) Sensory Grade o A (1-4) n B (5-7) 6 ° C(8-10) DFA of salmon fillets stored at different temperatures. Discrimination of overall quality based on sensory grades with electronic nose alone, color data alone and combination of both. (Experiment 2, Fall)

PAGE 151

131 When sensor readings and color data were analyzed separately, the classification rates were not as good as the combined data set (Table 5-5). The DFA models for the color data performed less effectively, with 78.6% and 66.7% correct classification. This could be due to the natural variability in flesh color of the six fillet replicates, or inability of panelists to quantify color. As explained by Schmidt and Cuthbert (1969), Skrede and Storebakken (1986) and Skrede et al. (1989), flesh color is greatly influenced by the composition of carotenoids in the diet. Color analysis showed that there was not only one color in the flesh of fresh salmon. There were several colors related to the red-orange appearance of the flesh. These colors were not present in the same levels in each replicate (Appendix B). Variable temperature storage studies demonstrated that the discriminant functions obtained using constant temperature experiments were able to predict the sensory grade of salmon fillets from electronic nose and color data with an accuracy of 78.4% and 82.1% for experiment 1 and experiment 2, respectively (Table 5-5). The functions used to perform the validation of the data were those obtained by pooling the three storage temperature data together. Figure 5-6 shows the DFA scatterplot of canonical scores for the three sensory groups of data from all constant storage temperatures combined. The graph also included the samples used in the validation set and their location in the three different sensory grades clusters. The graph shows the location of the missclassified samples. In experiment 1, from the 51 samples from the two variable temperature storage studies, 42 readings were correctly classified based on sensory grades given by panelists. This value is lower than the 92. 1% correct classification rate obtained in the model. In the

PAGE 152

132 second experiment, 69 out of 84 samples were correctly classified. The samples in the variable storage temperature conditions showed that the models obtained from the individual temperatures could be used to predict the overall quality of salmon fillets with a correct classification probability of 8 out of 10 samples analyzed. As expected, models for the electronic nose and color data alone had lower numbers of correctly classified cases. Higher classifications would be desirable, however the goal of this experiment was to prove that these two technologies have potential applications for quality evaluation. For these technologies to be used in commercial settings or by inspections agencies, a bigger database must be acquired. A large data set will create a more robust DFA model, which will probably have a higher classification accuracy. Salmon odor was also changing with storage time. At the lowest temperature (1.7°C), some panelists could not detect differences in odor between consecutive days. However, the Electronic nose was able to detect differences between the odor of salmon at different days of storage. DFA was used to calculate two discriminant functions that described the correlation between electronic nose readings and storage time for each storage temperature data. The correct classification rates for the discriminant functions ranged from 93 to 100% for both experiments (Table 5-6). Color data alone could not differentiate the differences in color among days very clearly. Classification rates averaged around 50-60%. Combining electronic nose and color data gave excellent discrimination by days (Table 5-6). Except for the 1.8°C storage temperature, all were 100% (in both experiments). Figures 5-7 and 5-8 show a clear discrimination between days in storage.

PAGE 153

133 In all three temperatures the cluster for day one is distant from the other days. All other days formed individual clusters with little or no overlap, meaning that there were distinct differences in the sensor readings for each day of storage. The models described above were obtained by using data from all 12 electronic nose sensors. DFA analysis showed that some sensors were not statistically significant in the models (Table 5-7). In feet, in experiment 1 there was only one sensor (T 262 ) which was significant in all models, and none for experiment 2. For the models that explained the correlation between electronic nose sensor readings and storage time, there were two sensors (T 27g and T 262 in experiment 1 ; T 4g3 and T 47g in experiment 2) that were significant for all three temperatures. These data are important when developing electronic noses for specific applications. However, reducing the number of sensors would also reduce the discriminating power of the electronic nose. There are a number of sensors to choose from; however, depending on the product, some will discriminate better than others, and those should be the ones used for a successful analysis. Color data from experiment 1 were fitted to equation 5-1, which assumes a first order reaction. Figure 5-9 shows the fits and Table 5-8 summarizes the calculated parameters for equation 5-1. Two colors were found to change monotonicalry with storage time. Color block no. 32, which was an orange-red color (RGB = 160, 32, 32), decreased with storage time, while color block no. 36, a brown color (RGB = 160, 96, 32), increased. Figure 5-9 shows that formation of the brown color gave a better fit than the disappearance of orange-red color. As storage temperature increased, the rate of brown color formation increased. In the case of the orange-red color, the disappearance

PAGE 154

Experiment 1, Summer 134 Experiment 2, Fall Sensory Grade • Validation Grade A Validation Grade B * Validation Grade C Sensory Grade • Validation Grade A Validation Grade B * Validation Grade C Figure 5-6. DFA model of salmon fillets overall quality correlated with electronic nose and color data combined, with the validation data set from variable storage temperature studies

PAGE 155

135 Table 5-6. Correctly classified cases obtained from the classification matrix for the DFA of electronic nose readings, color data, and combination of both when correlated with storage time. (Values are % of correctly classified samples) Storage time Data included in Storage Temperature Experiment 1: Summer Experiment 2 : Fall (days) DFA model 1.8°C (n=60) 7°C (n=42) 11.7°C (n=30) 1.8°C (n=480) 7°C (n=36) 11.7°C (n=24) e-nose 100.0 100.0 100.0 100.0 100.0 100.0 1 color 50.0 50.0 66.7 33.3 33.3 66.7 en-c 100.0 100.0 100.0 100.0 100.0 100.0 e-nose 100.0 100.0 100.0 100.0 100.0 100.0 2 color 33.3 50.0 83.3 33.3 0.0 83.3 en-c 100.0 100.0 100.0 100.0 100.0 100.0 e-nose 100.0 100.0 100.0 100.0 100.0 100.0 3 color 66.6 83.3 83.3 83.3 83.3 83.3 en-c 100.0 100.0 100.0 83.3 100.0 100.0 e-nose 100.0 100.0 100.0 100.0 83.3 100.0 4 color 50.0 83.3 100.0 16.7 16.7 100.0 en-c 100.0 1000.0 100.0 100.0 100.0 100.0 e-nose 83.3 83.3 100.0 100.0 100.0 5 color 33.3 66.7 100.0 66.7 83.3 en-c 83.3 100.0 100.0 100.0 100.0 e-nose 83.3 100.0 100.0 100.0 6 color 33.3 83.3 50.0 100.0 en-c 83.3 100.0 100.0 100.0 e-nose 100.0 100.0 100.0 7 color en-c 50.0 100.0 83.3 100.0 66.7 100.0 e-nose 83.3 83.3 8 color en-c 83.3 100.0 83.3 100.0 e-nose 100.0 9 color en-c 50.0 100.0 e-nose 83.3 10 color en-c 66.7 100.0 e-nose 93.3 97.6 100.0 97.9 97.2 100.0 Overall color 51.7 69.4 86.7 54.2 52.8 83.3 en-c 96.7 100.0 100.0 97.9 100.0 100.0 n number of electronic nose readings used to obtain the DFA functions

PAGE 156

-15 16 12 8 4 c ,o unct 0 u. -4 -8 -12 Storage temperature: 1.8°C 136 5 10 15 Function 1 Storage temperature: 7°C -10 -5 5 10 Function 1 15 20 25 Storage temperature: 11.7°C Storage Time 30 Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Day 9 Day 10 % 1 <0> 30 Storage Time ° Day 1 D Day 2 ° Day 3 A Day 4 • Day 5 Day 6 Day 7 1 ! i : — 0 0 J ©
PAGE 157

Storage temperature: 1.8°C 137 B B a B -12 12 8 -1 0 -4 -8 -12 -16 25 20 15 10 5 0 -5 -10 -15 -20 -80 Figure 5-8. -30 -20 -10 Function 1 Storage temperature: 7°C 10 -30 -20 -10 0 Function 1 Storage temperature: 11.7°C 10 20 8 -60 ^o -20 0 Function 1 20 40 60 Storage Time Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Day 7 Day 8 Storage Time Day 1 Day 2 Day 3 Day 4 Day 5 Day 6 Storage Time Day 1 Day 2 Day 3 Day 4 DFA of electronic nose and color data vs storage time of salmon fillets (Experiment 2, Fall)

PAGE 158

Table 5-7. Statistical significance (pvalues) from DFA of the electronic nose sensors contribution to the prediction of group membership Exp. Sensor Sensory evaluation Storage time iypc 1.8°C 7°C 11.7°C (Ml temp. 1 1.8°C 7°C 11.7°C 0.000 * 0.987 0.036 * 0.034 * 0.000 * 0.007 * 0.105 T 297 0.499 0.538 0.501 0.513 0.308 0.549 0.839 0.008 * 0.151 0.199 0.054 0.004 * 0.305 0.107 u ^279 0.002 * 0.271 0.380 0.000 * 0.000 * 0.009 * 0.191 T278 0.445 0.211 0.039 * 0.788 0.000 * 0.000 * 0.003 * ^264 0.000 * 0.240 0.568 0.001 * 0.000 * 0.308 0.002 * periment ^263 0.059 0.342 0.004 * 0.115 0.001 * 0.237 0.000 * ^262 0.001 * 0.000 * 0.000 * 0.000 * 0.000 * 0.001 * 0.000 * X W 1*261 0.034 * 0.289 0.081 0.000 * 0.000 * 0.012* 0.247 T26O 0.493 0.426 0.135 0.134 0.509 0.248 0.115 ^259 0.376 0.893 0.521 0.133 0.389 0.505 0.410 ^258 0.160 0.548 0.611 0.028 * 0.000 * 0.414 0.041 * T401 0.577 0.215 0.207 0.519 0.537 0.005 * 0.038 * ^298 0.066 0.758 0.087 0.172 0.641 0.196 0.107 T 297 0.366 0.768 0.081 0.107 0.009 * 0.962 0.090 ^483 0.917 0.068 0.003 * 0.009 * 0.000 * 0.031 * 0.003 * Fall T478 0.117 0.226 0.005 * 0.121 0.000 * 0.000 * 0.003 * a ^464 0.017 * 0.019 * 0.153 0.041 * 0.016 * 0.000 * 0.095 T M63 0.001 * 0.010* 0.672 0.926 0.005 * 0.001 * 0.827 & ^462 0.043 * 0.000 * 0.609 0.000 * 0.201 0.001 * 0.008 * T46I 0.191 0.054 * 0.044 * 0.025 * 0.665 0.009 * 0.058 T46O 0.364 0.852 0.777 0.674 0.004 * 0.002 * 0.911 T459 0.222 0.005 * 0.728 0.000 * 0.111 0.002 * 0.890 T4S8 0.027 * 0.000 * 0.163 0.000 * 0.000 * 0.000 * 0.052 * significant at the p-level < 0.05

PAGE 159

139 Table 5-8. Parameters for equation 5-1, fitted to color data of salmon stored at different temperatures (Experiment 1 , Summer) Storage Color 32, RGB = 160, 32, 32 orange red Color 36, RGB = 160, 96, 32 brown Temp. Color 0 (%) k(l/days) R 2 Color 0 (%) k(l/days) R 2 1.8°C 56.507 -0.061 0.44 21.527 0.084 0.69 7.0°C 69.965 -0.217 0.60 18.545 0.218 0.86 11.7°C 83.995 -0.241 0.62 11.200 0.379 0.92 of the color also changed with storage temperature, however for 7.0°C and 1 1.7°C the rates of change were similar. Using the Arrhenius approach and the parameters from Table 5-8, the energy of activation for equation 5-2 was calculated (Table 5-9). Using equation 5-2, data from the variable temperature studies were used to validate the model, and predict the changes in color based on storage temperature and time. Figure 5-10 shows that predicted values for the brown color were close to the measured values, however for the orange-red color there were more deviations. This model showed that formation of the brown color (a quality defect) can be predicted based on the time-temperature history of storage of the sample. Once you predict the amount of color, then you can predict what will be the grade given by the inspectors by using data presented in Figure 5-11, which was fitted to equation 5-3. Sensory scores showed good correlation with the amounts of the brown and orange-red color present in the samples. The second order polynomial equation had r 2 = 0.70 and 0.90 for the orange-red and brown colors, respectively. These fits were obtained by combining the color data from the three storage temperatures. Therefore, color machine vision system can quantify the amounts of colors present in a sample in an

PAGE 160

140 objective and reproducible manner, and correlate it to sensory scores to predict an inspector's evaluation. L*a*b* color values were measured using the color machine vision system. Results showed that L* value (lightness) tended to increase with storage time without following a specific pattern. At 7°C and 1 1 .7°C, the L* value was high at the end of the experiment. This increase could be due to the formation of slime, which gave a pale and lighter color. The a* and b* values tended to decrease (Figures 5-12 and 5-13). The L*a*b* values from experiment 1 and 2 were different. In experiment 1, L* was higher than that of experiment 2. This was due to the increase in fat content, which gave a more pale orange-red color. L*a*b* values showed low correlations with sensory score. In experiment 1 the correlation of L*a*b* values with sensory scores were 0.47, -0.59, and Table 5-9. Parameters used for the linear fit to calculate energy of activation and kg for color changes in salmon stored at different temperatures (Experiment 1, Summer) Storage Color 32, RGB = 160, 32, 32 orange red Color 36, RGB = 160, 96, 32 brown Temperature 1/Temp (1/°K) ln(k) 1/Temp (1/°K) ln(k) 1.8°C 0.00364 -2.801 0.00364 -2.473 7.0°C 0.00357 -1.530 0.00357 -1.525 11.7°C 0.00351 -1.423 0.00351 -0.971 Intercept 37.959 41.228 Slope -11156.74 -11997.97 r 2 0.838 0.988 E a -92761.59 -99755.91 ko -3.059xl0 16 8.034xl0 17

PAGE 161

141 Color no. 32 RGB = 160, 32, 32 orange red 100 r — 90 123456789 10 ^ n 7 ° c Storage time (days) Color no. 36 RGB = 160, 96, 32 brown 4 5 6 7 Storage time (days) 10 Storage Temperature I. 8°C 7.0°C II. 7°C Figure 5-9. Color changes vs time for salmon stored at different temperatures

PAGE 162

142 I O & O U 65 55 45 35 25 First set: color 36 (RGB 160, 96, 32) brown 1 ta i I J js^ff^^Z.b. 1 i I i I i — *n. > : 12 9 6 3 I 5 b _2 o 65 55 45 35 25 First set: color 32 (RGB 160, 32, 32) orange red O Experimental coloiY, Predicted color Y, Storage temperature Y 12 9 6 3 0 y o ' — I I £ Storage time (days) Figure 5-10. Measured and predicted colors of salmon fillets stored under variable storage temperature conditions i

PAGE 163

143 10 1 8 •rf 6 o o & 2 to g 09 Color No. 32 RGB = 160, 32, 32 (orange red) y= 10.070 -0.1 16 x0.0005 x 2 r* = 0.75 o N. o o o 0 G o o X. « > . o o ooo o 10 20 30 40 50 Color (% of total area) 60 70 10 1 8 •o 6 o o 00 p s o Color No. 36 RGB = 160, 96, 32 (brown) y = -5.540 + 0.343 x 0.002 x 2 r 2 = 0.90 0 O x o >^ o CO O Oy/ O / 0 o y&Q oo o 1 , 20 30 40 50 60 Color (% of total area) 70 80 Figure 5-11. Correlation of sensory scores with color from the flesh of salmon fillets

PAGE 164

144 Storage temperature: 1.8°C 55 50 45 40 35 30 I I _ i 4 5 6 7 Storage Time (days) Storage temperature: 7.0°C 55 50 45 40 35 30 3 4 Storage Time (days) Storage temperature: 11.7°C 55 50 I | 45 ja a d 40 8 8 U 35 30 10 -T 2 3 Storage Time (days) >^ LabL Lab a Labb ^ LabL Lab a Labb ""
PAGE 165

145 Storage temperature: 1.8°C 3 4 5 Storage Time (days) Storage temperature: 7.0°C 3 4 Storage Time (days) Storage temperature: 1 1 .7°C 2 3 Storage Time (days) ^ LabL Lab a ' , ».. Labb ^ LabL -*s Lab a *"*s Labb ^ LabL Lab a "*s Labb Figure 5-13. L*a*b* values from flesh of salmon fillets during storage at different temperatures (Experiment 2, Fall)

PAGE 166

146 -0.16 respectively, while in experiment 2 they were 0.15, -0.52, and -0.69. These results showed that L*a*b* values alone can not be used as predictors of quality of salmon fillets. Conclusion This study showed that the combination of electronic nose sensor readings and color data correlated well with the overall quality of salmon fillets with storage time and with scores from sensory evaluation by using DFA as the pattern recognition technique. These results could be used to develop methodologies to assist in the objective and repeatable quality evaluation of salmon. This method has potential in industrial and regulatory application where rapid response, no sample preparation, no requirements for chemicals, and no technical expertise to run the system are required. However, further work is needed to accumulate extensive data sets that could be used to predict the sensory grade of salmon from different species, origins, season of harvest, age of fish, growing environment (aquaculture or wild), etc. It is expected that including all these variables will affect the correct classification rate. It is also crucial to test the transportability of the model from one electronic nose system to another. This study used two electronic noses from the same manufacturer, however with different sensors, and data could not be transported between the two machines. This is critical in demonstrating the feasibility of using electronic nose-based inspections and evaluations in commercial settings. In color machine vision, the differences between hardware are the main concern for the transportability of the data. In this case color calibration must be performed to be able to compare images taken under different hardware conditions. Color standards should be

PAGE 167

147 used to adjust color settings in the acquired images. In these experiments discrimination ability of DFA did not depend on the hardware (frame grabbers). Both sets of data gave good classifications. Also, both sets of data had similar color blocks. However, the amount of each color block was different due to differences in color of the flesh, from one season to the other, rather than differences due to hardware.

PAGE 168

CHAPTER 6 EVALUATION OF DECOMPOSITION ODOR IN RAW AND COOKED SHRIMP: CORRELATION BETWEEN ELECTRONIC NOSE READINGS, ODOR SENSORY EVALUATION, AND AMMONIA LEVELS Introduction One of the main quality aspects of raw shrimp is smell. Current inspection of shrimp in local and international markets relies on sensory evaluation, where an inspector smells the shrimp for signs of decomposition. Generally when shrimp has a fishy, spoiled or putrid odor, it is rejected. Compared to other muscle foods, seafood products are significantly higher in low molecular weight non-protein nitrogen compounds (Finne, 1982). This results in their unique, delicate and different odors and flavors. These compounds are also responsible for the rapid deterioration of fresh seafood by serving as readily available substrates for typical spoilage microorganisms, which convert them into obnoxious smelling volatile bases (Finne, 1992), such as trimethylamine (TMA), dimethylamine (DMA), ammonia (NH 3 ) or hydrogen sulfide (Stansby, 1962). Deterioration of the odor of freshly caught seafood is caused by facultative aerobic bacteria, (Gorga and Ronsivalli, 1988), bacterial enzymes, and naturally occurring enzymes (Finne, 1982). To express the quality of shrimp in terms of chemical parameters, many reports suggested the determination of a variety of chemicals. Changes in pH 148

PAGE 169

149 (Bethea and Ambrose, 1962), TMA (Ruiter and Weseman, 1976; Shamshad et al., 1990; Fatima et al., 1988), DMA (Ruiter and Weseman, 1976), total volatile nitrogen (TVN), (Ruiter and Weseman, 1976; Cobb and Vanderzant, 1975; Malle and Poumeyrol, 1989), indole (Chang et al. 1983; Niola and Valletrisco, 1986), inosine monophosphate and hypoxanthine (Fatima et al., 1981), and total volatile nitrogen/amino nitrogen ratio (Cobb and Vanderzant, 1975) have been either used or suggested as quality indices for shrimp. According to the Code of Federal Regulations (CFR, 1998), freshly caught shrimp flavor and odor are best described as mild and pleasant, typical of the sea and seaweeds, while spoiled shrimp has an ammonia smell. Shrimp held on ice for less than one week will have little or no objectionable odor (Gorga and Ronsivalli, 1988; Campbell and Williams, 1952; Matches, 1982; Shamshad et al., 1990). According to the FDA (Federal Register, 1981), shrimp decomposition could be classified in three groups: Class I (fresh aroma and no odor identifiable as decomposition, indole level < 25 ug/100 g), Class II (slight odor of decomposition, the odor is persistent and readily perceptible to the experienced examiner, indole level > 25 ug/100 g) and Class III (strong odor of decomposition which is persistent, distinct and unmistakable, indole level > 50 ug/100 g). Analysis for indole is time consuming and requires complicated laboratory equipment such as HPLC and chemical reagents that can not be implemented by the industry in a regular quality control laboratory. Moreover, other chemical analyses are hardly used in the industry due to lengthy procedures and the use of hazardous chemicals. Volatile compounds can also be detected by sensors. An electronic nose is an array of sensors that have potential for food applications (Corcoran, 1993). The

PAGE 170

electronic nose has been used in evaluation of shrimp odor (Balaban and Luzuriaga, 1996), monitoring of haddock and cod freshness (Olafsson et al, 1992), recognition of fish storage time (di Natale et al., 1996), quality estimation of ground beef (Winquist et al., 1993), monitoring the flavor and aroma of beer and its raw materials (Pearce et al, 1993; Tomlinson et al., 1995; Zimmermann and Leclercq, 1995), classification of grains (Borjesson et al., 1996), aroma profiles of swiss cheese (Harper et al., 1996), volatiles of fresh squeezed orange juice (Bazemore et al, 1996), among other studies. Although the electronic nose is rapid and objective in quantifying odors, little work has been published on the correlation of electronic nose sensor outputs to sensory results in seafood products The shrimp industry is one of the most important seafood industries worldwide. The United States (U.S.) imported $2.9 billion dollars worth of shrimp in 1997, accounting for 37% of the value of total seafood imports (USDC, 1998). It is estimated that more than 75% of the shrimp consumed in the U.S. is imported from all over the world. Quality assessment relies on sensory evaluation in local and international markets, and is subjective and difficult to quantify. In the U.S., regulatory agencies conduct decomposition workshops to train the industry and other inspectors to detect odors of decomposition in shrimp and other seafood species by using sensory analysis. Therefore, this valuable commodity could benefit from improved inspection methods. Therefore, the electronic nose technology could be a promising tool for the odor evaluation of shrimp, because it is a fast, objective and simple methodology. However, databases must be developed for this technology to be used by industry and government agencies. The database should include different species of shrimp from around the world, different

PAGE 171

151 seasons, wild and aquacultured samples, raw and cooked products, among other factors. Database development will include a standard methodology for sample preparation (intact vs. chopped), conditioning (analysis temperature, equilibration time, etc.) and an analytical procedure for the electronic nose. The database should be transferable among different machines for this technology to be fully implemented. The overall objective of this study was to evaluate the decomposition odor of raw and cooked shrimp using an electronic nose, rapid ammonia analysis and sensory panels. The specific objectives were: 1) to evaluate the odor of shrimp samples with an electronic nose and correlate these data with results from sensory analysis; 2) to measure ammonia levels in shrimp samples and correlate results with sensory evaluation; 3) to assess the differences in odor between species of shrimp in the raw and cooked stage; and 4) to determine effects of different sample preparations (intact vs. chopped) on odor evaluation with an electronic nose. Materials and Methods Shrimp Samples Two sets of samples were used in this study. The first set was raw headless shellon shrimp of different species and origins (Ecuadorian white Penaeus vannamei . Mexican white P. setiferus. Mexican brown P. aztecus .. Thai pink P. duorarum and Thai tiger P. monodpn), that were obtained from 2 decomposition workshops conducted in 1996 and 1997 by the Food and Drug Administration (FDA) at the University of Florida.

PAGE 172

152 Immediately after the shrimp samples were evaluated by FDA experts, they were placed in plastic bags and frozen. Samples were kept frozen for less than a week. Then, they were placed in sealed plastic bags and thawed under running tap water. The samples were then kept in a cooler (2°C) for less than 2 hrs, then removed and equilibrated to room temperature for 1 hr and presented to the electronic nose. Immediately after this, the sample was analyzed for ammonia. Both intact and chopped shrimp samples were presented to the electronic nose. Samples were chopped by hand using a kitchen knife. The particle size was approximately 3 mm in diameter. The second set of samples consisted of raw Ecuadorian white (P. vannamei) . Honduras pink ( P. duorarum) and Thai tiger ( P. monodoh) frozen headless shell-on shrimp. Shrimp size were 88-1 10 count/Kg. Samples were purchased from Lombardi's Seafood (Orlando, FL) and arrived in 2.2-Kg frozen blocks. Samples were thawed under running tap water. Two Kg of each specie were placed in plastic bags and stored at refrigeration temperature (2°C) for 14 days. Replicates were taken at random from the plastic bag and evaluated every other day for odors in the raw and cooked forms by trained sensory panelists, electronic nose and ammonia analysis. Samples were removed from the refrigerator 1 hour prior to the electronic nose analysis. Samples to be cooked were removed from the refrigerator, and stored at room temperature for 30 min. Then they were placed in boiling tap water for 2 min. Different species were cooked separately. Cooking was done 1 hour before the electronic nose readings. This allowed for samples to equilibrate to room temperature.

PAGE 173

153 Sensory Evaluation For the first set of samples, a group of FDA experts from the decomposition workshops (4 in 1996 and 2 in 1997) evaluated the odor of the raw shrimp samples on a pass-fail scale. This scale is a simplification of the three classes of decomposition odors described by FDA in the Federal Register. Class I is a 'pass', while Class II and Class III are 'fail'. However, sometimes samples were classified as borderline-pass or borderlinefail during the decomposition workshops. Therefore, for the purpose of data analysis in this study, three sensory groups were chosen: pass, borderline and fail. During sensory analysis, the FDA experts sometimes broke the shrimp tissue to obtain a better odor signal. Therefore, shrimp samples for the electronic nose were presented intact and chopped, to detect differences in the odor profile based on sample preparation. For the second set of samples, the odor of raw shrimp was evaluated by a 10member trained sensory panel consisting of professors and graduate students, 24 45 years of age, from the Food Science and Human Nutrition Dept. at the University of Florida. Samples were evaluated every other day. At each sampling day, a random sample was collected from the 2 Kg bag of shrimp, and presented to the panelists. A sample of 150-200 g of shrimp were placed in a glass container, covered with aluminum foil, and equilibrated to room temperature. Panelists were asked to evaluate the sample based on the pass-borderline-fail scale. Two samples were presented each day, per specie, for both raw and cooked shrimp. Samples used for sensory analysis were not the same

PAGE 174

154 samples as the ones measured by the electronic nose, but were selected at random from the same 2 Kg bag of shrimp. Electronic Nose Measurements An electronic nose (e-NOSE 4000 model, Neotronics Inc., Flowery Branch, GA) equipped with twelve conducting polymer sensors (sensor types: 483, 478, 464, 463, 462, 461, 460, 459, 458, 401, 298 and 297) was used to quantify the sensor responses to odor changes in shrimp samples. Six replicates were analyzed by the electronic nose for each shrimp sample. In the first set, some samples had only four or five replicates, because of the restricted amount of shrimp collected from the FDA workshops. Replicates were taken at random from the plastic bags where shrimp was stored. Each replicate was approximately 50 g (five headless shell-on shrimp). Each replicate was taken out of the cooler 1 hour prior to the analysis, to let the shrimp temperature equilibrate to room temperature (21 .6 22.6°C). The sample was put in a 140-ml beaker, and placed in the sampling vessel of the electronic nose. The day the experiment started, the electronic nose was calibrated following the manufacturer's recommendation using a 75% v/v propylene glycol solution (100% solution from Fisher Scientific, No. P-355-20, Fair Lawn, NJ). Every day prior to the experiments the electronic nose was turned on and compressed air (CGA Grade D, Strate Welding Supply Inc, Jacksonville, FL) was passed through the sensors for at least 30 minutes. For each replicate the vessel was purged with compressed air for 2 minutes to eliminate any extraneous odor. Then the sensor head was purged for 4 minutes with compressed air. During these 4 minutes, the sample volatiles were

PAGE 175

155 equilibrating in the headspace of the vessel. Sensor response data was acquired for 4 minutes. Analysis time for each sample took 10 minutes. Readings at 4 minute exposure of the sensors to the samples were used for data analysis. At the end of the day, electronic nose sensors were cleaned with compressed air for at least 30 minutes. Electronic nose data can be obtained from Dr. Murat Balaban, Food Science and Human Nutrition Department at the University of Florida (filenames: "\\thesis\text files\Chap 6\FDA enose data.txt" and "\\thesis\text files\Chap 6\Storage enose data.txt". Ammonia Analysis Ammonia levels in the shrimp samples were measured with an ion-selective electrode (Model 95-12, Orion Research, Inc., Boston, MA) connected to an Expanded Ion Analyzer (EA 920, Orion Research Inc., Beverly, MA). Samples were measured for ammonia immediately after the electronic nose evaluation. The 50 g replicates were sprayed with 4 ml of 5N NaOH (Cat. No. LC24450-4, Lab Chem Inc., Pittsburgh, PA) and analyzed for ammonia in the headspace of an air-tight box, following the procedure described by Luzuriaga et al. (1997a). The ammonia electrode was calibrated prior to, and during the experiments using 10 ppm, 100 ppm and 1000 ppm ammonia solutions (Orion Cat No. 951007, Orion Research, Inc., Beverly, MA). Data Analysis Sensor readings and sensory data were analyzed in Statistica for Windows ('98 edition, StatSoft Inc., Tulsa, OK) using discriminant function analysis (DFA) as reported

PAGE 176

156 by other researchers (Corcoran, 1993; Gardner and Hines, 1997; Gardner and Bartlett, 1992). Discriminant function analysis was used to construct predictive functions to help in classifying data into the three odor groups (pass, borderline, fail). The 12 sensor outputs were reduced to 2 discriminant functions. These functions were used to map the data in two dimensional plots and observe separation between groups. Correct classification rates and the coefficients for each function were calculated. Results and Discussion The electronic nose was able to predict the degree of decomposition (pass, borderline, fail) of shrimp with accuracies close to those of the FDA inspectors. By using DFA for both intact and chopped shrimp samples, the twelve sensor outputs were reduced to two discriminant functions (Table 6-1 and 6-2) to calculate coordinates of points which were mapped on the two-dimensional plots. For the six species of shrimp (intact) collected during the decomposition workshops, DFA was capable of separating the data into the three levels of decomposition, with correct classification rates above 85% and in some cases 100% (Table 6-3). Two dimensional plots generated from the discriminant functions showed a very clear separation of the three levels of decomposition. Figures 6-1 to 6-5 show results from the DFA for the different shrimp species. Mexican brown, Thai pink and Mexican white shrimp showed perfect separation of the clusters. The other species had some

PAGE 177

157 overlap of the clusters. As expected, the borderline samples overlapped with the pass or fail clusters. This overlap could be due to missclassiflcation by the inspectors. During sensory evaluation of shrimp samples, sensory experts recommended breaking the tissue of the shrimp to get a better odor signal. This recommendation applied especially to borderline samples. When this recommendation was applied to the analysis of shrimp samples with the electronic nose, the system detected subtle differences among intact and chopped shrimp, as seen in the correct classification rates (Table 6-3). It was expected that if the sample were chopped, the volatiles in the sampling vessel of the electronic nose would reach equilibrium faster, which can be reflected as a higher response from the sensors and better discrimination. Correct classification rates did not change in most of the cases when analyzing chopped shrimp versus intact shrimp, except for the Mexican white and Ecuadorian white shrimp, which decreased to 80.6% from 94.4% and 90.5% from 97.6%, respectively. Figures 6-1 to 6-5 show the separation of clusters in space from the DFA for the different shrimp species (chopped). Decomposition odors are mainly encountered on the surface of the shrimp, where bacterial action takes place. When samples were chopped, the aroma from the interior flesh of the shrimp was also exposed to the sensors. Therefore, the electronic nose was picking up other odors not from decomposition odors such as ammonia, DMA, TMA and other chemicals, which are derived from bacterial breakdown of proteins and other nitrogenous compounds. The presence of flesh odors probably caused the lower classification rates for the chopped samples.

PAGE 178

T3 0> O o on CO a 1 in on I I S m o i oo (U CO o c o c "to o -a I £ o o , «8 o — 0 *** O »3 a * 1 I £? " > 23 p-s ° J2 (U CO * c H c I CD o O I a O a S£3 o O D O T3 < 6 Q * so O Os SO On 00 SO (N rn SO CO l> vo Os 00 (N so CO co 00 foo CN o O in 1 co° co i in •r> 1 Os Os' i © 1 © i i sd r> Os SO co CN tT © CN in SO so >n 00 CN o H* as CN as l> 1 00 1 CN i od co i CO CN CN O in p* co co o H co d © 1 © (N i © (N SO ir> SO co CN 1 >n •* 1 O o m a. © o in Os (N © m CO SO Os 00 Os CN SO Os 00 co so Os CN so Os F> i (N Os i o © CN CN CN in SO CN co CN 1 © OO ©' (N 1 © uo l> t' i co od CN i od i jo co fs© SO 00 v> SO rn SO Os co 00 00 SO © r~~ co r-~ OS H o od CN SO i SO m CO co so sd CN co co sd CN i co >n i H co r fN • r1 os 1 in T Os CN i (N i Os t CO 1 Os od CN CO od 00 H tr> oo as oo OO Os © Os 00 CN Os CN CO Os r~© © 1CN Os co (N ci © co i ©' i sd CS 1 wo" 1 © © co SO i CN i co CO CI oo H m 00 Os fOs r> co Os m 00 © CN 1 1 Os (N sd od © 1 Os co i sd I CN CN i in OS H >n © CN «n SO in © rOs © *t r~ 00 OO CO rSO rSO Os © 1 so 1 Os CO (N (N i CN co CN SO sr> co Os Os i co i oo Os H >n o 00 CN 00 (N r00 (N O CO 00 CN Os V) 1> Os 00 CO © co O 1 o> od sd m cn i od uo Vl wo co Os CN i i in TT O m H 00 o CN co 00 r-~ © © Os SO cl> t
PAGE 179

T3 4J O = o o M M OA .s •s u E o 3 £ S o jg u o I u fc o o • 09 U O c/i O __ 35 ~ C Q. .2 5 P s. O d 'I c u o U o i , 8 I 09 i .a c .o O I o o o . < e Q 4a IN i NO OJ S3 'o J5 o o CN NO 00 m NO rON co o -r ON NO >/-) Coi © 1 CN vi NO ON oo co 1 r-" co i CN NO CN oo »T> ON ro ON OO fN NO CO P-; ON o ON ON c^ CN NO o H vb 1 CO •O i O o NO 1 CN i 00 NO O v> o t-00 ON O 00 cn 00 oo NO o ON 00 oo OO ON vq co rs H © i r! co NO On OO On ON co (N 00 CO 00 o o ON NO ON NO "1 CN NO ON o cn) 00 o t~CO O 00 O CN NO 00 r*CN o o CN d ON 1 o co NO ci CN CO o >o 00 CN r-~ NO NO o NO NO CO i ON 1 CN o 1 NO 00 Ov (N H r~ NO rON CN r~ON o o V~t NO o NO CN CN o ON ON *o co CN i ON CO fN CN 1 ON I d co 1 ON CN 00 CN 1 CN i © ON CN ON O 00 ro CO 00 OO o v> O CO CN oo < c 3 u. O JB x u s d u § uO is x' d 'S. 'm H i G 'S. '3 § Urn u 1 3 Q LU ft c 3 U. O i 3 o a i f x' rN c 3 U. O — ' 1 >< -J s i 3 Lb 1 13 Vi 0 5 ri d c 3 U. U s 159 f— ii u CO

PAGE 180

160 Table 6-3. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with sensory grades of shrimp samples collected from decomposition workshops c o la Correct Classification Rate (%) mple prepar Sensory scores All Shrimn Species Shrimp Specie Mex. Thai Thai Ecuad. Mex. Diesel GO Brown Pink Tiger White White White n 162 24 36 24 42 42 48 Pass 95.24 100.00 100.00 100.00 100.00 94.44 79.17 i Borderline 10.71 100.00 100.00 75.00 83.33 100.00 83.33 i Fail 74.00 100.00 100.00 100.00 100.00 91.67 91.67 Fail-Diesel 100.00 Overall 74.07 100.00 100.00 95.65 97.61 94.44 85.42 n 129 16 35 42 36 48 •o Pass 63.23 100.00 100.00 83.33 83.33 95.83 u fi. Borderline 52.38 100.00 100.00 100.00 83.33 100.00 Choj Fail 70.00 100.00 100.00 100.00 75.00 100.00 Fail-Diesel 100.00 Overall 63.57 100.00 100.00 90.48 80.56 97.92 n = number of electronic nose readings used to obtain the DFA functions

PAGE 181

161 Mexican brown (intact) Function I Figure 61 . DFA of raw Mexican brown shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops

PAGE 182

162 Figure 6-2 DFA of raw Thai pink shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops

PAGE 183

163 Thai tiger (intact) 3 2 1
PAGE 184

164 Ecuadorian white (chopped) 3 | . , — — , , , -8 -4 0 4 8 12 16 ° Fail Function 1 Figure 6-4. DFA of Ecuadorian white shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops

PAGE 185

165 Mexican white (chopped) -4 -3 -2 -1 0 1 2 3 4 * Fa'' Function 1 Figure 6-5 . DFA of raw Mexican white shrimp odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops

PAGE 186

166 When the data for the different shrimp species were pooled together, the DFA calculated functions (Table 6-1 and 6-2) had correct classification rates of 74.1% and 63.6% for the intact and chopped samples respectively (Table 6-3). Figure 6-7 shows the DFA results for the combined shrimp species data sets. The graph shows a considerable amount of overlap among the three sensory clusters. This meant that a single model for decomposition odors of shrimp (different species) will not give suitable results with this electronic nose. This suggested that different species of shrimp had different odors. These differences in odors could be due to environmental conditions where the shrimp were harvested, microflora present in the shrimp at the time of the analysis, or differences in chemical composition of the shrimp. Therefore, it is recommended that decomposition levels be predicted by using the individual models for each shrimp specie. Moreover, it is recommended to use intact shrimp during electronic nose evaluations, since it eliminates sample preparation time, and in some cases correct classification could be higher. From the decomposition workshop, a sample of Ecuadorian white shrimp contaminated with diesel fuel was obtained. The electronic nose was able to discriminate very easily the contaminated sample (Figure 6-6). In both the intact and chopped sample, the shrimp contaminated with diesel was 100% correctly classified (Table 6-3). However, the overall correct classification rate for the chopped sample (97.9%) was higher than that of the intact (85.4%). This contaminated sample demonstrated the possibility of using the electronic nose to detect contaminants in shrimp, possibly even other chemicals such as sulfites or phosphates. These two chemicals are commonly used in the shrimp industry,

PAGE 187

167 Figure 6-6. DFA of raw Ecuadorian white shrimp odor based on sensory grades and electronic nose readings (including shrimp contaminated with diesel). Samples obtained from FDA workshops

PAGE 188

168 Figure 6-7. DFA of all shrimp species odor based on sensory grades and electronic nose readings. Samples obtained from FDA workshops

PAGE 189

169 900 800 700 I 600 £ > — ' 1 500 2 400 a o mm 300 < 200 100 0 c 1 n= ...J \ n-80 T n=47 Pass Borderline Sensory Grade Fail J_ Mean+SD Mean-SD Mean Figure 6-8. Sensory grades versus ammonia levels of raw shrimp samples (All species pooled together, samples obtained from FDA workshops) and the chemical analyses to detect them are time consuming and require expensive equipment. Ammonia levels were measured for almost every sample that was collected from the FDA decomposition workshops and analyzed in the electronic nose. Due to the restricted amount of samples for some species, ammonia analysis was performed in two or three replicates, the other replicates were chopped, and then analyzed with the electronic nose. There was a positive correlation between the ammonia levels and the degree of decomposition determined by the experts. Figure 6-8 shows the ammonia levels for all shrimp samples obtained from the decomposition workshops. Shrimp samples which were classified as 'pass' had levels lower than 240 ppm, with an average ammonia level of 159 ppm Shrimp samples classified as 'borderline' had an average ammonia level of 287 ppm, with levels between 150 ppm and 410 ppm. Samples that were decomposed or described

PAGE 190

170 as 'fail' had an average of 518 ppm, with values ranging from 290 ppm to 750 ppm. The results from this study correlated well with the data of Cheuk and Finne (1984) and Luzuriaga et al. (1997a) where spoiled shrimp was considered to have ammonia levels above 230 ppm. The second set of experiments, in which samples were stored at refrigeration temperatures (2°C) for 14 days, showed very good correlations between electronic nose sensor readings and sensory panel results. The functions obtained from the DFA are shown in Table 6-4 and the correct classification rates in Table 6-6. When data were analyzed individually by raw and cooked species, the classification rates for electronic nose readings correlated with sensory results were above 95% (Table 6-6). Figures 6-9 to 6-1 1 show the scatter plots obtained with the discriminant functions for each individual shrimp specie. Clusters were well defined and there was good separation between the 'pass' and 'fail' samples. Similar graphs were obtained for both raw or cooked shrimp. When data from all species were pooled together to obtain a general model to predict sensory scores for shrimp samples, regardless of the specie, the classification rate was 84.4% in the raw stage and 89.6% in the cooked stage (Table 6-7). Corroborating the results of the first set of samples from the FDA workshops, the second set also showed differences in odor between species. Data were more scattered in the model, as seen in Figure 6-12. In both raw and cooked shrimp, the clusters for 'pass' and 'fail' shrimp samples were well defined with some overlap between clusters (Figure 6-12). However, the cluster for the 'borderline' samples completely overlapped with the 'pass'

PAGE 191

so a u 99 D g O c o s-*-» M in Const 2.51 0.81 14.65 0.83 -14.03 47.56 -8.71 43.48 4.49 -24.05 -70.48 -10.13 -9.91 -27.35 -13.43 -34.22 oo CN H n rs NO © 1 CN i Os in oo co rs rs co Os •n ©' NO 00 o" CN roo o co 1 o ON >o 1 i NO oo CO sq CO i oo ri i CN rNO CO o co ON 1 rs NO 00 iri in co ON i 00 1 o NO t~ i O CS H so o © co NO CO 00 in rs i CO no i co NO co rs © oo 00 00 o CN O o NO* CN i NO Os CO CN i O co i Os NO NO CO oo NO ON co oo CO H >n ©' co i OS CN ©' i OS rno rs co co ON v~> rs co o o 1 o CN co o CN CS Os co vi CN 00 V"> rs co NO i in rCN rs H ON CN i co o o rs OO «n 1 Os •<* <*' NO 00 ^o NO CO 1 co o to 1 CN CO i v~t NO CN >r> co co NO o n •» oo CO in i CN H 00 OS (N © ©' 1 NO 1 rs OS m NO 00 m O Os rs 00 1 co NO 1 NO 00 r-^ Os co O CO rs VS rs rs' CN «n in o Os >n CN oo CN H NO © ©' 1 s© CN CO 1 O NO ^o OS o co i/^ CO SO i o CO r-' •n NO in CN o OS rs oo rs On H OS *n ©' rs i 00 00 1 in I--; co 1 Os © 1 00 r00 Os OS rs' rs t o OS r> 00 co i o Os 00 "* rs co CN NO rs o m OS Os oo On H oo «n o NO o NO NO OS "*' 1 CO CO Os' o CO 1 rs o rs rs NO fOs CO 00 Os 1 rs 00 1 NO CO CN i * fooo P35JOOO AVI

PAGE 192

i .a o 0 Const -53.89 -44.58 35.34 o CO © 1 26.69 60.09 -24.26 -129.35 2.16 -17.42 42.14 -11.70 oo H On vi ci OO cs co © © NO 00 1 oo vi CN NO CO CN 00 CO 5o 00 VI r~ T CN CO V) O CN On On NO i Tt On O co OO cn CN CN vi oo V) 1 00 i o o o CN V) NO O 9 n H ON On CN CN oo i NO vi i On On © So 00 o ON 1 NO H t~CN i On F; NO CN o co NO 1 On On r-" © 00 •o CN co cn 00 vi CN VI O ON v> On r~t^. V) 1 NO rs H CN O ON co i cn co (N CN v> 00 NO* ro co On VN o -*r CN 1 NO CN i NO On • co i 00 v> NO 1 NO O vi 1 On O i NO 00 CN CN CN VI NO CN O ON H co o co CN v> vi CN o co 1 00 ON H NO TT O CN CN co" CN CN O NO o o CO NO O O I 00 1 00 00 Vi CO CO CO 1 o CN CN O H t-vi On Tt NO co i o rs NO 1 On CN i F~ co i CN On NO CN i On On CO NO On O t"»" ro i 00 vi
PAGE 193

173 Table 6-6. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with sensory grades for different species of shrimp stored at 2°C for 14 days Correct Classification Rate (%) Groups All Shrimp Species Pink Tiger White raw (n=115) cooked (n=115) raw (n=35) cooked (n=35) raw (n=40) cooked (n=40) raw (n=40) cooked (n=40) Pass 81.82 90.91 100.00 100.00 100.00 100.00 95.00 90.00 Borderline 88.00 92.00 100.00 100.00 100.00 100.00 100.00 100.00 Fail 85.71 85.71 100.00 100.00 100.00 100.00 100.00 100.00 Overall 84.35 89.56 100.00 100.00 100.00 100.00 97.50 95.00 n = number of electronic nose readings used to obtain the DFA functions and 'fail' clusters. This is expected, since borderline samples are in transition from fresh to spoiled or from 'pass' to 'fail'. Figure 6-13 showed that each shrimp specie used in this experiment had its own odor. Clusters for each one of the three shrimp species (pink, tiger and white shrimp) were clearly separated. This graph was obtained by using samples at day 0 of storage. The differences in odor at the beginning of the experiment could be due to many factors during growth, harvest, processing and packaging of the shrimp. Therefore, developing DFA models for shrimp decomposition should take into account some history of the samples analyzed, and a considerable amount of shrimp samples must be used to develop a predictive model. Shrimp odors were changing with storage time. Some panelists could not detect differences in odor between consecutive days. However, the electronic nose was able to detect differences between the odor of shrimp at the different days of storage. DFA was

PAGE 194

Figure 6-9. DFA of raw and cooked pink shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 195

175 Figure 6-10. DFA of raw and cooked tiger shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 196

Figure 6-11. DFA of raw and cooked white shrimp odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 197

177 Cooked Shrimp Sensory Score o Pass 0 Borderline -8 -6-4 -2 0 2 4 6° Fail Function 1 Figure 6-12. DFA of different species of shrimp (white, pink, tiger) odor based on sensory grades and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 198

Raw Shrimp Cooked Shrimp -50 -40 -30 -20 -10 Function 1 0 0 o \ on 10 20 30 Shrimp specie ° pink a white « tiger Figure 6-13. DFA of fresh raw and cooked shrimp samples based on electronic nose readings. Day 0 of storage study

PAGE 199

179 Table 6-7. Correct classification rates obtained from the DFA of electronic nose sensor readings compared with storage time for different species of shrimp stored at 2°C for 14 days Correct Classification Rate (%) Storage time Pink Tiger White raw (n=35) cooked (n=35) raw (n=40) cooked (n=40) raw (n=40) cooked (n=40) 0 100.00 100.00 100.00 100.00 100.00 100.00 2 100.00 100.00 100.00 100.00 100.00 100.00 4 100.00 100.00 100.00 100.00 100.00 100.00 6 100.00 100.00 100.00 100.00 100.00 100.00 8 100.00 100.00 100.00 100.00 100.00 100.00 10 100.00 100.00 100.00 100.00 100.00 100.00 12 100.00 100.00 100.00 100.00 100.00 100.00 14 100.00 100.00 100.00 100.00 100.00 100.00 Overall 100.00 100.00 100.00 100.00 100.00 100.00 n number of electronic nose readings used to obtain the DFA functions used to calculate two discriminant functions that explained the correlation between electronic nose readings and storage time for each specie in the raw and in the cooked stage (Table 6-5). The correct classification rates for the discriminant functions were 100% for all species (Table 6-7). Figures 6-14, 6-15 and 6-16 show a clear separation between days of storage for raw and cooked shrimp. The data for the different storage times (days) formed individual clusters with some overlap, meaning that there were distinct differences in the sensor readings for each day of storage. Even though there was 100% classification, the two dimensional plots showed some overlap. However, if the

PAGE 200

180 clusters can be seen in three or more dimensions (up to 7 dimensions, which is days in storage minus one) then the separation will be more apparent. Ammonia levels in raw shrimp increased exponentially during storage (Figure 617). Cooked shrimp had about half of the ammonia levels of raw shrimp. This was caused by ammonia volatilization due to heat, and leaching from the tissue to the water used for cooking. Ammonia levels in raw and cooked shrimp were highly correlated with sensory grades (Figure 6-18). Table 6-8 has the ammonia levels and sensory scores of the samples that were analyzed. Raw samples that were classified as 'pass' had an average ammonia level of 1 12 ppm, 'borderline' samples had 221 ppm and 'fail' had 538 ppm. Cooked samples classified as 'pass' had an average ammonia level of 74 ppm, 'borderline' samples had 1 1 8 ppm and 'fail' had 271 ppm. As with the shrimp samples from the decomposition workshops, the ammonia levels are good indicators of the level of decomposition. Conclusion Decomposition odors in shrimp can be measured objectively and rapidly with the electronic nose technology, giving results similar to those obtained from experienced inspectors. Different species of shrimp differ in odors, which sometimes are difficult to distinguish with the human nose. Further work is needed to develop larger data sets that could be used to predict the sensory grade of shrimp taking into account species, origins, sizes, growing environment (aquaculture or wild), etc. It is expected that including all these variables will improve the correct classification rates. Also, it is recommended to test the transportability of the data from one machine to another. This will help in

PAGE 201

181 Figure 6-14. DFA of raw and cooked pink shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 202

182 Raw Tiger Shrimp Storage lime (days) -16 -12-8 -4 0 4 8 12 * 14 Function 1 Figure 6-15. DFA of raw and cooked tiger shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 203

183 Raw White Shrimp V * /"xT^X ^LaJ ! \ A* / ° / • \ / ° / x* J\ /o / / ° / / ° / A ° "" J' l / / ° V y * -16 -12 -8 -4 0 4 8 12 * 14 Function 1 Cooked White Shrimp -18 -12 -6 0 6 12 18 * 14 Function 1 Figure 6-16. DFA of raw and cooked white shrimp odor based on storage time and electronic nose readings. Samples stored at 2°C for 14 days

PAGE 204

184 Figure 61 7. Ammonia levels of different species of raw and cooked shrimp during storage at 2°C. Lines represent exponential fits

PAGE 205

Raw Shrimp 900 800 0 700 E o. § 600 J | 500 | 400 0 1 300 o I 200 100 0 n = 35 1 i n = 25 n = 55 § X Pass Borderline Sensory Score Fail HI Mean+SD Mean-SD ° Mean Cooked Shrimp 900 800 C 700 s S 600 c 5 | 500 § 400 o § 300 | 200 100 0 n 35 n = 25 c n = 55 : ~5~ i J Pass Borderline Sensory score Fail _l_ Mean+SD Mean-SD D Mean Figure 6-18. Sensory grades versus ammonia levels of raw and cooked shrimp samples stored at 2°C for 14 days

PAGE 206

186 Table 6-8. Ammonia levels (ppm) and sensory scores of different species of raw and cooked shrimp stored at 2°C for 1 4 days Shrimp Species Sample Preparation Days Samples (ammonia level in ppm) Average (ppm) St. dev. Sensory Score 1 2 3 4 5 Pink Raw 0 92.9 118 115 119 126 114.18 12.56 Pass 2 111 120 124 130 111 119.20 8.29 Pass 4 167 148 192 157 197 172.20 21.51 Pass 6 215 234 229 250 275 240.60 22.94 Borderline 8 295 280 310 305 340 306.00 22.19 Fail 10 779 742 745 710 583 711.80 76.03 Fail 12 1230 1030 1020 1100 1120 1100.00 84.56 Fail Cooked 0 64.7 72 81.2 99.7 90.5 81.62 14.01 Pass 2 88 93 100 85 82 89.60 7.09 Pass 4 98 127 124 114 117 116.00 11.34 Pass 6 174 149 143 154 199 163.80 22.86 Borderline 8 196 169 200 220 199 196.80 18.21 Fail 10 250 360 297 339 347 318.60 45.03 Fail 12 505 537 615 675 598 586.00 66.84 Fail Tiger Raw 0 49.8 56.5 59.6 49.7 51.8 53.48 4.39 Pass 2 62 73 72.8 84 73 72.96 7.78 Pass 4 110 115 113 114 104 111.20 4.44 Pass 6 149 153 155 189 144 158.00 17.83 Pass 8 203 223 171 195 170 192.40 22.45 Borderline 10 185 205 191 244 240 213.00 27.49 Borderline 12 312 265 300 282 327 297.20 24.41 Fail 14 480 538 448 399 417 456.40 55.08 Fail Cooked 0 43 45.4 42 37.5 40.2 41.62 2.97 Pass 2 49 43.8 44.2 47.5 56.5 48.20 5.13 Pass 4 56.3 48.8 68.4 75.6 63.6 62.54 10.41 Pass 6 81.9 77 61.5 56 67 68.68 10.71 Pass 8 72.1 90 60 77.9 75.9 75.18 10.81 Borderline 10 87.9 113 120 121 110 110.38 13.39 Borderline 12 153 147 163 167 173 160.60 10.53 Fail 14 192 208 173 192 208 194.60 14.48 Fail White Raw 0 63.4 62.1 86.2 66.5 71.4 69.92 9.78 Pass 2 75 84.1 99.6 106 108 94.54 14.40 Pass 4 125 106 98.9 106 106 108.38 9.79 Pass 6 136 174 163 170 171 162.80 15.51 Pass 8 181 177 200 176 192 185.20 10.43 Borderline 10 284 212 285 238 368 277.40 59.45 Borderline 12 478 577 401 345 472 454.60 87.63 Fail 14 385 423 459 448 499 442.80 42.37 Fail Cooked 0 56.7 61.5 62.8 67.1 50.9 59.80 6.20 Pass 2 67.9 72.8 76.4 70.7 50.6 67.68 10.04 Pass 4 76.5 72.8 98.1 75.6 73.9 79.38 10.56 Pass 6 105 112 90.3 94.8 107 101.82 8.98 Pass 8 115 165 128 133 110 130.20 21.58 Borderline 10 112 116 103 117 100 109.60 7.70 Borderline 12 191 224 170 180 192 191.40 20.32 Fail 14 198 .,,287 216 254 280 247.00 39.05

PAGE 207

187 implementing this technology in industrial and regulatory settings, and applications where fast and reliable measurements are required. In this study, only DFA was used as the pattern recognition technique, however other techniques could be used to obtain better classifications. Artificial neural networks could probably take into account species, sample history and sensor readings to obtain a model for shrimp decomposition. The electronic nose has very similar architecture and properties compared to the biological olfaction system Properties such as odor delivery, nonspecific sensor/receptor response, and sensor/receptor preprocessing work in a similar manner in both systems (Pearce, 1997a). Additionally, both systems share similar limiting factors such as degeneration and poisoning, sensor/receptor drifts, and limited sensor/receptor sensitivity (Pearce, 1997b). These issues have to be addressed before the electronic nose can be implemented in an industrial setting to evaluate shrimp for its decomposition level.

PAGE 208

CHAPTER 7 USE OF THE ELECTRONIC NOSE TO DETECT CHEMICALS USED IN SHRIMP PROCESSING Introduction Shrimp has become an important commodity in the United States, where its annual per capita consumption (all preparations) has increased in the last 20 years from 0.73 to 1.23 Kg (U.S. Department of Commerce, 1998). Unfrozen raw shrimp has a short shelf life. During processing and commercialization, shrimp is subject to enzymatic actions that causes visual defects. It is prone to microbial deterioration. The tissue is subject to the adverse effects of ice during freeze-thaw cycles, because most shrimp are shipped in the frozen stage. Therefore, shrimp processors try to maintain the quality of their product by using different chemicals. Shrimp melanosis, commonly termed black spot, is a surface discoloration due to enzymatic browning that occurs postmortem. The endogenous shrimp enzyme, polyphenol oxidase, catalyzes the initial step in black spot formation (McEvily et al., 1991). Sulfiting agents have been used in shrimp since the 1950's to inhibit melanosis formation (Fieger, 1951). Currently, sulfites are mainly employed on commercial vessels during transport and handling of shrimp. However, some shrimp aquaculture facilities have also adopted the use of sulfites to treat pond harvests. Adverse reactions to sulfites 188

PAGE 209

189 with particular concern to asthmatics are well known (Taylor et al., 1986). Therefore, detection of sulfites in shrimp is important from a food safety point of view. A common analytical method for sulfite detection is the acidic distillation procedure known as MonierWilliams. This procedure is time consuming and laborious. Therefore, as reported by Taylor et al. (1986), other methods have been developed but have been found to have inaccuracies or limited uses. Recently, a rapid method that utilizes malachite green was found to be useful to detect the presence of sulfites in shrimp (Forbes, 1997). Phosphates are frequently used as processing aids or additives in a variety of foods. They are used to increase water binding capacity, improve emulsification and buffering capacity, and bind metal ions (Lindsay, 1985). In seafood products, their most common and controversial use is in frozen products. They dramatically reduce thaw drip, and when used properly, the retention of moisture improves texture and flavor because flavor components are not lost in the thaw drip (Finne, 1995). This is also true for cooked seafood products, therefore processors treat shrimp prior to cooking and freezing. Use of phosphates can be easily abused, leading to excessive water weight increases in raw seafood. Detection of phosphates in shrimp is not easy. Shrimp has naturally occurring levels of phosphates, which vary according to the species and harvest location. Besides, phosphates have strong interactions with the protein structure, which makes them difficult to quantitatively extract using non-destructive solvent systems. Moreover, they can be transformed to other forms (orthophosphates) making them difficult to quantify (Finne, 1995). Therefore, alternative or indirect methods need to be developed to determine if shrimp were treated with phosphates.

PAGE 210

190 Another chemical that is widely used in the food industry is bleach. Bleach is one of the most cost effective sanitizers available in the market. Several cases have been observed where shrimp producers used bleach solutions to treat decomposed shrimp. Bleach will act as an antimicrobial and will mask the odors of decomposition. Depending on the concentration of chlorine some residual odor can be detected. Therefore, sensors that can interact with chlorine can be used to find out whether shrimp has been washed with bleach. Newly developed sensor arrays, also known as electronic noses, have wide applications in the food industry as well as in other industries. The electronic nose has been used to detect the presence of a variety of chemicals in different products, such as detection of some additives used in the sparkling wine process (Viaux et al., 1996), detection of meat adulteration (Turhan et al., 1998), determination of organic compounds in contaminated soils (Getino et al., 1998), monitoring of the space shuttle air for selected contaminants (Ryan et al., 1998), detection of adulteration of peppermint oils with cheaper ingredients (Hanrieder, et al., 1998), among other studies. Shrimp processors do not have the resources, nor the time, to run complicated and laborious analytical tests. Therefore, the electronic nose can be an alternative for detection of these chemicals with the advantages of no sample preparation, no use of chemicals, fast results and ease of use of the equipment. The overall objective of this study was to determine the ability of an electronic nose and sensory panels to detect if shell-on pink shrimp was treated with sodium hypochlorite (bleach), sodium tripolyphosphate (phosphates) or sodium metabisulfite

PAGE 211

191 (sulfites). The specific objectives were: 1) to treat shrimp with different levels of these chemicals and measure the electronic nose sensor response at 0, 24 and 48 hrs after treatment, 2) to conduct an odor sensory panel to determine it panelists can detect the presence of these chemicals in treated shrimp, and 3) to measure microbial loads, ammonia levels, moisture content, water activity and pH of treated shrimp during the 48 hrs to understand changes in shrimp odor other than those due to the treatment with the chemicals. Materials and Methods Shrimp Samples Three batches (8.8 Kg each) of frozen pink shrimp ( Penaeus duorarum) were purchased from Lombardi's Seafood (Orlando, FL). Each batch was split in half (4.4 Kg), to replicate the study. The first batch of headless shell-on pink shrimp (55/66 count/Kg) was treated with bleach solutions. The second batch was headless shell-on pink shrimp (79/88 count/Kg), which was treated with phosphate solutions. The third batch of headless shell-on pink (44/55 count/Kg) was treated with sulfite solutions. Samples were thawed under running tap water. Three Kg of shrimp were taken at random from each half of the batch, treated with the different solutions, and stored at refrigeration temperature (2°C) for 48 hours. Samples were evaluated every 24 hours for differences in odor by sensory panelists and an electronic nose with 12 sensors. The study was repeated immediately after finishing the first series of experiments.

PAGE 212

192 Chemicals Used and Sample Treatments Bleach solutions (25, 50, 100 and 200 ppm) were prepared with distilled water from a concentrated solution of sodium hypochlorite 5.25% w/v. Shrimp were dipped in the solutions for 10 minutes. Enough solution was used to completely cover the shrimp, which gave approximately a 2:1 ratio of bleach solution : shrimp. Samples were drained by placing them in a strainer for 3 minutes. They were then placed in 1 gallon freezer Ziploc bags and stored in a refrigerator at 2.0°C. Phopshate solutions (2, 4 and 6% w/v) were prepared from sodium tripolyphosphate (85%, Cat. No. 21867-0025, Lot: A010959901, Acros Organics, NJ) and distilled water. Solutions were prepared the day before and stored in sealed glass volumetric flasks at 2°C. Shrimp were dipped in cold phosphate solution for one hour. Phosphate solutions were cold to prevent microbial proliferation in the shrimp samples. Samples were drained in a strainer for 3 minutes, placed in 1 gallon freezer Ziploc bags and stored at 2°C. Currently, shrimp processors use 2% and 4% phosphate solution dips to treat shrimp. Sulfite solutions (0.75, 1.25 and 2% w/v) were prepared from sodium metabisulfite (reagent ACS 97%, Cat. No. 41958-0010, Lot. A008455701, Acros Organics, NJ) and distilled water. Shrimp were dipped in the solutions for one minute and then drained in a strainer for 3 minutes. Shrimp were placed in 1 gallon freezer Ziploc bags and stored at 2°C. Present regulations for the treatment of shrimp are 1 min dip into a 1 .25% sodium metabisulfite solution (Federal Register, 1982).

PAGE 213

193 Electronic Nose Measurements An electronic nose (e-NOSE 4000 model, Neotronics Inc., Flowery Branch, GA) equipped with twelve conducting polymer sensors (sensor types: 483, 478, 464, 463, 462, 461, 460, 459, 458, 401, 298 and 297) was used to quantify the sensor responses to differences in odor of shrimp samples that were treated with different levels of chemicals. Electronic nose measurements were done immediately (0 hr), and 24 hrs and 48 hrs after treatment. Samples were stored at 2°C. Five replicates were analyzed by the electronic nose for each treatment. Replicates were taken at random from the plastic bags where shrimp were stored. Each replicate was approximately 50 g (4 headless shell-on shrimp). Each replicate was taken out of the cooler 50 min prior to the analysis, to let the shrimp temperature equilibrate to room temperature (22.5 to 23.5°C). The sample was put in a 140-ml beaker, and placed in the sampling vessel of the electronic nose. The day the experiment started, the electronic nose was calibrated following the manufacturer's recommendation using a 75% v/v propylene glycol solution (100% solution from Fisher Scientific, No. P-355-20, Fair Lawn, NJ). Every day prior to the experiments the electronic nose was turned on and compressed air (CGA Grade D, Strate Welding Supply Inc, Jacksonville, FL) was passed through the sensors for 2 hrs. For each replicate the vessel was purged with compressed air for 2 min to eliminate any extraneous odor. Then the sensor head was purged for 4 min with compressed air. During these 4 min, the sample volatiles were equilibrating in the headspace of the vessel. Sensor response data was acquired for 4 min. Analysis time for each sample took 10 min. Readings at 4 min

PAGE 214

194 exposure of the sensors to the samples were used for data analysis. At the end of the day, electronic nose sensors were cleaned with compressed air for 2 hrs. Electronic nose sensor data can be obtained from Dr. Murat Balaban at the Food Science and Human Nutrition Dept., University of Florida (filename: "\\thesis\text files\Chap7\chemicals enose.txt"). Sensory Evaluation The odor of raw treated shrimp was evaluated by a 16-member untrained sensory panel consisting of students, 22 35 years of age, from the Food Science and Human Nutrition Department at the University of Florida. However, results from only 12 panelists were used, because the other 4 panelists did not participate in every evaluation. A "difference form control" test was performed every day during the three days of the study. Panelists were asked to smell shrimp samples and detect if there was any difference in odor among the treated samples and the control. The control was untreated shrimp. Panelists measured the differences in a 100 mm scale (0 mm = no difference, 100 mm = very different). Samples were randomized and a hidden control was included in the test. At each sampling day, a random sample was taken out of the refrigerator 30 min. before the sensory analysis. Approximately 50 g of shrimp were placed in an opaque plastic cup and covered with a lid (125 mL Praire Packaging S-400, Bedford Park, IL). All panelists smelled the same samples. Sensory tests were carried out in both replicates. Samples used for sensory analysis were not the same samples measured by the electronic nose, but were selected at random from the same bag of shrimp.

PAGE 215

195 Ammonia Analysis Ammonia levels in the shrimp samples were measured with an ion-selective electrode (Model 95-12, Orion Research, Inc., Boston, MA) connected to an Expanded Ion Analyzer (EA 920, Orion Research Inc., Beverly, MA). Two samples were measured for ammonia immediately after the electronic nose evaluation. The 60-g replicates were sprayed with 4 ml of 5N NaOH (Cat. No. LC24450-4, Lab Chem Inc., Pittsburgh, PA) and analyzed for ammonia in the headspace of an air-tight box, following the procedure described by Luzuriaga et al. (1997a). The ammonia electrode was calibrated prior to, and during the experiments using 10 ppm, 100 ppm and 1000 ppm ammonia solutions (Orion Cat No. 951007, Orion Research, Inc., Beverly, MA). Ammonia levels in shrimp were reported as ppm. Ammonia was only measured for the phosphate and sulfite treated shrimp (Appendix C). Moisture Content and Water Activity Measurements Moisture content was measured in triplicate at days 1 and 3 (first and last day of the experiment) using the oven method (AO AC, #24.003, 1980). A 50-g sample of shell-on shrimp (approximately 4 shrimp) was chopped in a mincer chopper ("HandyChopper", model HC20 , Black & Decker, Shelton, CT ). A sample of approximately 5 g was placed in an aluminum weighing dish (50 mm diameter, Cat. No. 08-732, Fisher Scientific, Fair Lawn, NJ). The sample was placed in an oven at 104°C for 24 hrs. Moisture content was reported as percent wet basis (Appendix C).

PAGE 216

196 Water activity (aj was measured using a Rotronic Hygroscop DT (Rotronic, Huntington, NY). A 5-g piece of the chopped shrimp was placed in a plastic cup provided by Rotronic and placed in the a„, meter. One measurement per sample was obtained after approximately 45 to 50 minutes when equilibration was achieved. The temperature at which a„ was measured was 24.5° ± 0.5°C. pH Measurements A 20-g sample of shrimp was placed in a blender with 80 ml of distilled water. The sample was blended for 15 seconds, transferred to a 140-ml beaker and placed on a stirrer plate. The pH electrode (ROSS pH electrode, Model 81-02, Orion Research Inc., Beverly, MA) was connected to an Expanded Ion Analyzer, and was calibrated every day with pH 4.00 and 7.00 standards (Buffer solution pH 4.00, SB 101 -500 and pH 7.00, SB 107-500, Fisher Scientific, Fair Lawn, NJ). Measurements were done in duplicate (Appendix C). Microbial Analysis Aerobic plate counts were performed daily on all shrimp samples using aerobic plate count Petri film (3M Company, St. Paul MN). Dilutions were made using pre-filled sterile disposable diluent bottles of phosphate buffer (NutraMax Products, Inc., Gloucester, MA). A 20-g sample of shrimp was placed in a sterile blender with 1 80 ml of phosphate buffer. The sample was blended for 15 seconds. Aliquots from the blender were taken to the dilution bottles. From each dilution 1 ml was taken and plated in the

PAGE 217

197 Petri films, which were incubated at 32°C for 48 hr. Colonies were counted and reported as log cfu/g of shrimp. Data Analysis Sensor readings were analyzed in Statistica for Windows ('98 edition, StatSoft Inc., Tulsa, OK) using discriminant function analysis (DFA) as reported by other researchers (Corcoran, 1993; Gardner and Hines, 1997; Gardner and Bartlett, 1992). DFA was used to construct predictive functions to help in classifying data based on the concentration of the chemical used. The 12 sensor outputs were reduced to 2 discriminant functions. These functions were used to map the data in two dimensional plots and observe separation between groups. Correct classification rates and the coefficients for each function were calculated. Moisture content, microbial load, ammonia level and sensory data were analyzed in Statistica for Windows using one-way analysis of variance (ANOVA). ANOVA's were calculated for every chemical, at every time step (0, 24 and 48 hrs) and for each replicate. Comparison of means was done using the LSD (least significance difference) test. Results and Discussion Results from these experiments showed that shrimp stored for 48 hrs had some changes in their physical, chemical and microbiological properties. These properties may help understand how these changes could affect the odor of the shrimp, and therefore interpret the response of the electronic nose sensors. Conducting polymer sensors are

PAGE 218

198 sensitive to moisture (Bartlett et al., 1997; Hodgins, 1997; Shurmer, 1990), therefore moisture content and water activity of treated shrimp were measured immediately after treatment and after 48 hrs of storage. Data shown in Table 7-1 and Appendices C-l to C3 showed that moisture content of shrimp did not change throughout the 48 hrs of storage for each batch of shrimp. Moreover, moisture content change due to the different levels of chemical treatment was negligible. However, there were some differences in the moisture content of the three different batches. The batch treated with bleach had a lower average moisture content (76.3% ± 0.77 (std. deviation)), while the other two were higher. The one treated with sulfites had 79. 1% ± 0.90 moisture content, while the one treated with phosphate had 79.9% ± 0.67. Water activity is related to the partial pressure of water over the food, and therefore related to the amount of water vapor that will go in the gas phase. The water activity for treated shrimp (Table 7-2) shows minimal changes throughout storage and within treatments. Even though time constraints dictated only one water activity measurement per treatment, the data seem to have very little variation within each treatment. However, water activity for the three different batches of shrimp were slightly different. On average, the batch of shrimp treated with sulfites had the lowest water activity (0.988), bleach had 0.990, while the one treated with phosphates had a higher water activity (0.993). Differences in moisture content and water activity are minimum, therefore it is expected that the difference in electronic nose sensors profiles will be due to volatile components present in the sample, rather than to water vapor.

PAGE 219

199 The microflora present in the shrimp proliferated during storage, as expected (Table 7-3, Appendices C-4 to C-6). In general the microbial count increased by two to three log cycles during the 48 hrs of storage. Analysis of variance was performed for each chemical, on each analysis time (0, 24 and 48 hrs) and for each replicate to see if there was any significant difference in microbial counts due to the level of the chemical used for treatment. It was expected that bleach would have an effect on the bacterial load of shrimp. However, results were not significant (Table 7-3), meaning that on average the non treated shrimp had the same microbial load as shrimp treated with bleach. These results could be due to a wide variation of microbial loads present in the shrimp at the beginning of the experiment. In the case of phosphates, at 0 and 48 hrs there was no significant differences in the microbial loads due to the treatment with phosphate solutions. However, in both replicates at 24 hrs after treatment, the microbial load of the non treated shrimp was significantly higher than that of the treated shrimp (Table 7-3). Even though phosphate solutions are reported to have an antimicrobial effect (Lindsay, 1985; Finne, 1995), results from this experiment did not corroborate that. When shrimp were treated with sulfites, there were some changes in microbial counts. Immediately after treatment, shrimp treated with 1.25 and 2.0% sulfite solutions had lower microbial loads than the 0.75% treatment and the non treated shrimp. After 24 hrs of treatment there was no significant difference in the microbial counts due to sulfite treatment. However, after 48 hrs, non treated shrimp had significantly higher microbial counts than the treated shrimp.

PAGE 220

200 Table 7-1 . Moisture content of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite (average of three samples) Chemical concentration Moisture content (% wet basis) Replicate 1 Replicate 2 yj in o 48 hr<: HO 111 O ft hr« 4R hr«j Ho III s Bleach (ppm) 0 A vex 75 41 77 D1 / / ,\j i 75 76 76 54 St. dev. ft 8? ft 59 ft ftQ ft 25 Ave 75 65 1ft 0 'ft 76 ft8 75 57 St. dev. 1 Of* 1 . \J\J ft 16 ft 56 U.JO 50 Ave .rvvg. 75 95 76 16 77 1ft 76 71 St. dev. 0 17 ft 51 ft 11 ft 44 100 Ave 75 75 76 7 ft 76 1? 76 49 St. dev. ft 79 ft 47 ft 11 200 Ave. '"6' 11 96 77 66 75 ft7 St. dev. 0.23 0 06 0 42 ft 1ft Phosphate (%) 0 Avg. 80.26 80.10 79.78 80.74 St. dev. 0.27 0.40 0.29 0.27 2.0 Avg. 79.23 80.48 80.54 79.46 St. dev. 0.27 0.25 0.48 0.32 4.0 Avg. 78.64 80.14 80.00 79.62 St. dev. 0.48 0.23 0.70 0.18 6.0 Avg. 79.60 79.33 79.17 80.57 St. dev. 0.20 0.19 0.73 0.35 Sulfite (%) 0 Avg. 79.32 79.27 80.29 79.12 St. dev. 0.96 0.61 0.37 0.55 0.75 Avg. 80.49 78.48 79.24 78.82 St. dev. 0.12 0.58 1.13 0.63 1.25 Avg. 78.94 78.12 77.68 79.51 St. dev. 0.31 0.67 0.44 0.36 2.0 Avg. 79.87 79.37 79.39 78.36 St. dev. 0.29 1.25 0.48 0.30

PAGE 221

201 Table 7-2. Water activity (% relative humidity) of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite Water activity (% RH)* Chemical concentration Replicate 1 Replicate 2 Ohrs 48hrs Ohrs 48 hrs 0 98.9 99.1 99.0 99.0 Bleach (ppm) 25 99.1 99.1 98.9 99.1 50 99.0 99.1 99.0 99.2 100 98.7 99.2 98.9 99.2 200 99.0 99.0 98.9 99.2 0 99.2 99.4 99.3 99.4 Phosphates 2.0 99.3 99.4 99.2 99.3 (%) 4.0 99.3 99.4 99.2 99.4 6.0 99.3 99.4 99.4 99.3 0 98.9 98.8 98.8 98.7 Sulfites 0.75 98.7 98.9 98.8 98.8 (%) 1.25 98.8 98.8 98.9 98.9 2.0 98.9 98.7 98.9 98.8 * measurements were done at 24.5° ± 0.5°C.

PAGE 222

202 Table 7-3. Microbial load of shell-on pink shrimp treated with different levels of bleach, phosphate and sulfite (average of two samples) Chemical concentration Microbial Load ( log cfu/g of shrimp) Replicate 1 Replicate 2 Ohrs 24 hrs 48 hrs Ohrs 24 hrs 48 hrs Bleach (ppm) r\ U Avg. 4.323 5.782 8.708 4.886 5.104 7.695 St. dev. 0.033 0.066 0.048 0.071 0.005 0.069 £j Avg. 4.301 5.063 8.724 4.908 5.756 7.748 St. dev. 0.034 0.008 0.058 0.008 0.054 0.111 Avg. 4.358 4.842 7.989 5.116 5.255 7.919 St. dev. 0.008 0.004 0.016 0.007 0.027 0.007 i on 1UU Avg. 4.212 5.186 8.607 4.796 4.916 7.021 St. dev. 0.186 0.042 0.023 0.035 0.025 0.149 700 Avg. 4.051 4.217 7.848 4.796 5.354 7.550 St. dev. 0.019 0.056 0.092 0.064 0.030 0.060 Phosphate (%) 0 Avg. 5.817 6.900 a 7.138 6.185 6.810 s 6.699 St. dev. 0.026 0.011 0.139 0.012 0.015 0.025 2.0 Avg. 5.274 5.591 b 5.978 5.651 6.233" 6.954 St. dev. 0.042 0.112 0.165 0.180 0.032 0.048 4.0 Avg. 5.275 5.922 b 6.906 5.428 5.957" 6.000 St. dev. 0.011 0.011 0.004 0.009 0.004 0.260 6.0 Avg. 5.506 5.470" 6.204 5.943 5.836" 6.389 St. dev. 0.018 0.031 0.038 0.030 0.040 0.114 Sulfite (%) 0 Avg. 5.92T 7.200 7.760* 5.947 a 7.259 7.771 a St. dev. 0.016 0.057 0.136 0.023 0.012 0.021 0.75 Avg. 5.884 ab 7.148 7.278 b 5.890" 7.207 7.411" St. dev. 0.020 0.042 0.021 0.018 0.034 0.016 1.25 Avg. 5.824 bc 7.188 7.134" 5.817" 6.942 7.195" St. dev. 0.012 0.036 0.005 0.030 0.038 0.030 2.0 Avg. 5.773 c 6.895 6.929" 5.780" 6.803 6.952" St. dev. 0.070 0.019 0.058 0.013 0.112 0.079 Note: superscripts denote significant difference at the p<0.05

PAGE 223

203 The increase in microbial population may cause major changes in the odor of the shrimp. Bacteria breaks down protein and other nitrogenous compounds into volatile bases such as ammonia, trimethylamine, dimethylamine, etc. These volatile bases will give a putrid or spoiled odor. Therefore, electronic nose data from 0, 24 or 48 hrs cannot be compared together, because differences in odor will be caused by microbial action. Moreover, data from each replicate will also have differences due to fluctuations in bacterial loads. Based on these findings, all the analysis of electronic nose data and sensory data will be performed for each treatment, at each analysis time and for each replicate. pH of shrimp slightly increased throughout storage (Table 7-4, Appendices C-7 to C-9) from 7.3 to 7.6 for all treatments. There were some differences in the pH among the different levels of chemicals used within each replicate. However, changes were rninimum, and did not follow any specific trend. Therefore, it is expected that variations in pH at any given sampling time were due to natural variation of the shrimp tissue. Changes in pH were not as large as microbial loads. When pH of shrimp becomes alkaline, volatile amines start to volatilize more easily, and that is what is perceived as the fishy or putrid odors of decomposed shrimp. In this case, pH had very little effect on the rate of volatilization of these amine compounds. Ammonia levels were highly correlated with the microbial loads. As microbial loads increased, the ammonia levels also increased (Table 7-5). Appendices C10 and

PAGE 224

204 Table 7-4. pH of shell-on pink shrimp treated with different levels of bleach, phosphate, and sulfite (average of two samples) Chemical concentration PH Replicate 1 Replicate 2 Ohrs 24 hrs 48 hrs Ohrs 24 hrs 48 hrs Bleach (ppm) Avg. 7.46 7.39 7.61 7.51 7.50 7.70 St. dev. 0.01 0.04 0.04 0.01 0.07 0.03 Avg. 7.26 7.67 7.53 7.53 7.52 7.74 St. dev. 0.02 0.02 0.02 0.07 0.04 0.01 SO Avg. 7.27 7.53 7.51 7.49 7.53 7.69 St. dev. 0.01 0.03 0.01 0.04 0.01 0.03 1 \J\J Avg. 7.24 7.51 7.72 7.44 7.55 7.78 St. dev. 0.01 0.03 0.03 0.02 0.01 0.05 200 Avg. 7.33 7.37 7.72 7.30 7.40 7.69 St. dev. 0.04 0.04 0.02 0.01 0.02 0.06 Phosphate (%) 0 Avg. 7.16 7.50 7.41 7.41 7.35 7.44 St. dev. 0.06 0.01 0.02 0.02 0.01 0.01 2.0 Avg. 7.29 7.50 7.48 7.48 7.51 7.41 St. dev. 0.01 0.02 0.04 0.04 0.02 0.02 4.0 Avg. 7.21 7.40 7.46 7.46 7.50 7.44 St. dev. 0.01 0.01 0.01 0.01 0.01 0.02 6.0 Avg. 7.40 7.48 7.44 7.44 7.51 7.48 St. dev. 0.01 0.02 0.04 0.04 0.03 0.04 Sulfite (%) 0 Avg. 7.51 7.67 7.51 7.40 7.53 7.64 St. dev. 0.05 0.02 0.02 0.08 0.01 0.01 0.75 Avg. 7.57 7.62 7.46 7.48 7.65 7.68 St. dev. 0.03 0.01 0.04 0.01 0.02 0.01 1.25 Avg. 7.46 7.60 7.67 7.37 7.57 7.63 St. dev. 0.01 0.01 0.01 0.08 0.01 0.01 2.0 Avg. 7.47 7.55 7.48 7.39 7.55 7.59 St. dev. 0.01 0.01 0.02 0.04 0.01 0.02

PAGE 225

205 Table 7-5. Ammonia levels of shell-on pink shrimp treated with different levels of phosphate and sulfite (average of two samples) Chemical concentration Ammonia (ppm) Replicate Replicate 2 u nrs Z4 nrs AO U r „ h-o nrs u nrs 7/1 Urc* z4 nrs 4o nrs Phosphate \ /a ) 0 A \TCt /\vg. 48 ^n OO. JU c 1 ^n ^4 ^n J4. jU 74 nn /'f.UU 7o
PAGE 226

two samples. In some cases they were able to detect differences in odor of the highest levels of chemical treatment when compared to the control or untreated shrimp (Tables 76 to 7-8). In the case of bleach treated shrimp (Table 7-6 and Table C-12), immediately after treatment, panelists could detect differences in odor for the samples treated with 100 and 200 ppm bleach. After 24 hrs they could not detect major differences, but after 48 hrs the sample treated with 200 ppm of bleach had a significantly different odor from the untreated shrimp, which could be related to the lower microbial load in those samples, compared to the others. Table 7-6. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of bleach. Superscripts denote statistical Bleach concentration (ppm) Ohrs 24 hrs 48 hrs Rep 1 Rep 2 Rep 1 Rep 2 Rep 1 Rep 2 0 12.00" 12.75 a 14.42 16.25 s 22.92 s " 12.08 s 25 19.1 7* 8.92 a 12.67 19.67 3 " 15.58 s " 12.50 s 50 11.00 a 15.17 a 24.83 17.50 a 27.42"° 18.83 s 100 31.67" 47.17" 10.75 22.42 s " 13.75 a 19.67 s 200 49.92 c 48.00" 22.58 32.50" 38.67 c 40.33" Table 7-7. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of phosphate. Superscripts denote statistical Phosphate concentration (%) Ohrs 24 hrs 48 hrs Rep 1 Rep 2 Rep 1 Rep 2 Rep 1 Rep 2 0 7.17 a 5.83 s 8.42 7.92 s 8.42 s 7.08 a 2 9.67 a 6.92 s 11.67 7.25 s 14.33 s " 12.00 a 4 10.42 s 10.50 s 14.42 9.25 s 14.50"" 11.66 a 6 16.83" 16.42" 14.75 18.17" 16.08" 19.25"

PAGE 227

207 Table 7-8. Average sensory score given by the 12 panelists to shell-on pink shrimp treated with different levels of sulfite. Superscripts denote statistical significance among the means at the p-level of 0.05 Sulfite concentration (%) Ohrs 24hrs 48 hrs Rep 1 Rep 2 Rep 1 Rep 2 Rep 1 Rep 2 0 22.25 a 17.42 27.58 ab 29.67 18.92 7.00 a 0.75 25.17 3 37.42 18.42" 23.08 13.92 14.17 a 1.25 29.00 a 31.25 32.75 ab 32.42 15.42 14.75 a 2.0 52.17 b 35.50 36.75" 22.92 16.50 28.00 b Sensory results for shrimp treated with phosphate were somehow unexpected. Since sodium triporyphosphate is not a volatile compound, it was anticipated that panelists were not going to be able to detect differences from the untreated sample. However, in both replicates immediately after treatment, panelists detected differences in odor of the 6% phosphate treated shrimp compared to the untreated one (Table 7-7 and Table C-13). At 24 hrs and 48 hrs, only replicate two data had significantly different odors in the highest concentration. In general sensory data for phosphate treated samples had lower scores compared to bleach or sulfite treated shrimp. Lower scores could mean that odors from phosphate treated shrimp are closer to the odor of untreated shrimp. Panelists could detect some differences in sulfite treated shrimp (Table 7-8 and Table C-14). In replicate one, at time 0 hrs, the 2% sulfite treated shrimp had significantly different odor than the other samples. Similar results were observed in replicate two, but at 48 hrs of storage. All the other samples and storage times did not show major significant differences in the odor of treated shrimp versus that of the control.

PAGE 228

208 Electronic nose sensor data showed very good classification results for shrimp treated with different chemicals. Based on the results from physical, chemical and microbial data, it was concluded that electronic nose sensor data should be analyzed separately for every analysis time (0, 24 and 48 hrs) and for each replicate. Replicates for each analysis time were also combined to observe the degree of classification on pooled data. As demonstrated from previous results, shrimp properties were changing during storage. Some of these changes could influence the odor of shrimp, which in fact was demonstrated by the DFA on electronic nose sensor data. Table 7-9 and Figure 7-1 show the classification of sensor data for each analysis time. In all three chemicals, the classification of sensor data in one of the 3 analysis time categories (0, 24 and 48hrs) was very good. The overall correct classification rates for bleach, phosphate and sulfite treated shrimp were 92.7, 95.8 and 99.2%, respectively. This shows that the electronic nose was able to sense differences in odor at 0, 24 and 48 hrs. Since data were pooled together for both replicates, it can be concluded that shrimp odors in both replicates were similar, otherwise data would have been more dispersed and with lower classification rates. In Figure 7-1, each cluster is made up of the untreated shrimp plus the shrimp treated with the different levels of each chemical for both replicates. For all three chemicals, sensor data for shrimp samples analyzed immediately after treatment had 100% correct classification rates. However, as time passed, the 24 and 48 hrs samples had lower classification rates. This could be due to the different rates of spoilage in each treatment and consequently their effect on the odor, making it more complicated to differentiate between a 24 hrs or a 48 hrs sample.

PAGE 229

209 Bleach Rep 1 & 2 Time after treatment o Ohrs a 24 hrs o 48 hrs Phosphates Rep 1 & 2 Sulfites Rep I & 2 o Time after treatment ° Ohrs ° 24 hrs 48 hrs -2 0 2 4 6 8 10 12 Function 1 Figure 7-1. DFA results of the correlation of electronic nose readings and storage time at 2°C of shrimp treated with different chemicals

PAGE 230

210 Table 7-9. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by storage time (both replicates together). DFA Time after treatment (hrs) Bleach Phosphate Sulfite 0 100 100 100 24 98 92.5 100 48 80 95 97.5 Overall 92.7 95.8 99.2 The DFA for each chemical, at each time step and for each replicate showed high classification rates when sensor data were classified by the concentration of chemical used to treat shrimp (Tables 7-10, 7-1 1 and 7-12). Shrimp that was treated with bleach had classification rates ranging from 92 to 100%. Phosphate treated shrimp had classification rates of 95 to 100%, while sulfite treated shrimp ranged from 90 to 100%. These results demonstrate that DFA of sensor data is able to distinguish differences in shrimp treated with different levels of these three chemicals. However, when results for both replicates were pooled together, classification rates were not as good. The classification rates for combined replicates ranged from 57.5 to 80% (Tables 7-10, 7-1 1 and 7-12), demonstrating that the electronic nose sensors perceived slight differences between replicates in all three treatments (bleach, phosphate and sulfite). For each chemical, both replicates came from the same batch of shrimp, but from different 2.2 Kg blocks. Replicate two remained frozen for a few days longer, but once thawed the process was exactly the same as the one for replicate one. Therefore, differences in odor could be due to the fact that samples came from different packages, or it could also be due to sensor

PAGE 231

211 drift, however three days is a short period of time for that to occur. Data for replicate one and two for each storage time were further analyzed with Principal Component Analysis (PCA). This method was used to observe if data can be separated into different groups. Results showed that factor one had a very high eigenvalue compared to factor two (Table 7-13). This meant that PCA could not really detect the presence of two distinct groups in the data. Therefore, the low classification rates obtained in the pooled data are due to the way DFA tries to maximize the distance in space of the grouping factors, discriminating small differences among the two replicates. Figures 7-2, 7-3 and 7-4 show the DFA results for the bleach treated shrimp. The coefficients of the discriminant functions are summarized in Table 7-14. For each individual replicate, clusters are well formed with minimum overlap. As stated before, microbial loads were not significantly different among the treatments, therefore it is expected that discrimination was in fact due to the odor of the bleach on the surface of the shrimp. In the phosphate treated shrimp, Figures 7-5, 7-6 and 7-7 also showed good separation of cluster for each level of treatment. The coefficients for the discriminant functions are listed in Table 7-15. Some overlap existed, however, classification rates were above 95%. Discrimination of the different levels of phosphate treated shrimp was not expected, since sodium tripolyphosphate is not a volatile. However, phopshates have a number of functions in foods. Some of the differences in odor detected by the electronic nose could be due to the ability to bind water or to chelate metal ions. These could affect

PAGE 232

212 Table 7-10. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of bleach used to treat shrimp. DFA models were obtained for each measurement time Replicate Results for Figure Xjrrip after treatment (hrs) Bleach concentration (ppm) Overall 0 25 50 100 200 1 7-2 0 100 100 100 100 100 100 7-3 24 100 100 100 100 100 100 7-4 48 100 80 100 100 100 96 2 7-2 0 60 100 100 100 100 92 7-3 24 100 100 100 60 100 92 7-4 48 100 100 100 80 100 96 1 &2 7-2 0 60 70 60 80 60 66 7-3 24 80 50 80 60 60 66 7-4 48 60 80 60 70 60 66 Table 7-11. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of phosphate used to treat shrimp. DFA models were obtained for each measurement time Replicate Results for Figure Time after treatment (hrs) Phosphate concentration (%) Overall 0 2.0 4.0 6.0 1 7-5 0 100 100 100 100 100 7-6 24 100 100 100 100 100 7-7 48 100 80 100 100 95 2 7-5 0 100 100 100 100 100 7-6 24 100 100 100 100 100 7-7 48 100 80 100 100 95 1 &2 7-5 0 70 60 80 80 72.5 7-6 24 60 80 60 70 67.5 7-7 48 60 60 40 70 57.5

PAGE 233

213 Table 7-12. Correct classification rates obtained from the DFA of electronic nose sensor readings grouped by the concentration of sulfite used to treat shrimp. DFA models were obtained for each measurement time Replicate Results for Figure Time after treatment (hrs) Sulfite concentration (%) Overall 0 0.75 1.25 2 1 7-8 0 100 100 100 60 90 7-9 24 100 100 100 100 100 7-10 48 100 100 100 100 100 2 7-8 0 100 100 100 100 100 7-9 24 80 100 100 100 95 7-10 48 100 100 100 100 100 1&2 7-8 0 70 70 90 90 80 7-9 24 40 80 70 70 65 7-10 48 70 80 70 80 75 Table 7-13. Eigenvalues for the first two factors obtained by PC A for the combined data set of replicates for each chemical at each storage time Chemical PCA Storage time Treatment Ohrs 24 hrs 48 hrs Bleach Factor 1 11.34 10.05 11.25 Factor 2 0.36 1.63 0.49 Phosphate Factor 1 10.44 11.21 10.82 Factor 2 1.01 0.59 0.86 Sulfite Factor 1 9.99 10.15 11.18 Factor 2 1.13 1.56 0.58

PAGE 234

214 the rate of volatilization of other compounds in the shrimp, which were detected by the electronic nose sensors and to some degree by the panelists. The sulfite treated shrimp (Figures 7-8, 7-9 and 7-10) also showed formation of distinct clusters for each level of sulfite. Table 7-16 contains the coefficients used to construct the discriminant functions. Classification rates were lower when compared to the other two chemicals, therefore some overlap was expected. In sulfite treated shrimp, ammonia levels were different for the different levels of sulfite. Therefore it is expected that lower classification rates could be due to the differences in ammonia levels. The nose responded to the ammonia and the odor profiles changed slightly. Table 7-17 contains the pvalues from DFA of the electronic nose sensors contribution to the prediction of group membership for all DFA models obtained in these experiments. Some of the sensors were expected to be responsible for the overall goodness of discrimination. However, data showed that there was no one sensor that was statistically significant for all treatments. Sensor 262 appeared to be significant in some DFA models, however, not in all of them. All the other sensors were at least significant in one model, except sensors 297 and 458, which were not significant in any of the models. Conclusions This study concluded that panelists had difficulty in trying to identify if shrimp were treated with a chemical, basing their judgement on odor alone. However, the electronic nose sensors showed the ability to discriminate samples of shrimp treated with

PAGE 235

215 Bleach Exp. 2 0 hrs after treatment Bleach Exp. 1 & 2 0 hrs after treatment Concentration ° Control D 25 ppm ° 50 ppm a 100 ppm • 200 ppm Concentration ° Control D 25 ppm ° 50 ppm A 100 ppm • 200 ppm Figure 7-2. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 0 hrs of treatment

PAGE 236

216 Bleach Exp. 1 24 hrs after treatment -8 - ) [ : a A ° ;/ ^s^_^ c -4 -2 0 2 4 Funtion 1 Bleach Exp. 2 24 hrs after treatment -4-2 0 2 Funtion 1 Bleach Exp. 1 & 2 24 hrs after treatment Concentration o o A • Control 25 ppm 50 ppm lOOppm 200 ppm £ c — fl — )S \ \^'p 7 Concentration o o A • Control 25 ppm 50 ppm 100 ppm 200 ppm Concentration ° Control D 25 ppm ° 50 ppm a 100 ppm • 200 ppm -2 0 2 4 Funtion 1 Figure 7-3 . DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 24 hrs of treatment

PAGE 237

217 Bleach Exp. 1 48 hrs after treatment -10 Bleach Exp. 2 48 hrs after treatment 6 . -4 -2 0 2 Funtion 1 Bleach Exp. 1 & 2 48 hrs after treatment Concentration ° Control ° 25 ppra ° 50 ppm a 100 ppm g • 200 ppm • f # / < Concentration 0 Control ° 25 ppm ° 50 ppm A 100 ppm 6 • 200 ppm Concentration ° Control D 25 ppm 0 50 ppm A 100 ppm • 200 ppm Figure 7-4. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of bleach solutions used to treat shrimp. Results after 48 hrs of treatment

PAGE 238

218 Phosphates Exp. 1 0 hrs after treatment -10 ' -r ; H"]"' o \u / 1 0 o ; by -6 -4 -2 0 2 4 Funtion I Phosphates Exp. 2 0 hrs after treatment Phosphates Exp. 1 & 2 0 hrs after treatment Concentration o Control 2.0 % o 4.0 % a 6.0 % Concentration o Control a 2.0% o 4.0 % ^ 6.0 % Figure 7-5 . DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 0 hrs of treatment

PAGE 239

219 Phosphates Exp. 1 24 hrs after treatment Phosphates Exp. 2 24 hrs after treatment Phosphates Exp. 1 & 2 24 hrs after treatment Figure 7-6. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 24 hrs of treatment

PAGE 240

220 Phosphates Exp. 1 48 hrs after treatment Funtion 1 Phosphates Exp. 2 48 hrs after treatment Phosphates Exp. 1 & 2 48 hrs after treatment Concentration o o A Control 2.0% 4.0% 6.0 % Concentration ° Control 0 2.0 % ° 4.0 % a 6.0 % Concentration ° Control a 2.0 % o 4.0 % a 6.0 % Figure 7-7. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of phosphate solutions used to treat shrimp. Results after 48 hrs of treatment

PAGE 241

221 Sulfites Exp. 1 0 hrs after treatment /L f, \ : / A a. Concentration ^ — Concentration o Control 0.75 % 1.25% a 2.0 % p a \ J \ o/ fV— -°><^• Concentration ° Control 0.75 % o 1.25% 5 a 2.0 % Figure 7-8. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 0 hrs of treatment

PAGE 242

222 Sulfites Exp. 1 24 hrs after treatment ° \ \a ; J \ -12 -6 0 6 12 Funtion 1 Sulfites Exp. 2 24 hrs after treatment Sulfites Exp. 1 & 2 24 hrs after treatment 0 Funtion 1 Concentration ° Control a 0.75 % o 125% a 2.0 % Concentration o Control Q 0.75 % o 125% a 2.0 % Concentration o Control 0.75 % ° 1.25% ^ 2.0 % Figure 7-9. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 24 hrs of treatment

PAGE 243

223 Sulfites Exp. 1 48 hrs after treatment 5 4 3 N 2 e 2 1 I 0 -1 -2 -3 -12 e .2 0 -4 0 4 Funtion 1 Sulfites Exp. 2 48 hrs after treatment 0 2 Funtion 1 Sulfites Exp. 1 & 2 48 hrs after treatment Concentration o Control o 0.75 % o 1.25% 12 ^ 2.0 % Concentration o Control a 0.75 % o 1.25% ^ 2.0 % Concentration o Control 0.75 % o 1.25% a 2.0 % Figure 7-10. DFA results for the correlation of electronic nose sensor readings of shrimp and concentration of sulfite solutions used to treat shrimp. Results after 48 hrs of treatment

PAGE 244

a O c § Ii g SB u c o 8 3 3 c Q § £. 3 SI) u o i 43 C3 I i I P» i i i oo i r
PAGE 245

g D 9 O C E 8 8 g o d o ft H SI i 9 H 00 m on On O C-i rs •A ON O oo On oo On r<-> On NO r<-i On n© IT) 00 oo oo 1/-) no On OO oo NO IT) O^ CnI NO Cn| Cn| NO On On oo ON On T ON NO OO m (~ni CnI cn o ON On 00 ON oo in >n o (N-N NO OO OO •nI
PAGE 246

c C o oc in IT) o o vo 00 o vo q in I*; ITl VO vO O OV VD oo f«"l oo o-' ON ro o VO in q C CM 3 5) o i ro oo r 1 O i g
PAGE 247

227 © ir> 00 © O O o 3 s © © o o o 00 o © 00 C o 1 I II /"> 2 o so o ro in o o 00 o © © 0O o © o oo oo O "55 .& 11 S3 B 9 H s © o B dew Egg g 03 I U 00 oo re 00 1 1 I o E 3 c/5 §) • I— « n § •a .8 .S J2 i *

PAGE 248

228 different levels of chemicals, under the conditions of these experiments. The electronic nose also detected the change in odor of shrimp after 24 and 48 hrs of storage. From the results it could be seen that the changes in odor due to storage are more pronounced than that of the chemical treatments. This technique could be an alternative to detect the presence of sulfites, phosphates and bleach in shrimp. However, for this to be used, more data should be gathered to take into account odor differences due to shrimp harvesting locations, different species, processing and storage conditions, etc. More samples will bring more variability to the odor, and probably lower classification rates for the detection of these chemicals will be obtained. When shrimp are treated with bleach, phosphates and sulfites, some color change occurs. Therefore, it is expected that an integration of electronic nose sensor data and color analysis can provide better means of detection of treated shrimp. Moreover, in this study only DFA was used as the pattern recognition technique, however other techniques could be used to obtain better discrimination, making this technology a fast, simple an objective method to detect treated shrimp, which can be implemented by seafood buyers, processing facilities, and inspection agencies.

PAGE 249

CHAPTER 8 CONCLUSIONS AND RECOMMENDATIONS The findings from this research demonstrated that objective quality evaluation of shrimp and fresh raw salmon can be achieved by using color machine vision techniques for visual evaluation, and electronic nose sensors for odor assessment. The results showed that these two techniques can quantify the color and odor of a sample in a fast, reproducible and simple manner. These results will help develop new methodologies that will assist in the objective and repeatable quality evaluation of shrimp and salmon. These methods have potential in industrial and regulatory application where rapid response, no sample preparation, no requirements for chemicals, and no technical expertise to run the system are required. However, several factors should be addressed before these technologies can be fully implemented. The specific findings and recommendations from this research were: Visual Evaluation 1 . The Color Analysis software was able to determine the colors present in an image of a sample and quantify their amounts based on the percentage of the area of the sample covered by each color. Color was measured in the RGB color system by using a 64, 5 12 or 4096-color scheme. The program allowed the user to analyze 229

PAGE 250

230 the entire image or only regions of interest, making color evaluation more flexible and precise. 2. The software was able to calculate the overall L*a*b* color values of a sample by obtaining the L*a*b* values of each individual pixel and averaging them. 3. Color data obtained by the color machine vision system cannot be directly compared with color data obtained from tristimulus colorimeters. Color measurement is device dependent. Illumination sources, light reflection angles and color filters are different within colorimeters, and with the computer vision system used in this research. An approximation can be obtained by using the color calibration procedure implemented in the software, plus mathematical transformations to take into account the light source. Color calibration should also be performed to be able to compare images taken under different hardware conditions. Color standards should be used to adjust color settings in the acquired images. More complex calibration procedures have been reported in the literature but were beyond the scope of this project. These should be studied further and eventually implemented in the Color Analysis program. 4. Differences in the color measurement between different hardware are the main concern for the transportability of the data. In this case color calibration must be performed to be able to compare images taken under different hardware conditions. Color standards should be used to adjust color settings in the acquired images.

PAGE 251

5. The Color Analysis software was capable of defining and quantifying the colors present in the different species of shrimp. Discriminant function analysis (DFA) was used as a pattern recognition technique to differentiate shrimp species based on color profiles. Results showed that tiger, pink and white shrimp had different color profiles when using the 64-color block scheme, however, brown shrimp had similarities with both pink and white shrimp. 6. In general, brown, pink, tiger and white shrimp changed color during storage, which could be used as indicators of their quality. 7. The major color change in all shrimp species was the increase of black spots or melanosis. Melanosis was identified and quantified in all shrimp species. Development and formation of melanosis was easily tracked with this technique. 8. The Color Analysis software was able to quantify and measure color changes in raw salmon fillets during storage. Formation of a brown color and disappearance of an orange red color could be used as indicators of quality, and predictors of storage temperature and time of raw salmon fillets. 9. Color data for both salmon and shrimp showed a considerable amount of variation between samples. Therefore, when analyzing such products, a large number of replicates should be used to take into account this natural variation. 1 0. Further work is needed to accumulate more data from different species, origins, season of harvest, age, diets, and growing environments (aquaculture or wild), that will help build databases that can be used to develop models for color assessment or prediction.

PAGE 252

232 Odor Assessment 1 . The electronic nose sensor readings were able to discriminate odors of shrimp and salmon by using discriminant function as the pattern recognition technique. In both species, the sensor readings could detect a change in odor of the samples from one day to another, however, those differences were not always apparent to panelists. 2. A combination of electronic nose sensor readings and color data was able to correlate the overall quality of salmon fillets with storage time and with scores from sensory evaluation, by using DFA as the pattern recognition technique. Results for the combined data sets gave better classification than the individual data sets. 3. Decomposition odors in shrimp were measured objectively with the electronic nose technology, giving similar results to those obtained from experienced inspectors from the U.S. Food and Drug Administration (FDA). DFA was able to classify samples in three categories: pass, borderline and fail. 4. Electronic nose sensor data of chopped and intact shrimp demonstrated that the odors of decomposition are mainly present in the outside. Therefore, it was concluded that chopping the samples has no benefit for odor evaluation, and in fact it can dilute the odors of decomposition. 5. Electronic nose sensor readings showed that different species of shrimp gave different responses, both raw or cooked. This means that shrimp odor depends on

PAGE 253

233 the specie, which sometimes is difficult to distinguish with the human nose. Therefore, a single model for odor evaluation of all shrimp species may not give a good classification. 6. Panelists were not able to identify if shrimp were treated with low levels of bleach, sulfites or phosphates, basing their judgement on odor alone. However, the electronic nose sensors showed the ability to discriminate samples of shrimp treated with different levels of chemicals, under the conditions of these experiments. 7. In this study only DFA was used as the pattern recognition technique, however other more complex multivariate statistical techniques or artificial neural networks could be used to obtain better discrimination. These pattern recognition techniques should allow the researcher to take into account species, sample history, origin, or other data that is available at the time of analysis. That can help develop good predictive models for odor evaluation of shrimp and salmon. 8. Simple comparison of electronic nose sensor readings among two different machines from the same manufacturer was not possible in this study. Methodologies to convert data from one machine to another will be crucial to be able to transport a classification model between electronic noses. Otherwise, this technology will have difficulties to enter regulatory and commercial settings. Color machine vision and electronic nose are two different technologies that evaluate two different quality parameters. These two parameters are odor and color, two important properties of the overall quality of a food product. In foods products,

PAGE 254

234 consumers do not buy a product by looking only at color, odor, texture or flavor, but at the combination of all of them. Therefore, data from both of these instruments should be combined to automate, better classify and measure the overall quality of a food product, because quality cannot be measured well by one parameter alone. With new technological advances such as electronic tongues and texture sensors, sensory evaluation of food can be easily automated and results will be more objective and reproducible, with possibilities of developing standard methods of quality evaluation for use in worlwide markets.

PAGE 255

APPENDIX A COLOR ANALYSIS SOFTWARE Table A-l . Description of the 64-color block scheme used in the Color Analysis software Color block # RGB color system L*a*b* color system NBS* Name NBS* # R G B L b o 32 32 32 5.04 0 0 267 1 32 32 96 10.15 25.31 -39.86 Hppn hlnp UbtU UlUt 1 79 2 32 32 160 19.09 48.46 -69.22 vi\/iH hi hp VIVIU UIUC 1 If, 3 32 32 224 28.74 67.9 -94.48 vivifi hliif* VIVIU UIUC 1 If. 4 32 96 32 30.35 -34.38 31.15 UCCU yCIIUWIolI glCCll 1 ^7 5 32 96 96 31.73 -19.52 -5.82 nurlf Klincn or^^ti Ual ft. UlUloll t;l CCI 1 l<« 1 OJ 6 32 96 160 35.18 7.02 -42.88 cfrrwio nJiif* MlUllfc; UIUC 1 78 I/O 7 32 96 224 40.51 34.76 -75.01 V1V1U UIUC 1 If, 1 id g 32 160 32 54.51 -56.76 53.69 viviH v/pI 1 en crr/*&n VIVIU yClIUWlMI lit CCU 1 70 9 32 160 96 55.12 -49.27 26.76 Miuiig yciiuwibii green 1 1 1 10 32 160 160 56.76 -31.83 -9.39 liiint nil iicn nrAon ll^Ill U1U1MI glCCM 11 32 160 224 59.56 -7.78 -44.33 Ulllllalll UIUC 1 77 ill 12 32 224 32 77.21 -76.39 73.07 viviu yciiowisn green 1 7Q 13 32 224 96 77.56 -71.93 53.84 viviu yciiowibn green 1 7Q 14 32 224 160 78.52 -60.58 21.82 vivid green 1 to 15 32 224 224 80.21 -42.71 -12.58 hrilliant hlnish orppn ui iiiiuin uiuioii gi ecu 1 so 16 96 32 32 16.93 30.3 18.26 deep reddish brown 41 17 96 32 96 19.54 38.49 -24.2 very deep reddish purple 239 18 96 32 160 25.26 53.5 -58.82 vivid violet 205 19 96 32 224 32.85 70.25 -87.53 vivid purplish blue 194 20 96 96 32 34.21 -8.88 36.24 moderate olive 107 21 96 96 96 35.39 0 0 dark gray 266 22 96 96 160 38.41 18.55 -37.54 strong purplish blue 196 23 96 96 224 43.2 40.99 -70.5 vivid purplish blue 194 24 96 160 32 56.28 -42.32 55.91 strong yellow green 117 25 96 160 96 56.86 -36.22 29.34 strong yellowish green 131 26 96 160 160 58.42 -21.57 -6.74 light bluish green 163 27 96 160 224 61.11 -0.51 -41.79 brilliant blue 177 28 96 224 32 78.23 -67.47 74.34 vivid yellowish green 129 29 96 224 96 78.58 -63.42 55.26 vivid yellowish green 129 30 96 224 160 79.51 -53.05 23.35 brilliant green 140 31 96 224 224 81.18 -36.52 -11.04 brilliant bluish green 159 32 160 32 32 31.97 51.92 38.2 vivjd, red 11 235

PAGE 256

236 Table Al, continued Color block # RGB color system L*a*b* color system NBS* Name NBS* # K O D T L a u 0 33 1 AA lOU 51 OA yo 11 1C 55.20 SS Al DD.05 -\.y / deep purplish red 256 34 i aa loU i 1 51 i An loU 1A £1 50.32 £A 11 04.3 / -4U.UJ vivid purple 216 35 i An lOU 11 52. 11 A 221 A 1 A 1 Hi .01 "7 A 1 1 /O. / 1 T> TO /z. /y vivid violet 205 36 1 AA 1 ou OA yo 1 1 52 Al A 4z.o zj.y l 4o.o 1 strong brown 55 37 1 AA 10U OA OA yo Al A G 4J.45 2S.5 1 iz.o grayish red 19 38 1 AA lOU OA i An 10U A C 11 13. / / 10 OS JO.OJ -ZD. 40 strong purple 218 39 i An lOU OA yo 11 A 221 AQ CO S/i n i j4.U 1 <0 Q/l -jy.oH vivid violet 205 40 i An lOU i An lOU 11 JZ AA "7A OU. 10 1 A 11 -14. JJ deep greenish yellow 100 41 i An loU l An lou OA yo A 1 11 01.2/ Tf Ol SD.6 1 light olive 106 42 i An loU i An 1 OU i An loU £.1 AA oz.oo o U A u medium gray 265 43 i An low i An loU 11 A 111 ac no oj.uy 1 £ *70 15. /y -3S.Z5 brilliant violet 206 44 i An lOU 11 A 221 11 52 on oa su.yo 1 A O 1 -4o. y4 "7*7 AO vivid yellow green 115 45 i An lOU 11 A 221 OA yo 0 1 1 0 5 1 .26 Al 11 -4 j. / 1 jV.Uz brilliant yellow green 116 46 i An I ou 11 \ 221 i An Q1 1 "7 5Z. 1 / K 1 1 -3 D. 51 Zl.Sy brilliant yellowish green 130 47 i An low 11 A 221 11 A 221 01 1A o5. 11 11 CO -2 1 .38 -O.V4 very light bluish green 162 48 11 A 221 11 a 11 52 A6. 11 40. 15 ~7n c /U.j jD.44 vivid reddish orange 34 49 11 A 221 1 1 52. OA yo Al Z 4 /.D TO AC zU.U / vivid red 11 50 11 A 221 11 52 lOU TO 1 /I /5. 14 1 o. 15 vivid purplish red 254 51 11 A 221 11 11 A 111 CO oc OA 00 00.55 vivid purple 216 52 yo 10 JZ SI Ad JJ.OH J l.j / *\G Q J7.7 vivid reddish orange 34 53 224 96 96 54.27 53.81 28.86 vivid red 11 54 224 96 160 55.94 60.08 -9.04 vivid purplish red 254 55 224 96 224 58.81 70.09 -44.59 vivid purple 216 56 224 160 32 67.81 16.76 69.94 strong orange yellow 68 57 224 160 96 68.25 19.35 45.83 moderate orange 53 58 224 160 160 69.42 26.08 10.6 moderate pink 5 59 224 160 224 71.49 37.09 -24.88 brilliant purple 217 60 224 224 32 85.58 -19.19 83.28 vivid grenish yellow 97 61 224 224 96 85.87 -16.82 65.31 brilliant greenish yellow 98 62 224 224 160 86.68 -10.55 34.21 pale greenish yellow 104 63 224 224 224 88.12 o o white 263 * National Bureau of Standards Color Dictionary (ISCC-NBS, 1955).

PAGE 257

APPENDIX B COLOR DATA FOR SALMON Table B1 . Color data of samples of salmon fillets stored at different temperatures (first experiment: Summer) Storage temp. (°C) Storage time (days) Color blocks (% of the total area, average of 6 samples) 16 32 36 48 52 53 57 1.8° 1 7.59 53.10 25.84 0.19 10.18 1.15 1.34 2 2.90 55.33 23.16 0.56 14.60 1.50 1.47 3 0.86 37.84 34.93 0.39 20.83 1.72 2.96 4 2.93 55.18 23.55 0.58 14.03 1.50 1.65 5 0.66 31.68 36.50 0.34 23.97 2.39 3.90 6 0.68 32.43 37.22 0.34 23.09 2.12 3.46 7 2.98 50.30 28.25 0.73 14.63 1.34 1.38 8 0.81 28.53 47.01 0.06 19.39 1.95 1.94 9 2.72 38.75 41.17 0.21 14.18 1.61 0.86 10 2.56 27.70 55.25 0.02 10.03 2.20 1.35 7.0° 1 2.42 48.45 28.73 0.41 15.27 1.70 2.71 2 2.64 54.18 25.14 0.64 13.97 1.21 2.03 3 0.83 33.04 41.28 0.22 19.81 1.41 3.17 4 2.18 49.69 32.95 0.25 12.36 1.21 1.17 5 0.37 15.79 57.68 0.05 21.94 2.10 1.80 6 0.76 6.65 70.22 0.00 14.19 4.32 2.83 11.7° 1 9.49 57.86 21.98 0.39 8.30 0.81 0.85 2 3.51 61.67 19.43 0.91 12.53 0.88 0.86 3 1.99 43.06 40.80 0.09 12.21 1.00 0.67 4 4.55 45.81 41.81 0.04 6.46 0.56 0.35 5 1.22 5.15 77.97 0.00 10.42 2.47 1.31 Variable temperatur e study # 1 1 2.38 49.89 29.64 0.49 13.11 2.02 2.22 2 1.91 48.85 28.69 0.53 16.04 1.75 1.84 3 0.67 32.56 43.64 0.08 18.33 1.78 2.60 4 0.89 37.33 37.12 0.12 18.02 3.10 2.64 5 0.61 28.02 47.08 0.09 20.82 1.91 1.25 6 1.04 21.76 55.76 0.01 17.36 2.46 1.23 7 4.27 29.07 55.70 0.01 7.82 1.71 0.66 Variable temperatur e study #2 1 10.69 62.32 16.08 0.72 8.95 0.69 0.44 2 5.63 64.32 15.87 1.09 11.61 0.81 0.61 3 1.00 53.11 21.30 1.14 21.33 0.97 1.05 4 1.35 52.01 21.30 1.08 23.09 0.83 0.31 5 0.53 54.10 19.42 1.97 22.41 0.88 0.63 6 1.73 48.60 26.06 0.92 20.48 1.15 0.89 7 4.35 51.89 28.62 1.31 11.82 1.24 0.55 8 0.60 24.27 49.73 0.14 20.58 3.19 1.31 9 2.20 33.07 48.08 0.21 9.70 3.89 1.92 in 6 59 21 89 55 15 001 4.10 3 81 3.81 237

PAGE 258

238 Table B-2. Color data of samples of salmon fillets stored at different temperatures (seconc experiment: Fall) Storage temp. (°C) Storage time (days) Color blocks (% of the total area, average of 6 samples) 16 32 36 48 52 53 1.8° 1 1.91 28.59 0.23 58.70 7.10 3.19 2 1.35 25.85 0.33 58.25 9.73 4.14 3 4.46 16.95 0.09 66.84 7.51 3.80 4 3.15 28.83 0.09 56.85 6.99 3.89 5 3.52 52.96 0.22 35.47 4.50 3.29 6 1.10 47.62 0.76 37.90 9.61 2.96 7 0.23 44.87 7.32 22.04 21.89 3.60 8 0.12 26.76 25.16 8.68 33.12 5.46 7.0° 1 0.18 47.77 0.52 37.99 9.04 4.18 2 0.25 48.68 0.61 37.66 8.76 3.84 3 0.12 22.69 0.25 60.83 10.87 4.93 4 0.27 52.84 0.60 34.63 7.84 3.70 5 0.25 53.43 2.67 26.86 14.41 2.33 6 0.14 25.47 25.61 5.99 38.13 4.32 11.7° 1 0.39 49.55 1.00 34.76 8.13 5.28 2 0.47 59.57 0.80 27.31 6.28 4.99 3 1.50 41.13 1.21 41.18 9.35 5.47 4 0.36 23.24 21.84 6.38 40.72 7.11 Variable temperature study # 1 1 1.36 62.29 0.95 22.74 7.73 4.65 2 0.58 33.78 0.42 47.66 12.15 5.10 3 0.34 49.03 0.28 37.86 6.14 6.11 4 0.91 65.90 2.09 19.67 6.80 4.35 5 0.63 58.92 4.44 23.14 10.17 2.61 6 0.15 39.02 26.75 9.94 19.52 4.42 7 0.12 16.39 56.37 1.00 17.22 7.19 Variable temperature study #2 1 0.18 52.31 0.98 37.45 4.84 3.87 2 0.33 36.94 0.45 50.60 7.41 3.74 3 0.17 39.16 0.35 50.36 5.71 4.06 4 0.11 48.56 0.59 39.47 6.16 4.83 5 0.89 69.51 2.36 17.02 7.62 2.49 6 0.12 42.14 18.17 12.58 23.05 3.64 7 0.04 14.61 37.33 1.02 37.57 8.84

PAGE 259

APPENDIX C DATA FOR SHRIMP TREATED WITH DIFFERENT CHEMICALS Table C-l . Moisture content of shell-on shrimp treated with different levels of bleach Bleach (ppm) Sample Moisture content (% wet basis) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) Ohrs 48 hrs Ohrs 48 hrs 0 1 75.28 77.26 75.84 76.72 2 76.29 77.44 75.67 76.02 3 74.67 76.34 75.76 76.87 Avg. 75.41 77.01 75.76 76.54 St. dev. 0.82 0.59 0.09 0.46 25 1 76.64 76.21 76.24 75.00 2 74.53 75.92 76.08 76.11 3 75.77 76.67 75.91 75.59 Avg. 75.65 76.26 76.08 75.57 St. dev. 1.06 0.38 0.16 0.56 50 l 76.11 75.97 77.41 76.38 2 75.78 76.95 77.56 77.22 3 75.96 76.18 76.93 76.59 Avg. 75.95 76.36 77.30 76.73 St. dev. 0.17 0.51 0.33 0.44 100 1 76.00 76.25 76.36 76.27 2 75.82 76.49 75.57 76.92 3 75.43 76.06 76.42 76.29 Avg. 75.75 76.26 76.12 76.49 St. dev. 0.29 0.22 0.47 0.37 200 1 77.09 76.40 77.38 75.10 2 77.52 76.37 78.15 75.06 3 77.16 76.48 77.47 74.91 Avg. 77.26 76.42 77.66 75.02 St. dev. 0.23 0.06 0.42 0.10 239

PAGE 260

240 Table C-2. Moisture content of shell-on shrimp treated with different levels of phosphate Phosphates (%) Sample r Moisture content (% wet basis) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) 0 hrs A f* 1 48 hrs 0 hrs A O 1 48 hrs 0 1 80.57 80.27 79.98 80.88 2 80.16 79.65 79.45 80.43 3 80.05 80.38 79.91 80.90 Avg. 80.26 80.10 79.78 80.74 St. dev. 0.27 0.40 0.29 0.27 2.0 1 78.93 80.57 81.08 79.83 2 79.33 80.67 80.38 79.28 3 A A 79.44 80.19 80.15 79.27 Avg. 79.23 80.48 80.54 79.46 St. dev. 0.27 0.25 0.48 0.32 4.0 1 79.12 79.94 80.42 79.63 2 78.65 80.40 80.39 79.43 3 78.16 80.10 79.19 79.80 Avg. 78.64 80.14 80.00 79.62 St. dev. 0.48 0.23 0.70 0.18 6.0 1 79.80 79.33 79.84 80.48 2 79.61 79.52 78.40 80.27 3 79.40 79.14 79.27 80.96 Avg. 79.60 79.33 79.17 80.57 St. dev. 0.20 0.19 0.73 0.35

PAGE 261

Table C-3. Moisture content of shell-on shrimp treated with different levels of sulfite Sulfites (%) Sample Moisture content (% wet basis) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) A I 0 nrs 48 nrs U nrs 4o nrs 0 1 78.73 79.46 OA Cn 80.62 ly. IS 2 OA /I A 80.44 *7A T/I 79.76 TA AA 79.90 no cn 3 no OA 78.80 TO CA 78.59 OA 1 A 80.34 ~IQ AO /8.98 Avg 79.32 ta n 79.27 OA OA 80.29 in i t St dev A A/C 0.96 A £ 1 0.61 a in 0.3 / A 0.75 1 OA O 80.36 70 A/T 78.06 TO AT /8.0/ /8.JJ 2 OA C 1 80.53 TO Oil 78.26 OA 1 1 80.33 TO /1A /8.4U 3 OA C A 80.59 79.14 TA "5 1 79.33 79.55 Avg OA /I A 80.49 no a o 78.48 TA T H 79.24 78.82 St dev 0.12 0.58 1.13 0.63 1.25 1 78.62 78.69 77.53 79.87 2 79.24 78.30 78.18 79.16 3 78.96 77.38 77.34 79.49 Avg 78.94 78.12 77.68 79.51 St dev 0.31 0.67 0.44 0.36 2.0 1 80.15 78.50 79.91 78.54 2 79.89 78.81 78.96 78.51 3 79.57 80.81 79.30 78.01 Avg 79.87 79.37 79.39 78.36 St dev 0.29 1.25 0.48 0.30

PAGE 262

242 Table C-4. Microbial load of shell-on shrimp treated with different levels of bleach Bleach (ppm) Sample Microbial Load ( log (cfu/g of shrimp) ) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) 0 nrs 24 nrs A O 1 , 48 nrs A 1 , 0 nrs 24 nrs A O 1 48 nrs 0 i i C 0£ 5.826 6.672 a on 4.833 C 1 AT 5.107 7.740 L A 1 A£ 4.346 5.732 O HA A 8.740 A AT A 4.934 C 1 AA 5.100 7.643 Avg. 4.323 C TOO 5.782 8.708 A OOZT 4.886 C 1 A A 5.104 *7 /"AC 7.695 ot. uev. A AH 0.033 0.066 A A /I O 0.048 A AT 1 0.071 A A A C 0.005 A A/*" A 0.069 25 i i 4.276 C ACT 5.057 8. 681 4.903 5.792 *7 /"/'"'> 7.663 z 4.324 C AZTO 5.068 O I/O 8.763 A A 1 A 4.914 5.716 7.820 Avg. 4.301 C A/T O 5.063 8.724 4.908 C '"ICS' 5.756 7.748 m. dev. A A'? /I 0.034 A AAO 0.008 A AC O 0.058 A AAO 0.008 A AC /I 0.054 A 1 1 1 0.111 50 i 1 A 1 O A GAZ 4.o45 O AAA 8.000 c in 5.121 5.236 *7 A1 yl 7.914 L 4.364 4.839 7.978 5.111 5.274 7.924 Avg. 4.358 4.842 7.989 5.116 5.255 7.919 oi. aev. 0.008 0.004 0.016 0.007 0.027 0.007 100 1 4.324 5.155 8.623 4.820 4.934 6.903 2 4.061 5.215 8.591 4.771 4.898 7.114 Avg. 4.212 5.186 8.607 4.796 4.916 7.021 St. dev. 0.186 0.042 0.023 0.035 0.025 0.149 200 1 4.064 4.255 7.908 4.839 5.375 7.380 2 4.037 4.176 7.778 4.748 5.332 7.672 Avg. 4.051 4.217 7.848 4.796 5.354 7.550 St. dev. 0.019 0.056 0.092 0.064 0.030 0.206

PAGE 263

243 Table C-5 . Microbial load of shell-on shrimp treated with different levels of phosphate Phosphate (%) oampie Microbial Load ( log (cfu/g of shrimp) ) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) U nrs Z4 nrs 4o nrs U nrs Z4 nrs 4» hrs 0 i i j.oi J o.ayz I.ZZj o.l /o o.ozU 6.716 z < TOO o.yuo /.uzy o.l yj o./yy 6.681 Avg. con C. C\(\(\ /.lie C 1 O? o.loj o.olO o.oyy ot. aev. u.uzo A A1 1 U.U1 1 U.U1Z U.UZj 2.0 i i 5.243 5.663 5.845 5.760 6.210 6.987 Z 5.303 5.505 6.079 5.505 6.255 6.919 Avg. 5.274 5.591 5.978 5.651 6.233 6.954 Ol. UtV. 0.042 0.112 0.165 0.180 0.032 0.048 4.0 1 5.267 5.914 6.908 5.435 5.954 6.146 2 5.283 5.929 6.903 5.422 5.959 5.778 Avg. 5.275 5.922 6.906 5.428 5.957 6.000 St. dev. 0.011 0.011 0.004 0.009 0.004 0.260 6.0 1 5.493 5.491 6.230 5.964 5.863 6.301 2 5.519 5.447 6.176 5.922 5.806 6.462 Avg. 5.506 5.470 6.204 5.943 5.836 6.389 St. dev. 0.018 0.031 0.038 0.030 0.040 0.114

PAGE 264

244 Table C-6. Microbial load of shell-on shrimp treated with different levels of sulfite Sulfites (%) Sample Microbial Load ( log (cfu/g of shrimp) ) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) a i U nrs 24 nrs 48 nrs U nrs 24 nrs 48 nrs 0 i i C A1 £ "7 O 1 O /.2Ja /.653 C. AAO 5.962 T I^A /.25U /. /85 L 5.939 "7 ICO 7. 1 58 H O A C 7.845 C A1A 5.930 7.267 7.756 Avg. 5.927 7.2U0 7.76U C A/1 T 5.947 T O CA 7.259 7.771 ot. oev. A A1 C U.U5 / U.136 A AT3 U.U23 A A 1 1 U.U12 A AO 1 U.U21 0.75 1 1 5.898 7.117 7.292 5.877 7.230 7.422 L 5.870 7.176 7.262 5.903 7.182 7.400 Avg. 5.884 7.148 7.278 5.890 7.207 7.411 oi. aev. 0.020 0.042 0.021 0.018 0.034 0.016 1.25 1 5.816 7.212 7.137 5.838 6.914 7.215 2 5.833 7.161 7.130 5.795 6.968 7.173 Avg. 5.824 7.188 7.134 5.817 6.942 7.195 St. dev. 0.012 0.036 0.005 0.030 0.038 0.030 2.0 1 5.721 6.908 6.886 5.789 6.875 6.892 2 5.820 6.881 6.968 5.770 6.716 7.004 Avg. 5.773 6.895 6.929 5.780 6.803 6.952 St. dev. 0.070 0.019 0.058 0.013 0.112 0.079

PAGE 265

245 Table C-7. -on shrimp treated with different levels of bleach Bleach (ppm) Sample Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) u nrs z4 nrs 4o nrs A Ufa u nrs Z4 nrs 4o nrs 0 i i 7 AC /AO 7 1A /.Jo 7 /.Do 7 ^ /.j 7 AS IAj 7 77 /. IZ 7 Z 7 A Z /.4j "7 /I 1 /.4 1 7 A/1 /.04 1 .jZ 7 ^ I.J J 7 AC /.Oo A \m /\vg 7 /IA /.40 7 lO i.jy 7 A1 7 ^ 1 /.J 1 7
PAGE 266

246 Table C-8. pH of shell-on shrimp treated with different levels of phosphate Phosphate (%) Sample _P hi Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) u nrs 7/1 U^c 24 nrs 4o nrs u nrs z4 nrs AO U_ 0 4o nrs 0 i i lAZ 7 /IO /Ay 7 "JO 1 ,5y n in f.3y 7.34 7 Z 7 7 /.Z "7 C 1 I.J 1 1 /17 I A 2. /.4z TIC /.3b n a z 7.45 /\vg. 1 1 A /. lo "7 CA 1 A 1 /.41 *7 A 1 /.41 TIC /.35 H A A /.44 ol. Ucv. A u.uo A A1 U.UI A A7 U.Uz A A7 U.Uz A A 1 U.UI A A 1 U.UI 2.0 1 1 7 78 7 48 /.4o 7 £.1 /.J 1 /.J 1 7 /IO /.4y 7 /I7 /.4z 7 z 7.3 7.51 7.45 7.45 7.52 7.39 A \rct /\vg. 7.29 7.50 7.48 7.48 7.51 7.41 0.01 0.02 0.04 0.04 0.02 0.02 4.0 1 7.2 7.39 7.46 7.46 7.49 7.45 2 7.22 7.41 7.45 7.45 7.51 7.42 Avg. 7.21 7.40 7.46 7.46 7.50 7.44 St. dev. 0.01 0.01 0.01 0.01 0.01 0.02 6.0 1 7.4 7.46 7.47 7.47 7.49 7.51 2 7.39 7.49 7.41 7.41 7.53 7.45 Avg. 7.40 7.48 7.44 7.44 7.51 7.48 St. dev. 0.01 0.02 0.04 0.04 0.03 0.04

PAGE 267

247 Table C-9. pH of shell-on shrimp treated with different levels of sulfite Sulfites (%) Sample p] HI Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) 0 hrs 24 hrs A ft 1 48 hrs 0 hrs 24 hrs A ft 1 48 hrs 0 1 7.47 7.68 7.49 7.34 7.52 7.64 2 ^ C A 7.54 7.65 7.52 7.45 7.54 7.63 Avg. 7.51 7.67 7.51 7.40 7.53 7.64 St. dev. 0.05 0.02 0.02 0.08 0.01 0.01 0.75 1 1 7.55 7. 61 7.49 "7 A f\ 7.49 7.66 7.67 2 7.59 7.63 7.43 7.47 7.63 7.68 Avg. 7.57 7.62 7.46 7.48 7.65 7.68 St. dev. 0.03 0.01 0.04 0.01 0.02 0.01 1.25 1 7.45 7.60 7.68 7.31 7.58 7.64 2 7.47 7.59 7.66 7.42 7.56 7.62 Avg. 7.46 7.60 7.67 7.37 7.57 7.63 St. dev. 0.01 0.01 0.01 0.08 0.01 0.01 2.0 1 7.46 7.55 7.49 7.41 7.54 7.57 2 7.48 7.54 7.46 7.36 7.56 7.60 Avg. 7.47 7.55 7.48 7.39 7.55 7.59 St. dev. 0.01 0.01 0.02 0.04 0.01 0.02

PAGE 268

248 Table C-10. Ammonia levels of shell-on shrimp treated with different levels of phosphate Phosphate Sample Ammonia (ppm) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) (J nrs z4 nrs A O 1 „ 48 nrs A I (J nrs 24 nrs 48 nrs 0 1 54 AO 08 /5 C 1 51 62 79 43 05 SO c o 58 86 OA 80 /\vg. 48.51/ 00. 5U O 1 CA 81.5U c A CA 54.50 74.00 79.50 <\t rlev Ol. UCV. 7 78 Z. 1Z Q 1 Q y. 1 y A QK 4.7J 1 A Q7 A T 1 U. / 1 2.0 1 1 48 61 87 53 66 82 51 66 93 56 70 81 Ave 49.50 63.50 90.00 54.50 68.00 81.50 St. dev. 2.12 3.54 4.24 2.12 2.83 0.71 4.0 1 35 69 75 58 69 82 2 43 63 88 57 70 78 Avg. 39.00 66.00 81.50 57.50 69.50 80.00 St. dev. 5.66 4.24 9.19 0.71 0.71 2.83 6.0 1 45 67 86 63 68 73 2 49 74 82 58 73 86 Avg. 47.00 70.50 84.00 60.50 70.50 79.50 St. dev. 2.83 4.95 2.83 3.54 3.54 9.19

PAGE 269

249 Table C-l 1 . Ammonia levels of shell-on shrimp treated with different levels of sulfite Sulfites (%) Sample Ammonia (ppm) Replicate 1 (Time after treatment) Replicate 2 (Time after treatment) U nrs 24 nrs A O ... , 4o nrs 0 nrs 24 nrs AO "U_ 48 nrs 0 i i 151 250 A 1 A 410 1 C 1 151 300 280 Z 1 54 26U inn 320 1 1 n 149 330 Avg. 1
PAGE 270

250 Table C-12. Sensory data for shrimp treated with different levels of bleach solutions Time after treatment Panelist Bleach solutions (ppm) Replicate 1 Replicate 2 0 25 50 100 200 0 25 50 100 200 Ohrs 1 2 5 15 20 72 0 1 8 25 36 2 0 24 0 40 68 0 0 0 52 30 3 10 0 10 25 40 1 0 10 69 56 4 0 5 15 35 20 22 10 4 41 59 5 21 88 0 68 95 34 18 41 82 75 6 0 0 10 20 40 17 2 1 30 47 7 10 10 20 31 35 21 21 2 41 41 8 5 3 5 15 28 0 5 5 10 40 9 70 78 0 83 60 37 47 47 52 47 10 0 5 25 15 45 0 3 43 65 18 11 14 10 32 28 72 0 0 21 42 53 12 12 2 0 0 24 21 0 0 57 74 24 hrs 1 0 8 18 23 48 10 15 12 25 47 2 5 0 38 29 19 42 17 0 6 28 3 2 0 8 28 38 0 0 51 19 22 4 36 24 0 0 0 7 3 7 3 3 5 79 60 70 0 10 20 3 47 64 32 6 0 2 4 11 23 0 40 3 9 52 7 0 8 26 14 0 57 57 32 46 53 8 18 5 10 0 20 0 2 18 40 73 9 0 16 16 0 16 5 5 5 5 5 10 2 14 44 0 48 0 55 24 47 63 11 0 0 20 15 44 15 15 5 5 12 12 31 15 44 9 5 39 24 6 0 0 48 hrs 1 0 3 18 23 25 0 4 8 1 30 2 16 0 46 6 31 10 5 35 15 45 3 0 0 0 12 41 0 0 0 0 33 4 4 4 19 15 19 10 4 25 15 35 5 64 39 22 0 91 3 37 24 60 72 6 10 15 28 15 34 28 15 4 32 28 7 12 25 12 25 18 5 12 18 39 12 8 53 44 33 8 71 9 13 3 7 61 9 13 13 40 26 40 5 10 25 15 52 10 68 19 49 26 10 45 20 40 35 47 11 4 20 46 9 36 0 15 20 9 45 12, _2L _5_.. _L6 48 30 . 15 24 8 24

PAGE 271

Table C-13. Sensory data for shrimp treated with different levels of phosphate solutions Time after treatment Panelist Phosphate solutions (%) Replicate 1 Replicate 2 0 2 4 6 0 2 4 6 Ohrs 1 15 21 22 30 12 12 17 17 2 17 0 10 29 0 10 0 26 3 1 2 12 26 1 3 10 8 4 4 15 4 15 6 6 25 11 5 19 27 10 14 8 10 19 22 6 10 15 12 16 5 7 6 1 7 12 10 5 26 15 10 1 35 8 6 11 17 11 12 7 21 12 9 26 10 11 17 6 6 12 12 10 0 0 15 0 0 0 0 14 11 0 0 4 10 5 10 0 25 12 0 5 3 8 0 2 15 14 24 hrs 1 0 16 32 3 4 15 0 24 2 1 15 11 22 2 4 9 13 3 4 6 14 0 5 3 1 9 4 15 22 22 30 12 12 17 17 5 11 10 6 8 14 12 1 37 6 7 12 18 11 12 7 25 1 1 7 26 13 28 18 8 5 9 16 8 4 9 6 11 5 8 2 8 9 3 5 3 9 5 1 8 18 10 18 5 21 16 4 0 17 25 11 0 12 5 4 12 10 8 15 12 12 15 7 45 12 10 14 25 48 hrs 1 2 5 14 10 3 15 2 14 2 0 16 33 3 4 11 0 24 3 1 14 11 17 2 3 8 12 4 5 23 14 14 5 15 9 25 5 14 41 7 28 6 17 29 1 6 21 6 6 21 9 9 7 22 7 4 4 14 4 6 3 5 9 8 15 23 22 31 12 14 19 22 9 0 5 14 19 2 14 6 7 10 3 7 18 12 14 10 15 31 11 5 7 6 15 10 15 19 25 12 31 21 15 19 12 18 21 39

PAGE 272

252 Table C-14. Sensory data for shrimp treated with different levels of sulfite solutions Time after treatment Panelist Sulfite solutions (%) Replicate 1 Replicate 2 0 0.75 1.25 2 0 0.75 1.25 2 Ohrs 1 0 25 58 71 5 18 42 68 2 0 32 51 49 0 40 48 29 3 9 42 57 60 6 36 50 54 4 8 8 8 19 8 44 23 8 5 39 88 6 64 3 89 59 25 6 12 7 5 48 15 10 4 36 7 43 36 36 51 48 48 48 38 8 5 10 25 49 5 5 15 25 9 13 7 7 4 12 10 8 15 10 54 6 22 83 23 76 4 0 11 0 26 58 44 0 29 59 44 12 84 15 15 84 84 44 15 84 24 hrs 1 28 5 38 48 23 6 14 39 2 52 1 1 36 41 0 49 19 3 11 17 27 37 2 3 19 36 4 11 12 18 25 0 5 7 14 5 0 39 20 90 65 29 86 5 6 10 5 14 25 2 7 3 24 7 2 18 65 42 10 18 35 39 8 51 9 16 44 52 62 51 24 9 19 11 16 23 4 4 4 7 10 30 10 38 20 12 60 41 30 11 77 24 100 11 100 38 76 18 12 40 70 40 40 45 45 4 20 48hrs 1 6 12 15 4 8 12 25 14 2 50 6 0 6 0 51 0 43 3 6 14 31 8 4 19 26 44 4 21 3 4 12 3 3 9 9 5 10 15 0 24 3 15 22 10 6 16 0 1 5 1 1 10 19 7 49 57 57 48 19 11 19 47 8 10 6 25 12 17 10 21 18 9 15 17 31 35 10 25 14 21 10 25 25 2 10 5 3 15 35 11 4 12 14 13 6 6 10 40 12 15 0 5 21 8 14 6 36

PAGE 273

REFERENCES Affeldt Jr., H.A. 1991. An industrial machine vision system for fruit grading. Sunkist Growers, Inc. Ontario, CA. Presented during the 1991 international summer meeting of the ASAE. June 23-26, 1991. Albuquerque, NM. Aishima, T. 1 99 1 . Aroma discrimination by pattern recognition analysis of responses from semiconductor gas sensor arrays. Journal of Agricultural Food Chemistry 39: 752-756. Alford, J. A. and Fieger, E.A. 1952. The non-microbial nature of the black spots on iced packed shrimp. Food Technology 6:217-219. Arnarson, H., Bengoetxea, K., and Pau L.F. 1988. Vision applications in the fishing and the fish product industries. International Journal of Pattern Recognition and Artificial Intelligence 2(4): 657-671. Balaban, M.O. and Luzuriaga, D.A. 1996. Automated quality evaluation of shrimp. Food Quality 5: 56-66. Balaban, M. O., Yeralan, S., and Bergmann, Y. 1994. Determination of count and uniformity ratio of shrimp by machine vision. Journal of Aquatic Food Product Technology 3(3): 43-58. Ballantine, D.S. and Wohltjen, H. 1989. Surface acoustic wave devices for chemical analysis. Analytical Chemistry 61(11): 704A-706A. Bartlett, P.N. 1994. Metal oxide and conducting polymer odour sensors for an electronic nose. Proceedings of the 1st Symposium on Olfaction and Electronic Nose. Sep 26-27, 1994. Toulouse, France. Bartlett, P.N., Elliot, J.M., and Gardner, J.W. 1997. Electronic noses and their application in the food industry. Food Technology 51(12): 44-48. Bazemore, R., Rouseff, R., and Sims, C. 1996. Comparison of fresh squeezed orange juice analysis by electronic nose, sensory panel and GC FID. Proceedings of the 253

PAGE 274

254 3rd International Symposium on Olfaction and Electronic Nose. Nov 3-6, 1996. Miami, FL. Bazemore, R., Rouseff, R, Nordby, H., Goodner, K., and Jella, P. 1997. Discrimination of grapefruit juice varietal differences by an electronic nose equipped with metal oxide sensors. Seminars in Food Analysis 2: 239-246. Beaudoin, J. 1997. In search of the true hue. Seafood Business. Jan/Feb: 30-34. Berrie, P.G. 1997. Fuzzy logic in the evaluation of sensor data. In: Handbook of Biosensors and Electronic Noses. Medicine, Food, and the Environment . Kress-Rogers, E., editor. CRC Press. Boca Raton, FL. pp. 469-500. Bethea, S. and Ambrose, M.E. 1962. Comparison of pH, trimethylamine content and picric acid turbidity as indices of iced shrimp quality. Commercial Fisheries Review 244(3): 7-10. Billmeyer, F.W. and Saltzman, M. 1981. Principles of Color Technology . Wiley. New York. Borjesson, T., Eklov, T., Jonsson, A., Sundgren, H., and Schnurer, J. 1996. Electronic nose for odor classification of grains. Cereal Chemistry 73(4): 457-461. Botta,J.R. 1995. Evaluation of Seafood Freshness Quality . VCH. New York. Burger, P. and Gillies, D. 1989. Interactive Computer Graphics. Functional, Procedural and Device-Level Methods . Addison Wesley. Wokingham, UK. Campbell, J.L. and Williams, O.B. 1952. The bacteriology of Gulf Coast shrimp. IV. Bacteriological, chemical, and organoleptic changes with ice storage. Food Technology 6(4): 125-126. Cardarelli, A. J., Tao, Y., and Bernhardt, J.L. 1998. High resolution machine vision for non-destructive internal inspection of damaged rice grain. ASAE Annual International Meeting, Jul 12-16, 1998. Orlando, FL. Paper No. 98-6024. CFR. 1998. "Code of Federal Regulations. Title 50, Wildlife and Fisheries" Part 265. The Office of the Federal Register, National Archives and Records Aclministration, USGPO. Washington DC. Chang, O., Cheuk, W.L., Ranzell, N., Martin, R., and Finne, G. 1983. Indole in shrimp: effect of fresh storage temperature, freezing and boiling. Journal of Food Science 48: 813-816.

PAGE 275

255 Chang, Y.C. 1994. A color calibration procedure for machine vision. Ph.D. Thesis. University of Illinois at Urbana Champaign. Urbana, IL. Chervin, C, Franz, P., and Birrell, F. 1996. Calibration tile slightly influences assessment of color change in pears from green to yellow using the L, a, b space. HortScience 31(3): 471. Cheuk, W. L. andFinne, G. 1984. Enzymic determination of urea and ammonia in refrigerated seafood products. Journal of Agricultural and Food Chemistry 32(1): 14-18. Cobb, B.F., Alaniz, L, and Thompson C. Jr. 1973a. Biochemical and microbial studies on shrimp: volatile nitrogen and amino nitrogen analysis. Journal of Food Science 38: 431-435. Cobb, B.F. and Vanderzant, C. 1975. Development of a chemical test for shrimp quality. Journal of Food Science 40: 121-124. Cobb, B.F., Vanderzant, C, Thompson, C. A., and Custer, C.S. 1973b. Chemical characteristics, bacterial counts and potential shelf life of shrimp from various locations on the northwestern Gulf of Mexico. Journal of Milk and Food Technology 36(9): 463-468. Collins, J. andKelley, C. 1969. Alaska pink shrimp, Pandalus borealis : effects of heat treatment on color and machine peelability. Fishery Industrial Research, U.S. Dept. of the Interior 5(5): 181-189. Corcoran, P. 1993. Electronic odour sensing systems. Electronics and Communication Engineering Journal 5(5): 303-308. Delaure, J, Gretsch, C, Liardon, R., and Visani, P. 1996. Detection of aroma above a coffee powder. Limits and perspectives of electronic sensors. Proceedings of the 3rd International Symposium on Olfaction and Electronic Nose. Nov 3-6, 1996. Miami, FL. Diehl, K.C., Awa, T.W., Byler, R.K., van Gelder, M.F., Koslav, M., and Hackney, C.R. 1990. Geometric and physical properties of raw oyster meat as related to grading. Transactions of the ASAE 33: 1270-1274. Di Natale, C, Brunink, J.A.J., Bungaro, F., Davide, F., d'Amico, A., Paolesse, R., Boschi, T., Faccio, M., and Ferri, G. 1996. Recognition of fish storage time by a

PAGE 276

256 metalloporphyrins-coated QMB sensor array. Measurement Science and Technology 7: 1103-1114. Dodd, G.H., Bartlett, P.N., and Gardner, J. W. 1992. Odours the stimulus for an electronic nose. In: Sensors and sensory systems for an electronic nose . Gardner, J.W. and Bartlett, P.N., editors. Kluwer Academic Publishers, NATO ASI Series. Amsterdan, Netherlands, pp. 1-12. Dore, I. 1990. Salmon. The Illustrated Handbook for Commercial Users . VanNostrand Reinhold. New York. Dore, I. and Frimodt, C. 1987. An Illustrated Guide to Shrimp of the World . Osprey Books. Huntington, NY. Elster, R.T. and Goodrum, J.W. 1991 . Detection of cracks in eggs using machine vision. Transactions of the ASAE 34 (1): 307-312. Fatima, R., Farooqui, B., and Qadri, R.B. 1981. Inosine monophosphate and hypoxanthine as indices of quality of shrimp (Penaeus merguensis) . Journal of Food Science 46: 1125-1127, 1131. Fatima, R., Khan, M. A, and Qadri, R.B. 1988. Shelf life of shrimp ( Penaeus merguiensis) stored in ice (0°F) and partially frozen (-3°F). Journal of the Sciences of Food and Agriculture 42(3): 235-247. Federal Register. 1981. Defect action level for decomposition in imported canned and cooked frozen shrimp. 46:39221. USGPO. Washington, DC. Federal Register. 1982. Sulfiting agents: proposed affirmation of GRAS status with specific limitations. 41:29956. USGPO. Washington, DC. Fieger, E.A. 1951. Cause and prevention of blackspot on shrimp. Ice and Refrigeration. 120: 49-50. Finne, G. 1982. Enzymatic ammonia production in penaeid shrimp held on ice. In: Chemistry and Biochemistry of Marine Food Products. Martin, RE., Flick,G.J., Hebard, C.E., and Ward, D.R., editors. AVI. Westport, CT. p: 323-332. Finne, G. 1992. Non-protein nitrogen compounds in fish and shellfish. In: Advances in Seafood Biochemistry, Composition and Quality. Flick, G. and Martin, R., editors. Technomic Publication. Lancaster, PA. p: 393-401.

PAGE 277

257 Finne, G. 1995. Phosphates: production aids for food processors. SCOPE, a Technical Bulletin from Silliker Laboratories. Chicago Heights, IL. January, 1-5. Former, B. and Meyer, T.E. 1997. Number by Colors. A Guide to Using Color to Understand Technical Data . SpringerVerlag. New York. Forbes, I. 1996. Sodium metabisuMte and sodium sulfite blends to control shrimp melanosis and reduce sulfur dioxide. Master's Thesis. University of Florida. Gainesville, FL. Francis, F.J. and Clydesdale, F.M. 1975. Food Colorimetry: Theory and Applications . AVI. Westport,CT. GalbiatLL.J. 1990. Machine Vision and Digital Image Processing Fundamentals . Prentice Hall. Englewood Cliffs, NJ. Gardner, J.W. and Bartlett, P.N. 1994. A brief history of electronic noses. Sensors and Actuators, B: Chemical Sensors and Materials 18-19: 21 1-220. Gardner, J.W. and Bartlett, P.N. 1996. Performance definition and standardization of electronic noses. Sensors and Actuators, B: Chemical Sensors and Materials 33: 60-67. Gardner, J.W. and Hines, E.L. 1997. Pattern Analysis Techniques. In: Handbook of Biosensors and Electronic Noses. Medicine, Food, and the Environment . Kress-Rogers, E., editor. CRC Press. Boca Raton, FL. pp. 633-652. General Electric. 1993. GE Lighting Spectrum 9200 Lamp Catalog. 21 st Edition. General Electric. Cleveland, OH. Getino, J., Gutierrez, J., Ares, L., Robla, J.I., Horrillo, M.C., Sayago, L, Fernandez, M.J., and Rodrigo, J. 1998. Gas sensor arrays for contaminated soils. Proceedings of the 5th International Symposium on Olfaction and Electronic Nose. Sept. 27-30, 1998. Hunt Valley, MA. Ghate, S.R., Evans, M.D., Kvien, C.K. and Rucker, K.S. 1993. Maturity detection in peanuts (Arachis hypogaea L.) using machine vision. Transactions of the ASAE 36 (6): 1941-1947. Giorgianni, E.J. and Madden, T.E. 1998. Digital Color Management . Addison Wesley Longman. Reading, MA.

PAGE 278

258 Gonzalez, R. C. and Wintz, P. 1977. Digital Image Processing . Addison Wesley. Reading, MA. Gorga, C. and Ronsivalli, L. J. 1988. Quality Assurance of Seafood . VanNostrand Reinhold, an AVI Book. New York. Guedalia, I.D. 1997. Visual based quality sorting of fruit. Proceedings from the Sensors for Nondestructive Testing. Measuring the Quality of Fruits and Vegetables. Northeast Regional Agricultural Engineering Service Cooperative Extension. Feb 18-21, 1997. Orlando, FL. Gunasekaran, S., Cooper, T.M., and Berlage, A.R. 1988. Soybean seed coat and cotyledon crack detection by image processing. Journal of Agricultural Engineering Research 41: 139-148. Haard, N.F. and Lee, Y.Z. 1982. Hypobaric storage of Atlantic salmon in a carbon dioxide atmosphere. Canadian Institute of Food Science and Technology Journal 15(1): 68-71. Halland, H. and Njaa, L.R. 1988. Ammonia (NH 3 ) and total volatile nitrogen in preserved and unpreserved stored whole fish. Journal of the Science of Food and Agriculture 44: 335-342. Hamaoka, T. and Sasaki K. 1992. Development for a system for judging the freshness of raw oysters from Hirsoshima using fuzzy reasoning. Japanese Journal of Fuzzy Theory and Systems 4(1): 65-73. Hanrieder, D., Lauer, F., and Hirschfelder, M. 1998. Investigations on peppermint oils using electronic noses. Proceedings of the 5th International Symposium on Olfaction and Electronic Nose. Sept. 27-30, 1998. Hunt Valley, MA. Harper, W. J., Sohn, S., and Jou, K.D. 1996. The role of fatty acids in the aroma profiles of Swiss cheese as determined by an electronic nose. Proceedings of the 3rd International Symposium on Olfaction and Electronic Nose. Nov 3-6, 1996. Miami, FL. Hatano, M., Takahashi K., Onishi A. and Kameyama,Y. 1989. Quality standardization of fall chum salmon by digital image processor. Nippon Suisan Gakkaishi. 55(8): 1427-1433. Hearn, D. and Baker, P.M. 1994. Computer Graphics . Ch. 15, Color Models and Color Applications. 2nd Edition. Prentice-Hall. Englewood Cliffs, NJ. p: 565-582.

PAGE 279

259 Hebard, C.E., Flick, G.J. and Martin, R.E. 1982. Occurrence and significance of trimethylamine oxide and its derivatives in fish and shellfish. In: Chemistry and Biochemistry of Marine Food Products . Martin, R.E., Flick, G.J. and Ward, D.R., editors. AVI. Westport, CT. p: 149-304. Heinemann, P. H., Morrow, C. T., Crassweller, R. M., Sommer, H. J., Marrazo, W. N., and He, B. 1 997. Machine vision based station for grading "golden delicious" apples. Proceedings from the Sensors for Nondestructive Testing. Measuring the Quality of Fruits and Vegetables. Northeast Regional Agricultural Engineering Service Cooperative Extension. Feb 18-21, 1997. Orlando, FL. Hetzroni, A. and Miles. G.E. 1994. Color calibration for RGB video images. Presented during the 1994 international summer meeting of the ASAE. June 19-22, 1994. Kansas City, MO. Hodgins, D. 1995. Use of the electronic nose to evaluate citrus juice aroma volatiles. Paper 85-3. Institute of Food Technologists Annual Meeting. June 3-7, 1995. Anaheim, CA. Hodgins, D. 1997. The electronic nose: Sensor array-based instruments that emulate the human nose. In: Techniques for Analyzing Food Aroma . Marsili, R., editor. Marcel Dekker. New York. pp. 331-371. Hodgins, D. and Simmonds, D. 1995. Sensory technology for flavor analysis. Cereal Food World 40(4): 186-191. Hollingworth, T.A. Jr. and Throm, H.R. 1982. Correlation of ethanol concentration with sensory classification of decomposition in canned salmon. Journal of Food Science 47: 1315-1317. Huang, Y.W., He, L., and Ghates, K.W. 1996. Qualities of fresh and previously frozen marinated shrimp. Proceedings of the 20 th Annual Conference of the Tropical and Subtropical Seafood Science and Technology Society of the Americas. November 5-8, 1995. Humacao, Puerto Rico, pp: 218-232. Huss, H.H. 1988. Fresh Fish Quality and Quality Changes. Food and Agriculture Organization of the United Nations. Danish International Development Agency. Rome, Italy. Hutchings, J.B. 1994. Food Colour and Appearance . Blackie Academic & Professional. Glasgow, Scotland.

PAGE 280

260 ISCC-NBS. 1955. The ISCC-NBS method of designating colors and a dictionary of color names. National Bureau of Standards. Circ. 553. November. Jin,Z. and Harper, W.J. 1996. Differentiation of enzyme-modified cheese slurries by electronic noses. Paper 89C-24. Institute of Food Technologists. Annual Meeting, June 22-26, 1996. New Orleans, LA. Kassler, M., Corke, P., and Wong, P. 1993. Automatic grading and packing of prawns. Computers and Electronics in Agriculture 9: 319-333. Kelleher, S.D. and Zall, R.R. 1983. Ethanol accumulation in muscle tissue as a chemical indicator of fish spoilage. Journal of Food Biochemistry 7: 87-92. Kennish, J.M. and Kramer, D.E. 1987. A review of high pressure liquid chromatographic methods for measuring nucleotide degradation in fish muscle. In: Seafood Quality Determination . D.E. Kramer and J. Liston, editors. Elsevier Science. Amsterdam, pp 209-219. Krutz, G.W. and Precetti, C.J. 1991. Automated cracker inspection using machine vision. Presented during the 1991 International winter meeting of the ASAE. December 17-20, 1991. Chicago, IL. Kuehn, R.L. and Luxenberg, H.R. 1968. Display Systems Engineering . McGrawHill. New York. LeBlanc, R.J. and Gill, T.A. 1984. Ammonia as an objective quality index in squid. Canadian Institute of Food Science and Technology Journal 17: 195-201. Li, J. 1990. Oyster hinge line detection using digital image processing. Presented during the 1990 International summer meeting of the ASAE. June 24-27. Columbus, OH. Liao, K., Paulse, M.R., and Reid, J. F. 1994. Real-time detection of colour and surface defects of maize kernels using machine vision. Journal of Agricultural Engineering Research 59: 263-271. Lindsay, R.C. 1985. Food additives. In: Food Chemistry. Second Edition . Fennema, O.R., editor. Marcel Dekker. New York. pp. 629-687. Ling, P.P., Searcy, S.W., and Grogan, J. 1988. Adaptive thresholding techniques for shrimp images. Presented during the 1988 international summer meeting of the ASAE. June 26-29. Rapid City, SD.

PAGE 281

261 Ling, P.P. and Searcy, S.W. 1989. Feature extraction for a vision based shrimp deheader. Presented during the 1989 international winter meeting of the ASAE. December 12-15. New Orleans, LA. Lucas, Q. and Castan, T. 1995. Study of pure hop and blends with an electronic nose. Proceedings of the 2nd International Symposium on Olfaction and Electronic Nose. Oct 2-3, 1995. Toulouse, France. Luzuriaga, D.A. 1995. Development and Testing of an Automated Quality Evaluation Device for Shrimp. Master's Thesis. University of Florida. Gainesville, FL. Luzuriaga, D.A., Balaban, M. O., and Hasan, R. 1997a. Automated headspace ammonia analysis of shrimp. Journal of Aquatic Food Process Technology 6(3): 29-52. Luzuriaga, D.A., Balaban, M.O. and Yeralan, S. 1997b. Analysis of visual quality attributes of white shrimp by using machine vision. Journal of Food Science 62(1): 113-118, 130. Malle, P. and Poumeyrol, M. 1989. A new chemical criterion for the quality control of fish: trimethylamine / total volatile basic nitrogen (%). Journal of Food Protection 52(6): 419-425. Matches J. R. 1982. Effects of temperature on the decomposition of Pacific coast shrimp ( Pandalus iordani) . Journal of Food Science 47: 1044-1047, 1069. Matrox. 1997. Command Reference. Matrox Imaging Library Version 5.00. Manual No. 1 05 1 2-MS-0500. Ottawa, Canada. Maul, F., Sargent, S.A., Huber, D.J., Balaban, M.O., Luzuriaga, D.A, and Baldwin, E.A. 1 999. Non-destructive quality screening of tomato fruit using "electronic nose" technology. Proceedings of the Florida State Horticultural Society 1 10:188-194. McEvily, A. J., Iyengar, R, and Otwell, S. 1991. Sulfite alternative prevents shrimp melanosis. Food Technology 9: 80-86. Miller, B. and Delwiche, M.J. 1988. A line scan color vision system for peach grading. Transactions of the ASAE 3 (4): 1484-1490. Miller, S. 1987. Color TV for robot fruit pickers. Agricultural Research. 2:14-15. Miller, W.M. and Drouillard, G.P. 1997. On-line blemish, color, and shape analysis for Florida citrus. Proceedings from the Sensors for Nondestructive Testing.

PAGE 282

Measuring the Quality of Fruits and Vegetables. Northeast Regional Agricultural Engineering Service Cooperative Extension. Feb 18-21, 1997. Orlando, FL. Moncrieff, R.W. 1961. An instrument for measuring and classifying odours. Journal of Applied Physiology 16: 742-748. Montgomery, W.A., Sidhu, G.S., and Vale, G.L. 1970. The Australian prawn industry. Food Preservation Quarterly 30: 21-27. Mulchandani, A., Luong, J.H.T., and Male, K.B. 1989. Development and application of a biosensor for hypoxanthine in fish extracts. Analytical Chimica Acta 22 1 : 215222. Newman, D. 1998. Automation of quality evaluation of raw tuna. Master's Thesis. University of Florida. Gainesville, FL. Nickelson, R. 1992. Bacteria of significance in the international trade of shrimp. SCOPE, a Technical Bulletin from Silliker Laboratories. Chicago Heights, IL. 7 (2): 1-7. Niola, I. and Valletrisco, M. 1986. II contenuto di indolo come parametro di valutazione per la quanta dei gamberi. Industrie Alimentari 1 : 30-34. Noh, B.S. and Ko, J.W. 1997. Discrimination of the habitat for agricultural products by using electronic nose. Food Engineering Progress 1(2): 103-106. Ohashi, E., Takao, Y., Fujita, T, Shimizu, Y., and Egashira, M. 1991. Semiconductive trimethylamine gas sensor for detecting fish freshness. Journal of Food Science 56: 1275-1278, 1286. Okamura, N, Delwiche, M., and Thompson, J. 1993. Raisin grading by machine vision. Transactions of the ASAE 36 (2): 485-492. Olafsson, R, Martinsdottir, E., Olafsdottir, G., Sigfusson, S.I., and Gardner, J.W. 1992. Monitoring offish using tin oxide sensors. In: Sensors and Sensory Systems for an Electronic Nose . Gardner, J.W. and Bartlett, P.N., editors. Kluwer Academic NATO, ASI Series. Amsterdan, Netherlands, pp. 257-272. OtwelL S. and Marshall, M. 1986. Studies on the use of sulfites to control shrimp melanosis (blackspot). Florida Sea Grant College. Technical Paper No. 46. Pau, L.F. and Olafeson, R. 1991. Fish Quality Control bv Computer Vision . Marcel Dekker. New York.

PAGE 283

263 Paulsen, M.R. and McClure, W.F. 1986. Illumination for computer vision systems. Transactions of the ASAE 29(5): 1398-1404. Pearce, T. C. 1997a. Computational parallels between the biological olfactory pathway and its analogue 'The Electronic Nose': Part I. Biological olfaction. BioSystems 41(1): 43-67. Pearce, T.C. 1997b. Computational parallels between the biological olfactory pathway and its analog the electronic nose. Part II. Sensor-based machine olfaction. Biosystems 41(1): 69-90. Pearce, T.C, Gardner, J.W., FrieL S., Bartlett, P.N., and Blair, N. 1993. Electronic nose for monitoring the flavour of beers. Analyst 1 18(April): 371-377. Perkins, C. 1992. The Advanced Seafood Handbook. Seafood Business Magazine Rockland, ME. Persaud, K. and Dodd, G.H. 1982. Analysis of the mammalian olfactory system using a model nose. Nature 299: 257-271. Powers, J.J. 1989. Applying multivariate statistical methods to enhance information obtainable from flavor analysis results. In: Flavor Chemistry of Lipid Foods . Min, D.B. and Smouse, T.H., editors. American Oil Chemists' Society. Champaign, IL. pp. 434-462. Precetti, C.J. 1995. RGB Color Space. [On-line, Internet]. World Wide Web: HTTP://pasture.ecn.purdue.edu/~precetti/htmls/rgbspace.html. Ravelo, L.M. 1990. Shrimp Processing: An Integrated View. Master's Thesis. University of Florida. Gainesville, FL. Regenstein, J.M., Schlosser, M.A., Samson, A., and Fey, M. 1982. Chemical changes of trimethylamine oxide during fresh and frozen storage of fish. In: Chemistry and Biochemistry of Marine Food Products . Martin, R.E., Flick, G.J. and Ward, D.R., editors. AVI. Westport, CT. pp: 137-148. Rehkugler, G.E. and Throop, G.A. 1989. Image processing algorithm for apple defect detection. Transactions of the ASAE 32: 267-272. Rigney, M.P., Brusewitz G.H., and Kranzler G.A. 1992. Asparagus defect inspection with machine visioa Transactions of the ASAE 35 (6): 1873-1878.

PAGE 284

264 Robertson, A.R. 1992. Color perception. Physics Today December: 24-29. Robie, DJ. 1997. Comparison of two electronic noses: three tobacco applications. Seminars in Food Analysis 2, 247-256. Rossotti, H. 1983. Colour . Princeton University Press. Princeton, NJ. Ruiter, A. and Weseman, J.M. 1976. The automated determination of volatile bases (trimethylamine, dimethylamine and ammonia) in fish and shrimp. Journal of Food Technology 11:59-68. Ryan, M.A, Homer, M.L., Buehler, M.G., Manatt, K. S., Lau B., and Jackson, S. 1998. Monitoring space shuttle air for selected contaminants using an electronic nose. Proceedings of the 5th International Symposium on Olfaction and Electronic Nose. Sept. 27-30, 1998. Hunt Valley, MA. Saito, A. 1969. Color in raw and cooked Atlantic salmon ( Salmo salar). Journal of Fisheries Research Board of Canada 26(8): 2234-2236. Sarkar, N.R. 1 991 . Machine vision for quality control in the food industry. In: Instrumental Methods for Quality Assurance in Foods. Fung, D.Y. and Matthews, RE., editors. Marcel Dekker Inc. New York, pp: 167-187. Sarkar, N. and R. R. Wolfe. 1984. Computer vision based system for quality separation of fresh market tomatoes. Paper presented at the 1984 Winter Meeting of the ASAE. December 11-14. New Orleans, LA. Savagon, K.A. and Sreenivasan, A. 1978. Activation mechanism of pre-phenolase in lobster and shrimp. Fish Technology 15: 49. Schmidt, P.J. and Cuthbert, R.M. 1969. Color sorting of raw salmon. Food Technology 23: 232-234. Searcy, S. W. and Reid, J. 1989. Machines see red... and so much more. Agricultural Engineering 11/12: 10-15. Shamshad, S.I., Kher-Un-Nisa, Riaz, M., Zuberi, R, and Qadri, R.B. 1990. Shelf life of shrimp (Penaeus m erguiensis) stored at different temperatures. Journal of Food Science 55: 1201-1205, 1242. Shiers, V.P. and Farnell, P.J. 1995. The electronic nose: aroma profiling in the food industry. Food Technology International Europe 168-171.

PAGE 285

265 Shih, T.Y. 1995. The reversibility of six geometric color spaces. Photogrammetric Engineering and Remote Sensing 61(10): 1223-1232. Shurmer, H.V. 1990. An electronic nose: a sensitive and discriminating substitute for a mammalian olfactory system IEE Proceedings 137-G(3): 197-204. Sites, P.W. and Delwiche, M.S. 1988. Computer vision to locate fruit on a tree. Transactions of the ASAE 3 1 : 257-263, 272. Skrede, G., Risvik, E., Huber, M., Endersen, G., and Blumlein, L. 1990. Developing a color card for raw flesh of astaxanthin-fed salmon. Journal of Food Science 55(2): 361-363. Skrede, G. and Storebakken, T. 1986. Characteristics of color in raw, baked and smoked wild and pen-reared Atlantic salmon. Journal of Food Science 51(3): 804-808. Skrede, G., Storebakken, T., and Tormond, N. 1989. Color evaluation in raw, baked and smoked flesh of rainbow trout ( Onchorhynchus mvkiss ) fed astanxanthin or canthaxanthin. Journal of Food Science 55(6): 1574-1578. Springett, M. 1991. Current trends in flavour analysis. Food Manufacture 66 (7): 32-34. Stansby, M.E. 1962. Speculations on fishy odors and flavors. Food Technology 4:2832. Steinbrecher, K.J. 1973. Co llaborative study of the determination of ammonia as an index of decomposition in crab meat. Association of Analytical Chemists 56(3): 598-601. Storey, R.M., Davis, H.K., Owen, D., and Moore, L. 1984. Rapid approximate estimation of volatile amines in fish. Journal of Food Technology 19: 1-10. Strachan, N.J.C. 1993. Length measurements of fish by computer vision. Computers and Electronics in Agriculture 8: 93-104. Strachan, N.J.C, Nesvadba, P., and Allen A. R. 1990. Fish species recognition by shape analysis of images. Pattern Recognition 23(5): 539-544. Tan, T., Lucas, Q., Moy, L., Gardner, J., and Barlett, P. 1995. The electronic nose a new instrument for sensing vapours. LC-GC International 8(4): 218-225.

PAGE 286

266 Taylor, S.L., Higley, N.A., and Bush, R.K. 1986. Sulfites in foods. Uses, analytical methods, residues, fate, exposure assessment, metabolism, toxicity, and hypersensitivity. Advanced Food Research. 30: 1-76. Tomlinson, J. 1995. Application of digital aroma analyzers in the brewing industry. Picking and electronic nose. Principles and practice of this new sensing technology. Leatherhead Food RA. Surrey, UK. pp: 63-65. Tomlinson, J.B., Ormond, H.L., and Sharpe, F.R. 1995. Electronic aroma detection in the brewery. Journal of the American Society of Brewing Chemists 53(4): 167-173. Turhan, M., Balaban, M.O., Turhan, K.N., and Luzuriaga, D.A. 1998. Potential use of electronic nose technique for detection of meat adulteration: separation of porkbeef mixtures. Fleischwirtschaft International 6: 26-28. U.S. Department of Commerce. 1998. Fisheries of the United States, 1997. Current Fishery Statistics No. 9700. Silver Spring, MD. Viaux, L., Magne, V., and Robillard, B. 1996. The electronic nose to help determination of the technical specifications of some additives and technological aids used in the sparkling wine process. Proceedings of the 3rd International Symposium on Olfaction and Electronic Nose. Nov 3-6, 1996. Miami, FL. Ward, D.R., Finne, G., and Nickelson II, R. 1979. Use of specific-ion electrode (ammonia) in determining quality of shrimp. Journal of Food Science 44 (4): 1052-1054. Weber, K. and Poling, J. 1996. Study of pure hops with an electronic nose. Proceedings of the 3rd International Symposium on Olfaction and Electronic Nose. Nov 3-6, 1996. Miami, FL. Werlein, H.D. and Watkinson, B.M. 1997. Evaluation of sensory quality of processed foods by means of the electronic nose. Seminars in Food Analysis 2: 215-220. Whitfield, F. B. 1988. Chemistry of off-flavors in marine organisms. Water Science Technology 20 (8/9): 63-74. Winquist, F., Hornsten, E.G., Sundgren, H., and Lundstrom, I. 1993. Performance of an electronic nose for quality estimation of ground meat. Meas. Sci. Technol 41493-1500.

PAGE 287

267 Wong, K., Barlett, F., and Gill T.A. 1988. A diagnostic strip for semiquantitative determination of trimethylamine in fish. Journal of Food Science 53: 1653-1655. Wyszecki, G. and Stiles, W.S. 1982. Color Science. Concepts, and Methods, Quantitative Data and Formulae. 2nd Edition . Wiley. New York. Zimmermann, D. and Leclercq, C. 1995. Electronic nose for monitoring the flavour of special malts. Proceedings of the 2nd International Symposium on Olfaction and Electronic Nose. Oct 2-3, 1995. Toulouse, France.

PAGE 288

BIOGRAPHICAL SKETCH Diego Luzuriaga was born in 1970, in Quito, Ecuador. He attended the American School of Quito. After graduation he obtained a scholarship to study agriculture at Escuela Agricola Panamericana, Zamorano (Honduras). Diego obtained the Agronomo degree and graduated as the best student in his class. He won the award for the best worker and became a member of the Honor Society of Agriculture, Gamma Sigma Delta. Diego went back to Ecuador and worked at LABOLAB, a food analysis laboratory. With his interest in food science, he transferred to the University of Florida where he earned his bachelor of science degree in agriculture in 1993. In August of 1993 he married Elena Bastidas, and both of them started their Ph.D. programs. Since 1993, Diego has worked with Dr. Balaban in several food engineering projects and been a teaching assistant in the food processing course. During his studies he received several awards for academic achievement, scholarships, and memberships in honor societies, and won three graduate student paper competitions from the Institute of Food Technologists. In 1995 Diego received his master of science degree in food science and human nutrition with a minor in Agricultural and Biological Engineering. Diego expects to receive his Doctor of Philosophy degree in Food Science and Human Nutrition in August of 1999. Then he plans to work in the food industry. Later in his career he would like to be a member of the academy teaching food processing and engineering. 268

PAGE 289

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Balaban, Chair Professor of Food Science and Human Nutrition I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Sean F. O'Keefe Associate Professor of Food Science and Human Nutrition I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. ^^ Kenneth M. Portier Associate Professor of Statistics I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in^cope and quality, as a dissertation for the degree of Doctor of Philosophy. Ad ) A ^ Charles A. Sims Professor of Food Science and Human Nutrition I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Professor of Agricultural and Biological Engineering

PAGE 290

This dissertation was submitted to the Graduate Faculty of the College of Agriculture and to the Graduate School and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. August 1999 J^^^/Ch •"TJean, College of Agriculture Dean, Graduate School