<%BANNER%>

Mass and Size Estimation of Citrus Fruit by Machine Vision and Citrus Greening Diseased Fruit Detection Using Spectral A...

Permanent Link: http://ufdc.ufl.edu/UFE0044825/00001

Material Information

Title: Mass and Size Estimation of Citrus Fruit by Machine Vision and Citrus Greening Diseased Fruit Detection Using Spectral Analysis
Physical Description: 1 online resource (61 p.)
Language: english
Creator: Shin, Junsu
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: estimation -- image -- machine -- mass -- processing -- spectroscopy -- vision
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Citrus is the major fruit crop in Florida. Citrus industry occupiesa significant portion of Florida’s agricultural economy. There have been manyefforts to minimize producing costs, improve productivity and increase profit.Precision agriculture emerged as a solution to such efforts. A machine visionbased imaging and a visible-near infrared spectroscopy are examples ofprecision farming technology widely used in agricultural sectors. A machine visionsystem for estimation of citrus fruit mass, fruit count, and fruit size duringpostharvest processing was investigated towards the development of an advancedcitrus yield mapping system. Such yield mapping system enables the citrusgrowers to efficiently manage the in-grove spatial variability factors such as:soil type, soil fertility, moisture content, etc., and can help increase yieldand profits. Thus, a machine vision system was developed and installed in acitrus debris cleaning machine, which removes debris from mechanicallyharvested loads. An image processingalgorithm was developed to identify fruit from images of the postharvest citrusfrom a commercial citrus grove. For fruit detection, logistic regression modelbased pixel classification algorithms were developed. A mass calibrationprocess was conducted, and fruit mass was estimated, which turned out to bereasonably good. The highest coefficient ofdetermination (R2) value between the measured fruit mass and the estimated fruit mass was observed to be 0.945 and the root mean square error was 116.1 kg. A H-minima transform based watershed algorithmwas used to separate the joined fruit and enabled an estimation of fruitcounting and fruit size. Fruit mass estimation using the fruit size informationwas also conducted and these results were compared with that of the massestimation based on fruit pixel area. This research further exploredthe application of visible-near infrared spectroscopy for HLB detection incitrus fruit. In the study, the possibility of identifying HLB disease incitrus fruit using spectroscopy was investigated in a laboratory setup. Citrusfruit samples (101 healthy and 101 HLB infected) were collected from a citrusgrove in Lake Alfred, Florida during June and July 2012. Spectral reflectance(400 to 2500 nm) of the fruit samples were measured using a spectrophotometer.The reflectance and its first derivative were analyzed using discriminabilityanalysis and the candidate wavelengths were selected. Wavelength features forclassification were chosen by stepwise discriminant analysis. Logistic regressionmodel and linear support vector machines (SVM) were used to classify HLBinfected citrus fruit. Both models yielded more than 95% overall accuracy whentrained with the first derivative. The classification results indicated thatthe first derivative data contained more discriminate features than theoriginal reflectance.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Junsu Shin.
Thesis: Thesis (M.E.)--University of Florida, 2012.
Local: Adviser: Lee, Won Suk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044825:00001

Permanent Link: http://ufdc.ufl.edu/UFE0044825/00001

Material Information

Title: Mass and Size Estimation of Citrus Fruit by Machine Vision and Citrus Greening Diseased Fruit Detection Using Spectral Analysis
Physical Description: 1 online resource (61 p.)
Language: english
Creator: Shin, Junsu
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2012

Subjects

Subjects / Keywords: estimation -- image -- machine -- mass -- processing -- spectroscopy -- vision
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Citrus is the major fruit crop in Florida. Citrus industry occupiesa significant portion of Florida’s agricultural economy. There have been manyefforts to minimize producing costs, improve productivity and increase profit.Precision agriculture emerged as a solution to such efforts. A machine visionbased imaging and a visible-near infrared spectroscopy are examples ofprecision farming technology widely used in agricultural sectors. A machine visionsystem for estimation of citrus fruit mass, fruit count, and fruit size duringpostharvest processing was investigated towards the development of an advancedcitrus yield mapping system. Such yield mapping system enables the citrusgrowers to efficiently manage the in-grove spatial variability factors such as:soil type, soil fertility, moisture content, etc., and can help increase yieldand profits. Thus, a machine vision system was developed and installed in acitrus debris cleaning machine, which removes debris from mechanicallyharvested loads. An image processingalgorithm was developed to identify fruit from images of the postharvest citrusfrom a commercial citrus grove. For fruit detection, logistic regression modelbased pixel classification algorithms were developed. A mass calibrationprocess was conducted, and fruit mass was estimated, which turned out to bereasonably good. The highest coefficient ofdetermination (R2) value between the measured fruit mass and the estimated fruit mass was observed to be 0.945 and the root mean square error was 116.1 kg. A H-minima transform based watershed algorithmwas used to separate the joined fruit and enabled an estimation of fruitcounting and fruit size. Fruit mass estimation using the fruit size informationwas also conducted and these results were compared with that of the massestimation based on fruit pixel area. This research further exploredthe application of visible-near infrared spectroscopy for HLB detection incitrus fruit. In the study, the possibility of identifying HLB disease incitrus fruit using spectroscopy was investigated in a laboratory setup. Citrusfruit samples (101 healthy and 101 HLB infected) were collected from a citrusgrove in Lake Alfred, Florida during June and July 2012. Spectral reflectance(400 to 2500 nm) of the fruit samples were measured using a spectrophotometer.The reflectance and its first derivative were analyzed using discriminabilityanalysis and the candidate wavelengths were selected. Wavelength features forclassification were chosen by stepwise discriminant analysis. Logistic regressionmodel and linear support vector machines (SVM) were used to classify HLBinfected citrus fruit. Both models yielded more than 95% overall accuracy whentrained with the first derivative. The classification results indicated thatthe first derivative data contained more discriminate features than theoriginal reflectance.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Junsu Shin.
Thesis: Thesis (M.E.)--University of Florida, 2012.
Local: Adviser: Lee, Won Suk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2012
System ID: UFE0044825:00001


This item has the following downloads:


Full Text

PAGE 1

1 MASS AND SIZE ESTIMATION OF CITRUS FRUIT BY MACHINE VISION AND CITRUS GREENING DISEASED FRUIT DETECTION USING SPECTRAL ANALYSIS By JUNSU SHIN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL F ULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2012

PAGE 2

2 2012 Junsu Shin

PAGE 3

3 To my loving wife for all her support

PAGE 4

4 ACKNOWLEDGMENTS I would like to thank my major advisor, Dr. Won Suk Daniel Lee, for his support and encouragement, both in my research and in my life. I would also like to give thanks to my other supervisory committee members, Dr. Reza Ehsani and Dr. Arunava Banerj ee for their advice and suggestions during my research. I would also like to thank CREC staffs, Lioubov Polonik and Cindy Basnaw, for helping me collect c itrus fruit samples My family and friends were a great help and sustained through the good and bad times of this educational journey. Above all however, I would like to thank my wonderful and caring wife. I would surely not have pursued and finished graduate study without her by my side.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 8 LIST OF ABBREVIATIONS ................................ ................................ ............................. 9 AB STRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 GENERAL INTRODUCTION ................................ ................................ .................. 12 Background ................................ ................................ ................................ ............. 1 2 Precis ion Agriculture ................................ ................................ ............................... 12 Citrus Harvesting ................................ ................................ ................................ .... 13 Citrus D ebris C leaning M achine ................................ ................................ ............. 13 Citrus Greening Disease ................................ ................................ ......................... 14 2 POSTHARVEST CITRUS MASS AND SIZE ESTIMATION ................................ ... 16 Introduction ................................ ................................ ................................ ............. 16 Objective ................................ ................................ ................................ ................. 19 Materials and Methods ................................ ................................ ............................ 19 H ardware S ystem for M achine V ision ................................ .............................. 19 Software D esign and A lgorithms ................................ ................................ ...... 20 Image acquisition and pre processing ................................ ....................... 21 Pixe l classification using logistic regression model ................................ .... 23 Morphological operations and filtering ................................ ....................... 24 Highly saturated area recovering ( HSAR) ................................ .................. 26 Mass calibration ................................ ................................ ......................... 27 Fruit separation using H minima transform based watershed transform .... 28 Fruit diameter estimation and mass estimation ................................ .......... 30 Results and Discussion ................................ ................................ ........................... 30 Image P rocessing and A nalysis ................................ ................................ ....... 30 Mass C alibration and E stimation ................................ ................................ ...... 32 Fruit S ize E stimation and C ounting ................................ ................................ .. 34 Mass E stimation B ased on the E stimated F ruit D iameter ................................ 36 Conclusion ................................ ................................ ................................ .............. 37 3 SPECTRAL ANALYSIS AND IDENTIFI CATION OF HLB INFECTED CITRUS FRUIT ................................ ................................ ................................ ..................... 40

PAGE 6

6 Introduction ................................ ................................ ................................ ............. 40 Objective ................................ ................................ ................................ ................. 42 Materials and Methods ................................ ................................ ............................ 43 HLB A ssociated C haracteristics of C itrus P eel ................................ ................. 43 Fruit C ollection and S pectral M easurement ................................ ...................... 44 Data A nalysis and F eature S election ................................ ............................... 45 Spectral derivative analysis ................................ ................................ ........ 45 Discriminability analysis ................................ ................................ ............. 45 Stepwise discriminant analysis ................................ ................................ .. 47 Classification ................................ ................................ ................................ .... 47 Logistic regression ................................ ................................ ..................... 48 Linear Support Vector Machines ................................ ................................ 48 Results and D iscussion ................................ ................................ ........................... 50 Spectral R eflectance and its F irst D erivative ................................ .................... 50 Data A nalysis and F eature S election ................................ ............................... 51 Di scriminability ................................ ................................ ........................... 51 Determination of optimal wavelengths ................................ ....................... 52 Classification ................................ ................................ ................................ .... 53 Conclusion ................................ ................................ ................................ .............. 55 4 SUMMARY AND FUTURE WORKS ................................ ................................ ....... 56 LIST OF REFERENCES ................................ ................................ ............................... 58 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 61

PAGE 7

7 LIST OF TABLES Table page 2 1 Field experiment summary: measured fruit mass and the number of images acquired. ................................ ................................ ................................ ............. 22 2 2 Results of regression analysis on the three mass calibration sets. ..................... 32 2 3 Summary of the field experiment results. ................................ ........................... 33 2 4 Potential fruit counting and diameter distribution. ................................ ............... 35 2 5 Results of regression analysis between the mass and the diameter of fruit sample in the calibration sets. ................................ ................................ ............ 37 2 6 Summary of the mass estimation results based on fruit diameter. ..................... 37 3 1 Summary of f ruit diameter measurement s ................................ .......................... 44 3 2 Candidate wavelengths ................................ ................................ ...................... 52 3 3 Optimal wavelengths chosen by stepwise discriminant analysis ........................ 53 3 4 Classification accuracy for the two classification models ................................ .... 54

PAGE 8

8 LIST OF FIGURES Figure page 1 1 Schematic diagram of citrus debris cleaning machine. ................................ ....... 14 1 2 Healthy and HLB infected citrus fruit ................................ ................................ 15 2 1 Machine v ision hardware setup ................................ ................................ ......... 21 2 2 Image processing algorithm block diagram. ................................ ....................... 21 2 3 Histograms of fruit and non fruit samples ................................ ......................... 25 2 4 Problem of filling holes operation ................................ ................................ ...... 26 2 5 Highly saturated area recovering (HSAR) algorithm ................................ ......... 27 2 6 H minima transform based watershed segmentation results with several h values ................................ ................................ ................................ ................ 29 2 7 Summari z ing the image processing results ................................ ....................... 31 2 8 Result of regression analysis between the measured fruit mass and the estimated fruit mass. ................................ ................................ .......................... 34 2 9 Fruit separation result with watershed transfo rm ................................ ............... 35 3 1 Citrus fruit samples ................................ ................................ ........................... 44 3 2 R eflectance data from two healthy and two HLB infected citrus fruit .................. 50 3 3 First derivative reflectance from two healthy and two HLB infected citrus fruit ... 50 3 4 Discriminability of the original reflectance data ................................ ................... 51 3 5 Discriminability of the first derivative ................................ ................................ ... 52 3 6 Selected wavelength points near local maxima or minima ................................ 53

PAGE 9

9 LIST OF ABBREVIATION S AYMS Automated Yield Monitoring System ECHO Extraction and Classification of Homogenous Objects GIS G eographic I nformation S ystem GPS Global Positioning System HLB Huanglongbing or citrus greening HSAR Highly Saturated Area Recovering HSV Hue, Saturation and Value NASS National Agriculture Statistics Survey PCA Principal Components Analysis PCR Polymercase Chain Reaction PDF Probability Density Function R 2 Coefficient of determination ROI Region of Interest RMSE Root Me an Square Error RGB Red, Green and Blue SSE Error sum of squares SVM Support Vector Machines YCbCr Luminance, chrominance in blue and chrominance in red YIQ Luminance, in phase chrominance and quadrature chrominance

PAGE 10

10 Abstract of Thesis Presented to the Grad uate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Engineering MASS AND SIZE ESTIMATION OF CITRUS FRUIT BY MACHINE VISION AND CITRUS GREENING DISEASED FRUIT DETECTION USING SPECT RAL ANALYSIS By Junsu Shin December 2012 Chair: W on Suk D aniel L ee Major: Agricultural and Biological Engineering Citrus is the major fruit crop in Florida. C itrus industry occupies a significant portion of Florida s agricultural economy. There have been many efforts to minimize producing costs improve productivity and increase profit. Precision agriculture emerged as a solution to such efforts. A machine vision based imaging and a visible near infrared spectroscopy are examples of precision farming technology widely used in agricultural sectors. A machine vision system for estimation of citrus fruit mass, fruit count, and fruit size during post harvest processing was investigated towards the development of an advanced citrus yield mapping system. Suc h yield mapping system enables the citrus growers to efficiently manage the in grove spatial variability factors such as: soil type, soil fertility, moisture content, etc., and can help increase yield and profits. Thus, a machine vision system was develope d and installed in a citrus debris cleaning machine, which removes debris from mechanically harvested loads. An image processing algorithm was developed to identify fruit from images of the postharvest citrus from a commercial citrus grove. For fruit detec tion, logistic regression model based pixel

PAGE 11

11 classification algorithms were developed. A mass calibration process was conducted, and fruit mass was estimated, which turned out to be reasonably good. The highest coefficient of determination ( R 2 ) value betwee n the measured fruit mass and the estimated fruit mass was observed to be 0.9 45 and the root mean square error was 116.1 kg. A H minima transform based w atershed algorithm was used to separate the joined fruit and enabled an estimation of fruit counting an d fruit size. Fruit mass estimation using the fruit size information was also conducted and these results were compared with that of the mass estimation based on fruit pixel area. This research further explored the application of visible near infrared spec troscopy for HLB detection in citrus fruit. In the study, the possibility of identifying HLB disease in citrus fruit using spectroscopy was investigated in a laboratory setup. C itrus fruit samples (101 healthy and 101 HLB infected) were collected from a ci trus grove in Lake Alfred, Florida during June and July 2012. Spectral reflectance (400 to 2500 nm) of the fruit samples were measured using a spectrophotometer. The reflectance and its first derivative were analyzed using discriminability analysis and the candidate wavelengths were sele cted Wavelength features for classification were chosen by stepwise discriminant analysis. Logistic regress ion model and linear support vector machines (SVM) were used to classify HLB infected citrus fruit. Both models yiel ded more than 9 5 % overall accuracy when trained with the first derivative. The classification results indicated that the first derivative data contained more discriminate features than the original reflectance.

PAGE 12

12 CHAPTER 1 GENERAL INTRODUCTION Background F lorida is the primary citrus producing state in the United States, supplying over 80% of the total citrus produced in the country. C itrus industry remains a major part of Florida s agricultural economy. The citrus industry generates more than $9 billion in economic activity in Florida. However, the Florida citrus industry is currently under pressure from low priced Brazilian imports. Brazil with cheap labor and abundant land surpassed Florida years ago as the world s top citrus grower and provider of concen trated orange juice. The increase in profitability of citrus production has been an issue for citrus growers in Florida to be competitive in the market. Precision farming emerged as a solution to improve yields and profit s Precision Agriculture Precision agriculture as called site specific management is a technology to achieve the improvements in productivity, efficiency and quality for citrus production. The introduction of precision farming into crop production was made by the integration of a number of information management technologies. These technologies include yield monitoring, remote sensing, geographic information system (GIS), G lobal P ositioning S ystem (GPS) and variable rate application. Through the precision agriculture citrus growers are able to identify the level of in grove spatial variability, such as yield, tree size, soil type, soil fertility, water content, and many other factors that affect the productivity. Yield mapping is a valuable tool to manage such spatial variabilit y and to impl ement site specific crop management. Citrus mass estimation is an important factor in predicting citrus yield map Since manual measurement of fruit mass of individual

PAGE 13

13 citrus trees is time consuming and laborious, indirect yield estimation techniques are r equired Citrus Harvesting The common method of citr us harvesting is hand harvesting but the hand harvesting is a labor intensive task involving large number of workers depending on the grove size. In order to improve production and decrease costs associa ted with hand harvesting, a cost effective mechanical harvesting machine has been developed and used. Its usage has been increased during the past several years. One of the mechanical harvesting machines commonly used in the fields is a canopy shake and ca tch harvester. Although mechanical harvester brings many benefits to citrus growers, the harvester still has its drawbacks. Mechanical removal of leaves, twigs and bra n ches along with fruit during harvesting result s in more debris being delivered to proces sing plants. Debris should be separated from fruit at a later stage. Citrus D ebris C leaning M achine A prototype for a citrus debris cleaning machine was developed to filter out debris in the grove immediately after harvesting by a mechanical harvester. Th e machine is mainly composed of a hopper, a de trasher, load cells and a conveyor belt. The fruit and debris are unloaded from a truck named goat as shown in Fig ure 1 1 Manual opening and closing of a gate installed underneath of the hopper control s t he feeding of fruit and debris in to the de trasher. The de trasher consists of a set of pairs of pinch rollers rotating opposite directions, and filters out leaves and twigs which are collected underneath the de trasher as the fruit and debris pass through the de trasher. At the end of the conveyor belt, the cleaned fruit load without any debris is transported back to another empty truck. Load cells were used to measure the mass of the material loaded

PAGE 14

14 in the hopper. The load cells were located on the four corners of the hopper. The measured mass i s displayed on digital screen (Model 715, Avery Weigh Tronix, Fairmont, MN USA ). The mass of the collected debris is measured using a weighing scale (XI 60K, IP 65, Denver Instrument, Bohemia, NY USA ). The fruit mass is determined by subtracting t he mass of the collected debris from the mass measured by the load cells Fig ure 1 1. Schematic diagram of citrus debris cleaning machine Citrus Greening Disease However, in recent years the citrus industry has been threatened by citrus greening disease also known as Haunglongbing (HLB). Haunglongbing is a destructive and rapidly spreading disease of citrus. The diseased tree will decline in its health and life time. Fruit from the infected trees are small and lo psided in shape and taste bitter. As shown in Figure 1 2, the shape of healthy citrus fruit is symmetric, whereas HLB infected fruit has non symmetric shape. Goat Truck De Trasher Hopper End Conveyor Belt Housing for Machine Vision System

PAGE 15

15 A B Figure 1 2 Healthy and HLB infected citrus fruit A) h ealthy f ruit B) HLB infected fruit Since there is no cure once a tree becomes infected, the spread of HLB is prevented only by removing the infected trees. The disease damages not only economic value of fruit, but also the whole citrus industry. Si nce symptoms of HLB disease resemble those of nutrient deficiencies such as iron or zinc deficiency the identification of HLB infected trees and fruit is a difficult task only depending on field observations.

PAGE 16

16 CHAPTER 2 POSTHARVEST CITRUS M ASS AND SIZE ESTIMAT ION Introd uction Florida is the primary citrus producing state in the United States, supplying over 80% of the total citrus produced in the country. The increase in profitability of citrus production has been an issue for citrus growers to be competitive in the mark et. Precision farming is a technology to achieve the improvements in productivity, efficiency and quality for citrus production. Through this technology, citrus growers are able to identify the level of in grove spatial variability, such as yield, tree siz e, soil type, soil fertility, water content, and many other factors that affect the productivity. Yield mapping is a valuable tool to manage such spatial variabilit y and to implement site specific crop management. Image processing based machine vision tec hnology has been employed in many yield monitoring and mapping applications. The widespread use of the machine vision technology in the agricultural sector is due to its capability of recognizing size, shape, color, texture and numerical attributes of the objects (Chen et al., 2002). Recently, Aggelopoulou et al. (2011) developed an image processing based algorithm for early yield estimation in an apple orchard. The algorithm forecasts tree yield by analyzing the texture of the tree image at full bloom. Sa fren et al. (2007) presented a multistage algorithm that estimate d the number of green apples in hyperspectral images of apple trees. The algorithm utili z ed princip al components analysis (PCA) and extraction and classification of homogenous objects (ECHO) as well as machine vision techniques. Another type of vision system for fruit yield estimation was attempted by Zaman et al. (2008). They investigated the feasibility of estimating ripe blueberry fruit yield using a

PAGE 17

17 digital camera and compared it with meas urement s of fruit yield acquired by hand raking. Zaman et al. (2010) implemented an automated yield monitoring system (AYMS) utili z ing a digital color camera, differential G lobal P ositioning S ystem, custom software, and a ruggedi z ed laptop computer. They a chieved highly significant correlation between measured and predicted fruit yield ( coefficient of determination ( R 2 ) =0.99, root mean square error ( RMSE ) =277 kg ha 1 ) In addition to yield mapping and monitoring application s machine vision systems have b een studied in many other agricultural applications includ ing robotic harvesting, fruit grading and fruit defect detection. Recently a machine vision algorithm (Hannan et al., 2009) was developed to recognize oranges in various light conditions and cluste rs for automated harvesting. Bulanon & Kataoka (2010) reported machine vision based fruit detection system for robotic harvesting of Fuji apples. A n umber of machine vision systems have been developed to inspect fruit quality and characteristics. These inc lude systems for the apple defect detection (Zou et al., 2010), automated strawberry grading (Xu & Zhao, 2010), banana quality inspection (Mansoory et al., 2010), tomato classification (Laykin et al, 2002) and the defect detection in citrus peel (Blasco et al., 2007). Zou et al. (2010) proposed a three color camera based classification system, that capture d the whole surface of apple fruit, for detecting defects in the fruit by segmenting and counting regions of interest (ROIs) corresponding to fruit blemis hes. A strawberry grading system developed by Xu & Zhao ( 2010) divided fruit into four grades using the shape, size and color information obtained from an image processing technique. Mansoory et al. (2010) used Fuzzy C means segmentation algorithm to ident ify banana from image s A tomato classification (Laykin et al., 2002) system was develop ed based

PAGE 18

18 on the quality parameters such as color, shape, color homogeneity, defects and stem detection acquired from color image analysis. Blasco et al. (2007) proposed a region oriented segmentation algorithm for the defect detection in citrus peel. Unlike the pixel based segmentation which would require supervised learning and frequent training sessions, the proposed method was developed as an unsupervised learning alg orithm stressing the contrast between different objects of interest in the image. A group of researchers at the University of Florida has been working for some years on citrus yield monitoring and mapping. Annamalai et al. (2004) investigated a machine vis ion system to identify citrus fruit and to estimate fruit yield in real time. This system carried out the yield estimation based on the images on citrus trees before the harvesting. Chinchuluun et al. (2009) used machine vision to develop citrus counting a nd size measurement system for a canopy shake and catch harvester. Their machine vision system was installed on the canopy shake and catch harvester and was tested, but the vision system was not examined in a field harvesting scenario. Maja & Ehsani (2010) developed a load cell based citrus yield monitoring system for different citrus mechanical harvesting machines. The system utili z ed a GPS receiver and a mass flow sensor to create a yield map. The highest R 2 value between the true mass and the estimated m ass was 0.97 but the average percentage error was 9. 2 % for high flow rate s and 3.6% for smaller loads A nother study was conducted at the University of Florida to estimate debris mass from mechanical harvesting. Bansal et al. (2011) investigated an autom ated machine vision system for estimating debris in the citrus canopy shake and catch harvester during harvesting and reported an R 2 of 0.78 and an RMSE of 0.02 kg between the

PAGE 19

19 actual and estimated debris mass The debris materials include non citrus objec ts such as leaves and twigs which are collected along with citrus fruit during harvesting. Diseased debris is a primary source of spreading citrus diseases such as citrus canker. Hence, it is imperative to have an automated debris disposal system. In an ef fort to build such system, a citrus debris cleaning machine that collects debris postharvest was developed Previous yield monitoring systems were developed such that measurement took place during or before harvesting. This type of measurement could reduce estimation accuracy since debris or non fruit objects might be include d Hence, measurement after the removal of unnecessary objects would increase yield estimation accuracy. For such a system image acquisition after the cleaning process would be require d Objective The objective of this research was to develop a real time machine vision system for citrus mass and size estimation in the postharvest citrus debris cleaning machine To achieve fruit detection a supervised learning algorithm was developed a nd a modified version of the watershed algorithm was proposed. The fruit detection algorithms were developed such that they could form a basis for developing an advanced citrus yield mapping system in future research. Materials and Methods H ardware S ystem for M achine V ision A machine vision hardware system was developed consist ing of a CCD color camera (Bobcat GigE VGA, Imperx Inc., Boca Raton, FL USA ) (Fig. 2 1 a), two of white Exolights (MetaWhite Metaphase Technologies Inc., Bensalem, PA USA ), an incr emental encoder (CI20 CoreTech Stegmann Inc., Dayton, OH USA ) (Fig. 2c),

PAGE 20

20 and a data acquisition card (DAQCard 6036E, National Instruments, Austin, TX USA ). A camera with high frame rates (206 fps) feature was chosen for the acquisition of clear high qu ality images. The size of the images acquired from the camera was 640 480 and the resolution was 0.78 1.16 mm pixel 1 To synchroni z e with the conveyor belt for image acquisition, an incremental encoder was installed on a rotating axis of the conveyor. The average speed of the conveyor belt was 0.67 m sec 1 The above hardware was installed in a citrus debris cleaning machine described previously The cleaning machine was operated by a power take off from the tractor and the tractor battery was used as a p ower source. In order to remove the effect of variations in sunlight, a housing that covers the camera and the two lights was designed (Fig. 2 1 b) The housing box was made of cardboard. The appropriate height of the camera was determined such that the ent ire area inside the housing could be acquired, and high quality images could be obtained with uniform illumination. For the uniform light distribution, high quality light was chosen which could provide shadow free illumination and diffused light. All hard ware setups are shown in Fig ure 2 1 S oftware D esign and A lgorithms An algorithm to estimate citrus mass based on machine vision was designed includ ing image rectification, image segmentation based on logistic regression model, morphological operations, hi ghly saturated area recovering (HSAR) and mass calibration algorithms. These are explained individually in detail later in the section. The block diagram representing the flow of the image processing algorithm is shown in Fig ure 2 2 The calculations invol ved in the image processing algorithm were implemented in MATLAB Version R2010b (The MathWorks Inc, Natick, MA, USA).

PAGE 21

21 A B C Fig ure 2 1 Machine vision hardware setup A ) CCD camera installed on the machine B ) h ousing cover and C ) e ncoder Fig ure 2 2 Image processing algorithm block diagram Image acquisition and pre processing The machine vision system acquired images of citrus fruit moving over the conveyor belt of the de trasher in the cleaning machine. Si nce the vision system wa s located at the end of the de trasher, the citrus debris was filtered by the de trasher and was most unlikely included in the images captured by the system. In order to avoid missing and overlapping between sequential images, the i ncremental encoder was Image acquisition Image rectification Fruit detection and seg mentation using logistic classifier Morphological operations & filtering Total pixel area calculation Highly saturated area recovering (HSAR) H minima transform based watershed separation Camera Housing Encoder

PAGE 22

22 installed on the rotating axis of the conveyor to synchronize with the speed of the conveyor belt in the de trasher. The images were recorded in the field and were post processed. Field experiments were conducted three times at a com mercial citrus grove (Lykes Bros. Inc., Fort Basinger, FL USA ). The experiments took place on May 19, May 31, and June 14, 2011. Table 2 1 summari z es the field experiments with the measured fruit mass and number of images in each test. A total of 4176, 49 48 and 8554 valid images were obtained from the first, second and third experiments, respectively. Each experiment was divided into a number of sets and each set represents different yield amount and harvesting conditions. The first set of images in each f ield experiment was used for training and developing classification algorithms. From the training image sets, fruit and non fruit sample images were manually cropped and assembled separately. Then, those cropped images were used for fruit and non fruit pix els sampling. Table 2 1 F ield experiment summary : measured fruit mass and the number of images acquired. Set number Measured fruit mass (kg) Number of images 1st experiment 1 1,721.4 1132 2 587.4 545 3 739.4 599 4 1,984.5 1900 2nd experiment 1 1 ,492.3 2084 2 979.8 917 3 1,628.4 1947 3rd experiment 1 2,004.9 2640 2 1,217.9 1787 3 1,510.5 2394 4 1,614.8 1733 The images captured from the camera we re not available for the direct use due to the distortion from the camera lens They ha d to be rectified by means of the camera

PAGE 23

23 calibration process. T o rectify images, models for both the camera s geometry and lens distortion wer e derived. These two models along with custom software we re used to correct intrinsic deviations and lens distortions. The software program wa s written using OpenCV C++ library (Bradski & Kaehler, 2008). All of images taken during the field tests were rectified using this program. This rectif ication process can be plugged in to the algorithm proposed in this research such that all the processing can be done in real time. Pixel classification using logistic regression model Classifying pixels into fruit or non fruit is regarded as the binary classification problem. For the binary classification, logistic regression model is utilized. Logistic regression is quick to train and easy to implement. In addition, the model runs rapidly so it is suitable for real time processing. The logistic regression model is defined by Eq. 2 1 ( 2 1) where (2 2 ) A function in Eq. 2 2 is the logistic sigmoid function (Bishop, 2006). The variable in Eq. 2 1 represents the feature vector. A weight vector represented by the variable is determined by the gradient ascent rule satisfying maximum likelihood condition. The outcome of this pixel classification is in the form of a binary image. The value zero (0) indicates a black pixel, and the value one (1) represents a white pixel in the binary output image. T he white pixel region denotes where fruit resides in an image, but the black pixel area denotes background (non fruit).

PAGE 24

24 To find distinctive feature vectors for the classification, the training images were converted from red, green, and blue (RGB) color space to various type of color spaces, such as h ue, saturation, and value (HSV) ; luminance, in phase chrominance and quadrature chrominance (YIQ) ; and luminance, chrominance in blue, and chrominance in red (YCbCr). The histogram analysis was performed in each color space. As shown in Fig ure 2 3 fruit and non fruit pixels occupy separate places with little overlapping in the histogram of hue ( H ) saturation ( S ) chrominance in blue ( Cb ) and chrominance in red ( Cr ) colo r spaces. Hence, these four color components we re chosen as the feature vector. It is noted that more color feature vector could have been chosen, but then the feature data would have contained redundant data. The feature vector is described as below. ( 2 3) Morphological operations and filtering Morphological operations including erosion, dilation and opening were applied to make a correction on segmentation errors and to remove noise from the segmented image. For the morphological operations, a disk shaped structural element of a size of three pixels was used This size was chosen empirically. Also, the geometrical information on the segmented part such as the ratio of major axis length to minor axis length was used to filter out the false segm entation. Assuming that a single fruit has an

PAGE 25

25 ellipse shape, the major axis and the minor axis are defined as the longest and shortest diameter of the ellipse, respectively. A B C D Fig ure 2 3 Histograms of fruit and non fruit samples A ) hu e B ) saturation C ) chrominance in blue and D ) chrominance in red After morphological operations, some holes remained in the segmented image which should be filled. As an easy trial, a filling holes operation could be used to fill the m However, the pro blem with the filling holes is that it incorrectly fills void spaces surrounded by fruit as well (Fig ure 2 4 ). It was observed that most of the holes remained in the segmented image due to the highly saturated area on the surface of fruit. A new algorithm which is explained later in the highly saturated area recovering

PAGE 26

26 section was developed to recover the highly saturated area. With this algorithm, the error due to the filling holes operation can be avoided. A B Fig ure 2 4 Problem of filling hol es operation A ) t est image with void space that should not be affected by filling holes operation B ) s egmented image showing the problem of filling holes operation Highly saturated area recovering (HSAR) Some part s of the fruit image and the background (non fruit) image we re highly saturated due to the light emitted from the lamps. The highly saturated areas may cause an error in the classification process and h ence th ey were excluded from the training sample for the logistic regression model. This mean s that the classification model does not identify the very bright areas on fruit in an image as fruit. Therefore, a h ighly saturated area recovering (HSAR) algorithm wa s developed to detect and recover highly saturated areas surrounded only by fruit pixels Fig ure 2 5 shows an example of this algorithm. The steps involved in the HSAR algorithm were: 1) Find all highly saturated areas by the thresholding operation (Fig ure 2 5 ( C )). 2) Extract pixels in circumference around the areas found in step (1) by the combina tion of dilation and logical AND operation (Fig ure 2 5 ( D )). 3) Look up the extracted pixels and see if they are part of fruit pixels using the fruit color (Fig ure 2 5 ( E )). 4) If they are fruit pixels, add the identified areas to the classification result by log ical OR operation (Fig ure 2 5 ( F )). This space should not be fi lled. This space should not be filled.

PAGE 27

27 The detected region is added to the classification result so that entire fruit pixels are found. The red rectangle in Fig ure 2 5 ( B ) indicates the highly saturated regions. Those regions are not categori z ed as fruit in the classification step. Later, those are recovered by HSAR algorithm as shown in Fig ure 2 5 ( F ). A B C D E F Fig ure 2 5 Highly saturated area recovering (HSAR) algorithm A ) o riginal test image B ) s egmented binary image without HSAR C ) c andidate highly saturated areas D ) p ixels around the areas E ) a ctual highly saturated areas and F ) r ecovered highly saturated area. Mass calibration While conducting each of the field experiments, a total of 40 fruit samples with varying sizes and mass es were taken in order to calibrate the pixel area of fruit with respect to actual mass. The pixel area for each fruit sample was found out from the Not classified as fruit Recovered by HSAR

PAGE 28

28 binary images obtained from manual cropping using an image editing software (GIMP, GNU Image Manipulation P rogram). The mass of the individual fruit sample was measured using a weighing scale (Adventurer, Ohaus Corporation, Pine Brook, NJ USA ). A regression analysis was conducted to find a relationship between pixel area and actual mass. A linear model was ass umed in the analysis. Hence, the model has the form of Eq. 2 4. ( 2 4) Fruit separation using H minima transform based w atershed transform In order to count the number of fruit and to estimate the fruit diameter, neighboring fruit which joined together in the output binary image need to be separated T o separate the touching fruit into individual fruit s, a w atershed transform ation wa s conducted on the inverse distance transform of the complement of the output binary images which w ere obtained from the image processing algorithm. However, it should be noted that the w atershed separation yields over segmentation results because every local minimum forms its own catchment basin which comprises one segmented area after the transform. T o minimi z e the over segmentation effect, local minima that are too shallow are eliminated using H minima transform (Jung & Kim, 2010). The H minima transform is a powerful tool to suppress local minima whose depth is lower than a given threshold constant h The H minima transform is defined by Eq. 2 5. The operator in the Eq. 5 represents the morphological reconstruction by erosion of Here, denotes the inverse d istance map of the binary image. ( 2 5)

PAGE 29

29 Fig ure 2 6 depicts the results of H minima transform based w atershed segmentation with several different h values. As shown in Fig ure 2 6 the number of segmented regions is changed directly by the constant h value. As the constant h increases, the number of the segmented regions decreases. A B C D E F G Fig ure 2 6 H minima transform based w atershed segmentation results with several h values A ) o riginal image B ) b inary image C ) i nverse distance map, D ) h = 0, E ) h = 2, F ) h = 20, and G ) h = 30

PAGE 30

30 Fruit diameter estimation and mass estimation When the calibration image sets were acquired, the diameter of each fruit sample was measured using a digital cal l iper. Based on th e diameter measurement, the diameter of the segmented fruit in image can be estimated. Since the measurement was conducted on only the second and the third experiments, the diameter estimation can be performed on only those two experiments. The calibration sets for the second and the third field experiments include the mass and diameter of the individual fruit sample s Using an equation obtained from a regression analysis, the estimated diameter can be mapped to fruit mass. The mapping equation has the form of Eq. 2 6 ( 2 6 ) Results and Discussion Image P rocessing and A nalysis The main finding of this work is the development of an image processing algorithm to perform the detection of citrus fruit in an image to estimate fruit mass. Pixel area corresponding to fruit was computed based on the binary image obtained from the image processing algorithm. The c ore part of the image processing algorithm is the logistic regression model based image segmentation, designed for classifying pixe ls as fruit or non fruit. Fig ure 2 7 summari z es the whole process for the segmentation. Pixels in highly saturated region of fruit were not categori z ed as fruit pixels by the logistic classification model as shown in Fig ure 2 7 (c) since they were not consi dered as fruit pixels in the classification model training. Fig ure 2 7 (d) show s the result image after the morphological operations and filtering. In the step shown in Fig ure 2 7 (e), the highly saturated regions were recovered by the HSAR algorithm so that the whole regions

PAGE 31

31 representing fruit in an image were detected. Fig ure 2 7 (f) depicts the result of the fruit separation using H minima transform based w atershed transform. A B C D E F Fig ure 2 7 Summari z ing the image processing results A ) o riginal image B ) r ectified image C ) s egmented image using logistic regression model D ) image a fter morphological operations and filtering E ) image a fter HSAR and F ) image after H minima transform based w atershed separation Execution time of the image processing algorithm written in MATLAB for a single image ranged between 0.512 and 0.676 s The processing time could be reduced

PAGE 32

32 significantly if the algorithm is implemented in machine level programming language such as C and C++. Most of the errors found in the segmentation procedure we re due to regions that share similar color characteristic s with fruit. The HSAR algorithm was developed in an effort to avoid those errors, but it detects only the highly saturated areas, which are very bright regions However, dark colored regions can cause the segmentation errors as well as the very bright regions. Some fruit ha d dark colored skin. Th e se we re very hard to distinguish from the dark colored non fruit regions, such as the image of the worn out floor of the conveyor belt. Thus, the unwanted area could be classified as fruit pixels. This would result in a false positive classification error in the mass estimation step. Mass C alibration and E stimation Table 2 2 shows the results of regression analysis on th e mass calibration sets obtained from the three experiments. The constants and in Table 2 2 are define d in Eq. 2 4. These two constants we re used in mapping pixel area to estimated mass. Table 2 2 Results of regression analysis on the three calibration sets Experiment number Error sum of squares (SSE, kg) Coefficient of determination ( R 2 ) Root mean square error (kg) 1 0.0055 0.982 0.0121 0.0000636 0.0892 2 0.0322 0.924 0.0291 0.0000686 0.0273 3 0.0303 0.929 0.0282 0.0000718 0.0659 The lowest coefficient of determination ( R 2 ) values were observed to be above 0.92 between the measured fruit mass and the pixel area, which implies that the fruit mass can be estimated from the pixel area with high degree of accuracy. The image processing algorithm was applied to all images acquired from the field experiments and binary images were generated. Then, the entire pixel area

PAGE 33

33 corresponding to citrus frui t in the experiment set was computed. The sum of pixel area was then mapped to estimated fruit mass using Eq. 2 4 Table 2 3 summari z es the estimation results. Table 2 3 Summary of the field experiment results Set Measured fruit mass (kg) Fruit pixel a rea (pixels) Estimated fruit mass (kg) Measured mass Estimated mass (kg) *Error (%) 1st experiment 1 1,721.4 25,105,593 1,597.6 123.8 7.2 2 587.4 10,317,474 656.5 69.1 11.8 3 739.4 12,053,362 767.0 27.6 3.7 4 1,984.5 29,128,644 1,853.7 130.8 6 .6 2nd experiment 1 1,492.3 22,631,009 1,552.1 59.8 4.0 2 979.8 12,377,930 848.9 130.9 13.4 3 1,628.4 23,265,168 1,595.6 32.8 2.0 3rd experiment 1 2,004.9 30,177,256 2,168.1 163.2 8.2 2 1,217.9 17,288,120 1,242.1 24.2 2.0 3 1,510.5 22,118,5 19 1,589.1 78.6 5.2 4 1,614.8 19,927,113 1,431.7 183.1 11.3 Regression analysis was conducted on the estimated fruit mass with respect to the measured mass obtained from the entire experiment sets. The highest R 2 between the measured fruit mass and the estimated fruit mass was 0.945. A root mean square error ( RMSE ) was 116.2 kg. Fig ure 2 8 depicts the result of the regression analysis. Here, an average of the measured fruit mass is introduced. The computed average wa s 1,407.4 kg. The RMSE (116.2 kg) wa s 8.2% of the average measured fruit mass, which means the mass was estimated reasonably well. There are several reasons exp laining this RMSE and the inconsistent errors. The encoder was used to synchroni z e with the speed of the conveyor belt of the cleaning machine, but the synchroni z ation did not operate ideally due to the sensor noise. This led to missed or

PAGE 34

34 overlapped images of the fruit. Also, the housing for blocking the sun light was another source of fruit missing error. A part of the housing partly blocked the camera view, because the cover was damaged due to it being flexible and the large amount of fruit pass ed through the housing. Thus, it introduced some error in capturing images. Fig ure 2 8 Result of regression analysis between the measured fruit mass and the estimate d fruit mass F ruit S ize E stimation and C ounting After only fruit regions were extracted from the image processing algorithm, the w atershed transform was applied to separate joined fruit into individual f ruit s The acquisition of individual fruit images enabled the number of fruit to be counted and the diameter of fruit to be estimated The w atershed transformation generated incorrect separation which led to over segmentation since the regional minima were utili z ed directly for separating the touching fruit. The excessive over segmentation in the w atershed separation was prevented using H minima transform. The appropriate

PAGE 35

35 constant h value was chosen empirically for the best segmentation. Fig ure 2 9 show s th e results of the w atershed separation with and without H minima transform. Table 2 4 summari z es the potential distribution of the fruit size and counting. As shown in Table 2 4, a majority of fruit had diameter between 6 cm and 8 cm. The average fruit size ranged from 6.4 cm to 7.0 cm. A B C D Fig ure 2 9 Fruit separation result with w atershed transform A ) o riginal image, B ) s egmented binary image, C ) a fter w atershed transform without H minima transform, and D ) a fter w atershed transform with H minima transform. The accuracy of the estimated number of fruit cannot be verified since counting fruit manually was impossible. Rather than specify ing the number of fruit, Table 2 4 gives an idea of the relative amount s of fruit among different fruit siz es. For example, set number 4 in the 3 rd experiment had more fruits of 5 6 cm than those of 7 8 cm among all dataset s which indicates the area where these fruit were harvested produced smaller fruit than other areas. T he fruit sizes mostly varied between 6 and 8 cm, and the second most harvested citrus ranged between 5 and 6 cm, and between 8 and 9 cm. This information could also suggest in grove spatial variability of fruit sizes.

PAGE 36

36 Table 2 4 Potential fruit counting and diameter distribution Set Number of fruit Average (cm) 5~6 cm 6~7 cm 7~8 cm 8~9 cm 9~10 cm Sum 2nd experiment 1 544 2016 2484 573 52 5669 7.0 2 605 1419 1190 166 18 3398 6.7 3 1059 3347 2131 216 44 6797 6.7 3rd experiment 1 994 3257 2660 515 64 7760 6.8 2 680 2199 1377 192 24 4472 6.6 3 962 3226 1637 119 15 5959 6.6 4 1325 2943 1219 137 28 5652 6.4 Mass E stimation B ased on the E stimated F ruit D iameter Using the fruit counting and diameter distribution described in the previous section, fruit mass estimation can be achiev ed. For the mass estimation, regression analysis between fruit diameter and fruit mass was conducted. Table 2 5 shows the results of the regression analysis and the parameters ( ) which define the mapping equation. Regression analysis between the estimated fruit mass and the measured fruit mass yielded an R 2 of 0.8112 and an RMSE of 182.6 kg. Table 2 6 summari z es the results of the fruit mass estimation based on the fruit diameter. As shown in the table, the estimation error ranges between 4 % and 29%. However, the average error was 13.2%, which implies that the result was estimated reasonably well with improvement. As mentioned earlier, the source of error could i nclude the synchronization problem and the housing blocking the part of the camera view. In addition to these problems, the large errors in fruit mass estimation using fruit diameter we re due to the errors of the w atershed separation. Two of the main probl ems in the w atershed algorithm we re over segmentation and under segmentation. Even if the H minima transform was used to prevent over segmentation, over segmented fruit images still existed as a result of the

PAGE 37

37 separation. Under segmentation occurr ed when un wanted regions we re included or some grouped parts with certain shape s we re not separa ted. Table 2 5 Results of regression analysis between the mass and the diameter of fruit sample in the calibration sets Experiment Error sum of squares (SSE, kg) Coefficient of determination ( R 2 ) Root mean square error (kg) 2 0.0179 0.958 0.0217 0.140 0.688 3 0.0222 0.947 0.0242 0.144 0.713 Table 2 6 Summary of the mass estimation results based on fruit diameter Set Measured fruit mass (kg) Estimated fruit mass (kg) Measured mass Estimated mass (kg ) Error (%) 2nd experiment 1 1,492.3 1,580.9 88.6 5.9 2 979.8 791.1 188.7 19.3 3 1,628.4 1,543.0 85.4 5.2 3rd experiment 1 2,004.9 1,926.0 78.9 3.9 2 1,217.9 1,018.4 199.5 16.3 3 1,510.5 1,314.7 195.8 12.9 4 1,614.8 1,152.3 462.5 28.6 Conclusion A machine vision system for citrus mass and size estimation during postharvesting was designed and implemented. The main hardware components of the machi ne vision system, such as the high frame rate camera, the lightning devices, the incremental encoder and the data acquisition card, were chosen for the acquisition of high quality images. In software system, an image processing algorithm capable of detecti ng and segmenting citrus fruit in an image was developed. For implementing more accurate image segmentation, a HSAR algorithm was developed. With the HSAR algorithm, the errors due to the filling holes operation were avoided. For the fruit size estimatio n and counting, H minima transform based w atershed transform was utili z ed. By choosing the

PAGE 38

38 appropriate constant h value, the w atershed separation minimi z ed over segmentation. Finally, the equations mapping the pixel area to the estimated mass were establis hed using the calibration sets obtained from the field experiments. The system was tested on a citrus debris cleaning machine at a commercial citrus grove. Images taken during the field experiments were converted to binary images using the developed image processing algorithm. The fruit mass, the number of fruit and the fruit diameter were estimated based on the output binary images generated from the image processing algorithms. The mass estimation was conducted in two ways: pixel area based estimation an d diameter based estimation. The pixel area based method yielded more reasonable result than the diameter based method. When using the pixel area based method, the highest coefficient of determination ( R 2 ) between the measured fruit mass and the estimated fruit mass was 0.945. Also, root mean square error ( RMSE ) was 116.2 kg. The RMSE was 8.2% of the average measured fruit mass. On the other hand, the diameter based method yielded an R 2 of 0.811 and an RMSE of 182.6 kg The research described in this study was conducted as a preliminary test towards an ultimate goal of developing a n advanced real time yield mapping system. Although the proposed system does not provide very site specific yield information such as tree by tree yield based on GPS coordinates yield information for different sections of tree rows is available under the current system configuration. T he system needs further improvements. In this study the pixel based segmentation was used to detect the fruit area in an image. Since this method makes use of only color information of a pixel, it is vulnerable to illumination change due to the

PAGE 39

39 outdoor condition and shadow between touching or occluded objects. Hence, a region based algorithm should be studied to ease such vulnerability. Incorporatio n of both pixel based and region based algorithms may perform better than the individual use of both algorithm s.

PAGE 40

40 CHAPTER 3 SPECTRAL ANALYSIS AND IDENTIFICATION OF HLB INFECTED CITR US FRUIT Introduction Citrus is the most important fruit crops in Florida a s it is the primary citrus producing state in the United States, supplying over 63 % of the total citrus produced in the country (NASS, 2012 ) C itrus industry remains a major part of Florida s agricultural economy. The citrus industry generates more than $9 billion in economic activity in Florida. However, recently it has been adversely affected by citrus greening disease also known as H u a nglongbing (HLB) H u a nglongbing is a serious disease of citrus and some citrus relatives. T he diseased tree will decline in its health and life time Fruit from the infected trees are small and lopsided in shape and taste abnormally bitter Hence the disease unfavorab ly affects the quality of juice Since there is no cure once a tree becomes infected, the spread of HLB is p revented only by removing the infected trees. The citrus grower s do not remove HLB infected trees, but try to manage fruit quality without eliminating those infected trees. It is significantly valuable for the growers to detect HLB disease at early stage. Si nce symptoms of HLB disease resemble those of nutrient deficiencies such as iron or zinc deficiency the identification of HLB infected trees and fruit is a difficult task only depending on field observations. Current methods for detection of HLB disease in citrus plants include visual inspection by trained personnel and DNA test using Polymerase Chain Reaction (PCR) methods (Jagoueix et al. 1996 ) Visual inspection is highly subject to human error and the disease may be present for up to several years b efore symptoms are visible. PCR has proven to be the best method available for detection of HLB disease. It is the only determinant method of detecting the disease. However, it is costly and time consuming

PAGE 41

41 Recently, spectroscopy and imaging technique s ha v e been widely used in agricultural applications such as food quality inspection and plant disease detection. Their popularity is due to the fact that the spectroscopy a nd imaging techniques are fast and inexpensive Also, the techniques provide a non destr uctive way of sensing the change inside the object and it is because s pectral reflectance varies when the chemical components in the surface or subsurface of crop canopy change. For this reason, there have been many effort s to detect HLB infected citrus pl ant using spectroscopy and imaging techniques. For example, Li et al. (2012) and Kumar et al. (2010) used multispectral and hyperspectral airborne images of citrus groves to detect HLB infected trees. Airborne spectral features obtained from the multispect ral and hyperspectral images of the citrus greening were analyzed and further utilized to distinguish HLB infected trees from healthy ones Pereira et al. (2011) investigated the potential use of laser induced fluorescence imaging technique to monitor HLB disease in citrus plants. They used a diode pumped solid state blue laser at 473 nm for fluorescence excitation of citrus leaf samples. The fluorescence images were recorded with a CCD digital camera Ten color descriptors from fluorescence images were eva luated using a paired Student s t test. They reported that the descriptors yielded promising results to diagnose HLB disease. The applicability of mid infrared spectroscopy for detecting HLB disease in citrus leaves was explored (Sankaran et al. 2010 ). I n their study, spectral signature in the range 5.15 10.72 m was acquired from processed leaf samples using a portable mid infrared spectrometer. It was shown that the spectra of HLB infected citrus leaves could be distinguished from the spectra of healthy and nutrient deficient leaves using mid

PAGE 42

42 infrared spectroscopy. Mishra et al. (2012) applied visible near infrared spectroscopy identifying HLB infected citrus trees Spectral reflectance data of healthy and HLB infected citrus trees in the wavelength rang e from 350 to 2500 nm were measured using a visible near infrared spectroradiometer. Three techniques, such as k nearest neighbors, logistic regression and support vector machines, were used for the classification. They found out that using only one spectr al measurement per tree resulted in poor classification performance because of large variability in spectral reflectance of citrus canopy. They suggested that multiple measurements per tree increased the classification accuracy. Previous studies on HLB det ection using spectroscopy and imaging technique have been focused on detection of HLB in citrus trees or leaves. They have a benefit of detecting HLB at early stages of develo pment in citrus groves and assisting citrus growers to manage and control the dis ease. In addition to the HLB detection in citrus trees and leaves, it is also beneficial to identify HLB infected citrus fruit. Identification of HLB disease in citrus fruit could have a significantly positive impact in managing fruit quality In field de tection of HLB disease in early stage o f fruit development could bring a valuable tool that help s citrus growers manage the fruit quality. Such in field detection combined with GPS data could be used to build an infection map. Objective The goal of this re search was to investigate the possibility of identifying HLB disease in citrus fruit using visible near infrared spectroscopy. The specific objective s w ere to determin e opti m al wavelengths that are most responsive to HLB infected fruit and to develop a spe ctral method for HLB detection This study was conducted as a

PAGE 43

43 preliminary research with a laboratory setup before implementing an in field system with outdoor setup Materials and Methods HLB A ssociated C haracteristics of C itrus P eel Citrus peel is mainly comp rised of two layers. The outermost layer is called flavedo, whereas the inner layer of citrus peel is called albedo. The f lavedo is mostly composed of cellu losic material but also contains other components, such as essential oils, paraffin waxes, fatt y acids, pigments (carotenoids, chlorophylls, flavonoids), en z ymes and etc. Liao and Burns (2012) evaluated global gene expression in HLB infected fruit tissues using microarray containing flavedo, vascular tissue and juice vesicles from symptomatic, asym ptomatic and healthy fruit. They reported that flavedo carbohydrate content was substantially reduced in symptomatic fruit. Also, it was shown that symptomatic fruit flavedo had higher chlorophyll content and significantly lower car otenoid content than hea lthy fruit flavedo Rosales and Burns (2011) investigated carbohydrate and phytohormone alterations in HLB infected fruit. They demonstrated that starch and sucrose contents were numerically higher in immature flavedo of healthy fruit as compared with tha t of symptomatic fruit. However, mature fruit flavedo of symptomatic fruit had significantly lower starch and sucrose contents than that of healthy fruit. These chemical changes in citrus peel due to HLB infection could be identified by spectroscopy techni que since the peel is part of the light path which can affect the spectrum.

PAGE 44

44 Fruit C ollection and S pectral M easurement Citrus f ruit samples were collected from a citrus grove in Lake Alfred, F lorida during June and July 2012 The samples include d 101 healt hy citrus fruit and 101 HLB infected citrus fruit (Figure 3 1) A B Figure 3 1. Citrus fruit samples A) healthy fruit samples. B) HLB diseased fruit samples The variety of fruit was Valencia. The diameter of the samples was measured at right angle s using a digital caliper. Table 3 1 summarizes the measured diameters. As listed in Table 3 1, healthy samples have larger size than HLB diseased samples and the difference in two measurement s in right angles explains the misshapen shape of HLB infected f ruit samples as HLB samples have the greater difference than healthy samples. Table 3 1 Summary of fruit diameter measurement s Diameter (mm) Healthy samples HLB infected samples Average of the longest 48.3 46.7 Average of the shortest 48.0 46.0 Averag e of difference 0.2 8 0.6 9 Spectral data of the samples were measured using a spectrophotometer (CARY 500 UV Vis NIR, Varian Inc., Palo Alto, California, USA) Reflectance of each fruit sample was measured in the wavelength range of 4 00~2500 nm with a 1 n m increment.

PAGE 45

45 Two spectral datasets acquired from fruit samples w e re preprocessed before further analysis. For the preprocessing, m oving average filter was applied to the datasets in order to remove the short term fluctuations. Data A nalysis and F eature S el ection Spectral derivative analysis Derivative analysis is very useful in spectral data analysis since the spectral derivative could lead to the extraction of useful features from spectral data. Although high order derivative techniques have been utilized in some of previous studies (Wettle et al., Wilson et al., Hochberg and Atkinson), first and second order derivatives have been the most commonly used. In this research, the first order derivative of the original spectral data was employed. It provides inf ormation on the rate of change in reflectance with respect to wavelength. The first order derivative was approximated by finite difference and was calculated by the following equation: (3 1 ) where represents the reflectance at the wavelength of and is the difference between two consecutive wavelengths. D iscriminability analysis Since spectral reflectance was measured from 400 nm to 2500 nm with a 1 nm increment, there were 2101 spectral elements for each sample spectra. The number of variables was much larger than the number of samples. Then, it is very probable that there is high multi collinearit y within the reflectance dataset. Multic ollinearity in spectral ref lectance data means that several wavelengths are not independent of each other, but are highly correlated. This can be found easily in the case in adjacent wavelengths.

PAGE 46

46 Many of modeling methods are not straightforwardly applicable to high dimensional data with high multi collineari ty This leads to the necessity of reducing multi collinearity in the spectral data. In an effort to decrease th e multi collinearity and choose candidate wavelengths for further anal ysis, discriminability analysis was conducted The mathematical expression of the discriminability of two probability density functions (PDFs) characterized by the same standard deviation is defined by Duda et al. (2000) as: ( 3 2 ) Generally, a higher discriminability is desired as it implies a greater separation In this case, it is highly likely that the standard deviations could be different between healthy and HLB infected reflectance at the same wavelengt h. Hence, the Eq. (3 2 ) cannot be directly used. For PDFs with the different standard deviations, the following alteration to Eq. (3 3 ) was suggested (Kane and Lee, 2006) (3 3 ) Discriminability is a simple and reliable way to determine the candidate wavelengths that have potentially significant separation. These candidate wavelengths will be used as variables for building a regression model later. The candida tes are wavelengths at which discriminability is higher than a threshold value. In order to avoid

PAGE 47

47 multi collinearity between adjacent wavelengths, only local maxima or peaks with discriminability higher than threshold were chosen. The use of local maxima ca n be justified by that the local peaks could represent the adjacent wavelength group. The discriminability analysis was conducted on both the original reflectance data and the first derivative of reflectance data. Stepwise discriminant analysis A discrimin ant analysis with stepwise selection is used to assess spectral features for classification based on multiple variables and select optimal wavelength bands that best identified HLB infected citrus fruit The candidate wavelengths obtained from discriminabi lity analysis were used as input data to the stepwise discriminant analysis. The following set s of data were made to run the discriminant analysis. Set I : Wavelengths determined by the discriminability analysis on the original reflectance data. Set II : Wav elengths chosen by the discriminability analysis on the first derivative of reflectance data. Set III : Wavelengths chosen by the discriminability analysis on b oth set I and II The SAS procedure PROC STEP DISC was employed to conduct the stepwise discrimina nt analysis. Classification The datasets mentioned in the previous section were used as the input feature vectors for the classification algorithm. Logistic regression model and l inear support vector machines were used for classifying spectral features of citrus fruit samples (healthy and HLB infected fruit). The output classes computed from the classification model were the healthy and HLB infected fruit.

PAGE 48

48 Logistic regression Identification of citrus fruit as healthy or HLB infected can be defined as a binary classification problem. For the binary classification, logistic regression model is utilized. Logistic regression is quick to train and easy to implement. The logistic regression model is defined by Eq. 3 4 ( 3 4 ) where ( 3 5 ) A function in Eq. 3 5 is the logistic sigmoid function (Bishop, 2006). The variable in Eq. 3 4 represents the feature vector. A weight vector represented by the variable is determined by the gradient ascent rule satisfying maximum likelihood cond ition. The outcome of this classification is a binary number that indicates whether healthy or HLB infected The value zero (0) indicates a healthy fr uit and the value one (1) represents a HLB diseased fruit Linear Support Vector Machine s Support vector machine (SVM) was first introduced by Vapnik (1995) and originally intended to solve pattern recognition problems. Basic idea of support vector machin es is to build a classification model by mapping the data into a higher dimensional input space and constructing optimal hyperplane for linearly sep a rable patterns. Basically, SVM model was designed to solve binary classification problem in which the input data and the output classes are defined as and respectively. The hyperplane separates the positive from the negative examples in the training set. The points

PAGE 49

49 which lie on the hyperpla ne satisfy where is normal to the hyperplane. Then, is the perpendicular distance from the hyperplane to the origin, and is the Euclidean norm of This can be formulated as follows: ( 3 6 ) Two equations in Eq. ( 3 6 ) are combined into a set of inequalities as follows: ( 3 7 ) The points for which the equality in Eq. ( 3 6 ) h olds lie on both hyperplane and Hence, the margin between the two data sets pertaining to each hyperplane is simply The margin can be maximized by minimizing su bject to the constraint of Eq. ( 3 7 ). For the maximization, positive Lagrange multipliers for each of the inequality constraints in Eq. ( 3 7 ) are introduc ed The objective now is to minimize given by Eq. ( 3 8 ) with respect to the weight vector and maximize it with respect to the undetermined multipliers ( 3 8 ) The optimization problem can be formulated as follows: ( 3 9 ) This problem can be solved by standard quadratic programming techniques. Once the optimization is completed, it is determined on which side of the hyperplane a given test vector lies. In other words, it is classified to one cla ss ( 1 ) or to the other ( 1 ). The decision function is given by following equation. ( 3 10 )

PAGE 50

50 Results and D iscussion Spectral R eflectance and its F irst D erivative Figure s 3 2 and 3 3 depict the preprocessed reflectance and the fir st derivative reflectance of both two healthy and two HLB infected citrus fruit. Figure 3 2 R eflectance data from two healthy and two HLB infected citrus fruit Figure 3 3 First derivative reflectance from two healthy and two HLB infected citrus fru it

PAGE 51

51 Data A nalysis and F eature S election A whole dataset consisted of the original reflectance and the first derivatives computed from 202 citrus fruit spectra (101 healthy, 101 HLB infected). The dataset was randomly divided into two sets: training and vali dation sets. The training set included 67 healthy and 67 HLB infected, whereas the validation set contained 34 healthy and 34 HLB infected. Discriminability The discriminability of the original reflectance and the first derivative is shown in Figure s 3 4 a nd 3 5 The highest discriminability within the original reflectance data was 0.2 and the highest discriminability with the first derivative was 0.7. Comparing two graphs in Figure s 3 4 and 3 5 and the two highest value s in discriminability it is observed that the first derivative data contain more discriminate features than the original data. Figure 3 4 Discriminability of the original reflectance data

PAGE 52

52 Figure 3 5 Discriminability of the first derivative In order to select candidate wavelengths for further analysis, local peaks with discriminability higher than a threshold were computed. Threshold values for the original data and the first derivative were 0.14 and 0.25, respectively. Table 3 2 lists the selected wavelengths. Table 3 2 Candidate wa velengths Data Candidate wavelengths (nm) Original 443, 468, 491, 67 7 749, 78 1 794, 825, 85 3 887, 91 4 1056 1986, 2062, 2091, 2116, 221 1 2242 1st derivative 991, 1191, 1212, 1236, 1269, 1614, 1675, 1700, 1713 1781, 1842, 1939, 1970, 1982, 2326, 2 346 Determination of optimal wavelengths The candidate wavelengths obtained from the discriminability analysis were used to generate three different datasets as explained earlier The stepwise discriminant analysis was performed using these datasets to f ind the optimal wavelengths that could identify HLB infected citrus fruit. Table 3 3 shows the results of the stepwise

PAGE 53

53 discriminant analysis. It should be noted that the wavelengths 991 nm, 1191 nm, 1970 nm and 2346 nm are located near local minimum points while the wavelengths 1675 nm and 1842 nm are positioned close to local maximum points. Those points are drawn on the average reflectance plot of healthy and HLB diseased samples as depicted in Figure 3 6 Table 3 3 Optimal wavelengths chosen by stepwi se discriminant an alysis Set Selected wavelengths (nm) I 4 91 67 7 8 25 887, 1056, 2242 II 991, 1191, 1236, 1675, 1713, 1842, 1970, 2346 III 491, 677, 825, 887 ( the original ), 991, 1191, 1236, 1675, 1713, 1842, 1970, 2346 ( the first derivative ) Fig ure 3 6 Selected wavelength points near local maxima or minima Classification Table 3 4 summarizes the results of classification using logistic regression and linear SVM models. The table lists the classification accuracy when using three different

PAGE 54

54 datase ts. In the table, true positive accuracy is defined as the percentage of HLB infected fruit correctly identified as HLB infected and true negative accuracy is explained as the percentage of healthy fruit correctly classified as healthy. Table 3 4 Classif ication accuracy for the two classification models Accuracy Logistic Regression Linear SVM Set I (original) True positive (%) 67 79 True negative (%) 82 82 Overall accuracy 75 81 Set II (1st derivative) True positive (%) 94 100 Tr ue negative (%) 97 97 Overall accuracy 95 98 Set III (Both) True positive (%) 100 100 True negative (%) 100 100 Overall accuracy 100 100 It is seen that both logistic regression and linear SVM models trained with Set II and III achieve d more than 9 4 % accuracy in all cases. However, the accuracy with Set I was relatively lower in the two methods This confirms that the first derivative data possesses more features that can identify HLB infected fruit than the original reflectance data as previously mentioned It can be concluded that even the first derivative information (Set II) by itself proved to be an enough source of features for the classification to achieve high accuracy even though the combination of Set I and Set II yielded 100% accuracy in all cases The linear SVM demonstrated better performance given the same data than the logistic regression comparing all accuracies in Set I and II. Even so, the logistic regression made a prediction with 95% accuracy based on Set II and that is still quite good result. Both algorithms required almost same amount of computational power because p rocessing time of running one single classification for

PAGE 55

55 both algorithms was less than 4 msec Hence, the linear SVM is preferable model for classi fying HLB infected citrus fruit, because it performed better classification accuracy than the logistic regression model Conclusion As a preliminary research, the feasibility of identifying HLB disease in citrus fruit using visible near infrared spectroscopy was investigated in a laboratory setup. For this study, citrus fruit samples (101 healthy and 101 HLB infected) were collected from a citrus grove in Lake Alfred, Florida during June and July 2012. Spectral reflectance (400 to 2500 nm) of the fruit samples we re measured using a spectrophotometer. The reflectance and its first derivative were analyzed using discriminability analysis and the candidate wavelengths were selected as a result of the analysis Wavelength features for classification were chosen by ste pwise discriminant analysis. Logistic regress model and linear support vector machines were used to classify HLB infected citrus fruit. Surprisingly, both models achieved more than 9 5 % overall accuracy when trained with the first derivative. The classifica tion results indicated that the first derivative data contained more discriminate features than the original reflectance.

PAGE 56

56 CHAPER 4 SUMMA RY AND FUTURE WORK S T he main goal of the research presented in this paper was to investigate new posthar vest technique s that could be beneficial to citrus industry. The research is divided into two major parts. The first part discussed in C hapter 2 describes a machine vision based citrus mass and size estimation during post harvest ing The goal of this study was to develo p a real time machine vision system for citrus mass and size estimation in the postharvest citrus debris cleaning machine To achieve fruit detection a supervised learning algorithm was developed and a modified version of the watershed algorithm was prop osed. The system was tested on a citrus debris cleaning machine at a commercial citrus grove. Images taken during the field experiments were converted to binary images using the developed image processing algorithm. The fruit mass, the number of fruit and the fruit diameter were estimated based on the output binary images generated from t he image processing algorithms. The second part explained in C hapter 3 investigated the application of spectroscopy technique for identifying HLB infected citrus fruit. The goal of this study was to explore the possibility of detecting HLB disease in citrus fruit using visible near infrared spectroscopy. In order to find the optimal wavelengths that best distinguish HLB infected fruit from healthy ones, the discriminability analysis and the stepwise discriminant analysis were utilized. Two machine learning algorithms (logistic regression and linear support vector machines) were used to classify HLB infected fruit. The results suggested that the classification was very accurat e when using the first derivative data. The m ass estimation conducted in C hapter 2 was based on two dimensional information as it only relied on only fruit on image plane. When performing machine

PAGE 57

57 vision based mass estimation, v olumetric information along w ith fruit density could increase the estimation accuracy. It could be the possible case that even two fruit with the same size have different weight. It is suggested that stereo vision could achieve better estimation since it enables us to extract depth in formation out of images. In order to improve the machine vision based mass estimation, more efforts need to be made on resolving problems such as heating, light source control, juice extraction and citrus debris. In the field experiment, heating decrease d the performance of the machine vision system. It was significantly important to m aintain consistent light condition and it will enhance the output of a machine vision application. Juice and citrus debris were the factors that made the fruit detection mor e difficult. Solving those problems would augment the result of the mass estimation. In Chapter 3, the identification of HLB infected citrus fruit using spectroscopy was implemented in a laboratory setup. This study showed a potential use of spectral infor mation for identifying HLB infected citrus fruit. It is not hard to predict that future research will be to implement an in field system capable of identifying HLB diseased citrus fruit using spectral measurement system such as hyperspectral or multispectr al camera. Under field conditions in a citrus grove, sunlight variation and other environmental factors could be obstacles to be overcome unlike the laboratory conditions.

PAGE 58

58 LIST OF REFERENCES Aggelopoulou, A. D., Bochtis, D., Fountas, S., Swain, K. C., Ge mtos, T.A., & Nanos, G.D. (2011). Yield prediction in apple orchards based on image processing. Precision Agriculture 12, 448 456. Annamalai, P., Lee, W. S., & Burks, T. (2004). Color vision system for estimating citrus yield in real time. ASAE Paper No. 043054. ASABE St. Joseph. MI. Bansal, R., Lee, W. S., Shankar, R., & Ehsani, R. (2011). Automated trash estimation in a citrus canopy shake and catch harvester using machine vision. Applied Engineering in Agriculture 27(5), 673 685. Blasco, J., Aleixos, N., & Molto, E. (2007). Computer vision detection of peel defects in citrus by means of a region oriented segmentation algorithm. Journal of Food Engineering 81, 535 543. Bulanon, D. M., & Kataoka, T. (2010). Fruit detection system and an end effector for robotic harvesting of Fuji apples. Agricultural Engineering International: CIGR Journal 12(1), 203 210. Bishop, C. M. (2006). Pattern recognition and machine learning New York, NW: Springer. Bradski, G., & Kaehler, A. (2008). Learning OpenCV Computer vision with the OpenCV library Sebastopol, CA USA : O Reilly. Chen Y. R., Chao K., & Kim, M. S. (2002). Machine vision technology for agricultural applications. Computers and electronics in agriculture 36, 173 191. Chinchuluun, R., Lee, W. S., & Ehsani, R. (2009). Machine vision system for determining citrus count and size on a canopy shake and catch harvester. Applied Engineering in Agriculture 25(4), 451 458. Duda, R. O., Hart P. E., Stor k, D. G ( 2000 ) Pattern Classification. Second edition ed. New York, N.Y: John Wiley and Sons. Efron, B., Hastie, T., Johnstone, I., Tibshirani, R. (2004). Least angle regression. Annals of Statistics, 32, 407 451. Hannan, M. W., Burks, T. F., & Bulanon, D. M. (2009). A machine vision algorithm for orange fruit detec tion. Agricultural Engineering International: CIGR Ejournal 11, Manuscript 2181. Hochberg, E. J., Atkinson, M. J. (2000). Spectral discrimination of coral reef benthic communities. Coral Reefs, 19(2), 164 171.

PAGE 59

59 Jagoueix, S., Bove J. M., Garnier, M. (1996). Techniques for the specific detection of the two Huanglongbing (greening) liberobacter species: DNA/DNA hybridization and DNA amplification by PCR. 13th IOCV Conference. Jung C., & Kim C. (2010). Segmenting clustered nuclei using H minima transform based marker extraction and contour parameterization. IEEE Transactions on Biomedical Engineering 57(10), 2600 2604. Kane, K. E. and Lee, W. S. (2006). Spectral sensing of different citrus varieties for precision agriculture. ASABE Paper No. 061065. St. Joseph, Mich.: ASABE. Kumar, A., Lee, W. S., Ehsani, R., Albrigo, L. G., Yang, C., Mangan, R. L., (2010). Citrus greening disease detection using airborne multispectral and hyperspectral imaging. International Conference on Precision Agriculture. Laykin, S., Alch anatis, V., Fallik, E., & Edan, Y. (2002). Image processing algorithms for tomato classification. Transactions of the ASAE 45, 851 858. Li, X., Lee, W. S., Li, M., Ehsani, R., Mishra, A. R., Yang, C., Mangan, R. L. (2012). Spectral difference analysis and airborne imaging classification for citrus greening infected trees. Computers and Electronics in Agriculture, 83, 32 46. Liao, H. L. and Burn, J. K (2012). Gene e xpression in c itrus s inensis f ruit t issues h arvested from Huanglongbing infected t rees: c om parison with g irdled f ruit Journal of Experimental Botany http://jxb.oxfordjournals.org/content/early/2012/03/09/jxb.ers070. Maja, J. M., Ehsani, R. (2010) Development of a yield monitoring system for citrus mechanical harvesting machines. Precision Agri culture, 11, 475 487. Mansoory, M. S., Fardad, H., Enteshari, R., & Mansouri, Y. S. (2010) Isolating healthy bananas from unhealthy ones based on feature extraction and clustering method using neural network. Modern Applied Science 4(11), 51 60. Min, M. a nd Lee, W. S. (2005). Determination of significant wavelengths and prediction of nitrogen content for citrus. Transactions of the ASAE, Vol 48(2), 455 461. Mishra, A. R., Karimi, D., Ehsani, R., Lee, W. S. (2012). Identification of citrus greening (HLB) us ing a VIS NIR spectroscopy technique. Transactions of ASABE, 55 (2), 711 720. NASS. 2012. National Agriculture Statistics Survey. Pereira, F. M. V., Milori, D. M. B. P., Pereira Filho, E. R., Venancio, A. L., Russo, M. S. T., Cardinali, M. C. B., Martins, P. K., Freitas Astua, J. (2011). Laser induced fluorescence imaging method to monitor citrus greening disease. Computers and Electronics in Agriculture, 79, 90 93.

PAGE 60

60 Rosale, R., and Burns, J. K. (2011). Phytohormone changes and carbohydrate status in sweet o range fruit from Huanglongbing infected trees, Journal of Plant Growth Regulation, 30, 312 321 Safren, O., Alchanatis, V., Ostrovsky, V., & Levi, O. (2007). Detection of green apples in hyperspectral images of apple tree foliage using machine vision. Tran sactions of the ASABE 50(6), 2303 2313. Sankaran, S., Ehsani, R., Etxeberria, E. (2010). Mid infrared spectroscopy for detection of Haunglongbing (greening) in citrus leaves. Talanta, 83(2), 574 581. Thomasson, J. A., Sui, R., Cox, M. S., Al Rajehy, A. (2 001). Soil reflectance sensing for determining soil properties in precision agriculture. Transactions of the ASAE, Vol. 44(6), 1445 1453. Vapnik, V. (1995). The nature of statistical learning theory. Springer Verlag: New York, 1995. Wettle, M., Ferrier, G. Lawrence, A. J., Anderson, K. (2003). Fourth derivative analysis of Red Sea coral reflectance spectra. International Journal of Remote Sensing, 24:19, 3867 3872. Wilson, C., Matthews, F., Greasham, R. L., Will, M., Copeland, R. A. (1989). Application of fourth derivative absorption spectroscopy to protein quantitation during purification. Analytical Biochemistry, 182, 141 145. Xu L., & Zhao Y. (2010). Automated strawberry grading system based on image processing. Computers and Electronics in Agriculture 71, 32 39. Zaman, Q. U., Schumann, A. W., Percival, D. C., & Gordon, R. J. (2008). Estimation of wild blueberry fruit yield using digital color photography. Transactions of the ASABE 51(5), 1539 1544. Zaman, Q. U., Swain, K. C., Schumann, A. W., & Perciva l, D. C. (2010). Automated, low cost yield mapping of wild blueberry fruit. Applied Engineering in Agriculture 26(2), 225 232. Zou X., Zhao J., Li Y., & Holmes M. (2010). In line detection of apple defects using three color cameras system. Computers and E lectronics in Agriculture 70 129 134.

PAGE 61

61 BIOGRAPHICAL SKETCH Junsu Shin received his bachelor s degree in automotive engineering from the Kookmin University Seoul, Repulic of Korea in 1999. Then, he worked as a software engineer for several years. He m oved to the University of Florida, Gainesville, Florida, the United States to pursue his graduate studies. He completed his Master of Engi neering degree in agricultural and biological engineering in 2012.