<%BANNER%>

Volumetric Yield Monitoring System Using Laser Scanner for Citrus Mechanical Harvester

Permanent Link: http://ufdc.ufl.edu/UFE0042602/00001

Material Information

Title: Volumetric Yield Monitoring System Using Laser Scanner for Citrus Mechanical Harvester
Physical Description: 1 online resource (130 p.)
Language: english
Creator: Jadhav, Ujwala
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: citrus, harvester, laser, mechanical, volumetric, yield
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Yield monitoring system is one of the important components in precision agriculture. There is no commercial yield monitoring system available for citrus mechanical harvesters to date. Most of the large equipment manufacturer investments are focused on developing the efficiency of their machines but make little investments or none at all in the research and development of yield monitoring systems for these high value crops. A yield monitoring system using an impact plate has been developed that uses the impact of the fruit to correlate to its mass. The system performs well in both laboratory and field tests but due to the impact created by the oranges falling on the plate, the durability of the system is poor. A non-contact method of collecting yield data will be very desirable to either augment the performance of the current system or eventually replace it. A volume-based yield monitoring system was implemented as a means to estimate the mass of the fruit passing on the conveyor system on the mechanical harvester. A LIDAR (Light Detection and Ranging) technology was used to scan the cross sectional area passing on the conveyor system and send the distance related information to calculate the volume of fruits on the conveyor by integrating the data over time. The scanner used in the experimental work was a low-cost general-purpose SICK LMS-200 sensor. An algorithm was developed to analyze the data and calculate the volume of fruit passing on the conveyor. The system was calibrated and the calibration curve information was used to convert the volume into a corresponding mass. The system performance was tested in laboratory setting on two conveyor systems; one with inclined conveyor used in mechanical harvesters (which has flap height of 8 cm) and other with horizontal conveyor system used in the trash removal machine (which has flap height > = 3 cm). The tested conveyor speeds where from 0.586 m/s to 1.71 m/s. The conveyor system used on the mechanical harvester was tested for its maximum speed of 1.1 m/s and its performance was found encouraging with a coefficient of determination (R squared) of 0.98 and a RMSE was 1.69 kg while the conveyor system used on the trash removal machine performed very well with its maximum speed of 1.71 meters/sec, the coefficient of determination was 0.99 and the RMSE was 0.98 kg. The system performance was satisfactory for the complete range of the speeds used on both the conveyor systems. The results indicated the potential of using volumetric yield calculation method using LIDAR in yield monitoring of citrus in mechanical harvesters. Although this study was conducted for citrus fruits, the concept could be easily applied with little modifications to calculate the yield of other fruits and vegetables.
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 Ujwala Jadhav.
Thesis: Thesis (M.S.)--University of Florida, 2010.
Local: Adviser: Ehsani, M Reza.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042602:00001

Permanent Link: http://ufdc.ufl.edu/UFE0042602/00001

Material Information

Title: Volumetric Yield Monitoring System Using Laser Scanner for Citrus Mechanical Harvester
Physical Description: 1 online resource (130 p.)
Language: english
Creator: Jadhav, Ujwala
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: citrus, harvester, laser, mechanical, volumetric, yield
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Yield monitoring system is one of the important components in precision agriculture. There is no commercial yield monitoring system available for citrus mechanical harvesters to date. Most of the large equipment manufacturer investments are focused on developing the efficiency of their machines but make little investments or none at all in the research and development of yield monitoring systems for these high value crops. A yield monitoring system using an impact plate has been developed that uses the impact of the fruit to correlate to its mass. The system performs well in both laboratory and field tests but due to the impact created by the oranges falling on the plate, the durability of the system is poor. A non-contact method of collecting yield data will be very desirable to either augment the performance of the current system or eventually replace it. A volume-based yield monitoring system was implemented as a means to estimate the mass of the fruit passing on the conveyor system on the mechanical harvester. A LIDAR (Light Detection and Ranging) technology was used to scan the cross sectional area passing on the conveyor system and send the distance related information to calculate the volume of fruits on the conveyor by integrating the data over time. The scanner used in the experimental work was a low-cost general-purpose SICK LMS-200 sensor. An algorithm was developed to analyze the data and calculate the volume of fruit passing on the conveyor. The system was calibrated and the calibration curve information was used to convert the volume into a corresponding mass. The system performance was tested in laboratory setting on two conveyor systems; one with inclined conveyor used in mechanical harvesters (which has flap height of 8 cm) and other with horizontal conveyor system used in the trash removal machine (which has flap height > = 3 cm). The tested conveyor speeds where from 0.586 m/s to 1.71 m/s. The conveyor system used on the mechanical harvester was tested for its maximum speed of 1.1 m/s and its performance was found encouraging with a coefficient of determination (R squared) of 0.98 and a RMSE was 1.69 kg while the conveyor system used on the trash removal machine performed very well with its maximum speed of 1.71 meters/sec, the coefficient of determination was 0.99 and the RMSE was 0.98 kg. The system performance was satisfactory for the complete range of the speeds used on both the conveyor systems. The results indicated the potential of using volumetric yield calculation method using LIDAR in yield monitoring of citrus in mechanical harvesters. Although this study was conducted for citrus fruits, the concept could be easily applied with little modifications to calculate the yield of other fruits and vegetables.
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 Ujwala Jadhav.
Thesis: Thesis (M.S.)--University of Florida, 2010.
Local: Adviser: Ehsani, M Reza.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-12-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0042602:00001


This item has the following downloads:


Full Text

PAGE 1

1 VOLUMETRIC YIELD MON ITORING SYSTEM USING LASER SCANNER FOR CITRUS MECHANICAL HARVESTER By UJWALA JADHAV A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2010

PAGE 2

2 2010 Ujwala S Jadhav

PAGE 3

3 To my family for their love and support

PAGE 4

4 ACKNOWLEDGMENTS I would like to give my sincere thanks to all those people who have offered me support, encouragement and help in completing my thesis successfully. I would like to thank my advisor Dr. Reza Ehsani, for his continuous guidance, valuable inputs and support throughout my master s studies. It was a wonderful experience and I am deeply indebted for giving me opportunity to work with him My appreciation goes to Dr. Won Suk Daniel Lee for giving me guidance and direction during the course of my study and thes is completion. I am also grateful for the useful suggestions and discussion provided by my committee members Dr. John K. Schueller and Dr. Masoud Salyani I am thankful for the funding provided by the Citrus Initiative of Florida to complete my research wo rk. I would like to thank CREC members e specially Mr. John Henderson from the packaging house for providing me citrus fruits during my laboratory experiments. I am greatly indebted to m y colleagues Dr. Mari Maja, Dr. Joa Camargo, Dr. Sindhuja Sankaran Dm itry K Bhargav Prasad Ashish Mishra Sherrie Buchanon, Sajith Udumala Savary and Raghav Panchapakesan for helping me during my experiments and giving me a fun filled lab environment My special thanks go to my husband Vijay, who was a lways there to help and support me. During my research work, discussion with him helped me understand the basic principles of engineering. It would be incomplete if I do not thank my mom and family members without whose continuous love, support, patience a nd encouragement any of this would not have been possible

PAGE 5

5 TABLE OF CONTENTS ACKNOWLEDGMENTS ...............................................................................................................4 page LIST OF TABLES ...........................................................................................................................7 LIST OF FIGURES .........................................................................................................................8 ABSTRACT ...................................................................................................................................11 CHAPTER 1 INTRODUCTION ..................................................................................................................13 Citrus Industry in Florida ........................................................................................................15 Citrus Harvesting ....................................................................................................................17 Precision Agriculture and Yield Monitoring ..........................................................................18 Reasons for Yield Monitoring in a Mechanical H arvester .....................................................21 Objectives of This Study ........................................................................................................21 Report O rganization ................................................................................................................22 2 HISTORY AND LITERATURE REVIEW ...........................................................................23 Citrus History ..........................................................................................................................23 Citrus Yield Monitoring .........................................................................................................24 Yield Mo nitoring Techniques .................................................................................................26 3 MATERIALS .........................................................................................................................39 Conveyor Systems ..................................................................................................................39 Continuous Canopy Shake and Catch (CCSC) Harvester ...............................................39 Trash removal Machine ...................................................................................................40 Sensor and Communication ....................................................................................................40 Infrared (IR) Laser Sensor ...............................................................................................41 LIDAR .............................................................................................................................43 Labview Software ............................................................................................................45 4 VOLUMETRIC YIELD MEASUREMENT ..........................................................................47 Yield Monitoring Using Linear Array of IR Distance Sensors ..............................................48 Volumetric Yield Calculation Algorithm ........................................................................48 Overview of the Experiment ...........................................................................................50 Materials and Methods ....................................................................................................52 System Calibration ..........................................................................................................54 Laboratory Test ...............................................................................................................56 Results and Disc ussions ..................................................................................................57 Conclusion .......................................................................................................................57

PAGE 6

6 Yield Monitoring U sing LIDAR (SICK Sensor) ....................................................................58 Volumetric Yield Calculation Algorithm ........................................................................58 Error S ources ...................................................................................................................61 Method of Griddi ng Interpolation ...................................................................................66 Experiments on Horizontal Conveyor (Trash removal machine, f/h <0.5) ............................69 Overview of Experiment .................................................................................................69 Method and Materials ......................................................................................................70 Experiment A ...................................................................................................................71 Experiment B ...................................................................................................................75 Results and Discussions ..................................................................................................76 Experiments on Inclined Conveyor (CCSC, f/h ...........................................................81 Overview of Experiments ................................................................................................82 Ma terial and Methods ......................................................................................................82 Experiment A ...................................................................................................................83 Experiment B ...................................................................................................................86 Results and Discussi ons ..................................................................................................87 5 Y I E LD MONITORING INTERFACE ...................................................................................94 Importance of Yield Monitoring Interface .............................................................................95 Design and Development ........................................................................................................95 Conclusion and Discussion ...................................................................................................102 6 CONCLUSION AND FUTURE WORK .............................................................................103 APPENDIX: LMS SICK DATA COLLECTION ALGORITHM ..............................................106 Algorithm for Horizontal Conveyor System in Trash Removal M achine ............................106 Algorithm for Horizontal C onvey or System in Citrus Mechanical Harvester M achine ......115 LIST OF REFERENCES .............................................................................................................125 BIOGRAPHICAL SKETCH .......................................................................................................130

PAGE 7

7 LIST OF TABLES Table page 31 Angular resolution and maximum scanning angle .............................................................45 41 Laboratory test results on True volume and Computed volume ........................................56 42 Different parameters used the two experiments for volume calculation algorithm ...........71 43 Calibration data with speed 0.92 m/s with angle 180 and angular resolution 1 .............72 44 Calibration data with speed 1.03 m/s with angl e 180 and angular resolution 1 .............73 45 Calibration data with speed 1.14 m/s with angle 180 and angular resolution 1 .............74 46 Validation of the data after applying the calibration equations .........................................75 47 Calibration data with speed 1.7 m/s ...................................................................................77 48 Calibration data with speed 1.34 m/s .................................................................................78 49 Calibration data with speed 1.03 m/s .................................................................................79 410 Validation data and error with speed 1.71 m/s ..................................................................80 411 Validation data and error with speed 1.34 m/s ..................................................................80 412 Validation data and error with speed 1.03 m/s ..................................................................81 413 Different parameters used in volume calculation algorithm ..............................................82 414 Validation data and error with speed 1.1 m/s ....................................................................84 415 Validation data and error with speed 1.1 m/s ....................................................................85 416 Standard and the percentage error in the actual and predicted mass 1.1 m/s .....................86 417 Calibration data with speed 0.586 m/s ...............................................................................88 418 Calibration data with speed 0.785 m/s ...............................................................................89 419 Calibration data with speed 1.1 m/s ...................................................................................91 420 Validation data and error with speed 0.586 m/s ................................................................92 421 Validation data and error with speed 0.785 m/s ................................................................92 422 Validation data and error with speed 1.1 m/s ....................................................................93

PAGE 8

8 LIST OF F IGURES Figure page 11 U.S. citrus utilized production and value of production packing house door equivalents. (Source: USDA Citrus Fruits 2009 Summary Report) ..................................14 12 Components of grain yield monitoring system (Source: Basics of Yield Monitor Installation and Operation by Shearer et al., 1999) ...........................................................15 13 Production of U.S Citrus by States (Ref. Florida Department of Citrus, Citrus facts 2008) ..................................................................................................................................16 21 Different methods used for yield monitoring and yield estimation for specialty crops (Ref. Ehsani and Karimi, 2010) .........................................................................................26 22 Schematic of common yield monitoring systems for specialty crops (a), (b), (c), (d), (e) and (g) Load cell based yield monitoring systems, (f) Volumetric flow rate based yield monitoring system (Ref. Ehsani and Karimi, 2010) .................................................27 31 (a) Continuous Canopy Shake and Catch (CCSC) and (b) Conveyor portion of the CCSC in l aboratory ............................................................................................................40 32 Trash removal machine ......................................................................................................41 33 An IR laser sensor with connectors (Source: Sharp Electronic) ........................................42 34 Analog output voltage vs. distance to reflective object (Source: Sharp Electronic) .........42 35 SICK LMS 200 Sensor ......................................................................................................43 36 Fan shaped scan by an internal rotating mirror of SICK 200(Ref SICK Manuel) ............44 37 Labview program interfaces and set parameters for SICK sensor .....................................46 41 Schematic of IR sensors setup on the conveyor belt ..........................................................49 42 3D view of volume calculation model ...............................................................................49 43 An array of sixteen sensors mounted on conveyor belt .....................................................51 44 Citrus mechanical harvester and the possible sensor location ...........................................53 45 Different Angles with Different Distance (Ref.: Acroname Robotics Inc. manual) .........53 46 Sensor calibration test in the laboratory .............................................................................54

PAGE 9

9 47 Sensor output (v) vs. Length (cm) .....................................................................................55 48 R square and RMS vs. Polynomial order ..........................................................................55 49 Schematic of SICK sensor setup on conveyor ...................................................................59 410 Schematic of SICK sensor on conveyor ............................................................................60 411 Flaps on trash removal machine conveyor system ............................................................62 412 (a) and (b) Spatial distribution of the raw data without removal of flaps and interpolating the data by gridding method .........................................................................63 413 After removing the flaps in the data ..................................................................................63 414 Conveyor system with flaps to hold the fruits ...................................................................65 415 Spatial distribution of the data showing flaps alone on the conveyor belt ........................65 416 (a) and (b) The distortion in the fruit shape due to the collection and transformation of data from polar coordinate system to Cartesian co ordinate system ............................66 417 Fruits on conveyor belt ......................................................................................................67 418 (a) and (b) The radial spikes observed in the shape of the fruit before applying gridding interpolation .........................................................................................................67 419 After using gridding interpolation, uniformly spaced data points on the conveyor surface without radial spikes ..............................................................................................68 420 After using gridding interpolation, uniformly spaced data points on the conveyor surface without radial spikes ..............................................................................................68 421 The goat truck dumping the fruits on the top container of the trash removal machine .....69 422 SICK LMS 200 location on trash removal machine ..........................................................70 423 (a) Mass vs. Calculated volume calibration curve with speed 0.92 m/s and (b) Mass vs. Calculated volume with grid data interpolation technique, calibration curve with speed 0.92 m/s ....................................................................................................................73 424 (a) Mass vs. Calculated volume with speed 1.03 m/s and (b) Mass vs. Calculated volume with grid data interpolation technique with speed 1.03 m/s .................................73 425 (a) Mass vs. Calculated volume with speed 1.14 m/s and (b) Mass vs. Calculated volume with grid data interpolation technique, calibration curves with speed 1.03 m/s ...74 426 % Error vs. Mass plot with different speeds ......................................................................75

PAGE 10

10 427 Actual mass vs. Calculated volume calibration curve with speed 1.71 m/s ......................78 428 Actual mass vs. C alculated volume calibration curve with speed 1.34 m/s ......................79 429 Actual mass vs. Calculated volume calibration curve with speed 1.03 m /s ......................80 430 % Error based on actual mass with respect to the speed of conveyor ...............................81 431 LMS SICK 200 for laboratory test on the conveyor belt ...................................................83 432 Calibration plot of mass vs. Calculated volume of fruits calculated after correcting the data with gridding method and removing the flap error ..............................................85 433 Calibration curve with speed 0.586 m/s .............................................................................90 434 Calibration curve with speed 0.785 m/s .............................................................................90 435 Calibration curve with speed 1.1 m/s .................................................................................92 436 Error spread with different speeds .....................................................................................93 51 Yield monitor components (left), Yield monitor 8 touch Screen Panel (right) ...............94 52 Snap shot of the data from GPS and l oad cell ...................................................................96 53 Mechanical harvesting machines on top and bottom rows in the grove ............................97 55 Yield monitor interface with drivers input tab .................................................................98 56 Yield monitor interface with top row mechanical harvester information tab ....................99 57 Yield monitor interface with bottom row mechanical harvester information tab ..............99 58 Yield monitor interface with bottom row mechanical harvester information tab ............100 59 Yield monitor interface top row mechanical harvester data communication port setting tab .........................................................................................................................101 510 Yield monitor interface bottom row mechanical harvester data communication port setting tab .........................................................................................................................101 511 (a), (b) Snap shot of raw data for the top and bottom row harvesters, (c) The processed data after combining raw data from top and bottom row harvesters. ..............102

PAGE 11

11 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science VOLUMETRIC YIELD MONITORING SYS TEM USING LASER SCANNER FOR CITRUS MECHANICAL HARVESTER By Ujwala Jadhav December 2010 Chair: Reza Ehsani Major: Agricultural and Biological Engineering Yield monitoring system is one of the important components in precision agriculture. T here is no commercial yield monitoring system available for citrus mechanical harvesters to date. Most of the large equipment manufacturer investments are focus ed on developing the efficiency of their machin es but make little investment s or none at all in the research and development of yield monitoring systems for these high value crops. A yield monitoring system using an impact plate has been developed that uses the impact of the fruit to correlate to its mass. The system performs well in both laboratory and field tests but due to the impact created by the oranges falling on the plate, the durability of the system is poor A noncontact method of collecting yield data will be very desirable to either augment the performance of the current system or eventually replace it. A volume based yield monitoring system was implemented as a means to estimate the mass of the fruit passing on the conveyor system on the mechanical harvester. A LIDAR (Light Detection a nd Ranging) technology was u sed to scan the cross sectional area passing on the conveyor system and send the distance related information to calculate the volume of fruits on the conveyor by integrating the data over time The scanner used in th e experimental work was a low cost gen eral purpose SICK LMS 200 sensor An algorithm was developed to analyze

PAGE 12

12 the data an d calculate the volume of fruit passing on the conveyor. The system was calibrated and the calibration curve information was used to convert the volume into a corresponding mass The system performance was tested in laboratory setting on two conveyor systems; one with inclined conveyor used in m echanical harvesters (which has flap height of 8 cm) and other with horizontal conveyor system used in the t rash removal machine ( which has flap height 3 cm ) The tested conveyor speeds where from 0.586 m/s to 1.71 m/s The conveyor system used on the mechanical harvester was tested for its maximum speed of 1.1 m/s and its performance was found encouraging with a coefficient of det ermination ( R2) of 0.98 and a RMSE was 1.69 kg w hile the conveyor system used on the trash removal machine performed very well with its maximum speed of 1.71 meters/sec, the coefficient of determination R2 was 0.99 and the RMSE was 0.98 kg The system performance was satisfactory for the complete range of the speeds used on both the conveyor systems. Th e results indicate d the potential of using volumetric yield calculation method using LIDAR in yield monitoring of citrus in m echanical harvest ers. Although this study was conducted for citrus fruits, the concept could be easily applied with little modifications to calculate the yield of other fruits and vegetables.

PAGE 13

13 CHAPTER 1 I NTRODUCTION A yield monitoring system is one of the most important elements in precision farming (Pelletier and Upadhyaya 1999) It gives quality information and direct feedback to the growers about the yield variability within the grove and is considered to be the first step in the process of site specific crop management (Pelletier and Upadhy aya 1999) Site specific agriculture focuses on improving management to increase profitability and y ield monitoring is essential for the creation of accurate yield maps (Persson et al., 2004) The growers can use this precise information optimize farm profit and minimize the effects on the environment. Citrus fruit is among in Florida s most valuable agricultural product s worth $1.5 billion (packaginghouse door equivalent) during 200809 season (USDA Citrus Fruit s 2009 Summary Report). In 200809 season citrus production was 12 million tons as shown in Figure 1.1. Florida produced 71% ; California produced 26 % and Texas and Arizona produced the remaining 3% of the total U.S. citrus production ( National Agricultural Statistics Service, 2009). Figure 1 1 shows U.S. citrus utilized production and value of production packing house door equivalents. Mechanical harvesting of citrus fruits in Florida is becoming more prominent due to the high demands and high costs associated with labor for harvesting. Tree canopy shakers are being used for automated citrus fruit harvesting. Mechanical harvesters such as canopy shake and cat ch type, have the conveyor belt which carries the harvested fruits to the containers that are attached to Oxbo machine. Once the container gets filled, it is carried to the processing plants for further operation. A yield monitor is used to give the farmer an accurate measure of the amount yields vary within the grove. A yield monitoring system can measure and store the yield related information and also give the onthe go yield information to the operator. A typical yield monitor is a combination of severa l components as shown in Figure 12.

PAGE 14

14 Figure 11. U.S. citrus utilized production and value of production packing house door equivalents. (Source: USDA Citrus Fruits 2009 Summary Report) It consists of several different sensors, a data storage device, a console installed in the operators cab. These sensors measure the mass or volume flow of vegetables or grains (grain flow sensors), ground speed, separator speed and the grain moisture sensor Depending upon the sensors specification, they might be connect ed to ADC and to the direct digital input. Using various parameters being calculated from the different sensors, the yield of the grain is determined. In order to understand the interaction of the yield monitor, combine cab operator and the combine dynami cs, one should understand the function of various components. A yield monitor consists of a main sensor to measure the actual mass or volumetric flow of the grain, fruits or vegetables Yield monitors calculate the yield by dividing the crop volume or mass flow rate that is passing through the mechanical harvesting machine for the given time by the covered area (Ehsani and Karimi, 2010) The area covered by the mechanical harvester is calculated by using the speed related information collected from the ground speed sensors for a given time. The GPS receiver mounted on the mechanical harvester gives the position information which is integrated along with the yield information and then stored into the storage device for further processing.

PAGE 15

15 Figure 1 2. Components of grain yield monitoring system (Source: Basics of Yield Monitor Installation and Operation by Shearer et al., 1999) In addition to the yield variability growers can use this information wisely to manage the harvesting machine operation. T he s tored data can be used to extract the machinery management related information such as actual harvest time and total downtime. Also, the field efficiency, and the machine operator performance can be analyzed This performance related information can be use ful to the grove managers and it can help them improve the efficiency of the fruit harvesting operation. This will help to reduce the cost while mak ing suitable management decisions (Ehsani and Karimi, 2010) In this chapter, a brief discussion about the c itrus industry in Florida, its history and the need for yield monitoring are provided. After some background information, the objective s of this study will be defined and the organization of the thesis will be provided. Citrus Industry in Florida In production of citrus, Florida has been the nations largest producing state. During the last decade nearly three quarters of all U.S. citrus was grown in Florida ( Figure 13) (Forecasting

PAGE 16

16 Florida's Citrus Production, USDAs NASS 2009). Florida growers produce several types of citrus including oranges, grapefruit, and specialty fruits including Temple oranges, tangerines and tangelos. The most commonly grown varieties of Florida oranges are Navel, Hamlin, Pineapple, Amber S weet, and Valencia. The typical se ason for orange runs from October through June (Florida Department of Citrus, Citrus facts 2008) Figure 13. Production of U.S Citrus by States (Ref. Florida Department of Citrus, Citrus facts 2008) In market share, Florida is second to Brazil in globa l orange juice production and in the production of grapefruit Florida is the worlds leading producer (Florida Department of Citrus, Citrus facts 2008). Florida produces more than 70% of the nations supply of citrus, with majority export markets to Canada Japan, France and the U.K. More than 80% of U.S.As orange juice is made from Florida grown oranges and nearly 87% of Florida citrus is processed into orange and grapefruit juices (Florida Department of Citrus, Citrus facts 2008). In Florida, citrus groves cover nearly 569,000 acres of land and there are more than 74 million citrus trees. Most of the citrus gro ves are cultivated in the southern part of the Florida where there is low probability of a freeze. Florida citrus fruit is harvested when it is rip e. Once picked, the fruit does not continue to ripe. The picked or harvested fruits a re placed into canvas

PAGE 17

17 bags transferred to the tubs and then put into specialized vehicles called goat trucks Citrus produced for the fresh consumption is brought to th e packaging houses where it is washed, graded and packed. Citrus produced for juice extraction is transported to the processing plants (Florida Department of Citrus, Citrus facts 2008) The citrus industry generates close to $1 billion in tax revenues whic h helps in supporting schools, highways, and healthcare services (Florida Department of Citrus: Citrus Facts, 2008). The citrus industry gives employment to nearly 75,000 Floridians (Florida Department of Citrus: Citrus Facts, 2008) Citrus Harvesting Citrus harvesting is a labor intensive process and depending upon the size of the grove, the demand for workers to harvest the citrus fruits increases. Mechanical harvesting and many other improvements in harvesting of citrus fruits in Florida began in the mid 1950s. In the 1950s, the harvesting operations were mechanized reducing the labor required by at least two thirds (Whitney 1995) The acreage along with the yield of Florida citrus was steadily increasing and finding the labor to harvest was getting difficult and expensive. T he Florida Department of Citrus, United States Department of Agriculture and the University of Florida started research development on a citrus mechanical harvesting system to aid in the r emoval of fruits from the trees thereby reducing the number of hand harvesters needed. In early 1960s, the focus was on the development of mass harvesters. Some of the initial efforts to mechanize the picking of fruits were towards the development of mach ines that could duplicate manual harvesting. The goal was to develop a system that could be used for both process and fresh market oranges. Research was done on trunk shakers, limb shakers, air shakers, and canopy shakers. Such a system could be very useful to the Florida citrus industry by reducing the picking/harvesting costs. In 1970s, air shakers and the canopy shakers were

PAGE 18

18 designed and implemented. These systems apply force to the canopy and limb portion of the tree. Hedden and Coppock ( 1971) used canopy or foliage and limb shakers to do series of tests and came to the conclusion that the foliage shakers had better selectivity. Despite of this during the late season harvest a considerable amount of young fruit w as removed and also the fruits were bruised during the following year. Whitney ( 1968) Whitney and Patterson ( 1972) developed the air shaker which did not come in contact, but this system was only efficient when used along with the abscission chemical but in that case it caused defoliation and peel damage too. In 1990s Peterson ( 1998) designed and developed a prototype similar to the current canopy shaker. The removal efficiency of these harvesters ranged from 80 to 90% which could be improved by finding the optimal operating parameters and changing the configuration of the system. Now using these machines, the fruit ca n be picked off the tree, more research was needed to collect or catch these fruits. The mechanical pickers were classified into two categories. In the first category, fruits were dropped onto the ground and then deposit ed by machines or manual workers int o bins and to goat trucks and later transferred to hauling trailers. In the other category, the mechanical harvesters had catch frame which caught the harvested fruits and deposited directly into goat trucks. These harvest machines with catch frames were designed and developed with the mechanical pickers ( Coppock ( 1967 ) ; Coppock and Hedden ( 1968) ; Churchill et al., ( 1976) ; Coppock ( 1976) ; Sumner and Churchill ( 1977) ). The main disadvantage of these methods is the damages caused to the fruit due to bruising when falling from the tree. Sometimes these bruise s are too severe for fruits that are intended for processing. Precision Agriculture and Yield Monitoring In precision farming each of the crop production inputs e.g. fertilizer, limestone, herbicide, insecticide, seed, etc. are managed on a site specific basis which helps in reducing waste and increase profits while maintaining the quality of environment ( Kuhar, 1997) In

PAGE 19

19 precision farming small units of land are treated instead of treating the whole field A grower analyzes, applies fertilizers pesticid es maps, and treats these small units of a field considering infield spatial variability Using precision farming techniques a grower can monitor the crop yield more precisely; a grower can control the application of fertilizers, pesticides, etc. In order to apply precision farming techniques, a farmer needs to be able to measure and record yield at different locations in the field as the crop is being harvested or processed. This site specific yield data then used to generate a yield map showing yield var iations for each unit of a field. In applying precision farming techniques for grain production, yield monitoring systems have been used. In grain harvesters or a combine, yield monitors are installed and used to gather the real time yield data as the grai n is weighed or measured on the harvester. When the yield information is related to the geographic information system (GIS) of that particular location, this gives advantage to the farmers in being able to know the productivity of each unit of the field, t hrough which units with low yield can be targeted, insect infestations can be outlined, different soil types can be identified and plotted, diseases can be identified, and fertilizer quantities can be adjusted. Precision agriculture is a management technology that responds to the spatial and temporal variability found on agricultural landscapes. Using precision agriculture technique, one can determine the yield variability in the field, determine cause of low yie ld, come up with solutions based on economic justification, implement new techniques, and repeat the procedure in cyclic approach. These techniques could be used to improve economic and environmental sustainability in crop production. Major technologies and techniques used for precision farming such as Global Positioning S ystem (GPS), yield monitoring and mapping, soil testing, remote sensing (RS), GIS and mapping, variable rate technology (VRT), and information technology have enabled the farmer to visualize the entire field in a way, that could help manage the agricultural operations

PAGE 20

20 efficiently and improve overall productivity. Using these technologies, the farmer can effectively manage the crop throughout its life cycle, starting from preparing soil, sow ing seeds, applying fertilizers/pesticides and finally measuring yield during harvesting based on each individual plant, which helps reduce the waste of resources due to in field variability. Yield monitoring and mapping plays a major role and is the firs t step to implement site specific crop management on a field. A yield monitoring and mapping system measures and records the amount of crop harvested at any point in the field along with the position of the harvesting system. The position and yield data co uld be used to generate yield map using mapping software. Yield maps are useful to identify variability within a field. A yield variability may be due to seasonal change in weather pattern or rainfall over several years or improper distribution of irrigation/drainage facilities for field and excessive/defici t application of farm inputs. Various yield monitoring and mapping systems have been researched and commercializ ed for various crops over the two decades. Whitney et al. ( 1998) and Schueller et al. ( 1999) developed the first yield monitor system for citrus industry. Schueller and Bae ( 1987) and Searcy et al. ( 1989) extensively studied and opted y ield mapping system during grain harvesting Examples of yield mapping and monitoring for other crops include cotton ( Wilkerson et al., 2001; Roades et al., 2000) sugarcane ( Benjamin 2002) potatoes ( Campbell et al., 1994) citrus ( Annamalai, 2004; Ehsani et al. 2009; Maja and Ehsani 2009a ) tomatoes (Pelletier and Upadhyaya 1999) silage (Lee et al., 2002) and tuber (Persson et al., 2004) Being able to evaluate the entire farm graphically, with respect to the yield and other associated field characteristics would tremendously help farmers to more meticulously know the variability in the field and thus help them to make important decisions in an efficient manner.

PAGE 21

21 Reasons for yield monitoring in a mechanical h arvester Currently, there is no yield monitoring system available for Citrus Mechanical Harvesters up to date. C anopy shake and ca tch harvesters and tree canopy shakers are the automated citrus fruit harvesting machines used by citrus growers in recent years. These machines lack the reliable method of yield monitoring. C anopy shake and catch harvesters have the conveyor mechanism tha t carries the citrus fruits and transports to a level where the fruits roll along a ramp into the container. A system can be implemented on the citrus carrying conveyors, which can measure the amount of citrus fruit volume pass ing along the conveyor and can convert this volume into mass to get the citrus mass carried per unit time in the collection bin. Yield monitoring systems for mechanical harvesting machines create real time data for maximum efficiency (Ehsani, 2007) Yield mon itoring is the process of measuring fruit yield for a given location in the citrus grove. The yield variability for a particular section or part of the grove is unknown to the growers, yield monitoring can provide and document the yield variability at a smaller scale and can lead to a way to manage the needs of each individual tree rather than treating the entire block of trees uniformly. Y ield monitoring system would not only provide information of the mass loaded into the container but also provide fruit yield of the area and approximate individual tree yield Objectives of This Study The main objectives of this research was to 1. Study and evaluate different sensors for volumetric yield calculation 2. Design a volumetric yield measurement system using the m ass flow sensor using LIDAR (Light Detection And Ranging) technology for citrus mechanical harvester and Trash removal machine 3. D evelop an algorithm for the laser based sensors that can estimate the volume of the fruits on the conveyor belt 4. Develop a yield measurement model to relate the volume to the mass of the fruits

PAGE 22

22 Report O rganization Chapter 2 is about the history of yield monitoring of citrus different techniques used for yield monitoring of specialty crops and the literature review related to this study. Chapter 3 explains about the materials used during the laboratory experiments involved in this study. Chapter 4 describes the volumetric yield measurement technique and the experiments performed. The performance of different sensors was analyzed in the laboratory conditions The analysis and validation are explained in this chapter. Chapter 5 discusses the yield monitoring interface development for the impact plate type yield monitor. Chapter 6 provides main conclusions from the study and indicates possible future research direction

PAGE 23

23 CHAPTER 2 HISTORY AND LITERATURE REVIEW Citrus History Citrus fruit can be of different types such as orange tangerine, lemon, lime, and grapefruit. The evergreen citrus species grow and produce fruit under varied climate conditions, ranging in the latitude over 40o North to 40o South from equatorial hot humid climates through warm subtropical and even cooler maritime climates (Spiegel Roy and Goldschmidt 1996) The exact origin of citrus fruit is not clearly known but the various species of citrus are native to the subtropical and tropical regions of SouthEast Asia including South China, northeastern India and Burma, and have spr ead from there to other regions of the world by least 4000 BC (Spiegel Roy and Goldschmidt 1996) The genus spread was very slow from one part of the world to the other. The first member of the group Citron, mentioned about 310 BC by Theophr astus was cultivated and known for several hundred years to European civilization. The sour and sweet oranges and lemon introduced after Citron several centuries apart. Sweet oranges had been grown in China for many centuries and then it became known to Europeans. Many older references to oranges may be found in ancient Chinese manuscripts and documents (Spiegel Roy and Goldschmidt 1996) By the middle of sixteenth century, citrus was succeeded to establish in the West Indies and Brazil. There is no any exact date when the citrus was brought to California, the first citrus seeds were introduced to California when the mission of the Francisc ans was established in San Diego in 1769. Citrus was first introduced to Florida sometime between 1513 and 1565 by the Spanish voyagers Spanish explorers, most likely Ponce de Leon, planted the first orange trees around St. Augustine, Florida (Florida Dep artment of Citrus, 2008). A first grapefruit seed was planted in Florida around 1809 (Hume 1926) Florida's unique sandy soil and subtropical

PAGE 24

24 climate proved to be ideal for growing the seeds that the early settlers p lanted and since then has flourished. Citrus Yield Monitoring Yield monitoring systems are available in market for grains but very few yield monitoring systems are available for specialty crops. Specialty crops are high valued commodities that need special care and are more sensitive to the growth conditions. Lack of yield monitoring systems limits the site specific crop management and considered as one of the bottlenecks in applying precision agriculture techniques for specialty crops (Upadhyaya et al. 1999) Citrus yield monitoring system was not developed for several decades. Miller and Whitney ( 1999) and Schueller et al. ( 1999) developed the first yield monitoring and mapping system for citrus and evaluated the system for site specific crop management for citrus in Florida. Citrus yield was measured by installing three electronic sensor systems on a specialized vehicle called a g oat, which is used to transfer the fruits from grove to the roadside containers. A GPS unit (GOAT, Geo Focus, Gainesville, FL) was attached to the sensor systems to record the location of bins as it was picked by the goat truck. This GPS unit was used to c oordinate all the operations of the goat truck. A crop harvest tracking application (GeoFocus, LLC, Gainesville, FL) was installed on the unit to record the pa llet bin or tub pickup location. This application was also programmed and modified later to incor porate two channel A/D data acquisition capability for A/D voltage measurements. This unit was used by the goat truck operator and the data logging process was triggered by a button push. This initiates the recording of GPS location and the A/D voltage acq uisition corresponding to the bin or tub. A LCD display on the unit indicated the acquisition process status and number of events recorded. The data then transferred to the computer through RS 232 port connection for further post processing. The LCD displa y was mounted on the goat truck such a way that it was accessible by the truck operator. It was found that this system

PAGE 25

25 sometimes produced incorrect maps due to the fact that the driver failed or forgot to record the location of the tub. Salehi et al. ( 2000) developed a system to overcome the problem in the previous system. An automatic triggering system was developed to record the location of the bin or tub. The pressure switch was used to detect the load on the dump c ylinder and on the boom lift cylinder. The position switch detected whether the tipping head was located over the truck bulk bin and then the system identifies that the truck was in the process of picking the bin or tub for collecting the fruit and data ga thering circuit was activated for a given time using the timer and relay circuit for collecting the DGPS data. However, the automatic triggering system also encountered problem in recording the tub locations, which could be the problems related with the de lay timer, pressure switch settings and hardware connections. Annamalai (2004) and Ehsani et al. (2009) developed yield estimation techniques by counting individual fruits or vegetables on the citrus trees. Maja and Ehsani ( 2009b) developed a load cell based yield monitor for citrus mechanical harvesters. Two load cells were attached to a carbon fiber plate and mounted at the end of the co nveyor system of a shaker machine. The carbon fiber plate was used to measure the force created by the rolling oranges on the conveyor. The system performed very well in field conditions with coefficient of determination of 0.97 with an average error of 7.8%. Along with the harvested fruits the conveyor carries the foreign objects also called as trash, such as citrus tree limbs, leaves, stems, and trunks. Because of the impact produced by the fruits and the trash, it cannot withstand for longer duration of time in the field and the system failed. The large diversity in the type and harvesting methods of specialty crops have provided little incentive for commercial companies to invest in developing yield monitoring systems for specialty crops.

PAGE 26

26 Yield Monitorin g Techniques Yield monitoring is the process of measuring crop yield per unit area and integrating this yield along with the GPS coordinate information. In earlier time, the yield monitoring attempts have been done in manual harvesting by counting and loca ting the containers at particular locations in the field (Schueller et al., 1999; Whitney et al., 2001) There are a number of ways to measure the yield of crops. Most of the methods developed over the years have involved weighing the crop after it has been threshed, separated and cleaned. This method is oldest method and it is still being used. The second method is batchtype yield monitor, which weighs grains in the grain ta nk of combines operator on a monitor. This method is considered forerunner of modern, site specific yield monitoring. The third and last method is instantaneous crop yield monitoring which measures and records yield onthe go. The most commonly used methods in developing yield monitoring systems for specialty crops are mass/volume flow rate sensors, direct weighing, and cumulative weighing. These methods are used for yield estimation and yield monitoring of specialty crops. The se methods can be categorized as shown in Figure 21 and 22. Figure 21. Different methods used for yield monitoring and yield estimation for specialty crops (Ref. Ehsani and Karimi, 2010)

PAGE 27

27 Figure 22. Schematic of common yield monitoring systems for specialty crops (a), (b), (c), (d), (e) and (g) Load cell based yield monitoring systems (f) Volumetric flow rate based yield monitoring system (Ref. Ehsani and Karimi, 2010) Yield monitoring of grains and other crops vary in different methods of measuring grain or crop flow. There are some popular yield monitoring systems which use different flow sensors in

PAGE 28

28 the path of clean grain/fruit flow. The grain flow sensor is typically mounted at the top of the clean grain elevator or conveyor belt in harvesters. Grain or fruit flow can be sensed by placing the impact plate in the path of grain or fruits and then the force applied by impact of grains or fruits are measured. In mechanical harvesters developed for specialty crops, the mass or volume measurement of harvested crop i s done at the end of the conveyor belt before the fruit get transferred into containers ( Figure 2 2a and 2 2b). Yield monitors for grain crops based on mass flow have been developed for the measurement of specialty crops Either an optical sensor or a load cell has been used for such measurements. The impact plate is mounted on the load cell which measures the force impacted and coverts this load into an electrical signal. The strain gauge bonded with the load cell converts the load into an electrical signa l. A very slight deformation of the load cell causes the measureable change in the resistance offered by the strain gauge to electrical current flow. Ehlert (2000) tested a rubber coated bounce plate with a force measu ring instrument for potato harvesters by placing it in different positions in the discharge trajectory of a conveyor belt running at 3 different speeds. The vibrations in actual potato harvesters were simulated by moving the measurement device at five diff erent frequencies and four amplitudes. The result showed a linear relationship between the mass flow and the force. For many parameter combinations, the coefficient of determination was more than 0.99 and in some cases; standard error was less than 0.083 k g/s which indicates that the bounce plate provides a basis for measurement in potato harvesters for yield mapping. Lee et al. (2002) developed a silage yield mapping system using DGPS, load cells, moisture sensor and bluetooth for wireless data transmission. A shear beam type load cells were installed on the four corners of the bottom of the trailer silage box. To compare the load cell weight and the actual weight, the axle of a silage trailer without the box was measured on a platform scale. In order to compare the

PAGE 29

29 weight measured by a platform scale and load cell, the weight of a trailer axle was subtracted from a total trailer weight. The error was calculated between the two measurements and was found to be in the range of 0.37~1.96%, which was very small. Benjamin (2002) developed a technique for yield monitoring of sugarcane. They evaluated t he system performance based on the variety of sugarcane, maturity levels and flow rates. The yield sensor predicted the sugar cane yield with a slope of 0.90 and coefficient of determination ( R2) of 0.96. The average error was 11.05%. Results showed that for the simple weight scale system, sugarcane maturity level and flow rate (induced by two different travel speeds of the combine) did not significantly ch ange the yield monitor readings but the effect of sugarcane variety was significant. The percentage err or ranged from 0 to 33% and 14 out of 118 tests showed an error of higher than 20%. Molin and Menegatti (2004) developed a similar system for sugarcane harvesters. Average error in field evaluations was from 3.5% to 8.3%. Several different methods used to estimate weight from a weigh scale, these methods ranged from inclusion and exclusion of slat speed, tilt sensor data, and different processing algorithms. The s tandard deviation ranged from 4 to 10% on absolute val ue error numbers. Bora et al. (2006) developed a mass flow citrus yield monitor under laboratory conditions. They found mass flow yield monitor is most accurate when the mass flow rates that are used in the calibration phase are in the same range in the actual measurements. In impact type mass flow sensor based system harvested crop grains hit the impact plate and the output measurement corresponding to the impact and velocity used to calculate the mass of the crop fl ow The coefficient of restitution of the crop should be accurate for the accuracy of impact plate yield monitors. The coefficient of restitution is a constant, representing the ratio of its velocity before and after an impact. If the coefficient of restit ution of a crop significantly depends on its maturity, moisture content then the impact type yield monitors will not result in

PAGE 30

30 accurate measurements (Pelletier and Upadhyaya 1999) Grift et al. (2006) developed three methods for yield monitoring system for citrus fruits. They found the first method which was based on large sensor array was inaccurate in measuring the lengths of clumps and spacing due to defocus problems. The second method based on the interruption of a low placed single laser beam showed promising results for higher mass flow densities. A third method which used gutters that forced fruits in single file was deemed most reliable and robust for application of tree canopy shakers. Magalhes and Cerri (2007) designed and implemented a sugarcane yield monitor consisting mass flow sensor which was mounted on a weigh plate. The w eighing plate was set up on the upper part of the har vester side conveyor, just before the sugarcane billets are dropped in an infield trailer. Additional sensors were used to measure the conveyor inclination angle and speed. Performance reports indicated that the maximum error of yield measurement did not exceed 6.4% and the mean error detected without any compensation was 4.3%. Qarallah et al. ( 2008) designed an impact based yield sensor consisting of two load cells, an acrylic plate and a polyurethane cushion for measuring the individual weights of onion bulbs. The precision of the sensor was maintained with a 30 mm cushion resulting in a relative err or of less than 2.0% during validation. The accuracy of measurement was independent of the orientation of the onion bulb on the conveyor belt or their orientation as they hit the impact plate. In batch weighing method, the fruits get divided into batches which are weighed before they get dropped into hauling truck or containers ( Figure 2 2 e). Heidman et al. (2003) introduced similar kind of yield monitor for pistachio. This system was field tested on the commercial pistachio harvester and found the 0.97 as the coefficient of determination between the true and the measured yield. Abidine et al. ( 2003) and Cerri et al. ( 2004) described the development and improvement of yield monitor for tomato. In this, a weigh bucket was mounted on the boom

PAGE 31

31 elevator of the tomato harvester. An impact plate was mounted on the four load cells continuously weighed the tomatoes in the bucket. Field test analysi s showed that the coefficient of correlation ( R2 > 0.99) between yield monitor load cell output and the true weight of the harvested tomato was highly correlated Several techniques were implemented on cumulative weighing and yield monitoring system for sp ecialty crops. In cumulative weighing, harvested crop is collected in a cont ainer on the harvesting machine ( Figure 2 2 g) and the system measures the total weight of the harvested crop over time. Behme et al. (1997) developed a system with a drawbar load cell and two weigh axle load cells to estimate the bale weight in a round baler. The output from the three load cells were summed to measure the weight of the bale. Field experiments showed an error between 2.5% and 5.2%. Wild and Auernhammer (1999) developed a weighing system for local yield monitoring of forage crops in round balers. The mounted weighting system was based on load cell in the drawbar coupling and strain gauges in t he axle similar to the work of ( Behme et al., 1997) The static tests and weighing during vehicle stops could be carried out with errors of less than 1%. Weight measurement onthe go needs further investigation to el iminate the sources of errors. Vellidis et al. ( 2001) developed and tested a peanut yield monitoring system over a 3 year period. This system was evaluated by 11 users and they were able to use the resulting yield maps to evaluate management practices. This yield monitor uses load cells for instantaneous load measurement and was accurate between 2 3% on a trailer basis and 1% on basket load basis. The error due to sloping terrain was found to be negligible. The noise due to vibrations of the harvester and uneven field surface were measured and dealt with using appropriate filtering. Rains et al. ( 2002) developed similar yield monitoring system for pecans. Yield was measured by

PAGE 32

32 weighing the pecan wagon as it is loaded by the harvester in the field. They conducted field expe riments and developed a simple linear equation to estimate the net pecan yield from the gross weight of the pecan and the foreign material. The total pecan yield measurements were closely correlated with the gross weight of pecans for the individual pecan trees ( R2 = 0.84). One of the challenges was separating the foreign material picked up by the harvester from the collected pecans. In volumetric flow rate measurement system, volume or volumetric flow rate is measured and then converted into the corresponding weight or mass of flowing fruits ( Figure 2 2f). Different optical laser scanners are being used to measure the volume of fruits carried on the conveyors and then convert the measured volume into mass of the grain or fruits. These systems needs calibration first and then use the calibration curve to get the weight or mass of the fruits on the conveyor. Along with the optical techniques such as laser scanners, computer vision systems are also widely used for the volume or volumetric flow rate measurement Persson et al. ( 2004) tested an optical sensor for the tuber yield monitoring. The laboratory and field test showed that the errors in size determination by the sensor were in the order of 1% to 2% on average. The m ean of optical sensor from the load cell in the field experiments showed the deviation of 3.2%. The laboratory tests were performed on the spherical objects and the sensor was very accurate in measuring the size and the number of objects. The field tests w ere performed on the potato harvester which showed that the number of pixels in the tuber image and the tuber weight has a coefficient of correlation 0.91. The possible errors were due to the differences in the tuber orientation with respect to the camera and moisture content. Gogineni et al. ( 2002) developed vision based sweet potato yield and grade monitor. The estimation of weight was based on multiple linear regression and neural networks, while grade classificati on was based on linear

PAGE 33

33 discriminant analysis and neural networks. In order to identify the sweet potatoes from soil clods, the images captured by the camera were transformed from RGB to YUV color space, where pixels were characterized by their brightness and color characteristics. Once the algorithm detected the sweet potatoes in the image; the other parameters such as area, polar moment of inertia, rectangular width and height, and major and minor axes were determined and using linear regression and neural networks, the weight was calculated. The multiple linear regression method was more accurate ( R2 real field conditions. The coefficient of determination between the number of sweet potatoes det ected by the image analysis algorithm and manual count had a R2 network method and R2 ant analysis method. They suggested that the reduced accuracy could be due to muddy field conditions and poor lighting in the boundaries of the field of view of camera. Bora et al. ( 2006) developed similar type of vision based system for citrus yield estimation. The system consisted of three CCD progressive scan digital color camera and an algorithm was developed for counting the number of fruits and measuring the size from the images. The system was tested in laboratory conditions and the test analysis showed that the system performed well in identifying the fruits that appeared as full in the image. However, when the fruit image was not complete, the recognition error rates were very high. The overall fruit detection accuracy was 85.5%. Hofstee and Molema (2003) have estimated the volume of individual potato tubers using machine vision on the conveyor of a harvester. Cundiff and Sharobeem (2003) developed a procedure to measure the mass flow and volumetric yield monitoring system for round balers. The system consisted of displacement sensors to measure the windrow cross section. This measurement were combined with the velocity and the volumetric f low rate was calculated, which was then multiplied by a mass factor to give the mass

PAGE 34

34 flow rate. Field experiments showed that the yield estimations were within 4% of the true value. Rains et al. (2005) tested the cotto n yield sensor on a peanut combine. A pair of optical sensor was installed in the deliver chute between air fan and basket. The optical sensors were strategically adjusted such a way that the sensors were not covered by the dust and were not hit by to the peanut flow. The peanut yield monitoring system performance was evaluated in three harvesting seasons in real field conditions. The absolute mean error for three different field harvested was 3 9%. They concluded that to maximize the performance of the yie ld monitor, the yield monitor needs to be recalibrated every time the field conditions or harvester adjustments change. Thomasson and Sui (2004) also developed, fabricated and tested an optical yield monitor for p eanut. Field evaluation on commercial peanut harvesters showed that this yield monitor was capable of estimating peanut yield with very high accuracy. The coefficient of determination between the two optical sensor outputs and true peanut weight was 8996 %. There were two major challenges that need to be addressed, peanut moisture content and nonlinear behavior of the sensor output for high mass flow rates. Konstantinovic et al. ( 2007) developed noninvasive ul tra wideband (UWB) ground penetrating radar system to detect sugar beets in soil. The system was successful in detecting the sugar beets with visual detectability from 90% to 96%. They were able to relate the reflected energy to the mass of individual sugar beets with a correlation over 80%. Laboratory tests showed that the detectability of beets buried in the soil depended on the size, the depth of the beets in the soil, the soil moisture content, and roughness of the soil surface. Beets which were small and deeper in the soil were more difficult to detect, and high moisture contents and rougher soil surfaces were less favorable. Price et al. ( 2007) developed a yield monitor for sugarcane harvesters where three optical sensors were mounted in the flooring of the harvester conveyor. This system was tested in laboratory conditions and the result showed

PAGE 35

35 the linear relationship with coefficient of correlation between estimated yield and true yield of 0.93. The system was found to be rugged in the field conditions and did not need frequent maintenance. Chinchuluun et al. ( 2007) developed a machine vision based system for citrus yield mapping and quality inspection. The image processi ng algorithm was implemented to count the number of citrus fruits passing on the harvester conveyor belt. The system was tested in laboratory and actual field conditions with coefficient of correlation 0.96 and 0.89, respectively. They concluded that the s ystem needs improvement in image acquisition (speed and quality of images). In yield estimation methods, yield of the crop is estimated before or after the fruits get harvested. These are the systems which use the vision based approach or other technologi es to count the number of fruits or to measure the volume of the fruits and estimate the yield of the crops but such systems require the density or the mass of individual fruit as constant. These systems use noncontact, nonintrusive measurement approach which could be a major advantage for certain crops. In volume estimation method, the system estimates the volume or the volumetric flow of the crop. The machine vision system was developed by Annamalai (2004) for yi eld estimation and mapping of the yield data in the real time in citrus grove. These systems estimate yield by counting the individual fruits on the tree or in the mechanical harvester. In this system, a color CCD camera was used to capture the citrus tree images and using an image processing algorithm, the number of fruits on the citrus trees were identified and counted. Using the test data, the yield prediction model was developed and the results were then compared with the hand harvested fruits. When the yield estimated using the prediction model and hand harvest were compared, a R2 of 0.46 was achieved. The reason for such poor relationship could be that a single camera was

PAGE 36

36 used to capture the image of the whole canopy and it was difficult to cover the c itrus canopy sides of the tree. In addition, the fruits which were hidden inside the canopy by leaves cannot be identified and are not visible in the image. Therefore, the algorithm was not able to count these fruits. Zam an et al. ( 2008) estimated the wild blueberry fruit yield using a digital camera. This digital color camera was used to take top view images of wild blueberry crop. An algorithm was developed to process these images and count the number of blue pixels in each image. The system was calibrated in field and the true blueberry yield was determined during hand harvesting. A linear regression analysis was performed to relate the percentage of blue pixels in each image to the blueberry yield data. The analysis showed a very high coefficient of determination of 0.98. The correlation between the two variables was used to predict the blueberry yield in another field. The model predictions were close to the true yield with R2 = 0.99. They concluded the potential of estimation of blue berry yield using digital photography within wild blueberry fields. Ehsani et al. ( 2009) developed two fruit counting techniques for citrus mechanical harvesting machines. In the first method, the arrival process of fruits was random whereas the second method relied on counting single fruits using five channels. The fruits were counted using l aser based photointerruption sensors installed at the channel exits. The system performed very well in laboratory conditions, counting individual fruits with an accuracy of 99.8%. The other method where the fruit arrivals at the sensor were assumed to be a Poisson process failed because laboratory experiments showed that this assumption was invalid. In the method or indirect yield estimation, the yield is estimated using related parameter(s) such as tree canopy size. Whitney et al. (2001) developed a DGPS based yield monitoring system for Florida Citrus along with the three fruit weighing systems for hand harvested citrus

PAGE 37

37 fruits. The results showed that the lift cylinder, scale unit utilizing the pressure transducer has the lowest errors ranging from 0.08 to 2.13% than that of the load cells, mounted on the truck frame and the loader boom. The load cell output was recorded every 0.1 s and the yield was computed. The real time prototype DGPS receiver eliminated th e post processing of GPS data and gave the position information accurate enough to locate the citrus bin loads and to plot their location in the map. Tumbo et al. (2002) designed a microprocessor based citrus yield moni toring system which automatically counts the number of tubs dumped in the goat truck. This system was implemented successfully with 100% accuracy. The microcontroller system was interfaced with the DGPS to record the position of each tub when dumped into t he goat truck with an accuracy of 98% of the time. Field evaluations showed that this system was accurate 89% of the time. Zaman et al. ( 2006) developed a system for estimating yield of citrus trees using the ultrasoni cally sensed tree canopy size. An ultrasonic sensor was used to measure the volume of tree canopy. The tree location was mapped with the weight of fruit tubs as they were picked by goat truck. The canopy volume and yield data showed a strong correlation ( R2 = 0.80). The regression model was used to estimate the yield based on the canopy volume for a new set of trees. The estimated and true values of fruit yield showed a correlation with R2 = 0.42. It was hypothesized that poor flowering or fruit set, fruit drop, and disease might be some of the factors that reduce the yield, but they not necessarily affect the canopy volume, which could have contributed to the low validation accuracy. Ye et al. ( 2008) used aerial hyperspectr al imagery to estimate the yield of individual citrus trees. From the images, the canopy features of individual trees were identified using pixel based spectral reflectance values at various wavelengths. The features were used to develop the yield predict ion models. Authors used the twoband vegetation index (TBVI) and multiple linear regression analysis to relate the spectral characteristics of the

PAGE 38

38 pixels of a tree canopy in the image to the yield of the tree. Analysis showed that the yield was highly cor related with the hyperspectral image from the period of fastest vegetation growth. They concluded that due to the alternate bearing of citrus trees, the size of tree canopy alone cannot be used to estimate the yield and canopy size should be used along wit h TBVI to predict the yield at individual tree levels. Hall and Louis ( 2009) reported the relationship between the grapevine canopy size and vegetation density and grape yield. They found statistically significant relati onships between the canopy density and grape yield in the same year. Also, they conclude that the canopy density at flowering stage in one year was correlated with the yield in the next year. This study suggested that canopy area and density information fr om the previous years can be useful in predicting next year grape yield.

PAGE 39

39 CHAPTER 3 MATERIALS The basic components for all the experiments were the same, differing only in their setup and how they were used. Experiments were carried on different type of conveyor systems. Two kinds of sensors were tested to evaluate their performance in measuring th e volume of citrus fruits on the conveyor. After evaluating their performance, one type of sensor was selected for further experiments. Different conveyor systems, sensor (s) and communications are explained in detail in this chapter. Conveyor Systems Cont inuous Canopy Shake and Catch (CCSC) Harvester Continuous shake and catch systems currently used in Florida citrus are manufactured by Korvan Industries, Inc. of Lynden, Washington, and OXBO International Corp. of Clear Lake, Wisconsin (Ref. http://edis.ifas.ufl.edu/hs239). The mechanical harvester (Figure 3 1) consists of whirls stacked horizontally and has an array of approximately 6 foot long, 1.5 to 2 inch diameter tines mounted to the whirls which are connec ted to a central drum. These tines go about 5 feet inside the tree canopy and shake it horizontally to remove the fruit. The machines ha ve the fruit catching frame on the ground below the canopy which catches and separates the fruit from leaves and stems reducing the amount of trash delivered to the processing plant. These fruits are then carried and forwarded through the conveyor system to the containers or directly to a goat type truck. The location where the conveyor system installed on the CCSC is shown in the Figure 3 1(a). This conveyor system as shown in the Figure 31(b) was used in laboratory for the experiments. The conveyor system was driven by a small hydraulic motor that operates the conveyor belt. The conveyor system has the area above the s urface where the laser sensors can

PAGE 40

40 be installed. Different types of sensor(s) were installed on the conveyor system to measure the volume of fruits on the conveyor system. The volumetric yield measurement system was installed above the fruit carrying conve yor at 3050 cm distance as shown in 31(b) system. (a) (b) Figure 3 1. (a) Continuous Canopy Shake and Catch (CCSC) and (b) Conveyor portion of the CCSC in laboratory Trash removal Machine A trash removal machine as s hown in the Figure 3 2 was developed by Dr. Ehsani and his team at the Precision Agriculture L aboratory, Citrus Research and Education Center, Lake Alfred, Florida. The t rash removal machine was built specifically for the application of trash removal from the harvested citrus fruits and also used for mounting the yield monitoring system. The volume measurement system was installed, calibrated and also tested on the t rash removal machine Sen sor and Communication Two types of sensors were tested to see the performance of the system with different speeds of the conveyor. The first type of sensor was an optical laser sensor, Sharp GP2Y0A02, and the other type was LIDAR based LMS SICK 200 laser s canner. Both these sensor s were

PAGE 41

41 used and tested in the laboratory conditions. An algorithm was developed to analyze the data collected from these sensor and then determine the fruit volume on the conveyor system. Several laboratory tests were performed and the data were analyzed to evaluate the performance of each sensor. The sensor properties and parameters are discussed as follows. Figure 3 2. Trash removal machine Infrared (IR) Laser Sensor In the laboratory test an array of 16 IR laser (Figure 3 3) s ensor s was installed on the aluminum plate. The Sharp GP2Y0A02 developed by Sharp Electronic Components has detection range of 20 cm to 150 cm All these sensors were connected to the PIC controller and then the output from the PIC controller was connected to the laptop. The analog to digital conversion (ADC) was done using a microprocessor chip with 10bit, 16 channel analog to digital converter. The operating supply voltage required for this sensor is 4.5 to 5.5 V

PAGE 42

42 Figure 33. A n IR laser sensor with connectors (Source: Sharp Electronic ) Figure 34. Analog output voltage vs. distance to reflective object ( Source: Sharp Electronic ) The above Figure 3 4 shows a typical output from these IR sensor detectors. As it shows in the graph, the output of these detectors within the stated range (10 cm 80 cm) is not linear but

PAGE 43

43 rather it is logarithmic. As per the specification given in the sensor manual, the calibration curve varies slightly from detector to detector and the sensors need to be calibrate d before use LIDAR The SICK LMS 200 laser scanner as shown in Figure 35 uses the LIDAR (Light Detection a nd Ranging) technology and was used for volumetric yield measurement of citrus fruits. Figure 3 5. SICK LMS 200 Sensor The scanner used in this experimental work was a general purpose LMS 200 model (SICK, Dsseldorf, Germany) The sensor will be c alled LIDAR hereafter The LIDAR sensor scans the space in polar co ordinate system and in twodimension and sends the distance related data in real time for further evaluation via a serial interface. The LIDAR sensor can aim its laser beam in a wide range as its head rotates horizontally and a mirror flips vertically. The laser beam

PAGE 44

44 is used to measure the distance to the first object on its path. The LIDA R system operates on the principle of measuring the time of flight of laser light pulses. A pulsed laser beam is emitted and reflected if it sees an object and the receivers scanner receives the reflection from the object. The time between transmission an d reception of the impulse is directly proportional to the distance between the scanner and the object (time of flight). The pulsed laser beam is deflected by an internal rotating mirror so that a fan shaped scan as shown in Figure 36 is made of the surro unding area (laser radar). The shape of the target object is determined from the sequence of impulses received (LMS 200 Manual). Figure 36. Fan shaped scan by an internal rotating mirror of SICK 200(Ref SICK Manuel) Fundamentally, the distance per indi vidual impulse is evaluated. This means that a distance is provided every 0.25, 0.5 or 1, depending on the angular resolution of the scanner (Table 31). Angular resolution is set using a software telegram. The LMS 200 model has accuracy of 150mm and 50mm standard deviation in a range up to 8m. In this study, an angular resolution of 0.5 or1 and a scanning angle of 180 were used.

PAGE 45

45 Table 31. Angular resolution and maximum scanning angle Angular resolution () 0.25 0.25 0.5 0.5 1 1 Max. scanning angle () (Symmetrical from the center) 100 180 100 180 100 180 Max. no. of measured values 401 721 201 361 101 181 Frequency Achieved (Hz) 13 18 13 18 26 36 26 36 68 75 68 75 The LMS 200 has a standard RS 422 serial port for data acquisition. Labview 8.6 (National Instruments Corporation, Austin, TX, USA) and MATLAB.8 software (The Mathworks Inc., Natick, MA, USA) were used for data acquisition and data processing, respective ly. A program in Labview software available online were used to interface the SICK sensor through RS 422 port and store the data on the disk in the form of .dat file. An algorithm was implemented in Matlab software to postprocess the data and for calculat ing the volume of the fruits carried onto the conveyor during the experiment. A 24 V DC input power supply was given to the sensor. Labview Software The Labview program as shown in F igure 37, developed for the SICK sensor data collection was used to set the parameters and to set the output file to store the data. The program interface has settings for the port to read the data from, LMS resolution, baud rate, maximum distance and the file path to store the data from the SICK sensor. The frequency of the SICK sensor depends on the LMS resolution. The data collection frequency increases as the resolution over the angle increases as shown in Table 31 such as for 0.25, 0.5, 1 degree interval ove r

PAGE 46

46 angle 180, 100, the frequency will be 18, 36, and 75 Hz respectively An average size of an orange fruit is about 8 cm ; therefore, to calculate the volume of a fruit, minimum of 3 data points are required along the conveyor. With the setting 180/1 (1 80 scanning angle and 1 resolution), the sensor gives 75 Hz frequency and to get the at least 3 data points on the fruit the conveyor speed can be increased up to 1.9 meters per second. Figure 37. Labview program interfaces and set parameter s for SIC K sensor

PAGE 47

47 CHAPTER 4 VOLUMETRIC YIELD MEASUREMENT Continuous canopy shake and catch type harvester s used to harvest the citrus tree and removes the fruit which are conveyed by the conveyor system to the fruit collecting containers or goat trucks. A load cell based yield monitoring system for citrus mechanical harvester was developed by (Maja and Ehsani 2009) that uses the impact of the frui ts on the plate and correlate the yield to its mass. The system performed well in both laboratory and field tests but due to the impact created by the citrus fruit falling on the plate, the durability of the system is low In the field the mechanical harvesters work continuously to harvest the fruits, and because of the c ontinuous impact of these fruit and the trash in the harvested fruit, the system failed. A noncontact method of measuring the yield of citrus fruits was very much desirable to either aug ment the performance of the current system or eventually replace it. Th is chapter describes test ing and analysis of different sensors on the conveyor system and to evaluate their performance with respect to the volumetric yield measurement of the fruits on the conveyor, in the laboratory. This chapter provides the background about the hardware system, the experimental setup used on different conveyor systems, and development of the volume measurement algorithm and its performance with respect to the amount of fruits on the conveyor. The conveyor systems on which the volumetric yield measurement system was tested had difference in their orientation and height of flaps present on them. The first section describes the performance of the IR sensors with respect to the yield calculation. This section begins with an algorithm used for volumetric yield calculation of fruits using the IR sensors then gives an overview of an experiment, calibration of the sensors and summarizes the performance of the system with resul ts and conclusion. In the second section, the chapter discusses about the experiments performed using the LIDAR technology using the SICK LMS 200 sensor. This

PAGE 48

48 section begins with an algorithm developed for volumetric yield calculation using LIDAR. The perf ormance of this system was tested on two conveyor systems; one on the mechanical harvester conveyor system and other on the t rash removal machine Yield Monitoring Using Linear Array of IR Distance Sensors The objective of this experiment was to assess th e performance of the infrared IR sensors on the conveyor system in the laboratory conditions. This section details the yield calculation algorithm implementation, overview of experimental setup, materials used during the experiment and the yield measuremen t results and discussion. Volumetric Yield Calculation Algorithm The volumetric yield calculation algorithm uses discrete point data captured using linear array of optical sensors mounted on the top of the conveyor. The number of sensors and the spacing b etween them ensures some minimum number of data points per fruit. By using the discrete data points, amount of citrus fruit volume passing on the conveyor system can be calculated. Since array of sensors would give only two dimensional cross sectional data (along conveyor width), information in third direction (along conveyor length) is achieved by acquiring data from sensors at some sampling frequency. The linear array of optical IR sensors was mounted on the aluminum plate as shown in Figure 41. The co ordinate system is chosen such that the X axis is along the conveyor with positive X direction being against the movement of the conveyor. Y axis is across the conveyor width and the Z axis is perpendicular to the conveyor surface with positive Z pointing upwards ( Figure 4 2). Each sensor was installed at a distance of 4.5 cm apart from each other. Distance from the left side and the right side conveyor wall to the respective nearest sensor was 3.5 cm The distance between two consecutive sensors represents dy value as shown in Figure 42. The distance from the sensor to the conveyor surface was 40 cm In operation, each of the infrared

PAGE 49

49 distance sensor measures distance from itself to the fruit surfa ce or conveyor, whichever is closer or appears first. This distance was subtracted from the total distance (40 cm) between the sensor and the conveyor surface to achieve z values as shown in the Figure 4 2. By correlating conveyor travel speed with the s ensor scanning frequency we can achieve data along the conveyor length and the distance between two consecutive data points that is referred to as dx. Figure 41. Schematic of IR sensors setup on the conveyor belt Figure 42. 3D view of volume calculation model

PAGE 50

50 As discussed in C hapter 3, the Sharp GP2Y0A02 IR sensor gives analog voltage output corresponding to the measured distance, so all the 16 sensors were calibrated before the actual experiment. The calibration equations thus achieved were used to convert the analog voltage data into distance in cm The unit cell volume is calculated as the volume of a parallelepiped having base of (dy dx) with a height of Zavg as shown in Figure 4 2. The measured height values (z values) were averaged (Equation 44) and used to calculate the unit cell volume (Equation 42). Summation of all such unit cell volumes along the conveyor width and the conveyor length gives the total volume of all the fruits /objects on the conveyor. TotalVolume = SumAll ( UnitCellVolume ) ( 41) UnitCell Volume = dx dy Zavg (4 2) dx = dt ConveyorSpeed ( 43) Zavg =Z 0 + Z 1 + Z 2 + Z 3 4 (4 4) w here, dt = 1/f, f = Sensor scanning frequency in Hz Z0, Z1, Z2, Z3 as shown in Figure 42 are the distance values from sensor to the object on the conveyor Overview of the Experiment A series of laboratory tests were conducted with different speeds and setup. A small conveyor system as shown in F igure 3 1 (b) similar to the actual fruit conveyor belt of an Oxbo self propelled, canopy shake and catch harvester, model 3220 was used to simulate the citrus flow during the harvesting. An array of sixteen sensors were installed on an aluminum plate each

PAGE 51

51 sensor apart at the distance 4.5 cm on the conveyor belt. The aluminum plate mounted away from the surface of conveyor was as shown in Figure 43. Figure 43. A n array of sixteen sensors mounted on conveyor belt The output terminal of all the sensors was connected to the ADC and then the data were transferred using the RS 232 output port to .dat file. The output voltage data from each sensor were combined together and a continuous string was sent and stored in a file on the laptop. The conveyor system in the laboratory has differe nt flow controls from 1 to 10 and with different speeds increasing from 1 to 10. The system was tested with four speeds of 0.35, 0.53, 0.72, 0.86 m/s The conveyor speed was measured and adjusted using the CDT 2000HD digital tachometer (ELECTROMATIC Equipm ent Co., Inc). The digital tachometer was set to measure the revolutions per minute (RPM). The conveyor rotating shaft diameter and the revolutions per minute were used to calculate the speed of the conveyor system. The tachometer was inserted

PAGE 52

52 into the conveyor shaft for four speeds and the RPM was recorded. When the system was ready, the conveyor system was operated, and the computer program was used to start data collection and stor e the data continuously in a file on the laptop. For each run, different volume of fruit was put on the conveyor and the ir respect ive volume was recorded into an E xcel file. Once the fruits passed the sensor plate on the conveyor, the computer program was stopped and the file was stored. This procedure was repeated for different set of fruits An algorithm was developed in Matlab software to post process the data stored into a file and calculate the volume on the conveyor. This experiment was performed to see the error between the actual volume of fruits passed on the conveyor and the volume calculated by the algorithm. The conveyor system was run by a hydraulic motor (Eaton Corporation, Eden Prairie, MN, USA). A hydraulic unit was used to power the system (Foster Manufacturing Corp., Racine, WI, USA) and allowed a wide range of flow and pressure rates. By changing the set point on the flow rate control valve, it was possible to change the speed of the conveyor system from zero up to 10 (maximum setting of the flow control) Materials and Methods The Sharp GP2Y0A02 ( Figure 3 3 ), an optical IR laser was selected for volume calculation of the fruits because of its operating range and the accuracy in detecting an object. As shown in the Figure 4 4, the possible location of the sensors is the area where the fruits carrying conveyor is located in a c itrus mechanical harvester. The distance available on top of the conveyor belt is not more than 50 cm and there is no scope to increase this distance, so the sensors whose detection range matches the available range w ere chosen. The operating range of this sensor was from 20 cm to 150 cm which resulted in a voltage of 0.5 to 2.5.

PAGE 53

53 Figure 44. Citrus mechanical harvester and the possible sensor location The Sharp GP2Y0A02 IR sensor use s triangulation method and a small linear CCD (cha rge coupled device) array to compute the distance from the object in the f ield of view. A pulse of infra red light is emitted by the emitter and when this light hits an object, the light reflects off an obj ect and it returns to the detector and creates a triangle between the point of reflection, emitter and the detector as shown in Figure 45. In absence of an object in the range, the light is not reflects back and the sensor does not detect any object Fi gure 45. Different Angles with Different Distance (Ref .: Acroname Robotics Inc. manual )

PAGE 54

54 As the distance to the object varies the angle in the triangle changes The receiver portion of the sensor transmits the reflected light onto various portions of the linear CCD array based on the angle. The CCD array determines what angle the reflected light came back at and therefore, it can calculate the distance to the ob ject and gives the voltage related to the distance. This sensor takes a continuous distance reading and returns a corresponding analog. System Calibration Sensor calibration was required because of the nonlinear relationship between the sensor output and the depth measurement. All the sensors were calibrated using static tests before the laboratory test. An array of sixteen laser sensors was installed on an a luminum plate as shown in the Figure 4 6. The output voltage for each of the sensor was recorded by keeping a flat plane surface of a wooden plank starting from the distance 15 cm to 65 cm at the increment of 1 cm each. For each of the distance, the analog voltage data were recorded into a file for 30 s For each distance, the analog voltage data w ere a veraged and stored. Figure 46. Sensor calibration test in the laboratory When the recorded voltage and the corresponding distance data were plot as shown in Figure 47, the calibration curve was nonlinear as expected behavior of the sensor. When the R

PAGE 55

55 square and RMS versus Polynomial order was plot as shown in Figure 48 a 3rd degree polynomial fit observed to be the best fit for the data. For each of the sensor a 3rd degree polynomial calibrat ion equation was generated and was used in the volume calculation algorithm. Figure 47. Sensor output (v) vs L ength (cm) Figure 48. R sq uare and RMS v s Polynomial order

PAGE 56

56 Laboratory Test In the laboratory test, a number of citrus fruits were grouped into a set so that each set was of different volume than the other set Volume of each set was calculated by measuring the diameter of each fruit and summing up the volumes of each fruit into a set The conveyor system was operated with speeds of 0.35, 0.53, 0.72, 0.86 m/ s and for each speed the fruit set was put on the conveyor and the data were collected. Each test was repeated 4 8 times to check the repeatability of the volume for each set. Then the data were processed using the volume calculati on algorithm and the fruit volume on the conveyor was calculated for each run. Table 4 1 shows the tr ue volume and the computed volume of the fruits. The mean computed volume is generated by averaging the computed volume of repeated runs. Table 41. Labor atory test results on t rue volume and c omputed volume True Volume x103(cm3) Compute Volume x103(cm3) %Error 3.22 3.71 15.17 3.22 4.35 35.26 3.22 3.47 7.72 3.22 5.94 84.67 5.78 6.33 9.42 5.78 5.75 0.45 5.78 7.55 30.59 5.78 5.81 0.60 5.78 5.61 3.01 2.26 2.07 8.52 2.26 2.27 0.32 2.21 1.91 13.76 2.21 1.40 36.88 4.48 3.83 14.53 4.48 3.60 19.68 4.48 3.99 10.85 4.48 5.75 28.57 270.00 6.22 97.73 270.00 55.00 79.95 270.00 4.96 98.19

PAGE 57

57 Results and Discussions The laboratory tests show that for speeds 0.35, 0. 53, 0.72, 0.85 m/s the percentage error is not consistent. To confirm the results, the system was recalibrated and percentage error between the true volume and the actual volume was found to be less as 1 percentage to high as more than 100%. The reason for such variation in the volume error can be the data collection frequency and the synchronization between the sensors. All the sensors are not synchronized with each other so the data collection from each sensor varies and because o f this the error in the calculated volume varies The result shows that as the speed increases the error has increased and authors reached to the conclusion that these sensors would not work for conveyor speed more than 1.1 m/s. For the conveyor with speed more than 1 m/s, the data collection frequency should be more than 75Hz, so the number of points on a single fruit would be at least 3. The frequency of the sensor is 21 Hz which will reduce the number of data points on a fruit ( 110/21 = 5.2 cm data at 5.2 cm) These sensors when kept close to each other, interfere and records wrong voltage and the distance between two sensors cannot be reduced which might be the cause of inconsistent error. Also, t hese sensors do not work in outdoor and different lighting conditions. Conclusion The IR sensors used for volume calcul ation were not suitable for yield monitoring in the mechanical harvesters as they interfere with each other which introduce an error in the distance measured by these sensors. T he conventional mechanical harvesters run with the speed higher than 1 m/s, so there was a need of sensors which has frequency more than 75Hz. There was a need to do some research on other sensors which can be used for volume calculation purposes and has data collection frequency more than 75Hz.

PAGE 58

58 Yield Monitoring using LIDAR (SICK Sensor) LMS SICK 200 sensor was tested on different conveyor belts in the laboratory. The objective of this section is twofold; first to explain the volumetric yield calculation algorithm implemented to post process the sensor data, different types of error sources in calculating the volume and the method of gridding interpolation to reduce the error in volume calculation. The second part details the experiments conducted to calibra te and validate the volume calculation algorithm. Volumetric Yield Calculation Algorithm The volumetric yield calculation algorithm uses discrete point data captured using SICK LMS 200 sensor mounted on the top of the conveyor. As discussed in C hapter 3, t he sensor scans the space in polar co ordinate system, in two dimension and captures the distance related information. After converting the data in Cartesian coordinate system, it represents the cross sectional information (along the conveyor width), information along the conveyor length was achieved by acquiring data from sensor at some sampling frequency. Figure 49 shows the conveyor system and the chosen three dimensional coordinate systems. The X axis is along the conveyor with positive X direction being against the movement of the conveyor. Y axis is across the conveyor width and the Z axis is perpendicular to the conveyor surface with positive Z pointing upwards as shown in the Figure 4 9. The angular resolution setting used for the sensor determin es the number of discrete data points along Y axis and the spacing between them is selected to ensure some minimum number of data points per fruit. By using the discrete data points along X, Y and Z axis, amount of citrus fruit volume passing on the conve yor system can be calculated. To calculate y & z distances, the LMS SICK polar co ordinate data were converted into the Cartesian co ordinate system. As shown in Figure 49, the sensor scanning direction is from

PAGE 59

59 right to left side of the conveyor wall in 180 angle with the angular resolution of 0.5 or 1 As discussed in the chapter 3, these parameters were set using the Labview program. The perpendicular distance z, between the point shown in the Figure 49 on the fruit and the point P, is equal to r1*sin given by the sensor and (horizontal in our case) and the current scanning location. Also, the distance between the sensor location and the point P (y) is equal to r1*cos Figure 49. Schematic of SICK sensor setup on conveyor The sensor gives 181 data points along the sensor scanning direction with the scanning angle set to 180 and angular resolut ion set to 1. Similarly we get 361 data points with setting of 180 and 0.5 angular resolution. For each of the data point, the values of z and y were calculated. Data outside the conveyor width was not useful and not considered for calculating the volume. If we just consider information equal to conveyor width then due to the nature of the scanning partial data for the fruits close to the side walls will be obtained Hence, appropriate side wall height was included in the data analysis. The average size of a fruit was considered to

PAGE 60

60 be 8 cm and hence side wall data equal to 8 cm height was included in the analysis. If we which data should be analyzed. The distance between the conveyor and the SICK sensor was set to 25 cm so that there would be a minimum distance for the fruits to pass on the conveyor system hence the height h was calculated (refer Figure 410 ) to be equal to 258 = 17 cm The 1 Figure 410. Schematic of SICK sensor on conveyor The distance between two consecutive data points along Y axis represents dy as shown in Figure 4 1 was calculated using the distance between two consecutive y values. The distance z, was subtracted from the total distance (25 cm) between the sensor and the conveyor surface to achieve z values as shown in the Figure 4 2. By correlating conveyor travel speed with t he sensor scanning frequency, we can achieve data along the conveyor length and the distance between two consecutive data points is referred to as dx. The unit cell volume and the total volume were calculated as explained in section 4.1.

PAGE 61

61 The calculated v olume was the volume generated for the different objects on the conveyor. This volume does not contain the volume of fruits alone but some other objects such as the conveyor flaps, conveyor side walls, foreign objects / trash. These error sources and the m ethod to reduce the error associated with them are discussed in the next section. Error S ources When the fruits get harvested, along with fruits other foreign objects get carried on the conveyor and then to the fruit carrying containers or goat trucks. F oreign objects include citrus tree leaves, stems and limbs. The volume calculation algorithm cant separate out these objects and the volume calculation adds up the error in the actual volume of the fruits. Also, the conveyor contains flaps and conveyor si de wall which add to the error in the volume. Flaps on Conveyor The LMS SICK sensor was mounted and tested on two machines. The first one was the t rash removal machine and the other was Continuous Shake and Catch type m echanical harvester (CCMH). These two machines have a conveyor system which differs in their orientation and the height of flaps present on them. The t rash removal machine has a horizontal conveyor with flap height of approximately 3 cm while the CCMH mach ine has inclined conveyor system with flap height of approximately 8 cm. Depending upon the height of the flap to fruit height ratio different methods were used to reduce the error produced by these flaps. Flap height to fruit height ratio < 0.5. In this method, the experiment was performed on the t rash removal machine which has horizontal conveyor system and has flaps with height 3 cm as shown in Figure 411. These flaps add an error in the fruit volume calculated by the volume calculation algorithm. If t he flap height to fruit height ratio is less than 0.5, which means the flap size is less than half the size of the fruit and the flaps can be easily removed from the data without removing the fruit related data. If the calculated z value is less than 3 cm (which is less

PAGE 62

62 than half of the fruit height) then the corresponding z value is set equal to zero. Since the sensor scans only top surface of the fruits, such an operation is not expected to remove the useful data points lying on the fruit surface. By usin g this technique most of the volume error produced by the flaps and the foreign objects which has less than 3 cm size was removed and the error was minimized. The error in volume produced by the foreign objects such as stems, leaves, limbs which lay on the conveyor and have height less than 3 cm can also be removed using this technique. Figure 411. Flaps on t rash removal machine conveyor system The flap correction technique was applied and implemented in the algorithm. When the data were plotted as show n in Figure 412, the conveyor flaps and the citrus fruits were visible on the conveyor system.

PAGE 63

63 Figure 412. (a) and (b) Spatial distribution of the raw data without removal of flaps and interpolating the data by gridding method After removing the flap error, the spatial distribution of the data as shown in Figure 413, shows just the fruits on the conveyor system, though the fruits appear to be distorted in the shape. Gridding interpolation technique was used to remove the distor tion in the fruit shape and correct the spatial distribution of the data. This technique is explained in detail in the later part of this chapter. Figure 413. After removing the flaps in the data

PAGE 64

64 Flap height to fruit height ratio >= 0.5. As discussed in C hapter 3, Continuous canopy shake CCMH shakes the citrus canopy causing fruit to fall from the tree and onto a catch frame. These fruits then carried by fruit carrying conveyor system of mechanical harvester and dropped to the containers or the goat tr uck. The conveyor is oriented in such a way that it makes 50 angles with respect to the horizontal. These conveyors have the vertical flaps (height = 7 to 8 cm) as shown in Figure 414 which hold the fruits on the conveyor and forward those to the end of the conveyor belt into the container. The flap height to fruit height ratio is 0.5 with the height of the flaps being equal to the average size of the fruits. Hence the earlier flap correction technique is not applicable as it will remove useful fruit da ta. A different technique was used to remove the error produced by these flaps and the conveyor side wall. When the algorithm calculates the volume, the data contain the information about the fruits as well as the flaps as shown in Figure 415 which needs to be removed when the actual volume is calculated. Therefore, to remove the error added by flaps and side wall of the conveyor, before collecting the data for actual fruits, data were collected by operating the conveyor empty for 30 s for each test speeds This data were processed and the amount of volume generated by the empty conveyor was referred as an error volume. The distance between two flaps was measured on the conveyor and from the test time and the speed; length of the conveyor covered in anal ysis was calculated. Using this information, the total number of flaps present on the conveyor was determined. The error volume contribution per flap was calculated and then removed from the actual test data. However the error volume produced by foreign objects (trash, leaves, and branches) cannot be removed using this technique.

PAGE 65

65 Figure 414. Conveyor system with flaps to hold the fruits Figure 415. Spatial distribution of the data showing flaps alone on the conveyor belt

PAGE 66

66 The sensor data in polar coordinate system when converted into Cartesian system; shows distortion in the shape of the fruits as shown in Figure 416. When we consider the cross sectional data (Y Z plane data) attained from the sensor, due to the polar swe eping nature of the scanning, once we encounter the end of the fruit, the coordinates of the next point on the conveyor surface lay considerable amount away from the previous data point on the fruit. These results in the spikes being observed directed tow ards the sensor mounting location. This distortion in fruit shape can add a huge error in volume calculation, so this data needs to be corrected. A method of gridding interpolation was used to correct this data and the error in volume was minimized. Figu re 416. (a) and (b) The distortion in the fruit shape due to the collection and transformation of data from polar coordinate system to Cartesian co ordinate system Method of Gridding Interpolation The radial spikes observed in the raw data leads to erroneous volume being calculated using the volume calculation method employed in this research. To take care of this, uniformly spaced Y vector is created having the same dimension as the original raw data and data gridding interpolation was performed using the Matlab native function griddata.

PAGE 67

67 Figure 417. Fruits on conveyor belt In the process griddata fits a cubic surface to the original raw data and then interpolates this surface at the points specified by the uniformly spaced X and Y arrays to produce Z at these newly formed data points. The cubic surface is created such that it always passes through the original raw data points. Figure 418. (a) and (b) The radial spikes observed in the sha pe of the fruit before applying gridding interpolation

PAGE 68

68 By using gridding interpolation we get uniformly spaced data points on the conve yor surface and no radial spikes were observed (refer Figure s 417, 4 18, 419 and 420). The volume calculated based on the gridding data points was observed to be more accurate than the volume calculated based on the original raw data. Figure 4 19. Aft er using gridding interpolation, uniformly spaced data points on the conveyor surface without radial spikes Figure 420. After using gridding interpolation, uniformly spaced data points on the conveyor surface without radial spikes

PAGE 69

69 Experiments on Horizo ntal Conveyor ( Trash removal machine f/h <0.5) Overview of Experiment The catch and shake type of mechanical harvester (CCMS) cant be stopped frequently during the field operation. These systems are very expensive ; so the growers use them efficiently wi th maximum utilization for harvesting the citrus fruits in the grove. To avoid interruption and downtime of these mechanical harvesters a system was built called Trash removal machine shown in t he Figure 3 2 and discussed in C hapter 3. The performance of the t rash removal machine was tested in the real field conditions during the summer of 2010 at Fort Basinger, FL. The top portion of this system as shown in Figure 421 was where the fruits from the goat truck were dropped for trash removal and volume measurement purpose using SICK sensor. Figure 421. The goat truck dumping the fruits on the top container of the t rash removal machine The SICK sensor was mounted as shown in Figure 4 22 on the top of the conveyor where the fru its were carried after removal of the stems, leaves and other trash from the trash removal machine

PAGE 70

70 Figure 422. SICK LMS 200 location on t rash removal machine Two experiments were performed on the t rash removal machine In the first experiment citrus f ruit loads from 4 to 41 kg were used to calibrate the system and then two loads of 11 and 22 kg were used to evaluate the performance of the system using the developed calibration equation. The volumetric yield calculation system performed very well with the percentage error less than 8%. In the second experiment, loads from 4 to 41 kg were used to calibrate the system and then system was validated with five different loads. In both the experiments SICK sensor was set to 180 angle with 1 angular resoluti ons. The parameters used for the volume calculation algorithm are shown in Table 42. These parameters were measured and recorded before the experiment was performed. Method and Materials The SICK sensor was mounted on an aluminum plate perpendicular to the conveyor surface as shown in Figure 4 22. The aluminum plate was welded on top and at the center of the conveyor such that the sensor will be at the center of the conveyor belt. The dist ance from the sensor to the conveyor surface was about 25 cm The sensor has supplied a power of 24 V and

PAGE 71

71 the output with R S 422 port was connected to a s ea level port to the computer. The sea level port converts the RS 422 input to USB port. Different par ameters such as the distance from the SICK LMS to the conveyor surface, width of the conveyor, the flap length, average distance between two consecutive flaps, distance from sensor midpoint to left and right side conveyor wall, frequency of the sensor and different speeds were recorded before the experiment as shown in Table 4 2. For this experiment around 90 kg of oranges were collected from the orange field present at the Citrus Research and Education C enter, Lake Alfred, FL. These fruits were divided int o seven bins with mass from 4 to 41 kg and 11 and 22 kg When the volumetric yield calculation system was ready, the conveyor was started using the hydraulic power and adjusted to a particular speed. Data were collected by pouring the fruits from each bin manually on the conveyor system. Table 42. Different parameters used the two experiments for volume calculation algorithm Experiment A Experiment B Date of experiment 16 Jul 10 22 Jul 10 Sensor Frequency ( Hz ) 75 75 Conveyor Speeds (m/s) 0.92, 1.03, 1.14 1.7, 1.3, 1.0 Angle / resolution 180 /1 180 /1 Conveyor width (cm) 63 63 Distance from sensor to right side conveyor wall 32 32 Distance from sensor to conveyor surface (cm) 25 25 Distance between two flaps (cm) 27 27 Experiment A Calibration The system was calibrated by putting the citrus fruits from each bin of different mass The t rash removal machine conveyor was operated for three different speeds of 0.92, 1.03, 1.14 m/s and for each speed a mass from 4 to 41 kg a t the increment of 4.5 kg was put onto the conveyor system. For each mass the data were collected three times to check the repeatability

PAGE 72

72 and the volume variation for that particular mass After performing the tests for calibrating the system, validation da ta were collected with mass 11 and 22 kg For each run the data were stored to respective .dat file on the computer and then this data were post processed using the Matlab program developed on the basis of the volume calculation algorithm explained in the earlier part of this section Tables 43, 44 and 45 shows the calibration data for speeds 0.92, 1.03 and 1.14 m/s As discussed, the experiment was repeated three times for each mass the volume for each mass were calculated and the average volume was used to plot the data. Table 43. Calibration data with speed 0.92 m/s with angle 180 and angular resolution 1 Mass ( kg) Calculated Volume x 103 ( cm3) With Grid data Calculated Volume x 103 ( cm3) 4.54 4.85 4.68 9.08 13.55 11.03 13.62 17.46 15.43 18.16 23.22 21.15 21.51 29.04 25.40 When the actual mass versus the calculated volume using grid data as shown in F igure 423 (b), 424 (b) and 425 (b), for respective speeds were plot, linear trend was observed between the mass and calculated volume with R2 = 0.99. Also, if the actual mass versus the volume calculated without using the grid interpolation technique as shown in Figure 423 (a), 424 (a) and 425 (a), were plot it show ed a linear trend with R2 0.98.

PAGE 73

73 (a) (b) Figure 423. (a) Mass vs. C alculated volume calibration curve with speed 0.92 m/s and (b) M ass vs. C alculated volume with grid data interpolation technique, calibration curve with speed 0.92 m/s Table 44. Calibration data with speed 1.03 m/s with angle 180 and angular resolution 1 Mass ( kg) Calculated Volume x 103 ( cm3) With Grid data Calculated Volume x 103 ( cm3) 4.54 5.18 4.87 9.08 12.77 10.61 13.62 17.83 15.68 18.16 23.06 20.42 21.51 27.96 25.64 (a) (b) Figure 424. (a) Mass vs. C alculated volume with speed 1.03 m/s and (b) M ass vs. C alculated volume with grid data interpolation technique with speed 1.03 m/s

PAGE 74

74 Table 45. Calibration data with speed 1.14 m/ s with angle 180 and angular resolution 1 Mass ( kg) Calculated Volume x 103 ( cm3) With Grid data Calculated Volume x 103 ( cm3) 4.54 5.55 5.02 9.08 12.06 10.65 13.62 17.82 14.94 18.16 26.44 23.94 21.51 30.06 25.87 (a) ( b) Figure 425. (a) Mass vs. C alculated volume with speed 1.14 m/s and (b) Mass vs. C alculated volume with grid data interpolation technique, calibration curve s with speed 1.03 m/s Validation Two validation tests were performed with mass 11 and 22 kg for each of the 3 speeds. When the calibration equations were used to predict the mass of fruits on the conveyor, the % error shows very promising results with percentage error less than 8%. Table 46 shows the calculated volume and percentage error betwe en the actual and calculated mass in kg Results and Discussion When the % E rror versus the mass of the fruits on the conveyor was plot ted as shown in Figure 4 26 with respect to the speed, it sh ows that as the mass increased, the error decreased The ca libration data and the validation data were not sufficient to prove the results so more tests are needed to confirm the error with respect to speed and mass of the citrus

PAGE 75

75 fruits on the conveyor system. Also, experiments with high speed need to be repeated to verify the performance of the system with speeds higher than 1.14m/s. Table 46. Validation of the data after applying the calibration equations Speed (m/s) Mass ( kg) Calculated Volume x103( cm3) With Grid Data Calculated Volume x103 ( cm3) With Grid data Calculated Mass ( kg) With grid %Error 0.92 11.35 14.67 12.95 11.09 2.32 0.92 22.70 27.56 24.26 20.78 8.47 1.03 22.70 27.69 24.74 21.31 6.13 1.14 11.35 16.29 13.46 11.03 2.84 1.14 22.70 28.86 25.87 21.20 6.60 Figure 426. % Error vs. M ass plot with different speeds Experiment B Calibration In this experiment the volumetric yield calculation system was recalibrated with mass from 4 to 41 kg at the increment of 4.5 kg The experiment was repeated three times for

PAGE 76

76 each mass and for 1.7, 1.3, 1.0 m / s conveyor speeds and the system performance was analyzed. The conveyor speed of 1.7, 1.3, 1.0 m / s were chosen to evaluate the system performance when the speed of the conveyor belt increases and reaches its higher speed of 1.7 m/s. Once the cal ibration data were collected, system was validated using with different mass citrus fruit bins on conveyor. After collecting the calibration data, it was analyzed using the volume calculation algorithm and the amount of fruit volume on the conveyor was det ermined. As shown in Table s 47, 48, 49 for each mass and speed the volume is calculated with and without applying the gridding interpolation. When the volume calculated using gridding interpolation technique and the mass were plot for different speeds a s shown in Figure s 427, 4 28 429, it showed the linear trend and coefficient of determination R2 equal to 0.99. The calibration equations for each speed were used for validating the data. Validation The calibration equations derived by plotting the act ual mass and the calculated volume for each speed were used to validate the next set of data. Table 4 10, 411, 412 shows the actual mass and the calculated volume using those calibration equations. Results and Discussions The validation data shows that as the speed was increased the root mean squared error was reduced. As shown in Table 410, 411, 412 for speed 1.71 m/s the RMSE error was 0.98 kg. This RMSE error was increased as the speed was decreased. For speeds 1.34 m /s, 1.03 m/s the RMSE error was 0.81 kg and 3.21 kg respectively. When the percentage e rror versus the mass of the fruits on the conveyor was plot as shown in Figure 4 30 with respect to the speed, it shows that as the mass increases error decreases, so fo r the mechanical harvesters with continuous stream of fruits the error might go down. Also, the

PAGE 77

77 plot shows that for speed s higher than 1.71 m/s, the system performed well with an error less than 8 % for mass greater than 20 kg and less than 5% for the spee d 1.34 m/s Table 47. Calibration data with speed 1.7 m / s Mass ( kg) Calculated Volume x103( cm3) With Grid data Calculated Volume x103( cm3) 4.54 5.52 5.18 9.08 12.25 10.93 13.62 19.78 16.74 18.16 25.62 21.94 22.70 33.93 29.35 27.24 37.96 32.45 31.78 45.12 37.61 36.32 51.68 44.53 40.86 58.53 49.77 4.54 6.95 6.07 9.08 13.70 11.55 13.62 20.28 17.59 18.16 27.29 22.85 22.70 33.23 29.37 27.24 39.32 33.83 31.78 45.45 37.18 36.32 52.26 44.38 40.86 57.95 48.84 4.54 7.39 6.22 9.08 13.58 11.65 13.62 20.85 18.36 18.16 28.04 23.93 22.70 33.75 28.93 27.24 38.78 32.74 31.78 42.33 35.85 36.32 55.05 48.96 40.86 60.09 51.60

PAGE 78

78 Figure 4 27. Actual mass vs. C alculated volume calibration curve with speed 1.71 m/s Table 48. Calibration data with speed 1.34 m / s Mass ( kg) Calculated Volume x103( cm3) With Grid data Calculated Volume x103( cm3) 4.54 6.05 5.17 9.08 13.72 12.67 13.62 19.17 16.66 18.16 24.19 20.51 22.70 31.53 27.09 27.24 34.96 31.54 31.78 43.91 37.54 36.32 48.27 42.11 40.86 53.04 44.88 4.54 6.05 5.17 9.08 11.86 10.95 13.62 20.05 17.52 18.16 26.38 23.22 22.70 29.79 25.71 27.24 38.75 34.17 31.78 43.12 36.38 36.32 48.30 41.51 40.86 52.05 45.88 4.54 6.77 6.33 9.08 12.53 11.13 13.62 20.12 18.02 18.16 26.17 22.68 22.70 32.30 28.00 27.24 39.32 34.26 31.78 39.33 33.30 36.32 46.65 39.68 40.86 45.93 39.65

PAGE 79

79 Figure 428. Actual mass vs. C alculated volume calibration curve with speed 1.34 m/s Table 49. Calibration data with speed 1.03 m/s Mass ( kg) Calculated Volume x103( cm3) With Grid data Calculated Volume x103( cm3) 4.54 5.18 4.87 9.08 12.77 10.61 13.62 17.83 15.68 18.16 23.06 20.42 21.51 27.96 25.64 27.24 30.21 24.37 31.78 36.78 31.73 36.32 47.14 40.05 40.86 48.54 40.86 27.24 30.72 26.74 31.78 38.66 33.85 36.32 39.09 33.68 40.86 45.69 39.24 27.24 30.48 25.88 31.78 39.33 33.30 36.32 46.65 39.68 40.86 45.93 39.65

PAGE 80

80 Figure 429. Actual mass vs. C alculated volume calibration curve with speed 1.03 m/s Table 410. Validation data and error with speed 1.71 m/s Actual Mass ( kg) (Y) With Grid data Calculated Volume x103 ( cm3) Calculated Mass x103 ( cm3) (Y') %Error (Y Y')^2 9.19 14.57 12.09 6.83 0.39 23.24 33.45 28.40 0.70 0.03 16.12 26.27 21.60 8.92 2.07 35.66 54.11 45.62 3.95 1.99 29.94 45.05 37.63 2.13 0.41 RMSE ( kg ) 0.98 Table 411. Validation data and error with speed 1.34 m/s Actual Mass ( kg) (Y) With Grid data Calculated Volume x103 ( cm 3 ) Calculated Mass x103 ( cm3) (Y') %Error (Y Y')^2 9.19 13.51 11.20 4.55 0.18 23.24 31.69 27.46 1.36 0.10 16.12 21.47 18.82 0.21 0.00 35.66 48.07 40.69 2.09 0.56 29.94 37.64 33.06 5.25 2.47 RMSE ( kg ) 0.81

PAGE 81

81 Table 412. Validation data and error with speed 1.03 m/s Actual Mass ( kg) (Y) With Grid data Calculated Volume x103 ( cm 3 ) Calculated Mass x103 ( cm3) (Y') %Error (Y Y')^2 9.19 12.88 10.51 12.07 1.23 23.24 29.22 25.81 8.88 4.26 16.12 21.78 19.38 17.94 8.36 35.66 48.65 41.91 15.24 29.55 9.94 38.30 33.49 9.70 8.43 RMSE (kg) 3.21 Figure 430. % Error based on actual mass with respect to the speed of conveyor Experiments on Inclined Conveyor (CCSC, f/h 0.5) As discussed in the section 4.2, different types of error sources introduce an error in volume calculation of fruits on conveyor. This section describes few methods that were used to minimize the errors. The conveyor system present on CCSC machines are inc lined and the flap to fruit size ratio was 0.5. The performance of volume measurement system was tested on the conveyor system used in the CCSC. This section will discuss es the experiments performed, material and method used, the calibratio n and validati on tests and the result s in detail.

PAGE 82

82 Overview of Experiments Two experiments were performed in the laboratory on a small conveyor system as shown in Figure 4 31. The conveyor system was used to simulate the flow of citrus fruits during the harvesting using CCMH. The volume measurement system was mounted on the conveyor system as shown in Figure 4 31. Different parameters used in both the experiments as shown in T able 413 were recorded before the experiment. Table 413. Different parameters used in volume calculation algorithm Experiment 1 Experiment2 Date of experiment 12 Jan 10 27 Jul 10 Sensor Frequency 24 Hz 75 Hz Conveyor Speeds (m/s) 1.1 0.587, 0.785, 1.1 Angle / resolution 180 /0.5 180 /1 Conveyor width (cm) 76 76 Distance from sensor to right side conveyor wall (cm) 36.9 36.5 Distance from sensor to conveyor surface (cm) 28.7 35.6 Distance between two flaps (cm) 40 40 The system was calibrated and then the calibration data were used to validate the next set of data. System performance was analyzed by the amount of error in calculated mass of fruits with respect to the actual mass of fruits on the conveyor system. Materi al and Methods The SICK sensor was mounted as shown in Figure 4 31 on the top of the fruit carrying conveyor present and then fruits were dropped on to the conveyor system.

PAGE 83

83 During the first experiment the SICK sensor was set on 180 scanning angle and 0.5 angular resolutions while the second experiment was performed with 180 scanning angle and 0.5 angular resolutions set using the Labview program. The conveyor system has different flow controls from 1 to 10. The speed of the conveyor was measured using Digital tachometer The tachometer was inserted into the conveyor shaft by setting the flow control on 10 and the RPM was recorded. The diameter of the conveyor shaft was 20 cm and was used to calculate the shaft revolutions per min and then calculate the speed of the conveyor system. Figure 4 31. LMS SICK 200 for laboratory test on the conveyor belt Experiment A Calibration In this experiment the system was calibrated for speed 1.1 m/s with heavy loads from 11 kg to 85 kg For each mass three tests were carried out, the average volume was

PAGE 84

84 calculated for each set of mass and this averaged volume was used to calibrate the system. Table 414 shows the calculated volume for the fruits on the conveyor for each run while Table 415 shows the averaged mass and the volume for each set of mass The calibration plot of actual mass and the calculated volume as shown in Figure 4 32 has linear trend between the actual mass and the calculated volume with good coefficient of correlation of 0.99. Table 414. Validation data and error with speed 1.1 m/s Actual M ass ( kg ) Calculated V olume x103( cm3) 11.80 31.00 11.78 34.99 11.76 40.01 23.31 77.37 23.24 62.39 23.24 75.97 34.82 102.72 34.57 94.11 34.57 107.18 48.19 107.18 48.03 135.81 48.12 145.09 62.18 102.03 62.08 126.00 61.97 139.63

PAGE 85

85 Table 415. Validation data and error with speed 1.1 m/s Actual M ass ( kg) Calculated V olume x103( cm3) 11.78 35.33 23.27 71.91 34.65 101.34 48.08 140.45 84.58 215.50 Figure 432. Calibration plot of mass vs. C alculated volume of fruits calculated after correcting the data with gridding method and removing the flap error Validation The same data were used for validation test and when the calibration equation was applied there was a significant error in the volume. Table 416 shows that the percentage error and the SEP is more if the method of gridding and flap correction was not used. In addition, this error may increase if the conveyor is running for hours. In an ideal field conditions, the mechanical harvester does not stop that frequently unless there are problems or the workers are taking break. Thus, the flap error should be removed continuously during yield prediction.

PAGE 86

86 Table 416. Standard and the percentage error in the actual and predicted mass 1.1 m/s Actual M ass ( kg ) Calculated Volume x103 (cm3) Calculated M ass ( kg) %Error 11.78 35.33 10.24 13.08 23.27 71.91 25.18 8.24 34.65 101.34 37.21 7.38 48.08 140.45 53.19 10.63 84.58 215.50 83.86 0.85 RMSE ( kg ) 2.8 Results and Discussions The volumetric yield monitoring system needs to validate for different set of data and also the flap error needed to decrease or removed from the data which will improve the performance of the system and reduce the amount error in yield prediction. The reason for the high error in volume might be the frequency which was 36 Hz. Experiment B In this experiment, the algorithm was modified to minimize the error produced by the flaps; also a new set of validation test was performed to see the performance of the volumetric yield measurement system. Also to increase the data acquisition frequency from 36 Hz to 75 Hz, the angular resolution was changed to 10 from 0.50. Calibration For the system calibration, the fruits were weighed from 4 to 41 kg and put into the bins then for each speed the fruit bins were dropped onto the conveyor and the data were collected. The system performance was tested for three different speeds. For each mass the experiments were performed three times, to see the repeatability in volu me calculation. Then the validation tests were performed with five different mass When the data were processed, the volume for the amount of fruits on the conveyor system was calculated. Table 417, 418 and 4 19 shows the data collected with 0.587, 0.785, and 1.1 m/s speeds. When the actual mass versus

PAGE 87

87 the volume calculated were plot it showed linear trend with the coefficient of correlation R2 = 0.98 as shown in Figure 4 33, 434, 435. The calibration equations with different speeds were applied on the validation set to check the performance of the volume measurement system. Validation System calibration shows a linear trend with coefficient of correlation 0.99 between the actual mass and the calculated volume as shown in Figure 4 33, 4 34 and 435. Th ese calibration equations were used to validate the system. A data set with different speeds was collected and the calibration equations were applied to calculate mass of fruits in kg on the conveyor system. Table 4 20, 421, 422 shows calculated mass fro m the amount of volume on the conveyor system with speeds 0.586, 0.785 and 1.1 m/s respectively. Results and Discussions The result show ed that for different speeds the error reduces as the amount of fruit on the conveyor increases. As shown in the Figure 4 36, for each speed the percentage error has reduced and it is less than 5% for the mass more than 20 kg This trend shows that in the actual field, percentage error will be less than 5% as the amount of load on the conveyor is more than 35 kg The algor ithm removes most of the flap error but it cannot completely remove it as the flap error is not consistent all the time. This flap error gets added if the conveyor runs without fruits for long time. In actual field conditions this situation may occur at th e start or at the end of the citrus tree harvesting or when the mechanical harvester is moving from one tree to an other. Also this condition may occur when the mechanical harvester starts harvesting and initially the conveyor is empty. This technique shows the potential use volumetric yield calculation using LMS SICK sensor in yield monitoring of citrus in m echanical harvester.

PAGE 88

88 Table 417. Calibration data with speed 0.586 m / s Actual Mass (kg) Calculated Volume x103 (cm3) 4.54 10.75 4.54 10.13 4.54 11.82 9.08 15.61 9.08 16.55 9.08 18.15 13.62 25.44 13.62 24.60 13.62 27.61 18.16 34.89 18.16 37.68 18.16 26.10 22.7 35.63 22.7 39.54 22.7 39.23 27.24 44.12 27.24 44.78 27.24 51.77 31.78 50.70 31.78 52.21 31.78 51.60 36.32 59.32 36.32 60.59 36.32 54.82 40.86 71.45 40.86 66.23 40.86 73.32

PAGE 89

89 Table 418. Calibration data with speed 0.785 m/s Actual Mass (kg) Calculated Volume x103 (cm3) 4.54 10.89 4.54 9.36 4.54 7.27 9.08 16.90 9.08 15.87 9.08 16.34 13.62 22.17 13.62 24.37 13.62 21.73 18.16 30.20 18.16 31.25 18.16 27.21 22.7 38.12 22.7 36.51 22.7 109.27 27.24 43.54 27.24 44.76 27.24 42.44 31.78 48.36 31.78 50.10 31.78 45.21 36.32 49.35 36.32 48.21 36.32 53.15 40.86 59.84 40.86 59.54 40.86 60.09

PAGE 90

90 Figure 433. Calibration curve with speed 0.586 m/s Figure 4 34. Calibration curve with speed 0.785 m/s

PAGE 91

91 Table 419. Calibration data with speed 1.1 m / s Actua l Mass (kg) Calculated Volume x103 (cm3) 4.54 13.51 4.54 15.08 4.54 9.67 9.08 17.25 9.08 16.39 9.08 17.03 13.62 26.21 13.62 23.05 13.62 28.45 18.16 34.04 18.16 68.25 18.16 45.13 22.7 38.68 22.7 37.51 22.7 38.11 27.24 44.07 27.24 40.55 27.24 42.56 31.78 52.07 31.78 47.88 31.78 56.84 36.32 57.92 36.32 61.22 36.32 59.26 40.86 66.64 40.86 63.64 40.86 64.22

PAGE 92

92 Figure 435. Calibration curve with speed 1.1 m/s Table 420. Validation data and error with speed 0.586 m/s Actual Mass (kg) Calculated Volume Calculated Mass (kg) % Error (Y Y')^2 6.67 12.48 5.77 13.47 0.81 11.58 20.49 10.80 7.62 0.61 19.84 32.51 18.33 6.75 2.29 26.92 47.68 27.83 3.36 0.82 38.25 64.16 38.15 0.24 0.01 RMSE (kg) 0.95 Table 421. Validation data and error with speed 0.785 m/s Actual Mass (kg) Y Calculated Volume x 103 (cm3) Calculated Mass (kg) Y' % Error (Y Y')^2 6.67 11.57 5.62 15.84 1.12 11.58 15.18 8.24 5.45 11.15 19.84 29.66 18.76 32.89 1.17 26.92 36.24 23.54 12.57 11.46 38.25 55.45 37.49 1.99 0.58 RMSE (kg) 2.25

PAGE 93

93 Table 422. Validation data and error with speed 1.1 m/s Actual Mass (kg) Y Calculated vol wt grid and flap correction x 103 (cm3) Calculated Mass (kg) Y' % Error (Y Y')^2 38.25 63.23 39.43 3.07 1.39 26.92 41.24 24.65 8.43 5.16 19.84 32.71 18.92 22.74 0.85 11.58 17.89 8.96 4.64 6.86 6.67 13.87 6.26 6.23 0.17 RMSE (kg) 1.69 Figure 436. Error spread with different speeds

PAGE 94

94 CHAPTER 5 Y I E LD MONITORING INTERFACE As discussed in C hapter 1, one of the components of yield monitor is the user interface and a console unit located in the combines cab as shown in F igure 5 1. This unit was integrated with the other yield monitor components such as the speed sensor, mass or volume of the grain/fruits flow sensor, grain moisture sensor, GPS / DGPS receiver unit. The console was also used to enter the operator specified information such as the fieldname, block number, grain/fruit type, calibration numbers, and correlation factors. Figure 51. Yield monitor components (left), Yield monitor 8 t ouch Screen Panel (right) The yield monitor interface unit provides the medi um to store all this information which will then be used for the post processing. This information was also used by the user interface onthe go to calculate the yield of the harvested grain/fruits. The console connected with the DGPS receiver stores the harvester position related information. These spatially indexed data were combined with the yield information and later used to produce the yield maps. Most yield monitors can display the instantaneous yield and the total amount of yield harvested at a given time and provides statistics for loads or batches of grain / fruits within an area of a field.

PAGE 95

95 Importance of Yield Monitoring Interface A load cell based yield monitoring system was developed by Maja et al., (2009) that use d load cell to measure the impact force and correlate it to the mass of citrus fruits getting harvested The load cell sensors were attached to a plate made up of a carbon fiber and was placed at the end of the conveyor belt of an Oxbo machine. Due to the large sets of data, all of the computation for yield was done off line. The real time data were actually stored in a flash disk and then was transferred to the PC once a month for processing. The yield data were then made available to the growers on a monthly basis. A real time data logging and display or interface would prove very beneficial for the growers and at the same time for the operator of the machine. Most of the times the yield monitoring systems are calibrated and once it is calibrated then the calibration information is used to calculate the amount of fruits harvested in kg The calibration information, such as the slope and the intercept is recorded in the yield monitor interface before the mechanical harvesters start to harvest the fruits in the grove. The yield m onitor interface is explained in detail in the next paragraph. Design and Development The yield monitor interface was developed for the load cell based yield monitoring system (Maja et al., 2009). The system can be used for any other yield monitoring syst em with minor changes in existing software. The load cell based system produced the analog voltage for the continuous impact produced by the harvested fruits on the plate and this analog output from the entire load cells were combined using the summing boa rd. Also, the position of the harvester was recorded using the GPS unit mounted on the mechanical harvester. The GPS information and the summed analog voltage was combined together and sent wirelessly to the yield monitor interface for further processing. Figure 5 2 shows the snapshot of the data sent wirelessly to the interface console.

PAGE 96

96 Figure 52. Snap shot of the data from GPS and load cell During the harvest season two mechanical harvesters move side by side in a row harvesting the citrus trees, co llecting the fruit and transferring the harvested fruit to the haul truck (sometimes referred to as the goat truck) as shown in F igure 5 3. The fruit collection frame of one of these two mechanical harvesters in a row is on top of the fruit collection frame of other mechanical harvester (bottom) in the second row. In mechanical harvesting operation the top row refers to the row where a mechanical harvester frame is on top of the another. The fruit get s harvested by these two harves ters, so the yield monitor interface can be installed on either one of the harvester. The yield related data and the GPS position information from the individual harvesters is transferred wirelessly to the interface.

PAGE 97

97 Figure 53. Mechanical harvesting machines on top and bottom rows in the grove Figure 5 4 Yield monitoring interface mounted in the Oxbo machine The yield monitor interface was developed using Labview software from National Instruments. This software was then installed on an 8 t ouch scre en panel (Comfile technology model CUPC P80) As shown in F igure 55, the interface shows six tabs on the interface such as the Drivers input, Yield monitor top, Yield monitor bottom, Combined, Setup 1, and Setup 2. The machine operator is responsible to input the information about the grove such as field, block, crop type, the goat truck capacity, the calibration data ( the slope and the intercept) and other information shown in F igure 55.

PAGE 98

98 Figure 55. Yield monitor interface with drivers input tab The calibration information is used to calculate the amount of fruit harvested in mass, kg as the mechanical harvester moves along the row and harvest the fruits. The Yield monitor top tab as shown in Figure 56, shows the information about the mechanical harvester on the top row. It shows the amount of fruits harvested by calculating the instantaneous yield using the calibration slope and the intercept constant specified by the operator in the driver s input tab. It calculates the total fruits harvested in terms of the cumulative yield and the number of boxes. The calibration information such as the slope and intercept can be computed by running a simple test on the Oxbo machine with a define d weigh of oranges and correlate it with the impact data from the control box. It can be done once before the harvesting starts and the calibrated slope and intercept can be used for the whole harvest season. One of the nice features of th e interface build was the alarm ing system. The operator can set the trailer capaci ty in one of the interface input and when the citrus trailer reach its capacity, the system alarms the operator that the trailer is full and a new trailer should be used by the goat truck. This feature can save the trailers from over loading its capacity an d eventually avoid the possibility of paying the fines for littering the fruits on the road.

PAGE 99

99 Figure 5 6. Yield monitor interface with top row mechanical harvester information tab The mechanical harvester bottom row as shown in F igure 57, is the same as for the top row. This interface collects the yield related information from the other mechanical harvester on the bottom side. Figure 5 7. Yield monitor interface with bottom row mechanical harvester information tab

PAGE 100

100 The next tab ( F igure 5 8) with combine yield shows the amount of fruits harvested by both of these top and bottom side mechanical harvesters. This tab gives the information about the yield and total number of boxes harvested combined by the two harvesters. Figure 5 8. Yield monitor interface with bottom row mechanical harvester information tab The setup1 and setup 2 tabs as shown in Figure 5 9 and 510 are used to set the RS 232 terminals for the wireless input from the GPS and the impact plate output ports from the tw o harvesters. Setup shows the additional alarm playing setting after the fruit trailer reaches to its specified capacity. The yield monitoring software processes the data from the two mechanical harvesters and creates two files. The first file is the raw d ata file and the other one is for the processed data. When the yield monitoring interface starts collecting the data, it creates a folder named YieldMonitor_Interfacedata on C: \ drive of the panel and creates two subfolder in the main folder called RawData and ProcessedData. Once these folders get created it creates another two files for the raw data received from the two mechanical harvesters and three processed data files

PAGE 101

101 after processing the individual data from the respective harvesters such as top an d bottom side mechanical harvesters and a third file which was the combination of the processed data. Figure 59. Yield monitor interface top row mechanical harvester data communication port setting tab Figure 5 10. Yield monitor interface bottom row m echanical harvester data communication port setting tab

PAGE 102

102 The sample files generated are shown in F igure 5 11. The F igure 5 11(a) and (b) show s the sample file raw data received from the top and bottom side mechanical harvester respectively. Figure 5 11 (c) shows the sample file generated after processing the raw data received from the mechanical harvesters. The se files are stored in the panel memory and used later to generate yield maps. (a) (b) (c) Figure 511. (a), (b) Snap shot of raw data for the top and bottom row harvesters, (c) T he processed data after combining raw data from top and bottom row harvesters. Conclusion and Discussion The yield monitor interface developed for the impact based system can be mo dified and used for the volumetric yield monitoring system. As presented earlier, it shows that there was information that needs to be placed in the program for this to work properly. One of the most important parts of it was the calibration, e.g. slope and intercept which can be made available by doing a simple test of known weighs. The y ield monitor ing interface also has very important feature that tell the operator of the harvester that the trailer which is located outside the harvested block is already full based on its current harvest and thus another trailer should be used. This saves both time and money for the grove owners as they will know that the trailer that transports their fruits from their grove to the processing plant is not overloaded.

PAGE 103

103 CHAPTER 6 CONCLUSION AND FUTURE WORK The overall objective of this study was to explore the use of noncontact method specifically LIDAR technology to measure volumetric flow on a conveyor system to predict the yield of citrus fruits. A methodology was developed to use the cross sectional inf ormation captured over specified amount of time to calculate the amount of mass and thus fruits harvested and collected on the citrus mechanical harvester and the t rash removal machine This objective was achieved by testing two types of laser sensors to m easure the cross sectional information The first sensor system consist ed of a linear array of IR sensors from which data were available in Cartesian co ordinate system while the second sensor was a single LMS SICK sensor which measures the data in polar coordinate system. An algorithm was developed to process the data collected by these sensors and generate the amount of volume passing on the conveyor system by doing time integration of the information A linear calibration curve was developed to calculat e the actual mass ( kg ) of the fruits and thus yield in terms of measured fruit volume on the conveyor system. Five different laboratory experiments were conducted to achieve this objective. The first experiment was performed by mounting a linear array of (Sharp GP2Y0A02) sixteen IR laser sensors on the conveyor system. The repeatability of the sensor was questionable with measured volumes showing errors anywhere from 10% to 100%. Sensors were found to interfere with each other when the distance between the m was less than 4 cm Sensors were also found to be sensitive to the ambient lighting conditions. Additional possible source of error could be sensor synchronization or rather lack of it. The highest possible sensor frequency of 23Hz is found to be in suffi cient for the conveyor speeds of more than 1 m /s, the number of data points per fruit were less than three which adds error in fruit volume calculation.

PAGE 104

104 Two experiments were conducted by mounting the LMS SICK laser sensor on the horizontal conveyor system of t rash removal machine The results were encouraging with error less than 5% for citrus fruits weighing more than 40 kg with conveyor speed of 1.34 m/s. The mass measurement error was found to be less than 8% with a conveyor speed of 1.71 m/s, the maximum possible setting of current t rash removal machine Additional two experiments were performed using the same sensor on an inclined conveyor system present in the mechanical harvester. This conveyor has flaps of height 8 cm to hold the fruits and t o carry them towards the fruit collecting bins or containers. These flaps introduce error in the volume measurement, which was corrected by using a 30 second dry run of conveyor as a reference. The volumetric yield calculation system was calibrated for the speeds of 0.58, 0.875 and 1.1 m/s and the linear calibration curve information was used to predict the citrus fruit mass in kg The results showed that for the maximum speed of 1.1 m/s, the coefficient of determination, R2 between the actual yield and the predicted yield was 0.989 and the RMSE was 1.69 kg for a measurement range of 4 to 41 kg The error is expected to reduce with increase in mass which increases the conveyor occupancy. The error in predicted mass was found to be less than 5% for the fruit mass of more than 20 kg It will be safe to conclude that in the actual field, percentage error will be less than 5% as the amount of load on the conveyor is normally more than 35 kg Thus the non contact measurement method using the LIDAR technology was successfully implemented and its usefulness in measuring volumetric flow of fruit and the yield was successfully demonstrated for citrus mechanical harvester as well as the t rash removal machine As future work, field trials can be conducted to ascertain the performance observed in the laboratory setting. Using more than one LMS SICK sensors can be explored to further reduce the

PAGE 105

105 error in predicted yield. The current volume calculation algorithm can be further developed for real time yield measurement.

PAGE 106

106 A PPENDIX LMS SICK DATA COLLECTION ALGORITHM This appendix gives an algorithm implemented in Matlab software to process the da ta collected when the LMS SICK s canner was mounted on the conveyor system. Two algorithms were implemented to process the SICK sensor data, one for the horizontal conveyor system in trash removal machine and the inclined conveyor system in the citrus m echanical harvester machine. Algorithm for Horizontal Conveyor System in Trash Removal M achine clear all close all % x axis is along conveyor % y axis is across conveyor % z axis is along depth, starting at conveyor face %% Load data from a txt file vdata = load ('Test1_Speed5_30s_Error.dat', ascii'); % Voltage data % time in seconds between when the sensor data collection star ts and the conveyor starts moving % corresponding data should be removed from the 'vdata' conveyorwaittime = 0; %% Required inputs drate = 75; % Sensor frequnecy in Hz convspeed = 1.032955665*1000; % mm/sec conveyor speed dsenseconv = 25*10; % mm distance between sensor and conveyor surface wallheight = 4*10; % conveyor side wall height (mm) to be considered in calculation senstortwall = 32*10; % mm distance between sensor mid and conveyor right side wall

PAGE 107

107 conveyorwidth = 63*10; % conveyor width in mm tot alangle = 180; % total angle in y z plane for which data is collected by the sensor angspace = 1; % anglespacing is the interval between 2 data points in y z plane flappitch = 40*10; % average distance between two flaps on conveyor % total_vol = 6.3566e+006 volume for 94 flaps when conveyorwaittime = 0; flapvolcorfact = 0; %7.9122e+007/99; % calculated volume per flap when conveyor was run empty (it will give volume due to flaps alone) % total_vol1 = 1.0883e+008 volume for 94 flaps when used grid data flapvolcorfact_griddata = 0; % calculated volume per flap when conveyor was run empty when the data is corrected using grid technique(it will give volume due to flaps alone) %% % delete initial data, when conveyor is not moving vdata(1:ceil(drate*conveyor waittime),:)=[]; %Size of vdata in c rows and d colums [l,m] = size(vdata); % gives size of data, 'c' rows and 'd' columns sensortorightwall = senstortwall*dsenseconv/(dsenseconvwallheight); % mm distance between sensor mid and conveyor right side wall se nsortoleftwall = (conveyorwidth*dsenseconv/(dsenseconvwallheight)) sensortorightwall; % mm distance between sensor mid and conveyor left side surface convsensheight = dsenseconv; rangle = 180/pi()*atan(sensortorightwall/convsensheight); % Angle setting fo r matlab analysis, it is scanning included angle

PAGE 108

108 langle = 180/pi()*atan(sensortoleftwall/convsensheight); % Angle setting for matlab analysis, it is scanning included angle anglesetting = langle + rangle; %% Erase side wall data % left side rightsidestart = 1; rightsideend = floor((totalangle/2 rangle)/angspace) 1; leftsideend = totalangle/angspace rightsideend+1; leftsidestart = floor(leftsideend ((totalangle/2 langle)/angspace)); vdata(:,rightsidestart:rightsideend) = []; vdata(:,leftsidestart:leftsid eend) = []; % vdata(:,end)=[]; %% Internal calculations % Size of vdata in c rows and d colums [c,d] = size(vdata); % gives size of data, 'c' rows and 'd' columns sensorspoints = d; % number of points across conveyor dt = 1/drate; % data frequency in secon ds % x axis data t = [0:dt:(c 1)*dt]; % time scale for i=1:length(t) x(i,:)=t(i)*convspeed*[ones(1,d)]; % distance between 2 area scans end

PAGE 109

109 % z axis and y axis data at=angspace*pi()/180; % reading at the rate of 'at' radians startang=(90 rangle)*pi()/180; for i=1:c angle = startang; % for every scan in y z plane, starting angle is 0 deg for j=1:d theta(i,j) = angle; % current angle z(i,j) = vdata(i,j)*sin(theta(i,j)); % z = r*sin(theta) y(i,j) = vdata(i,j)*cos(theta(i,j)); % y = r*cos(theta) angle=angle+at; % increment angle by 0.5 deg for next data point end end z = dsenseconv*ones(size(z)) z; % gives distance from conveyor surface to fruit top surface %% Plotting figure % surface da ta points plot % subplot(1,2,1), surface(x,y,z) xlabel('Along conveyor (mm)') ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : data points') axis equal

PAGE 110

110 axis tight shading interp view([ 40 20]) colorbar for i=1:c for j=1:d if z(i,j)<45 z(i,j)=0; end if z(i,j)>85 z(i,j)=80; end end end y = y min(min(y)); % this makes wall = 0, gives distance from conveyor side wall to fruit top surface %% Elemental volume (4 adjacen t points), linear interpolation [a, b] = size(x); for i=1:a 1 % dx(i) = x(i+1) x(i); % elemental length along x axis for j=1:b 1 dy(i,j) = (y(i,j) y(i,j+1)+y(i+1,j) y(i+1,j+1))/2; % avg elemental length along y axis zavg(i,j) = (z(i ,j)+z(i,j+1)+z(i+1,j)+z(i+1,j+1))/4;

PAGE 111

111 vol(i,j) = zavg(i,j)*dx(i)*dy(i,j); dxpoint(i,j) = x(i)+dx(i)/2; dypoint(i,j) = y(i,j)+dy(i,j)/2; end xslice_vol(i) = sum(vol(i,:)); % volume of all elements between 2 xslices end total_vol = sum(sum(vol(:,:))), % total volume % removing volume corresponding to flaps xrange = max(max(x)); % finds out what is the length of the conveyor we have considered in data analysis flapvolume = floor(xrange/flappitch)*flapvolcorfact; total_vol_flapcor = total_vol flapvolume, % total volume with flap correction %% Plotting figure % surface data points plot % subplot(1,2,1), surface(x,y,z) xlabel('Along conveyor (mm)') ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : da ta points') axis equal axis tight

PAGE 112

112 shading interp view([ 40 20]) colorbar %% Working on data achieved through grid data % data reshaping uniform x and y coordinates, interpolate z coordinate % Presently we have kept the data size same. increasing data s ize might improve results X1 = x; % keeping the original x data Y1 = ones(c,1)*[floor(max(y(:,1))): floor(max(y(:,1)))/(d 1):0]; % changing y data to get uniformly spaced y data % floor(y(:,1)) gives rounded down value of maximum element in the 1st column of y data Z1 = griddata(x,y,z,X1,Y1,'cubic'); % 'linear' interpolation of z data, we can use 'cubic' as well % elemental volume (4 adjacent points), linear interpolation [a, b] = size(X1); % check due to interpolation if we get any element of Z1 as NaN, if we do, % then replace that element with original z value, we can do this because % we have kept the size of x,y,z same when it becomes X1, Y1 and Z1 Z1check=isnan(Z1); % checks if any of the element of Z1 is 'NaN' and returns '1' else '0' at that element for i=1:a for j=1:b if Z1check(i,j)== 1 Z1(i,j)=z(i,j); end

PAGE 113

113 end end for i=1:a for j=1:b if Z1(i,j)<45 Z1(i,j)=0; end if Z1(i,j)>85 Z1(i,j)=80; end end end for i=1:a 1 % dx1(i) = X1(i+1) X1(i); % elemental length along x axis for j=1:b 1 dy1(i,j) = (Y1(i,j) Y1(i,j+1)+Y1(i+1,j) Y1(i+1,j+1))/2; % avg elemental length along y axis zavg1(i,j) = (Z1(i,j)+Z1(i,j+1)+Z1(i+1,j)+Z1(i+1,j+1))/ 4; vol1(i,j) = zavg1(i,j)*dx1(i)*dy1(i,j); dxpoint1(i,j) = X1(i)+dx1(i)/2; dypoint1(i,j) = Y1(i,j)+dy1(i,j)/2; end

PAGE 114

114 xslice_vol1(i) = sum(vol1(i,:)); % volume of all elements between 2 xslices end total_vol1 = sum(sum(vol1(:,:))), % total volume %removing volume corresponding to flaps X1range = max(max(X1)); % finds out what is the length of the conveyor we have considered in data analysis flapvolume1 = floor(X1range/flappitch)*flapvolcorfact_griddata; total_vol 1_flapcor = total_vol1 flapvolume1, % total volume with flap correction % plotting figure % % surface data points plot % subplot(1,2,1), surface(X1,Y1,Z1) % mesh(X1,Y1,Z1) xlabel('Along conveyor (mm)') ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : data points') axis equal shading interp view([ 40 20]) colorbar

PAGE 115

115 Algorithm for Horizontal Conveyor System in Citrus Mechanical Harvester M achine clear all close allangspace %% About the File % This file shows the walls properly than other versions in this folder % x axis is along conveyor % y axis is across conveyor % z axis is along depth, starting at conveyor face %% Load data from a txt file vdata = load ('test.dat', ascii'); % Voltage data conveyorwaittime = 0; conveyorwaittimee nd = 0; %% Required inputs drate = 75; % Sensor frequnecy in Hz convspeed = 1.057669527*1000; % mm/sec conveyor speed dsenseconv = 35.6*10; % mm distance between sensor and conveyor surface wallheight = 5*10; % conveyor side wall height (mm) to be conside red in calculation senstortwall = 36.5*10; % mm distance between sensor mid and conveyor left side wall conveyorwidth = 76*10; % conveyor width in mm dbar = 55; %in mm error data other than fruit data % time in seconds between when the sensor data collect ion starts and the conveyor starts moving % corresponding data should be removed from the 'vdata' totalangle = 180; % total angle in y z plane for which data is collected by the sensor

PAGE 116

116 angspace = 1; % anglespacing is the interval between 2 data points in y z plane flappitch = 40*10; % average distance between two flaps on conveyor % total_vol = 3.4908e+007 volume for 80 flaps when conveyorwaittime = 0; % calculated volume per flap when conveyor was run empty (it will give volume due to flaps alone) flapvolcorfact = 3.72E+07/80; % total_vol1 = 3.4006e+007 volume for 96 flaps when used grid data % calculated volume per flap when conveyor was run empty when the data is corrected using grid technique(it will give volume due to flaps alone) flapvolcorfa ct_griddata = 3.26E+07/80; %% % delete initial data, when conveyor is not moving vdata(1:ceil(drate*conveyorwaittime),:)=[]; [aa,bb]=size(vdata); vdata((aaceil(drate*conveyorwaittimeend)):end,:)=[]; %Size of vdata in c rows and d colums [l,m] = size(vdata); % gives size of data, 'c' rows and 'd' columns sensortorightwall = senstortwall*dsenseconv/(dsenseconvwallheight); % mm distance between sensor mid and conveyor right side wall sensortoleftwall = (conveyorwidth*dsenseconv/(dsenseconvwallheight)) sensortorightwall; % mm distance between sensor mid and conveyor left side surface convsensheight = dsenseconv;

PAGE 117

117 rangle = 180/pi()*atan(sensortorightwall/convsensheight); % Angle setting for matlab analysis, it is scanning included angle langle = 180/pi()*atan(s ensortoleftwall/convsensheight); % Angle setting for matlab analysis, it is scanning included angle anglesetting = langle + rangle; %% Erase side wall data % left side rightsidestart = 1; rightsideend = floor((totalangle/2 rangle)/angspace) 1; leftsideend = totalangle/angspace rightsideend+1; leftsidestart = floor(leftsideend ((totalangle/2 langle)/angspace)); vdata(:,rightsidestart:rightsideend) = []; vdata(:,leftsidestart:leftsideend) = []; % vdata(:,end)=[]; %% Internal calculations % Size of vdata in c rows and d colums [c,d] = size(vdata); % gives size of data, 'c' rows and 'd' columns sensorspoints = d; % number of points across conveyor dt = 1/drate; % data frequency in seconds % x axis data t = [0:dt:(c 1)*dt]; % time scale for i=1:length(t)

PAGE 118

118 x(i,:)=t(i)*convspeed*[ones(1,d)]; % distance between 2 area scans end % z axis and y axis data at=angspace*pi()/180; % reading at the rate of 'at' radians startang=(90 rangle)*pi()/180; for i=1:c angle = startang; % for every scan in y z plane, star ting angle is 0 deg for j=1:d theta(i,j) = angle; % current angle z(i,j) = vdata(i,j)*sin(theta(i,j)); % z = r*sin(theta) y(i,j) = vdata(i,j)*cos(theta(i,j)); % y = r*cos(theta) angle=angle+at; % increment angle by at de g for next data point end end z = dsenseconv*ones(size(z)) z; % gives distance from conveyor surface to fruit top surface %% Plotting figure % surface data points plot % subplot(1,2,1), surface(x,y,z) xlabel('Along conveyor (mm)')

PAGE 119

119 ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : data points') axis equal axis tight shading interp view([ 40 20]) colorbar for i=1:c for j=1:d if z(i,j)85 z(i,j)=85; end end end y = y min(min(y)); % this makes wall = 0, gives distance from conveyor side wall to fruit top surface %% Elemental volume (4 adjacent points), linear interpolation [a, b] = size(x);

PAGE 120

120 for i=1:a 1 % dx(i) = x(i+1) x(i); % elem ental length along x axis for j=1:b 1 dy(i,j) = (y(i,j) y(i,j+1)+y(i+1,j) y(i+1,j+1))/2; % avg elemental length along y axis zavg(i,j) = (z(i,j)+z(i,j+1)+z(i+1,j)+z(i+1,j+1))/4; vol(i,j) = zavg(i,j)*dx(i)*dy(i,j); dxpoint(i,j) = x(i)+dx(i)/2; dypoint(i,j) = y(i,j)+dy(i,j)/2; end xslice_vol(i) = sum(vol(i,:)); % volume of all elements between 2 xslices end total_vol = sum(sum(vol(:,:))), % total volume % removing volume corresponding to flaps xrange = max(max(x)); % finds out what is the length of the conveyor we have considered in data analysis flapvolume = floor(xrange/flappitch)*flapvolcorfact; total_vol_flapcor = total_vol flapvolume, % total volume with flap correction %% Plotting figure % surface data points plot % subplot(1,2,1), surface(x,y,z) xlabel('Along conveyor (mm)')

PAGE 121

121 ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : data points') axis equal axis tight shading interp view([ 40 20]) colorbar %% Working on d ata achieved through grid data % data reshaping uniform x and y coordinates, interpolate z coordinate % Presently we have kept the data size same. increasing data size might improve results X1 = x; % keeping the original x data Y1 = ones(c,1)*[floor(max(y(:,1))): floor(max(y(:,1)))/(d 1):0]; % changing y data to get uniformly spaced y data % floor(y(:,1)) gives rounded down value of maximum element in the 1st column of y data Z1 = griddata(x,y,z,X1,Y1,'cubic'); % 'linear' interpolation of z data, we can use 'cubic' as well % elemental volume (4 adjacent points), linear interpolation [a, b] = size(X1); % check due to interpolation if we get any element of Z1 as NaN, if we do, % then replace that element with original z value, we can do this because % we have kept the size of x,y,z same when it becomes X1, Y1 and Z1 Z1check=isnan(Z1); % checks if any of the element of Z1 is 'NaN' and returns '1' else '0' at that element

PAGE 122

122 for i=1:a for j=1:b if Z1check(i,j)== 1 Z1(i, j)=z(i,j); end end end for i=1:a for j=1:b if Z1(i,j)85 Z1(i,j)=80; end end end for i=1:a 1 % dx1(i) = X1(i+1) X1(i); % elemental length along x axis for j=1:b 1 dy1(i,j) = (Y1(i,j) Y1(i,j+1)+Y1(i+1,j) Y1(i+1,j+1))/2; % avg elemental length along y axis zavg1(i,j) = (Z1(i,j)+Z1(i,j+1)+Z1(i+1,j)+Z1(i+1,j+1))/4; vol1(i ,j) = zavg1(i,j)*dx1(i)*dy1(i,j);

PAGE 123

123 dxpoint1(i,j) = X1(i)+dx1(i)/2; dypoint1(i,j) = Y1(i,j)+dy1(i,j)/2; end xslice_vol1(i) = sum(vol1(i,:)); % volume of all elements between 2 xslices end total_vol1 = sum(sum(vol1(:,:))), % t otal volume %% removing volume corresponding to flaps X1range = max(max(X1)); % finds out what is the length of the conveyor we have considered in data analysis flapvolume1 = floor(X1range/flappitch)*flapvolcorfact_griddata; total_vol1_flapcor = total_vol1 flapvolume1, % total volume with flap % correction %% weight calculation % weight = 1.38E 06*total_vol1_flapcor 4.50E+00, % 0.5864 m/s % weight = 1.60E 06*total_vol1_flapcor 6.14E+00, % 0.785 m/s weight = 1.48E 06*total_vol1_flapcor 6.74E+00, % 1.1 m/s %% plotting figure % % surface data points plot % subplot(1,2,1), surface(X1,Y1,Z1) % mesh(X1,Y1,Z1)

PAGE 124

124 xlabel('Along conveyor (mm)') ylabel('Across conveyor (mm)') zlabel('Depth (mm)') title('Surface plot : data points') axis equal shading interp view([ 40 20]) colorbar

PAGE 125

125 LIST OF REFERENCES Abidine, A., S. K. Upadhyaya and J. Leal. 2003. Development of an electronic weigh bucket. In ASAE Paper No. 031043, St. Joseph, MI, USA : ASAE. Annamalai, P. 2004. Color vision system for estimating citrus yield in realtime. MS Thesis UFL: University of Florida, Agricultural and Biological Engineering Department. Behme, J. A., J. L. Schinstock, L. L. Bashford and L. I. Leviticus. 1997. Site specific yield for forages. In ASAE Paper No. 971054. St. Joseph, MI, USA : ASAE. Benjamin, C. E. 2002. Sugar cane yield monitoring system. MS diss. Baton Rouge, Louisiana: Louisiana State University, Department of Biological and Agricultural Engineering. Bora, G. C., R. Ehsani, K. Lee and W. Lee. 2006. Development of a test rig for evaluating a yield monitoring system for citrus mechanical harvesters. In ASAE Paper No. 701P0606, St. Joseph, MI, USA: ASAE. Campbell, R. H., S. L. Rawlins and S. F. Han. 1994. Monitoring methods for potato yield mapping. In ASAE Paper No. 941584, St. Joseph, MI, USA: ASAE. Cerri, D. G., S. K. Upadhyaya and J. Leal. 2004. Development of an electronic weigh bucket. In ASABE, Paper No: 041097, St. Joseph, MI, USA : ASABE. Chinchuluun, R., W. S. Lee and R. Ehsani. 2007. Citrus yield mapping system on a canopy shake and catch harvester. In ASAE Paper No: 073050, St. Joseph, MI, USA : ASAE. equipment. Agricultural Research Service, ARS S 84.New Orleans, LA Co ppock, G. E. 1976. Catching frame development for a citrus harvest system. Transaction of the ASAE 19(4): 751045. Coppock, G. E. 1967. Harvesting early and midseason citrus fruit with tree shaker harvest systems. Proc. Fla. State Hort. Soc. 80:98104. Coppock, G. E. and S. L. Hedden. 1968. Design and development of a tree shaker harvest system for citrus fruit. Transactions of the ASAE 11(3): 339342. Cundiff, J. S. and Y. Sharobeem. 2003. Development of yield monitor (Part one). In ASAE Paper No: 031036, St. Joseph, MI, USA : 2003. Ehlert, D. 2000. Measuring mass flow by bounce plate for yield mapping of potatoes. Precision agriculture 2(2): 119130. Ehsani, M. R., T. E. Grift, J. M. Maja and D. Zhong. 2009. Two fruit counting techniques for citrus mechanical harvesting machinery. Computers and Electronics in Agriculture 65(2): 186191.

PAGE 126

126 Ehsani, R. and D. Karimi. 2010. Yield monitors for specialty crops. Advanced engineering systems for specialty crops: A review of precision agricul ture for water, chemical, and nutrient application, and yield monitoring. Spec. Issue of Landbauforschung vTI Agriculture and Forestry Research 340: 31 43 FDOC. 20 10. M arketing, research and regulation of the Florida citrus industry: Bartow, FL.: Availabl e at: http://dev08.floridajuice.com Accessed 15 May 2010 Gogineni, S., J. G. White, J. A. Thomasson, P. G. Thompson, J. R. Wooten and M. Shankle. 2002. Image Based Sweetpotato Yield and Grade Monitor. In ASAE Paper No: 021169, St. Joseph, MI, USA : ASAE. Grift, T., R. Ehsani, K. Nishiwaki, C. Crespi and M. Min. 2006. Development of a yield monitor for citrus fruits. In ASABE Paper No: 061192, Portland, Oregon, USA: ASABE. Hall, A. and J. Louis. 2009. C hapter 3: Vineclipper: A Proximal Search Algorithm to tie GPS field locations to high resolution grapevine imagery. In Innovations in Remote Sensing and Photogrammetry, 361372. ed. W. CartwrightGartner, G.Meng, L. and Peterson, M., Springer Berlin Heidelb erg. Hedden, S. L. and G. E. Coppock. 1971. Comparative harvest trials of foliage and limb shakers in Valenciaoranges. In Proc. Fla. State Hort. Soc, 8892. Heidman, B. C., U. A. Rosa, P. H. Brown and S. K. Upadhyaya. 2003. Development of a pistachio y ield monitoring system. In ASAE Paper No. 031040, St. Joseph, MI USA: ASAE. Hofstee, J. W. and G. J. Molema. 2003. Volume estimation of potatoes partly covered with dirt tare. In ASAE Paper No:031001, Paper No: 031001. St. Joseph, MI, USA : ASAE. Hume, H H. 1926. The cultivation of citrus fruits. New York, NY.: The Macmillan Company. Konstantinovic, M., S. Woeckel, P. S. Lammers, J. Sachs and M. Minneapolis. 2007. Evaluation of a UWB radar system for yield mapping of sugar beet. In ASABE Paper No. 071051, Minneapolis, Minnesota: ASABE. Kuhar, J.E. 1997. The precision farming guide for agriculturists. Moline, IL.: John Deere Publ Lee, W. S., T. F. Burks and J. K. Schueller. 2002. Silage yield monitoring system. In ASAE Paper No: 021165, St. Joseph, MI, USA: ASAE. Labview. 2008. Labview for Windows. ver. 8.6.: National Instruments Corporation, Austin, TX, USA Magalhes, P. S. G. and D. G. P. Cerri. 2007. Yield Monitoring of Sugar Cane. Biosystems Engineering 96(1): 16.

PAGE 127

127 Maja, J. M. and R. Ehsani. 2009a. Development of a yield monitoring system for citrus mechanical harvesting machines. Precision Agriculture 113. Maja, J. M. and R. Ehsani. 2009b. Development of a yield monitoring system for citrus mechanical harvesting machines. Precision Agriculture 113. Miller, W. M. and J. D. Whitney. 1999. Evaluation of weighing systems for citrus yield monitoring. Applied Engineering in Agriculture 15(6): 609614. MATLAB. 2009. Matlab for Windows ver. 2009a : The Mathworks Inc., Na tick, MA, USA Molin, J. P. and L. A. A. Menegatti. 2004. Fieldtesting of a sugar cane yield monitor in Brazil. In ASAE Paper No. 041099, St. Joseph, MI USA: ASABE. Pelletier, G. and S. K. Upadhyaya. 1999. Development of a tomato load/yield monitor. Com puters and Electronics in Agriculture 23(2): 103117. Persson, D. A., L. Eklundh and P. A. Algerbo. 2004. Evaluation of an optical sensor for tuber yield monitoring. Transactions of the ASAE 47(5): 18511856. Peterson, D. L. 1998. Mechanical harvester for process oranges. Applied Engineering in Agriculture 14(5): 455458. Price, R. R., J. Larsen and A. Peters. 2007. Development of an optical yield monitor for sugar cane harvesting. In ASABE Paper No. 071049, Minneapolis, Minnesota, USA: ASABE. Qarallah, B., K. Shoji and T. Kawamura. 2008. Development of a yield sensor for measuring individual weights of onion bulbs. Biosystems engineering 100(4): 511515. Rains, G. C., D. L. Thomas and C. D. Perry. 2002. Pec an mechanical harvesting parameters for yield monitoring. Transactions of the ASAE 45(2): 281285. Rains, G. C., C. D. Perry and G. Vellidis. 2005. Adaptation and Testing of the Agleader Cotton Yield Sensor on a Peanut Combine. Applied Engineering in Agri culture 21(6): 979983. Roades, J. P., A. D. Beck and S. W. Searcy. 2000. Cotton yield mapping: Texas experiences in 1999. In Proceedings of the Beltwide Cotton Conference, 404 408. 404407 San Antonia, TX. Jan 48, 2000. Memphis, TN: National Cotton Council of America. Salehi, F., J. D. Whitney, W. M. Miller, T. A. Wheaton and G. Drouillard. 2000. An automatic triggering system for a citrus yield monitor. In ASAE Paper No. 001130, St. Joseph, MI USA: ASAE. Schueller, J. K., J. D. Whitney, T. A. Wheaton, W. M. Miller and A. E. Turner. 1999. Low cost automatic yield mapping in hand harvested citrus. Computers and Electronics in Agriculture 23(2): 145153.

PAGE 128

128 Schueller, J. K. and Y. H. Bae. 1987. Spatially attributed automatic combine data acquisition. Computers and Electronics in Agriculture 2(2): 119127. Searcy, S. W., J. K. Schueller, Y. H. Bae, S. C. Borgelt and B. A. Stout. 1989. Mapping of spatially variable yield during grain combining. Transactions of the ASAE 32(3): 826829. Spiegel Roy, P. and E. E. Goldschmidt. 1996. Biology of citrus. Cambridge, UK: Cambridge University Press. Sumner, H. R. and D. B. Churchill. 1977. Collecting and handling mechanically removed citrus fruit. Proc. Int. Soc. Citriculture 2413418. Thomasson, J. A. and R. Sui. 2004. Optical peanut yield monitor: development and testing. In ASABE Paper No. 041095, St. Joseph, Michigan, USA: ASABE. Tumbo, S. D., J. D. Whitney, W. M. Miller and T. A. Wheaton. 2002. Development and testing of a citrus yield monitor. Applied Engineer ing in Agriculture 18(4): 399403. Upadhyaya, S. K., U. A. Rosa, M. Ehsani, M. Koller, M. Josiah and T. Shikanai. 1999. Precision farming in a tomato production system. In ASAE Paper No. 991147, St. Joseph, MI : ASAE USDA. 2009. The USDA Economics, Statis tics and Market Information System: 20082009. National Agricultural Statistics Service. Washington, D.C.: USDA National Agricultural Statistics Service. Available at: www.nass.usda.gov. Accessed 15 May 2010. Vellidis, G., C. D. Perry, J. S. Durrence, D. L. Thomas, R. W. Hill, C. K. Kvien, T. K. Hamrita and G. Rains. 2001. The peanut yield monitoring system. Transactions of the ASAE 44(4): 775785. Whitney, J. D., Q. Ling, W. M. Miller and T. A. Wheaton. 2001. A DGPS yield monitoring system for Florida c itrus. Applied Engineering in Agriculture 17(2): 115119. Whitney, J. D. 1995. A review of citrus harvesting in Florida. Applied Engineering in Agriculture 413339. Whitney, J. D. 1968. Citrus fruit removal with an air harvester concept. Proc. Fla. State Hort. Soc 814348. Whitney, J. D. and J. M. Patterson. 1972. Development of a citrus removal device using oscillating forced air. Transactions of the ASAE 15(5): 849855. Whitney, J. D., W. M. Miller, T. A. Wheaton, M. Salyani and J. K. Schueller. 1999. Precision farming applications in Florida citrus. Applied Engineering in Agriculture 15(5): 399403. Whitney, J., Q. Ling, W. Miller and T. Wheaton. 2001. A DGPS yield monitoring system for Florida citrus. In 115 119.

PAGE 129

129 Whitney, J. D., T. A. Wheaton, W. M Miller, M. Salyani and J. K. Schueller. 1998. Site specific yield mapping for Florida citrus. Proc. Fla. State Hortic. Soc 111: 148150. Wild, K. and H. Auernhammer. 1999. A weighing system for local yield monitoring of forage crops in round balers. Co mputers and Electronics in Agriculture 23(2): 119132. Wilkerson, J. B., F. H. Moody, W. E. Hart and P. A. Funk. 2001. Design and evaluation of a cotton flow rate sensor. Transactions of the ASAE 44(6): Ye, X., K. Sakai, A. Sasao and S. Asada. 2008. Po tential of airborne hyperspectral imagery to estimate fruit yield in citrus. Chemometrics and Intelligent Laboratory Systems 90(2): 132144. Zaman, Q. U., A. W. Schumann and H. K. Hostler. 2006. Estimation of citrus fruit yield using ultrasonically sensed tree size. Applied Engineering in Agriculture 22(1): 3944. Zaman, Q. U., A. W. Schumann, D. C. Percival and R. J. Gordon. 2008. Estimation of wild blueberry fruit yield using digital color photography. Transactions of the ASAE 51(5): 15391544.

PAGE 130

130 BIOGRAPHICAL SKETCH Ujwala Jadhav was born in year 1982, to Mangal Jadhav and Subhash Jadhav, in Satara, Maharashtra, India. She attended Universit y of Pune, India from year 19992004 where s he earned her Bachelor of Computer Science degree in 2002 and Master of C omputer Science degree in year 2004. Later she wo rked as a software engineer in three reputed Indian companies and gained quality experience in software project development. S he joined University of Florida to pursue her higher education in year 2008 and worked in Precision Agriculture Lab on Citrus yiel d monitoring project for mechanical harvesters She will receive her concurrent degrees in c omputer science as well as agricultural and biological e ngineering in December 2010.