<%BANNER%>

Machine Vision Based Citrus Yield Mapping System on a Continuous Canopy Shake and Catch Harvester

Permanent Link: http://ufdc.ufl.edu/UFE0021209/00001

Material Information

Title: Machine Vision Based Citrus Yield Mapping System on a Continuous Canopy Shake and Catch Harvester
Physical Description: 1 online resource (78 p.)
Language: english
Creator: Chinchuluun, Radnaabaz
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: canopy, catch, citrus, harvester, machine, mapping, shake, vision, yield
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Yield mapping is a well known beneficial tool in precision agriculture. Traditionally citrus growers ignored in-field variability of soil properties, environmental conditions and citrus yield because they dealt with comparatively small scale groves. They treated whole-grove as a uniform single unit. Citrus growers always seek new production practices to increase profit by maximizing citrus yield. Because of lack of information of grove and expertise, growers sometimes cannot make optimal decisions. Hence, as a result of inappropriate management, growers are not able to maximize their profit. Fortunately competitive growers are able to adopt new technologies such as Global Positioning System (GPS), yield monitoring, remote sensing (RS), variable rate technology (VRT) and sensing systems. With advance of these technologies, farmers can manage large scale grove at smaller scale or individual tree basis. Consequently growers are able to increase yield, fruit quality and economic return. In this study, machine vision based citrus yield mapping and fruit size detection system was developed, which could be used on a continuous canopy shake and catch harvester. The system consisted of a 3CCD camera, four halogen lamps, a DGPS receiver, an encoder and a laptop computer. A total of 3,653 images were taken during an experiment on the test-bench at the Citrus Research and Education Center, Lake Alfred, Florida and 709 images were used for analysis. The system was also tested on a canopy shake and catch harvester at a Lykes grove located in Fort Basinger, FL. A total of 773 images were acquired as well. Fruit weight was measured in 14 test trials of image acquisitions during the test-bench experiment as well as in two test trials of image acquisitions during a field trial with a commercial canopy shake and catch harvester. A supervised image processing algorithms that could identify and inspect fruit qualities were developed. The number of fruit and total fruit areas were measured from the sets of color images using the developed algorithm. Finally, number of citrus fruit that was found by the image processing algorithm during the test-bench experiment was compared against actual fruit weight. The coefficient of determination was 0.962 between them. For a validation purpose of the canopy shake and catch harvester experiment, number of fruit was counted manually from a total of 60 images. Human counting was compared with algorithm counting. The coefficient of determination was 0.891 between the actual and estimated number of fruit.
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 Radnaabaz Chinchuluun.
Thesis: Thesis (M.E.)--University of Florida, 2007.
Local: Adviser: Lee, Won Suk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021209:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021209/00001

Material Information

Title: Machine Vision Based Citrus Yield Mapping System on a Continuous Canopy Shake and Catch Harvester
Physical Description: 1 online resource (78 p.)
Language: english
Creator: Chinchuluun, Radnaabaz
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: canopy, catch, citrus, harvester, machine, mapping, shake, vision, yield
Agricultural and Biological Engineering -- Dissertations, Academic -- UF
Genre: Agricultural and Biological Engineering thesis, M.E.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Yield mapping is a well known beneficial tool in precision agriculture. Traditionally citrus growers ignored in-field variability of soil properties, environmental conditions and citrus yield because they dealt with comparatively small scale groves. They treated whole-grove as a uniform single unit. Citrus growers always seek new production practices to increase profit by maximizing citrus yield. Because of lack of information of grove and expertise, growers sometimes cannot make optimal decisions. Hence, as a result of inappropriate management, growers are not able to maximize their profit. Fortunately competitive growers are able to adopt new technologies such as Global Positioning System (GPS), yield monitoring, remote sensing (RS), variable rate technology (VRT) and sensing systems. With advance of these technologies, farmers can manage large scale grove at smaller scale or individual tree basis. Consequently growers are able to increase yield, fruit quality and economic return. In this study, machine vision based citrus yield mapping and fruit size detection system was developed, which could be used on a continuous canopy shake and catch harvester. The system consisted of a 3CCD camera, four halogen lamps, a DGPS receiver, an encoder and a laptop computer. A total of 3,653 images were taken during an experiment on the test-bench at the Citrus Research and Education Center, Lake Alfred, Florida and 709 images were used for analysis. The system was also tested on a canopy shake and catch harvester at a Lykes grove located in Fort Basinger, FL. A total of 773 images were acquired as well. Fruit weight was measured in 14 test trials of image acquisitions during the test-bench experiment as well as in two test trials of image acquisitions during a field trial with a commercial canopy shake and catch harvester. A supervised image processing algorithms that could identify and inspect fruit qualities were developed. The number of fruit and total fruit areas were measured from the sets of color images using the developed algorithm. Finally, number of citrus fruit that was found by the image processing algorithm during the test-bench experiment was compared against actual fruit weight. The coefficient of determination was 0.962 between them. For a validation purpose of the canopy shake and catch harvester experiment, number of fruit was counted manually from a total of 60 images. Human counting was compared with algorithm counting. The coefficient of determination was 0.891 between the actual and estimated number of fruit.
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 Radnaabaz Chinchuluun.
Thesis: Thesis (M.E.)--University of Florida, 2007.
Local: Adviser: Lee, Won Suk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021209:00001


This item has the following downloads:


Full Text





MACHINE VISION BASED CITRUS YIELD MAPPING SYSTEM ON A CONTINUOUS
CANOPY SHAKE AND CATCH HARVESTER




















By

RADNAABAZAR CHINCHULUUN


A THESIS PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
MASTER OF ENGINEERING

UNIVERSITY OF FLORIDA


2007





































02007 Radnaabazar Chinchuluun



































To no God who abs ays blesses me and to no father, Chinchuluun OL hlil 11ninig. who abs ays takest~~~~ttttt~~~~tttt
care of me fr~om heaven.









ACKNOWLEDGMENTS

I want to express my deepest gratitude to Dr. Won Suk Lee for his enormous support and

guidance. He is the best advisor I have ever had in my life. He always provided me greatest

guidance in my life and study with immense kindness and patience. He opened my eyes by

introducing me to the precision agriculture and machine vision. I also thank my committee

members Dr. Thomas F. Burks and Dr. James J. Ferguson for their kindness and guidance, and

thank Dr. Reza Ehsani for his support and allowing me to collaborate in his research. All that I

have learnt during the course of my thesis would not have been possible without their

dedications.

I especially thank Dr. Kyeong Hwan Lee and Dr. Ganesh Bora for their support in my

experiment, and Mr. Michael Zingaro, Mr. Harmon Pearson for sharing me their hands

throughout my research. Mr. Greg L. Pugh shared his programming experience with me. I would

like to thank my friends Mr. Sanghoon Han, Vij ay Subramanian and Kevin Kane for their

encouragements and support.

Most importantly I would like to express my dearest appreciation to my family for their

dedication, support and love.












TABLE OF CONTENTS


page

ACKNOWLEDGMENT S .............. ...............4.....


LI ST OF T ABLE S ............ _...... ._ ...............7....


LIST OF FIGURES .............. ...............8.....


AB S TRAC T ............._. .......... ..............._ 10...


CHAPTER


1 INTRODUCTION ................. ...............12......... .....


Back ground ..........._.._........... ..... ...............12......
Precision Agriculture for Citrus................ ...............13
Citrus Harvesting ............... .... ...._ ._ ............. .............1
Motivation for Citrus Yield Mapping System ................ ...............14...............
Obj ectives ........._.. ..... ._ ._ ...............15....

2 LITERATURE REVIEW ........._.._ ..... ._._ ...............16....


Citrus Harvesting ................. ...............16........ ......
Robotic Harvesting ................. ...............16........ ......
Mechanical Harvesting ................. ...............18........ ......
Yield M apping .................. .. ....._.. ... ........... .............1
Machine Vision and Fruit Size Estimating System .............. ...............22....

3 IMAGE ANALYSIS .............. ...............25....


Color Model s ..................... ...............25.
The RGB Color Model .............. ...............25....
The HSI Color Model ............. ...... ...............25..
The YIQ Color Model .............. ...............27....
Image Classif cation .............. ...............28....
Color Classifieation ............_...... ...............29...

Bayesian classifier............... ...............2
Shape Classifieation .............. ...............3 1....

4 MATERIALS AND METHODS .............. ...............34....


Hardware Design of the System ............ .....__.....__ ....__ ...............34
System Operations .............. ......__ ...............35...
System Components' Interconnection............... ............3
Camera Parameters ............. .... ...__ ...............36...

Image Synchronization with Encoder .............. ...............38....












System Software Design............... ...............39.
System Experiments .............. ...............41....
Test-bench Experiment............... ......... .. .........4
Image acquisition during the test-bench experiment............... ...............4
Citrus Canopy Shake and Catch Harvester Experiment ......____ ..... ... ._ ..............42
Image Processing Analysis ................ ...............43........... ....
Fruit Parameters ........._.. ..... ._ ...............47.....
Yield Prediction Model ........._.._ ..... .___ ...............47....
Camera Calibration ........._.. ..... ._ ...............50.....


5 RESULT AND DISCUS SION ........._.._ ..... .___ ...............52...


Color Segmentation Result on Test-bench .............. ...............52....
Color Segmentation Result on the Harvester............... ...............5
System Execution Time ........._.._ ..... .___ ...............58....
System Limitation............... ...............5

6 CONCLUSIONS .............. ...............60....


APPENDIX


A PROGRAMMING CODE ........._.._ ..... .___ ...............62....


Encoder Thread Code .............. ...............62....

ImageWarp Thread Code......................... .. ..... ...............6
Image Processing ImageWarp Script in Real-time Mode .............. ...............64....
Image Processing ImageWarp Script in Post-mode .............. ...............65....

B MEASURED FRUIT PARAMETERS .............. ...............66....


LIST OF REFERENCES ....._._. ................ ......._._. .........7


BIOGRAPHICAL SKETCH .............. ...............78....











LIST OF TABLES

Table page



4-1. Conveyor belt specification ..........._...__........ ...............35...

4-2. Camera resolution ..........._...__........ ...............38....

4-3. Means of hue, saturation and I (YIQ) components of fruit and background classes .............45

5-1. Summary of the test-bench experiment results. .........._.._.. ........._. .........._._......__54

5-2. System execution time............... ...............59..

B-1. Fruit parameters .............. ...............66....










LIST OF FIGURES


Fiare page

3-1. RGB color cube. ............. ...............26.....

3-2. HSI color model representation. .............. ...............27....

3-3. Image classification design cycle. ............. ...............28.....

3-4. Watershed representation. A) Two touching oranges. B) Binary image of the image A.
C) Distance image. D) Topographic representation of the image (Source: Vincent
(1991)). E) Separated oranges............... ...............33

4-1. General overview of the system. A) Camera and four halogen lamps. B) System
housing. .............. ...............35....

4-2. Pictorial illustration of system operation. .............. ...............37....

4-3. System connection diagram ................. ...............37........... ...

4-4. Image acquisition of moving obj ect at the different shutter speeds. A) Auto shutter
speed. B) 1/2000 s. C) 1/2150 s. D) 1/2230 s. E) 1/2300 s ................. ............ .........38

4-5. The system software overview. ............. ...............40.....

4-6. Encoder thread al gorithm. ............. ...............41.....

4-7. Test-bench experiment. A) Feeding conveyor belt. B) Weight measurement of tested
fruit. ................. ...............42........ .....

4-8. Canopy shake and catch harvester experiment. ................. ...._._ ........... ........4

4-9. Color image of fruit and background. ............. ...............44.....

4-10. Images in different color models and their histograms. A) Hue component of HSI
color model. B) Hue histogram of fruit and background. C) Saturation component of
HSI color model. D) Saturation histogram of fruit and background. E) I component
of YIQ color model. E) I (YIQ) histogram of fruit and background. ............. ..... .........._.44

4-11. Image processing al gorithm ................. ...............46.......... ...

4-12. Discriminant function between fruit and background classes. .............. ....................4

4-13. Relationship between actual fruit weight and diameter ................. ......................._.48

4-14. Relationship between actual fruit weight and area. .............. ...............49....

4.15. Relationship between actual fruit weight and volume. .............. ...............50....










4-16. Camera perspective projection. ............. ...............51.....

5-1. Image segmentation result. A) Original image. B) Bayesian classifier result. C)
Morphological operation result. D) Watershed transform result ................... ...............53

5-2. Actual fruit weight versus the number of fruit by image processing algorithm:
regression analysis. ............. ...............54.....

5-3. Actual fruit weight versus fruit area: regression analysis ................. .......... ...............55

5-4. Fruit diameter versus actual fruit weight: regression analysis. ............. .....................5

5-5. Color segmentation result for field experiment. A) Original image. B) Bayesian
classifier result. C) Morphological operation result. D) Watershed transform result........56

5-6. Circular filtering result for field experiment 2. ............. ...............57.....

5-7. Regression analysis between human counting and algorithm counting. ............... ...............58









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 Engineering

MACHINE VISION BASED CITRUS YIELD MAPPING SYSTEM ON A CONTINUOUS
CANOPY SHAKE AND CATCH HARVESTER

By

Radnaabazar Chinchuluun

August 2007

Chair: Won Suk Lee
Major: Agricultural and Biological Engineering

Yield mapping is a well known beneficial tool in precision agriculture. Traditionally citrus

growers ignored in-field variability of soil properties, environmental conditions and citrus yield

because they dealt with comparatively small scale groves. They treated whole-grove as a uniform

single unit. Citrus growers always seek new production practices to increase profit by

maximizing citrus yield. Because of lack of information of grove and expertise, growers

sometimes cannot make optimal decisions. Hence, as a result of inappropriate management,

growers are not able to maximize their profit. Fortunately in "high tech" era, competitive

growers are able to adopt new technologies such as Global Positioning System (GPS), yield

monitoring, remote sensing (RS), variable rate technology (VRT) and sensing systems. With

advance of these technologies, farmers can manage large scale grove at smaller scale or

individual tree basis. Consequently growers are able to increase yield, fruit quality and economic

return.

In this study, machine vision based citrus yield mapping and fruit size detection system

was developed, which could be used on a continuous canopy shake and catch harvester. The

system consisted of a 3CCD camera, four halogen lamps, a DGPS receiver, an encoder and a

laptop computer. A total of 3,653 images were taken during an experiment on the test-bench at









the Citrus Research and Education Center, Lake Alfred, Florida and 709 images were used for

analysis. The system was also tested on a canopy shake and catch harvester at a Lykes grove

located in Fort Basinger, FL. A total of 773 images were acquired as well. Fruit weight was

measured in 14 test trials of image acquisitions during the test-bench experiment as well as in

two test trials of image acquisitions during a field trial with a commercial canopy shake and

catch harvester. A supervised image processing algorithms that could identify and inspect fruit

qualities were developed. The number of fruit and total fruit areas were measured from the sets

of color images using the developed algorithm. Finally, number of citrus fruit that was found by

the image processing algorithm during the test-bench experiment was compared against actual

fruit weight. The coefficient of determination, R2, WaS 0.962 between them. For a validation

purpose of the canopy shake and catch harvester experiment, number of fruit was counted

manually from a total of 60 images. Human counting was compared with algorithm counting.

The coefficient of determination, R2, WaS 0.891 between the actual and estimated number of

fruit.









CHAPTER 1
INTRODUCTION

The thesis addresses a machine vision based citrus yield mapping system coupled with a

continuous canopy shake and catch harvesting system.

Background

One of the maj or agricultural products in Florida is citrus. Citrus production contributes

enormous value to not only Florida state economy, but also the U.S. economy. Florida citrus

industry accounted approximately 67 percent of the total U. S. citrus production. According to the

USDA statistics that was published in 2005, Florida dealt with a total of 679 thousand bearing

acreages and produced 13,045 thousand tons of citrus in 2003-2004 season (Florida Agricultural

Statistics Service [FASS], 2005). A total of $891,500,000 value was added in the season. In

Florida most citrus fruits go to processing industry rather than fresh fruit industry. In addition to

juice processing, citrus processing industry produces important by-products such as citrus pulp

and meal, molasses, and D-limonene (Hodges et al., 2001). These by-products are used to make a

variety of products starting from cleaners, disinfectants to livestock feeds. In 2003 -2004 crop

season, Florida citrus processing industry produced more than 1.1 million tons of citrus pulp and

meal, 3 8,000 tons of molasses, and 16363.6 tons of D-limonene. The total value of these by-

products was $136 million. Managing such huge production in 2003-2004 season, Florida citrus

industry created 76,336 jobs, with 61,307 jobs in the processing sectors and 15,029 jobs in fresh

fruit. Moreover, citrus industry impacted to maj or industrial sectors such as construction, finance

and insurance, health and social sciences, retail trade, wholesale trade, professional-scientific and

technical services, real estate and rentals (Hodges et al., 2001). Based on these facts, it is

apparent that citrus industry becomes a vital part of Florida economy.









Precision Agriculture for Citrus

As a result of mechanizing agricultural grove management, most growers started to deal

with large-scale areas in which in-grove variability of nutrients, tree size, soil types, etc. highly

likely existed. Traditionally, growers dealt with comparatively small citrus groves. Thus small

groves usually have less in-field variability. At the same time, they were easily manageable by

citrus growers. As grove size increases, citrus growers could not deal with it without

technologies. However, precision farming, also known as site-specific crop management,

enables treating small areas of a large-scale field as separate management units using new

technologies. As becoming tools of the precision farming, the Global Positioning System (GPS),

Geographic Information Systems (GIS), Remote Sensing (RS), Variable Rate Application

(VRA), yield mapping and advanced sensors and information technologies benefit farmers by

managing agricultural operations with less human involvements, reducing overall costs, and

eventually improving overall effectiveness and productivity by considering in-field variability.

Moreover, these technologies would help growers collect large amount of production

information over past seasons of their groves.

One of the maj or goals of all citrus growers is to increase yield, fruit quality and profit by

applying advanced management practices to the groves. A yield mapping allows growers to

know where a site-specific management would be beneficial. Eventually the growers are able to

make appropriate management practices on production inputs if necessary.

Citrus Harvesting

Most citrus groves in Florida are harvested manually. Citrus harvesting is a labor intensive

process involving large number of workers depending upon the size of grove. At the same time,

cost of hand harvesting operations has steadily been increasing. Harvesting productivity is









declined depending on grove condition and height of tree since workers use a ladder and picking

bags.

Recently mechanical harvesting use is increasing. During 2002-2003 season, about 1.7 sq.

km of the 58.6 sq. km of oranges were mechanically harvested. Mechanical harvesting systems

can increase labor productivity by 5 to 15 times and reduce unit harvesting cost by 50% or more

(Brown, 2002). Generally harvesting of citrus fruit takes 3 5% to 45% of total production cost

(Sanders, 2005). It is therefore clear that mechanical harvesting is a maj or solution for growers

who can manage large scale citrus groves in today's competitive citrus industry.

Motivation for Citrus Yield Mapping System

A site-specific crop management would give opportunity to growers optimizing crop

production based on in-field variability. By identifying different factors for in-grove variability,

citrus growers would make cost effective and environmentally friendly decisions on crop

production by changing inputs such as soil property, limestone, herbicide, insecticide, etc. At the

same time, mechanical harvesting would improve citrus production extensively. Significant

efforts have been done to improve productivity and effectiveness of mechanical harvesting since

mid-1960s (Futch et al., 2005).

A numerous research on fruit external quality inspection systems based on machine vision

has been conducted (Leemans et al., 2002; Blasco et al., 2003). Machine vision systems are

usually implemented at processing plants, not at grove. They basically examine fruit for color,

size, blemishes and diseases, and sort them by given criteria. However, no research has been

done to measure citrus size at the grove and simultaneously record citrus yield in real-time.

Thus, the purpose of research in this thesis was to work towards the development of a

citrus yield mapping and fruit size estimating system on a continuous canopy shake and catch









harvester. This machine vision system allows growers to know not only yield variability, but also

fruit size variability.

Objectives

Because of motivations described previously, the obj ectives of this study were to build a

machine vision based citrus yield mapping and fruit size measuring real-time system coupled

with a continuous canopy shake and catch harvester. More specific obj ectives were:

* To build hardware components of a real-time citrus yield mapping system that is able to
count number of fruit, and measure size of fruit for individual tree or unit area on a citrus
canopy shake and catch harvester,

* To develop an image processing algorithm to recognize individual fruit and measure its
size, and

* To test the complete system in a commercial citrus grove.









CHAPTER 2
LITERATURE REVIEW

Citrus Harvesting

Manual citrus harvesting has been used for many years, and still most of groves have been

harvested manually in Florida. A hand harvesting requires a crew of people carrying ladders and

picking bags. They detach citrus fruits from stems, put them into the picking bags, which weigh

several pounds, and transfer them to a nearby tub. Finally tubs are collected by a goat truck. As

citrus tree height increases, harvesting productivity is declined since they use ladders to reach

fruit on top of the tree canopy. Another concern of using manual harvesting is that cost of

manual harvesting operations is high. Because harvesting constitutes about 35% to 45% of a total

cost of citrus production (Sanders, 2005), there is a need to improve citrus harvesting practice, to

be cost effective and productive.

To mechanize and automate citrus harvesting practice is the maj or goal of the citrus

growers in Florida. A numerous studies have been conducted to develop and improve such

systems that are eventually able to eliminate manual labor, increase productivity and deliver cost

effective practices to the growers.

Robotic Harvesting

Robotic harvesting has been studied since 1960s. During this time, a lot of work has been

done. Several research groups all over the world tried to develop automatic citrus harvesters.

Robotic harvesters were usually designed mimicking human arm movements. It involves a

sequence of operations. First robotic arm or manipulator detects fruit from tree canopy.

Detection is usually accomplished with help of imaging sensors by recognizing color or shape of

fruit. Consequently it positions the end-effector to the relative location of fruit to be ready to










grasp it. The final operation is fruit removal. It involves gripping fruit and detaching fruit from

stem .

The most common design of robot manipulator is a three degree-of-freedom (DOF). The

obj ective of gripping fruit is to locate its coordination in the three-dimensional (3D) coordinate

system. The position coordinates from the fruit detection system are used by robot's controller to

calculate the three joint positions of the three DOF needed to reach to the desired position. The

limitation of the three DOF is that it cannot reach desired position due to its lack of degree of

freedom. To remedy such limitation of robot manipulators, researchers started designing five to

six DOF arms (Hannan and Burks, 2004).

Mehta (2007) developed a control system of robotic manipulator using cameras. By using

two cameras, a "teach by zooming" (TBZ) approach was utilized. TBZ allowed to orient the

robot arm to the referenced coordinate in the Euclidean coordinate system. Fruit detection has

been done using color thresholding. However, fruit occlusion was maj or problem.

At the end of the robotic arm, several pneumatically actuated fingers or end-effectors are

usually mounted to grasp citrus fruits. Some of the robotic systems have pneumatic suction

instead of several fingers. The end-effectors fall into several types depending on fruit detaching

design: cutting, pulling, twisting or twisting/pulling. The most common design is a cutting end-

effector. Cutting tools include a circular micro-saw (Muscato et al., 2005) and a blade (Buie,

1965; Macidull, 1973; Yoshida et al., 1988). Other type of end-effector is a pulling end-effector.

Several pulling based robotic harvesting systems were designed. Pulling force usually applies

along the fruit stem axis. However, the pulling force can be reduced if the force is applied at a

900 angle to the axis of attachment, as is often done in hand harvesting. Pulling actions disturb

the surrounding branches and leaves of tree, so it makes harder for consequent picking since it is









not easy to catch moving fruit. Third type of robotic design is a twisting method. In the twisting

method, researchers usually use a rotating suction cup. This type of automatic harvesting systems

was conducted in Juste et al. (1992). They reported that the detachment rate was 64% to 67%.

The reasons of failures were caused by obstacles and mechanical failures.

Flood (2006) conducted research on an end-effector design. He started investigating

physical properties to see the safe grasping limits of the end-effector. He reported that although

end-effector did not meet all criteria, it was able to harvest, and further refinements needed to be

done. Although robotic harvesting is involved a lot obstacle, it will be one of the solutions in a

citrus production.

Mechanical Harvesting

Significant efforts have been devoted to improve productivity and mechanize the

harvesting of the Florida citrus crop (Futch et al., 2005). The usage of mechanical harvesting

system has been comparatively increasing in last several years because of increasing cost of hand

harvesting operations. Today, in Florida, two types of mechanical harvesters are being used

(Futch and Roka, 2005). One is a trunk shake harvester that shakes a tree trunk, consequently

fruit will drop to the ground and then dropped fruit is picked up manually by a crew. The second

one is a canopy shake and catch harvester. It shakes tree canopies causing fruit to fall onto catch

frame. Then fruit would be carried through the conveyor system to the goat-like trucks. A

canopy shake and catch harvester is the most effective and productive mechanical harvester in

Florida. It can harvest 200 to 400 trees per hour assuming it travels down the tree row at ground

speed of 0.8 to 2. 1 km per hour (Futch and Roka, 2005).

A numerous research has been devoted to improve productivity of these harvesting

systems. As reducing the fruit detachment force, harvesting efficiency would improve

significantly. Several research has been conducted to understand how variety, harvesting season









and abscission chemicals would affect fruit removal efficiency. Earliest research was started in

1967 to examine reattachment characteristics of oranges for the development of mechanical

harvesting and abscission chemicals. Whitney et al. (2001) also studied different abscission

chemicals and tested how they could improve the fruit removal efficiency of trunk shakers. In

general, fruit removal efficiencies were increased 10% to 15% when orange detachment forces

were reduced 50% to 80%, but they reported that abscission chemicals could increase fruit

removals up to 14% and 30% on trunk shake harvesting system. However, higher concentration

of chemicals such as prosulfuron and metsulfuron-methyl negatively affects the trees (Kender et

al., 1999). Another research was to eliminate manual picking of the trunk shake harvester.

OXBO International Corp. developed and evaluated fruit pick-up machine (Bora et al., 2006). As

a result of evaluation, its picking rate was 1 16. 1 to 195.9 kg per min and its picking efficiency

was 80% to 97% depending on grove conditions.

One of farmers' concerns is that shake harvesting systems may affect negatively to the

groves as they make bruises to the tree branches. However, Buker et al. (2004) studied the effect

of mechanical harvesting to tree health and yield reduction over ten years. They found that there

was no negative effect on overall fruit yield although the harvester made bark injuries to citrus

trees.

Yield Mapping

Yield mapping is one of the crucial precision technologies since in-grove variability of

nutrients, tree size, soil types, etc. always exists. It allows growers to know where a site-specific

management would be beneficial. A large number of yield monitoring and mapping systems

were studied and developed for various fruit and crops such as grains (Schueller and Bae, 1987,

Searcy et al., 1989; Borgelt and Sudduth, 1992), potatoes (Campbell et al., 1994), peanuts

(Vellidis et al., 2001), silage (Lee et al., 2005), cotton (Sui and Thomasson, 2001; Perry et al.,









2005), coffee (Balastreire et al., 2002) and citrus (Schueller et al., 1999; Whitney et al., 2001;

Annamalai et al., 2004; MacArthur et al., 2006; Grift et al., 2006; Kane and Lee, 2006;

Chinchuluun and Lee, 2006).

Yield monitoring for citrus can be done either early season when fruit is on trees

(Annamalai et al., 2004; MacArthur et al., 2006; Kane and Lee, 2006; Chinchuluun and Lee,

2006) or harvesting season when fruit is harvested (Schueller et al., 1999; Whitney et al., 2001;

Grift et al., 2006). In earlier research for citrus yield mapping (Whitney et al., 2001), yield was

defined by the number of tubs and their locations in a grove. Thus, yield was based on several

trees, not a single tree. In addition, manual harvesting was required to collect and put fruit into

tubs to create yield maps.

However, other researchers (Annamalai et al., 2004; MacArthur et al., 2006; Grift et al.,

2006; Kane and Lee, 2006; Chinchuluun and Lee, 2006) developed yield mapping systems that

used more advanced technologies such as machine vision and weighing systems. Annamalai et

al. (2004) initiated the idea of acquiring yield information when fruit was on trees. Their

machine vision system was required taking non-overlapping sequence of images of tree canopies

while the system was driven through grove. The goal was to take images of fruit that was visible

from outside of tree canopies based on an assumption that number of fruit growing outside

canopies was proportional to the total number of fruit growing in the tree. Image location was

recorded by a DGPS receiver as well. Yield information was predicted by the number of visible

fruit of tree canopy per unit area. After they grabbed images, they counted the number of fruit

from each image using their developed image processing algorithm. Eventually they found a

yield prediction model to find citrus yield indirect way. Hand harvested yield obtained from

growers was compared against the predicted yield and a coefficient of determination, R2, WaS









0.53. The reason that predicted yield had low correlation to hand-harvested yield was that one

camera sometimes could not cover whole tree height. Thus some fruit that was outside the

camera' s field of view could not be counted. Reganathan and Lee (2005) extended the citrus

yield mapping system by adding four ultrasonic sensors to estimate fruit size along with citrus

yield information. In addition, they developed three different color classification techniques

(Bayesian, Fisher's linear discriminant and neural network) to improve fruit segmentation from

background. They found that the neural network classifier outperformed for estimating the

number of fruit with a root mean square error of 2.6 fruit than Bayesian classifier and Fisher' s

linear discriminant with root mean square errors of 4.2 and 7.2 fruit, respectively. Chinchuluun

and Lee (2006) added one more CCD camera to the system to provide full tree height coverage.

In addition to color image classification technique, they implemented marker-based Watershed

algorithm to separate individual fruit since occlusions were a maj or problem. During field

experiment, images of a total of 42 trees were acquired and these trees were hand harvested as

well. The coefficient of determination between the algorithm predicted yield and actual

harvested yield was 0.64. MacArthur et al. (2006) conducted preliminary research on yield

mapping using a mini-helicopter. The system had latest technologies such as camera and wireless

technology to control the mini-helicopter remotely. However, most difficult problem was to take

clear images. Vibrations of the mini-helicopter platform made blurriness on image acquisition.

They suggested that the system needed a lot of further improvements.

A canopy shake-catch harvester gives an opportunity to adopt site-specific crop

management in the sense that the harvester would travel in every unit area and collect fruit from

every tree in real-time. In addition to the shake and catch system, the harvester has conveyor

system that carries out all fruit on the catch system to load trucks.









Grift et al. (2006) conducted research in al laboratory condition on possibility of

developing yield mapping for fruit counting system on the conveyor system of the harvester

using a dual photo sensor as well as a laser sensor. Passing fruit through conveyor system was

compared with flows of fertilizer particles since many fruits passing per second formed a mass

flow. They tested several approaches. Firstly, a dual photo sensor was placed on conveyor to

count individual fruits. Since fruit was not singularized, this method counted several fruit as one

fruit. Then they built a gutter system allowing single fruit would pass each line at a time. In each

line, laser sensors were mounted to count them. According to authors' discussion, most feasible

approach among their methods was the one used a singulation. They advised that further research

was needed to make them reliable.

In this thesis, a feasibility of machine vision based citrus yield mapping system that would

not only count number of fruits per unit area, but also estimate fruit size was investigated.

Machine Vision and Fruit Size Estimating System

In a machine vision system, recognizing the fruit along with its characteristics such as size,

external color, blemishes and presence of defects from images would be very useful. A numerous

machine vision oriented applications were investigated in precision agriculture and remote

sensing such as fruit grading by external and internal qualities, determining bruised areas,

finding calyx, classifying forest image, etc. (Koning et al., 1994; Shahin and Symons, 2001;

Whitney et al., 2001; Leemans and Destain, 2004)

Fruit quality inspection is an important step in fruit production and marketing. Fruit

external quality inspection and grading systems are usually based on color, bruises, stem, calyx,

blemish, disease and insect damage. A number of such systems were developed for apple

(Leemans and Destain, 2004), potato, lentil (Shahin and Symons, 2001) and citrus (Whitney et










al., 2001). Fruit size is also one of the most important factors that customers recognize. Koning

et al. (1994) developed machine vision systems to grade potato by its size.

A variety of classifiers were used to identify fruit from background in image processing.

Classifieation methods include neural network, Bayesian classifier and discriminant analysis

based on different features of fruit surface. Leemans and Destain (2004) used a Bayesian

classifier to detect defects on 'Jonagold' apple. They reported that most of the defects were

extracted except russets, since russet was sometimes confused with the transition area between

ground and blush color. Marchant and Onyango (2003) compared a multilayer feed-forward

neural network classifier with a Bayesian classifier for classifying color image pixels into plant,

weed and soil. They found that the Bayesian classifier outperformed the neural network in the

sense of total misclassification error. However, if the number of features increased, more than

five or so, the Bayesian classification was infeasible. Thus they recommended using Bayesian

classifier over feed-forward neural network when the number of features was few enough to

require a feasible amount of storage.

Another maj or obstacle for outdoor machine vision and robotic harvesting systems is fruit

occlusion. One of the commonly used methods for occlusion is a watershed segmentation

method developed by Beucher and Lantuej oul (1979). Casasent et al. (2001) segmented nuts

from X-ray images using the Watershed transform. Lee and Slaughter (2004) used the modified

Watershed transform for separating occluded tomato plant leaves. Besides color analysis in the

visible range, near infrared (NIR spectrum plays an important role in machine vision. Aleixos et

al. (2000) developed a citrus fruit inspection system using a machine vision in a controlled

environment. In the system, CCD and NIR cameras were used to estimate fruit size and shape,

detect blemishes and classify them by color. Although they tried to singularize fruit while









passing through an inspection system, individualizing mechanism did not work efficiently. Thus,

they allowed traveling several fruit at the same time. Another problem was fruit touching when

they tried to extract contours of individual fruit.









CHAPTER 3
IMAGE ANALYSIS

Color Models

The use of color in image processing is motivated by the fact that color is a powerful

descriptor that often simplifies obj ect identification and extraction from scene. Colors that

humans perceive in an obj ect are determined by the nature of light reflected from the obj ect. Red,

green and blue colors are called primary colors because human eye is more sensitive to these

three colors. Variation of colors is composed of these primary colors. To make specification of

color standards, several color models are created. A color model is a specification of a coordinate

system and a subspace within that system where each color is represented by a single point.

The RGB Color Model

The red, green and blue (RGB) color model is based on a Cartesian coordinate system.

Figure 3-1 shows a cube of RGB color model in which subspace of interest is defined. Each

color consists of primary spectral components: (R, G, B). For instance, red, green, blue, black,

and white colors are defined by (255, 0, 0), (0, 255, 0), (0, 0, 255), (0, 0, 0) and (255, 255, 255),

respectively. The scale that is from 0 to 255 along each axis defines brightness. The diagonal

from black to white represents gray levels with different brightness.

The HSI Color Model

Although RGB is the most suited color model for human perception, it cannot provide all

needs in image processing algorithms. The HSI has advantages in the sense that color (Hue)

component is separated from intensity (I), so it alleviates the image processing complexity. It

consists of hue (H), saturation (S) and intensity (I) components. Hue component defines color. It

is measured by degrees. Saturation is color purity represented by vector. The length of vector

defined purity. The last component is intensity that measures lightness. HSI color model is









represented by any shape such as a hexagon, triangle or circle. The choice of shape does not

matter since any of the shape can be warped one from another. Figure 3 -2 shows the

representation of the model.


Figure 3-1. RGB color cube.

Converting from RGB to HSI is straightforward using the following equations. Hue

component of each RGB pixel is obtained using the equation


H = (3-1)
360 8, if B > G,

where,


1 j G)> + (R B)
B = cos2
-G C)2+(R-B()(G-Bl) /2


Saturation can be defined as










S = 1 3 nin( R,G, B)
(R +G +B)

The intensity component is defined by


I = (R + G + B)


(3-2)





(3-3)


green


yle ozv


cyan


black


Figure 3-2. HSI color model representation.

The YIQ Color Model

The YIQ color model is intended to take advantage of human color response

characteristics. The eye is more sensitive to changes in the orange-blue (I) range rather than in

the purple-green range (Q). This is the same color model as HSI model in which brightness

information (Y) is separated from chrominance information (IQ). To convert from RGB to YIQ,










the following formulas are used. R, G, B and Y components range from 0 to 1, while I

component ranges from -0.5957 to 0.5957 and Q ranges from -0.5226 to 0.5226.

Y = 0.299R +0.587G +0. 114B (3 -4)

I = 0.5957 R 0.2744 G 0.3212 B (3-5)

Q = 0.2114 R 0.5226 G + 0.31 11B (3 -6)

Image Classification

Classifieation involves recognizing obj ects based on their certain features from image. The

first step of designing classification is to choose features of obj ect. Classifieation accuracy would

be improved if the feature values of an interested obj ect are more differentiated than feature

values of the rest of objects. This step requires general knowledge of the obj ect such as its color,

shape, texture, size, etc. Maj ority of conducted studies have used color as a feature for obj ect

classifications. Depending on color models, any components of color models can be chosen as

features of obj ect. Figure 3-3 shows general steps of classification.

Choose features



Choose model



Train classifier



Evaluate classifier



Figure 3-3. Image classification design cycle.

Next step involves designing and choosing model or classifier. There is no universal

classification technique. Goodness of model choice depends on a nature of problem. The process

of using data to determine discriminants of the classifier is referred to as training the classifier. In










this step, a set of feature values is collected. Different classifiers are discussed later in this

chapter. The Einal step is to evaluate whether chosen model really works in a given problem. The

classification can be divided into two types based on chosen features: color and shape

classification. In this chapter, different classification methods are discussed.

Color Classification

Classifieation can be of two types depending on availability of prior knowledge of an

object: supervised classification, and unsupervised classification.

In supervised classification, a prior knowledge of an obj ect is required. This prior

knowledge comes from training operations. For instance, in the orange classification problem, all

possible color values of obj ect should be sampled before classifier would be designed. However,

collecting all possible set of feature values is infeasible. Thus, some approximations are used in

pattern recognition systems, so that it estimates the unknown values of the model. In this

research, the following Bayesian classification technique was implemented as a supervised

classification.

Bayesian classifier

Bayesian decision theory is a well known and robust statistical method. First let' s

formulize a decision-based recognition approach. Let x be an instance of n dimensional feature

vector. Also let g, (x) be one of W decision or discriminant functions. For W pattern

classes w w,, w,,...p,ll, the basic problem of pattern recognition is to Eind W discriminant

functions g, (x), g,(x), g, (x),...., g, -(x) with the property that, if a vector x belongs to class w ,,

then

g, (x) > g, (x), j ,,.. ;itj (3 -7)









In other words, an unknown pattern x is said to belong to the ith class if, upon substitution of x

into all discriminant functions, g,(x) yields largest numerical value. Now let' s find discriminant

functions using Bayes' rule defined in Equation 3-8.

p(x | wl )P(wl)
P(w, | x) = (3-8)
p(x)
where
P(w, | x) = posterior probability,
p(x | w, )= likelihood or probability density function of class ws,
P(w,) = prior probability and

p(x) = p(xr |w,)P)(w,)


Bayesian classification rule requires that prior probability P(w,) and likelihood of each

class must be known before discriminant functions would be designed. The posterior probability

can be chosen as discriminant function in the sense that for given x vector, P(w:| x) is a

probability that x vector belongs to class w,. Later decisions can be made by comparing these

probabilities. If the summation of Equation 3-8 would be removed, Equation 3-8 becomes the

equation shown in Equation 3-9.

g, (x) = P(w, | x) = p(x | w, )P(w, ) (3 -9)

For the sake of simplification, let us multiply both sides of equal sign of Equation 3-9 by

logarithm, so the product of two terms becomes a sum of two terms. Thus, the final Baysian

discriminant function becomes the following.

g, (x) = In p(x | wl) + 1n P(wl) (3-10)

As mentioned before, sampling all possible feature values of each obj ect is infeasible, so that

unknown distribution of likelihood is approximated by Normal distribution (Duda et al., 2000).

If it is assumed that likelihood is a normal distribution, then multivariate normal density

p(x | w, ) in d dimensions is defined as









1 1 -1
p(x | w, ) = 127 ( )(-1
(2a)cl1~ 2 12 2 il ( U 3-1
Where,
d = dimension of x vector,
pL = mean vector of w, class and
S= d-by-d covariance matrix.
Mean and covariance matrix can be found from the sampling set. Thus, the discriminant function

is simplified as follows:


g, (x) = -(- ) (-p)-In 2K n nPi, (-2


Shape Classification

Besides color, shape is an important feature that obj ects can be recognized. There are two

group of techniques used in segmentation: those based on contour-detection and those involving

region growing. These are primarily used in gray-tone image segmentations.

One of the region growing tools is a watershed transformation, which is used in separating

partially occluded obj ects in gray images. It was first introduced as a morphological tool by

Digabel and Lantuej oul (1978). Since then, many variations were introduced. They were

reported by Beucher and Lantuej oul (1979) and Beucher and Meyer (1992). Many researchers

used Watershed segmentation algorithm in many applications by means separating touching

oranges, apples, nuts, plants from weeds, etc. Some of them have already been mentioned in

Chapter 2.

The transformation named watershed because in the segmentation, a gray level image is

considered as topographic surface. The value of each pixel in a gray level image represents an

"elevation" of that point. For instance, let us consider two touching oranges (Figure 3 -4-A) and

see how the Watershed transformation is applied.










Figure 3-4-B shows that two orange regions were extracted by using color classification.

Orange region is shown by white color and the background is shown by black color. A distance

image is calculated by finding shortest distance from any point that is inside of fruit region to the

nearest point of the fruit boundary. The distance image is shown in Figure 3-4-C where white

points are located furthest from the boundary. A region which has highest numerical values of

the distance image is represented as the deepest part of the topographic surface. Those deepest

regions are called catch basins. In Figure 3-4-D, two catch basins are shown. The deepest point

of each catch basins is called a minima. A boundary of these two basins is called a watershed

line or dam. The goal of Watershed transformation is to find these dams. Let us assume that

water continuously floods the different catch basins. During the flooding, two or more floods

coming from different minima may merge. To avoid these floods merged each other, dams are

constructed on the points of surface where the floods would merge. These dams or watershed

lines will separate the touching obj ects as shown in Figure 3-4-E.













Catch basmns





Caehd~ DI























Figure 3-4. Watershed representation. A) Two touching oranges. B) Binary image of the image
A. C) Distance image. D) Topographic representation of the image (Source:
Vincent (1991)). E) Separated oranges.









CHAPTER 4
MATERIALS AND METHODS

In this chapter, hardware and software designs of the citrus yield mapping and fruit quality

inspection system as well as conducted image analysis will be discussed.

Hardware Design of the System

A hardware system to acquire high quality images for a machine vision based yield

monitoring and fruit size measurement system was designed. A housing (99.06 cm x 96.52 cm x

43.42 cm, 0.63 cm thick) for the system was constructed using aluminum sheet to acquire high

quality images (Figure 4-1-B) because outdoor imaging is much more complex than indoor

imaging. The system consisted of a 3CCD progressive scan digital color camera (HV-F31,

Hitachi Kokusai Electric Inc, Woodbury, New York), four halogen lamps (Master Line Plus

50W GU5.3 12V 38D, Phillips Electronics, Atlanta, GA), a laptop (CF-51, Panasonic, Secaucus,

NJ), and a data acquisition card (DAQCard-6036E, National Instruments, Austin, TX). In Figure

4.1-A, only two components are shown. Polarizing filters (25CP, Tiffen Co. LLC, Hauppauge,

NY) were placed in front of each lamp as well as camera to remove glare from the lamps.

Camera had a lens (TF4DA-8, Fujinon Inc., Wayne, NJ) with focal length of 4 mm. As discussed

in Chapter 1, a citrus canopy shake and catch harvester was used in this research. The harvester

had conveyor system that could deliver harvested fruit to load trucks. On the conveyor system,

the machine vision based yield mapping system was installed. To simulate the citrus canopy

shake and catch harvester's conveyor system, a test-bench was designed (Figure 4-1-B). The

specification of the test-bench was exactly the same as the bench of the harvester as shown in

Table 4-1.









Table 4-1. Conveyor belt specification
Specification Scale
Width of the conveyor belt 86.3 cm
Speed of the shaft 200 RPM
Diameter of gear wheel 16 cm
Speed of the conveyor belt 167.6 cm/s

System Operations

A simple illustration of system operation is shown in Figure 4-2. The conveyor along with

fruits moves constantly at high speed (1.67 m/s).





























Figure 4-1. General overview of the system. A) Camera and four halogen lamps. B) System
housing.

Camera takes a sequence of non-overlapping and non-skipping images of the conveyor

belt. Camera parameters are adjusted properly. The speed of conveyor belt is sensed by a hall-

effect encoder, so that the system knows when exactly it takes the next image. Consequently










images are analyzed by the developed algorithm to count number of fruit and record size of each

fruit.

System Components' Interconnection

At the end of the shaft of conveyor system, an encoder wheel with 30 teeth was installed

(Figure 4-1-C). A hall-effect sensor was placed next to the encoder wheel to acquire the speed of

shaft. The Hall Effect sensor would generate sequential pulses. The number of pulses is

proportional to the number of gear wheel tooth passed per second. The data acquisition card was

used as an interface to the laptop computer.

Camera Parameters

To acquire clear images from the moving conveyor belt, different camera shutter speeds

were studied in this research. Shutter speed is the time for which the shutter is held open during

the taking of image to allow light to reach the camera image sensor. Although the camera has

auto shutter and auto aperture mode, these modes could not provide high quality images since the

conveyor system moves along with fruits at high speed. Thus, the shutter should be opened very

shortly to freeze fast-moving obj ect. On the other hand, since high shutter speed would allow

less light coming to imaging sensor, image would not be bright enough. Therefore, the camera

was tested by changing the shutter speed while moving the camera and taking obj ects in order to

trade-off sharpness and brightness of images. The shutter speed of the camera can be set in the

range of 4 to 1/100,000 second.



























~ovingdirection


Conveyor belt


Figure 4-2. Pictorial illustration of system operation.

Figure 4-4 shows images taken at the different shutter speeds (auto, 1/2000 s, 1/2150 s,

1/2230 s and 1/2300 s) while the camera was moving. Fruits in the first two images look blurry.

Data Image
Laptop 3CCD
Encoder acq isition coptracquisition cmr



DGPS
reciever


Figure 4-3. System connection diagram.

However, in the last three images, oranges can be seen clearly. An image in Figure 4-4-E is

generally looked fuzzy although oranges can be seen. Thus, as the result of test trials, the optimal

shutter speed was determined to be 1/2200 s and aperture was set to auto mode.










Image Synchronization with Encoder

Based on the camera' s field of view and image acquisition timing, the camera was installed

at the proper height to cover the frame width of 68.6 cm. A size of 800 x 600 pixels was chosen

as an optimal frame size of the camera (Table 4-2).

Table 4-2. Camera resolution
Frame size Max FPS Grabbing delay Actual frame
(ms) width (cm)
1024x768 24 bit 7.5 133.3 190.5
800x600 24 bit 15 66 68.6

The width of housing was 41.1 cm. It means that the camera must capture every 41.1 cm

width of the conveyor since housing was installed on the conveyor system. As a result of several

trials, the 41.1 cm distance of conveyor movement was equal to 30 teeth of the encoder wheel.

The speed of conveyor belt was 167.6 cm/s, whereas the width of housing was 41.148 cm. So,

image had to be acquired at every 41.1 cm /167.6 cm/s = 245 ms. Therefore, image acquisition

was synchronized with the conveyor belt using 30 teeth of the encoder. The accuracy of

conveyor belt movement is defined as 41.1 cm/30 teeth = 1.3 cm.















Figure 4-4. Image acquisition of moving obj ect at the different shutter speeds. A) Auto shutter
speed. B) 1/2000 s. C) 1/2150 s. D) 1/2230 s. E) 1/2300 s.





















C D














Figure 4-4. Continued


System Software Design

A multithreading system software was developed using Microsoft Visual C++ 6.0

MFC/COM and ImageWarp (BitFlow Inc., Woburn, MA). Since an image had to be taken in 245

ms assuming that the conveyor speed is 167.64 cm/s, the program had to be run fast enough to

read from encoder, acquire an image, read DGPS receiver and analyze images. General overview

of the multithread-based program is shown in Figure 4-5.

A thread is a path of execution through a program's code, plus a set of resources assigned

by operating system. A preemptive scheduler in the Windows operating system kernel divides

CPU time among active threads so that they appear to run simultaneously. Multiple threads

dramatically improve an application's efficiency and running time.









The system software consisted of three working threads each of which worked

independently from each other. Communication between these threads was accomplished by

creating events. An encoder thread would always read signal that comes from the hall-effect

sensor, and count period of square waves (Figure 4-6). When the counter reached at the encoder

counter limit, N, the thread created an event telling to the ImageWarp thread that it is a right time

to acquire next image. The counter limit N was determined as 30, based on camera view and

traveling distance of conveyor system.






Thread 1.Read Encoder counter
Encoder thread 2.If counter reaches N, it sends a
signal to the Thread 2 and End.
Thread.



1.Grab Image.
,,Thread2 ImageWarp thread 2.Process Image.
3.Save Images to the
Hard drive.


,,Thread3 GPS thread Read GPS data




Figure 4-5. The system software overview.

The system software was able to run in two modes: real-time and post processing. In real-

time mode, the ImageWarp thread would acquire a digital image and directly determine the

number of fruits and size of each fruits. Whereas, in post-processing mode, the thread would

acquire image and save it to the hard drive. The GPS thread would constantly read signal coming

from the DGPS receiver. It then parses row of characters to latitude, longitude and time. Finally

it saves GPS information in a raw file.









































Figure 4-6. Encoder thread algorithm.

System Experiments

The machine vision based citrus yield mapping and fruit size measurement system was

examined both in a laboratory on a designed test-bench and on a shake and catch harvester.

Valencia oranges were used throughout these experiments.

Test-bench Experiment

The machine vision based citrus yield mapping and fruit inspecting system was first tested

in a laboratory at the Citrus Research and Education Center (CREC) at the Lake Alfred, Florida

during February and March of 2007.










Image acquisition during the test-bench experiment

An algorithm for counting the number of fruit and measuring the size of fruit was

developed and tested on the test-bench. A total of 14 trials were conducted. In each trial,

different number of boxes of oranges was manually fed to the conveyor belt. Total fruit weight

was measured with a balance (PS60 Parcel Scale, Mettler Toledo, Columbus, OH) as shown in

Figure 4-7 and a sequence of fruit images was acquired for each trial. A total of 3,653 images

were acquired. Among these, 703 images were randomly used for classification.













A B

Figure 4-7. Test-bench experiment. A) Feeding conveyor belt. B) Weight measurement of tested
fruit.

For each image, fruit parameters such as the number of fruit, fruit diameter and fruit area

were extracted during image analysis. Consequently these parameters were added up for images

of every test in order to see how they were correlated with actual fruit weights.

Citrus Canopy Shake and Catch Harvester Experiment

The harvester experiment was conducted during March of 2007 at the commercial citrus

grove (Lykes Inc.) located in Fort Basinger, FL. A total of 773 images were taken during static

operation. Fruit was loaded into the goat like truck and its weight was measured using a balance

(RW-05S, CAS, Korea). The fruit weight was determined by subtracting only truck weight from

the measured total weight.































figure 4-t8. Canopy shake and catch harvester expenment.

Image Processing Analysis

The basic goal of image processing analysis was to recognize fruit pixels and estimate the

number of fruit and their size from images. Although the housing of the machine vision system

with artificial lighting was built to improve image acquisition, color similarity between fruit and

background, sunlight coming from the bottom, and fruit occlusions were maj or obstacles (Figure

4-9) to accomplish the obj ectives stated in Chapter 1.

The most visible feature to identify fruit from image was their color. Different color

components were carefully examined in various color models how fruit pixel values could be

isolated from background pixel values. The most isolated components were saturation and hue

components of HSI color model and I component of YIQ color model. Therefore, hue, saturation

and I components were studied further to be chosen as features in color image classification.






























Figure 4-9. Color image of fruit and background.

A total of 1,118,618 fruit pixels and 2,365,368 background pixels were collected from 60

random images to implement supervised color classification algorithm. These 60 images were

taken during the test-bench experiment. In color classification, any given pixel has to be assigned

to either fruit class or background class. The histograms along with images in different color

components are shown in Figure 4-10. Hue and I components provided better separation of two

classes (Figure 4-10-B, F). Their means are shown in Table 4-3.



12 -*Background









50 100 150 200 250 300
A ~Hue component B

Figure 4-10. Images in different color models and their histograms. A) Hue component of HSI
color model. B) Hue histogram of fruit and background. C) Saturation component of
HSI color model. D) Saturation histogram of fruit and background. E) I component of
YIQ color model. E) I (YIQ) histogram of fruit and background.













Sx 104
Fruit
4 5 --Background










x15



Fruit
**Background

10I









-%740 -20 0 20 40 60 80 100
E I component F


Figure 4-10. Continued


Table 4-3. Means of hue, saturation and I (YIQ) components of fruit and background classes
Classes Hue Saturation I (YIQ)
Fruit class 60.6 153.0 30.8
Backround class 108.6 127.7 -4.5


The image processing algorithm steps are shown in Figure 4-11. After images were


acquired, digital images were converted to HSI and YIQ color models. Then color classification


algorithm was followed. Different color classification approaches were studied. Although linear


threshold-based color classification was implemented for color classification, it did not give


good fruit recognition. Therefore, quadratic classification models were investigated rather than


linear one to accomplish the obj ectives stated in Chapter 1.




















Figure 4-11. Image processing algorithm.

As formulated in Chapter 3, Bayesian classifier was used in the color classification.

According to the Equation 3-10, fruit and background discriminant functions are defined as

Equation 4-1 and Equation 4-2, respectively.

K7,,,,(x) = In p(x | fruit) + In P( fruit) (4-1)
ghackgrouncfx) = In p(x | background) + ln P(backgrouna) (4-2)

Thus, the decision is made as


For a given x vector, frii f,,x ,~,,()(4-3)
background: othel~ nI ise~

The prior probabilities depend on the coverage of fruit pixel in an image. The coverage of

fruit pixel would change from image to image, hence the priors are not really known for any

individual test cases. Marchant and Onyango (2003) advised to choose equal prior probabilities

in these situations. However, in randomly sampled images, some images had no fruit, but some

of them had full coverage of fruit. Therefore, the prior probabilities in this study are defined as

follows:

Number of sampyled fruit pixels
P( fruit) =
total pixels

P(backgrounci) = 1- P( fruit) (4-4)











The priors was found as P(fruit) = 0.22 and P(background) = 0.88 using the fruit and

background samples. The quadratic discriminant function was found on I and Hue as shown in

Figure 4-12.



| Discriminant function

250-


200-


E 150-


100-




50-


I component x1 ig



Figure 4-12. Discriminant function between fruit and background classes.

Fruit Parameters

In image analysis, the following parameters were calculated for every individual fruit: area,

circularity, convexity, average diameter, maximum and minimum diameter and ellipticity. The

ImageWarp provided all functions to calculate these parameters.

Yield Prediction Model

To develop a citrus yield prediction model, a relationship between fruit weight and fruit

diameter was studied. The first assumption was that the correlation between them might be

linear. Welte (1990) studied Jonagold apple growth size during growing season, and found that

apple weight was directly correlated to apple diameter with regression coefficient of 0.9899.










Thus orange samples were collected from the CREC. Fifty three randomly selected

oranges' actual weight and actual diameter were measured. The diameter was referred to an

average of minor and maj or dimension of fruit since most fruits were not complete circle. As the

result of measurement (Figure 4-13), coefficient of determination was 0.971. Thus, the

relationship between them was assumed linear.



320


R2 = 0.971
Y = 7.86 x 359.18


o gd; o


120


72 77
Actual fruit diameter (mm)


Figure 4-13. Relationship between actual fruit weight and diameter.

Fruit area can be measured from images in 2-dimension (2D). If fruit is assumed as a

complete circle, then area (A) is determined as


A = 7iR2 = id2
4


(4-5)


Where R = radius of fruit, d = diameter of fruit.










If fruit weight was compared against with two-dimensional fruit area, then the correlation

between fruit weight and fruit diameter would not be linear, but it is quadratic. The regression

analysis was conducted with them and the coefficient of determination was found to be 0.972.


330




En280




S230




180


R2 = 0.972
Y = 0.05 X 71.36


3800


4300


4800


5300 5800 6300

Actual fruit diameter (mm2)


6800


7300


Figure 4-14. Relationship between actual fruit weight and area.

In 3-dimension, fruit shape is sphere. The weight (M) of any obj ect is defined by density (g) and

volume (V).

M~ = gV,

V = xR3~- d 3 (4 -6)
3 6

Thus, if it is assumed that density is uniform for all fruit, then fruit mass (M) is defined as


M= d3g
6


(4-7)










A regression analysis was conducted on fruit weight and fruit cubic diameter. The coefficient of

determination was 0.969.


R2 = 0.969
Y = 0.0004 X + 25.29


330




S280




S230




180


130
230000 280000 330000 380000 430000 480000 530000 580000 630000 680000

Actual fruit diameter m3)


Figure 4.15. Relationship between actual fruit weight and volume.

From above analysis, fruit weight and fruit area comparison showed better performance

than others.

Camera Calibration

To measure actual size of fruit from images, camera had to be calibrated. Since camera

was placed at constant height of 891.8 mm, camera intrinsic parameters were used to find the

actual obj ect size. In this calibration, a pinhole camera model was used (Forsyth and Ponse,

2001). The focal length of the camera's lens was 4 mm and the camera pixel size was 4.65-10-3

mm x 4.65-10-3 mm. The perspective projection figure can be represented in Figure 4-16, where









x defines an object size in an imaging plane, f defines the camera focal length, Z defines a

distance from camera to the obj ect and X defines a obj ect size in a real world coordinate system.





( ) X






Figure 4-16. Camera perspective proj section.

A ratio of similar triangles is defined as:

X[nm2] _Z[nm]
(4-8)
x[ pixel] f[nmm]

Each unit of the parameters is written in Equation 4-8. To make parameter units consistent,

pixel should be converted into millimeters. Since the camera pixel size is 4.65-10-3 mm x

4.65-10-3mm, the actual object size X is defined as:

Z 891.81479
X = x 4.65 10 nm2 4.65 103 -x vnm = 1.0367 x vnm (4-9)
f 4









CHAPTER 5
RESULT AND DISCUSSION

In this chapter, the result summaries of the test-bench as well as canopy shake and catch

harvester experiments will be discussed. The chapter starts with a summary of image

classification and processing. Then yield prediction results will be discussed later.

Color Segmentation Result on Test-bench

As mentioned in the previous chapter, a total of 14 test trials were conducted on the test-

bench at CREC, Lake Alfred, FL. The image processing algorithm was applied to the 609 images

that were taken during the experiment at CREC. The color segmentation algorithm was started

with image acquisition (Figure 5-1-A) and followed by Bayesian classification (Figure 5-1-B).

After Bayesian classification, morphological operation removed non-fruit pixels that came into

fruit regions. Finally only fruit regions were extracted (Figure 5-1-C). Since some neighboring

fruit regions joined each other, watershed transform separated them into individual fruit (Figure

5-1-D).

Finally, fruit parameters were extracted from the segmented digital images taken during 14

trials. The parameters included number of fruit, diameter and area of each fruit were determined

by the ImageWarp software. The fruit region was estimated by the number of pixels in a blob.

However, the fruit diameter is defined as


Diameter = 4- re (5-1)


Besides these parameters, fruit circularity, ellipticity, convexity and maximum/minimum

diameters were measured from each fruit in digital images. Some portion of 207-page results is

included in Appendix B.












-
''
.~ : ''' ~
;? L. :T i~'i:B
'' ::
..,.;
-~ J:' ~UiK 'I
~I 1
"~ "
.' .-s~j
~
r~ I:j,.c..; :~L~'~J(CID~;I~D4Db~~~\ i; ~f~r:: !;"
.~ :- 1. ':
:I~-: ;
,,
,e rrc;-

.. ~-. '''''`ert,., i~
''
~- ,~~~ ~~~ [ C:.:Y
j~l; .)i!llllll.l I:
i '' II'
i- ~-
I l:~::~;r;. Ti~* .tii C-(i:-i ::


B


**


Figure 5-1. Image segmentation result. A) Original image. B) Bayesian classifier result. C)
Morphological operation result. D) Watershed transform result.

Table 5-1 summarizes the test-bench experiment results by the image analysis. A


regression analysis was conducted on fruit parameters which were found by the image analysis.


First, a relationship between the actual fruit weight and the number of fruit found by image


processing algorithm was studied (Figure 5-2). The relationship was almost linear with the


coefficient of determination of 0.892. Moreover, a trend between the actual fruit weight and fruit


area measured by the algorithm was studied as well (Figure 5-3). The relationship between them


was even better with R2 of 0.962. The relationship between the actual fruit weight and fruit


diameter measured by the algorithm was also conducted (Figure 5-4). The coefficient of


determination was 0.963.





Table 5-1. Summary of the test-bench experiment results

Total Actual fruit Sum of fruit Algorithm
Test
images weight (kg) area (pixel) fruit counting


Sun of Fruit
Diameter
(pixel)
3424
5889
10471
11792
15798
15591
22040
20043
29588
26476
28024
37307
38888
33336


Sunt of Fruit
Diameter
(mm)
3549.78
6105.33
10855.7
12225.2
16378.3
16163.7
22849.6
20779.3
30674.9
27448.5
29053.5
38677.5
40316.5
34560.6


22.59
25.92
31.18
46.11
49.28
48.99
75.30
71.03
97.51
93.85
91.58
114.26
111.22
120.29


208915
338612
583687
660859
867118
884879
1245102
1151251
1677222
1524808
1572474
2073615
2202060
1847159


140.00


R2 = 0.892
Y = 0.13 X + 20.54


120.00



S100.00

800

60.00






0 0


20.00


0 100 200 300 400 500 600 700 800 900

Algorithm fruit counting

Figure 5-2. Actual fruit weight versus the number of fruit by image processing algorithm:
regression analysis.












140


120


-100


.? 80


4- 60


<' 4 0


R2 = 0.962
y = 5.25E-05 x + 8.14


o
O


0




Figure 5-3.


140-


120 -


S100-


-2 80-


S60 -


s 40-


20-


500000 1000000 1500000 2000000

Fruit area (pixel2)


Actual fruit weight versus fruit area: regression analysis.


2500000


R2 = 0.963
Y = 1.53 E-12 X + 44.55


0 5000 10000 15000 20000 25000 30000 35000

Algorithm fruit diameters (pixels)


40000 45000


Figure 5-4. Fruit diameter versus actual fruit weight: regression analysis.








Color Segmentation Result on the Harvester
The system was tested on a commercial canopy shake and catch harvester (Freedom Series

3220, Oxbo, Clear Lake, WI) at a commercial citrus grove (Lykes grove, Fort Basinger, FL).

Fruit weight was measured using a balance (RW-05S, CAS, Korea). The developed image

processing algorithm was applied to images taken during the experiment. The result is shown in

Figure 5-5.







A B


C LA D

Figure 5-5. Color segmentation result for Hield experiment. A) Original image. B) Bayesian
classifier result. C) Morphological operation result. D) Watershed transform result.
The Bayesian classifier worked relatively well on the images. As a result of morphological

operations, some fruit regions were incorrectly classified. Fruit regions became smaller than the


cr~~ni









original sizes. Consequently, incorrect segmentation made harder the individual fruit extraction

(Figure 5-5-D).

Breunig et al. (2000) investigated detecting outliers from large data sets based on a local

density. An outlier is defined as an isolated obj ect with respect to the surrounding neighborhood.

The decision for detecting outlier obj ects is the following:


(Object, if density of object in a circle >7h~erehold(52
Outlier, othernii\ ite2

Using this density clustering approach, fruit pixels from binary image was clustered. The

circle radius was chosen as two pixels based on several trials. The density threshold value (5

pixels) was chosen too. The result of density cluster or circular filtering is shown in Figure 5-6

from Figure 5-5-C.















Figure 5-6. Circular filtering result for Hield experiment 2.

Although fruit weight was measured twice, two measurements were not enough for the

analysis. Thus, the number fruit from 60 images was counted to compare it with the result of the

image processing algorithm. The regression analysis was also conducted on number of fruit that

was extracted by human counting and algorithm counting. The coefficient of determination was

0.891 (Figure 5-7).











-Human counting

60 R2 = 0.891 Computer algorithm counting


50-








2 20-


10-



1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59

-10-

Image number

Figure 5-7. Regression analysis between human counting and algorithm counting.



System Execution Time

The system was tested on a laptop computer with 2.1 GHz CPU and 64 MB RAM

specification. Two functions were developed to display actual execution time on a dialog. The

average execution time of 50 trials for acquiring an image, reading DGPS output and reading the

encoder was measured (Table 5-2). It was constant at 120 ms. The average time for color

segmentation was 138 ms. The execution time of morphological operation and Watershed

transform were not constant. They depended on the number of fruit in an image. The average

time of morphological operation and Watershed transform were 40 ms and 89 ms, respectively.

Thus, total execution time was 367 ms.










Table 5-2. System execution time
Average
Execution steps execution
time (ms)
Image, DGPS, encoder acquisition and saving an
image 120
Color segmentation 118
Morphological operation 40
Watershed transform 89
Total execution time 367



System Limitation

The machine vision system has not been tested at citrus grove when the continuous canopy

catch and shake system was used for harvesting. Several obstacles are expected to test it at citrus

grove. First of all, canopy catch and shake system creates vibration in which the machine vision

system components such as wire and interconnections may be malfunctioned. In addition, the

canopy shake and catch system creates dust while harvesting, so that it may effect to the image

acquisition by reducing image quality. Another issue is that some foreign materials will also be

harvested along with fruit, such as leaves or branches, which will pose some difficulty in

identifying fruit. Those leaves and branches may affect image segmentation negatively. Some of

the pesticide residues such as copper (greenish color) or disease such as rust mite could also

interfere with color segmentation.

Some other complications can be expected, but these are not known until the machine

vision system is tested at actual citrus grove.









CHAPTER 6
CONCLUSIONS

In this study, a machine vision based citrus yield mapping and fruit size measurement

system on a canopy shake and catch harvester was developed successfully. The advantage of this

system from previous citrus yield mapping systems is that it can record fruit size information in

real-time and without any manual labor involvement. The system was tested on a test-bench as

well as on a commercial canopy shake and catch harvester.

To develop and test the system, a test-bench with the same specification as the mechanical

harvester was designed at the Citrus Research and Education Center located at Lake Alfred, FL.

The system had to acquire high quality images of citrus fruit that was passing through a conveyor

system. Multithread based system software was developed using Visual C++ and ImageWarp

image processing software.

A supervised image processing algorithm was developed to count number of fruit and

measure the size of individual fruit. The fruit area, the number of fruit and fruit diameter were

measured using an image analysis algorithm from the set of images for the test-bench trial.

Actual fruit weight was also measured. The coefficients of determination of the sum of areas, the

number of fruit and the sum of fruit diameters against actual fruit weight were 0.962, 0.892 and

0.963, respectively. When relationships between actual fruit weight (W) and actual diameter (D)

of a single fruit were studied, a quadratic (W~D2) and linear (W~D) relationships showed the

highest linear trend with R2 of 0.972 and 0.971, respectively. In a test-bench trial, these two

relationships were still held. Thus, the image processing algorithm performed quite well on

images taken during the test-bench trial.

When the citrus yield mapping system was tested on a canopy shake and catch harvester,

the number of fruit was counted manually from 60 images. The coefficient of determination










between the number of fruit counted by image processing algorithm and the human counting was

0.891 for the harvester trial. The reason for the low performance of the system for the harvester

trial was due to the following:

1. Although illumination was controlled by utilizing housing for the system, the sunlight
came from under the conveyor system, so that image lighting was not always
uniform.
2. Due to this non-uniform lighting, the color segmentation algorithm could not perform
well.

To improve the performance of the system, a cover under the conveyor system should be

placed to block the sunlight coming into the system. The next step would be to test the system at

the citrus grove while the canopy shake and catch harvester is harvesting citrus fruit.









APPENDIX A
PROGRAMMING CODE

Encoder Thread Code

UINT WorkerThreadProc( LPVOID IParam)

EncoderReading m_oEncoderInst;
//InfoStruct *mrinfo = (InfoStruct*)1Param;
CComDemoDlg* m_dlg = (CComDemoDlg*)1Param;

unsigned long imNum = 0, LIMIT = 20;
unsigned int lastCount = 0, data=0, counter=0;

CString ss, ss2, ss3;

LIMIT = ::GetDlgItemlnt (m_dlg->m~hWnd, IDCENCODERLIMIT, NULL, true);
ss2.Format("LIMIT=%d", LIMIT);

: :PostMessage (m_dlg->m~hWnd, WMSEND_S TATUSME SSAGE, 1 0);

while (!STOPFLAG)

data = m_oEncoderInst.DAQmxRun();

counter = data lastCount;
ss.Format ("counter value=%d", counter);


AS SERT(m_dlg->m_plmWThread->m~hThread = NULL);
AS SERT(m_dlg->hlWEvent = NULL);
::SetDlgItemlnt (m_dlg->m~hWnd, IDCEDIT_COUNTER, (UINT)counter,
true);

// Read counter. When the counter reaches LIMIT, then it sets the hEncoderEvent event.
if (counter > LIMIT)

imNum++;
lastCount = data;
// Raised
if (!S etEvent(m_dl g->hEncoderEvent))

AfxMessageBox(_T("SetEvent failed "));
m_dlg-> Send Status(_T(" SetEvent failed "));

else









m_dlg-> Send Status(_T(" SetEvent for Encoder "));


//::PostMessage (m~info->hWnd, WM_COUNT, counter, imNum); //
wParam = counter value, IParam = imNumber


//-1 = Worker thread exiting...
//-2 = Busy. Script running...
//-3 = Script thread exiting...
: :PostMessage (m_dlg->m~hWnd, WMSEND_S TATUSME SSAGE, 1, 0);

return 0;


ImageWarp Thread Code


UINT WorkerlWProc(LPVOID IParam)


DWORD dwWaitResult;
int flagl, flag2;
// InfolWThread* templW
CComDemoDlg* m_dlg I
HANDLE hEvents[2];
flagl = 0; flag2 = 0;


(InfolWThread* )Param;
(CComDemoDlg*)1Param;


: :PostMessage (m_dlg->m~hWnd, WMSEND_S TATUSME SSAGE, 3 0);
while (!STOPFLAG)


hEvents[0]
hEvents[1]


m_dlg->hEncoderEv ent;
m_dlg->hlWEvent;


//m_dlg->SendStatus(_T(" IMThread "));
dwWaitResult = WaitForSingleObj ect (m_dlg->hEncoderEvent, INFINITE);


if (dwWaitResult


WAIT_OBJECT_0)


m_dlg->SendStatus(_T(" Inside of wait "));


if (!ResetEvent(m_dlg->hEncoderEvent) ||I !ResetEvent(m_dlg-


m_dlg->SendStatus(_T(" Failed on ResetEvent "));


>hlWEvent))










: :PostMessage(m_dlg->m~hWnd, WMIW, 0, 0);

else

m_dlg->SendStatus(_T(" Failed reading WaitForSingleObject "));




m_dlg->SendStatus(_T(" IMThreadExit "));
//-1 = Worker thread exiting...
//-2 = Busy. Script running...
//-3 = Script thread exiting...
:: SetDlgItemText(m_dlg->m~hWnd, IDCFIRST, _T("IW thread exiting..."));
//::PostMessage (m_dlg->m~hWnd, WM_1SEND_STATUSMESSAGE, -3, 0);
//CButton* th2 = (CButton*)::GetDlgItem(m_dlg->m~hWnd,IDC_CHCH)
//th2->SetCheck(0);

return 0;



Image Processing ImageWarp Script in Real-time Mode


closeAll()
setCurDir("C :\\RESEARCH\\CitrusYieldMapping\\07-0226-Ts~ae\"
Dim I as Integer
I= 1
do
resetParam()
setShowGrid(FALSE)
grablm(1)
grablm(100)

setS elect(1 00, SH RECT, SEL NEW, 124, 172, 89 1,622)
splitRGB (100, 102,101,103) R=102
splitYIQ(100, 103,105,104) I=105

threshold(102, 106,MPRESET,48.,236 .,TRUE,FALSE) Thresholding using red
scrap(106,107,0,8,0)
dilate(107,106, 1,PR VBAR)
fill(106, 107,1,0)
erode(1 07, 108, 1,PRCRO SS)
open(108,1 09,2,PRCIRCLE)
scrap(109,110,0,1 470,0) '110










threshold(105,106,MPRESET,132.,187.,TRUEFLE Thresh by I

scrap(106,107,0,8,0)
dilate(107,108, 1,PR VBAR)
fill(108, 109,1,0)
erode(1 09,1 11 1,PRCRO SS)
open(111,1 07,2,PRCIRCLE)
scrap(107,108,0,800,0)
andim(108,110, 112) 108 AND 110 = 112

separate(108,113,1,1)
scrap(113,114,0,400,0)
duplicate (114, 2)
resetParam()
selectParam("Count"',"Area", "Diameters","DMax", "DMin")
measObj ects(1 14,105)
savelm (100, strFormat ("image%d.tif", I))
savelm (114, strFormat ("Processedlmage%d.tif", I))
I= I+1
SendMessage(0)
loop


Image Processing ImageWarp Script in Post-mode


setCurDir("C :\\RESEARCH\\CitrusYieldMapping\\07-0226-Ts~ae\"
Dim I as Integer
I= 1
do
grablm(1)
grablm(100)
savelm (100, strFormat ("image%d.tif", I))
I= I+1
SendMessage(0)
loop
End















Table B-1. Fruit parameters
Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixel') Cruaiy Cne~ daee imtrDaee litc
(pixels) (pixels) (pixels)
1 6035 0.895 0.984 87.658 88.837 82.680 0.998
2 5222 0.845 0.961 81.541 85.703 75.644 0.995
3 4353 0.857 0.971 74.447 78.294 68.542 0.996
4 3722 0.833 0.960 68.840 75.432 60.407 0.994
5 5048 0.881 0.978 80.171 83.487 73.756 0.997
6 6012 0.910 0.984 87.491 90.449 81.000 0.999
7 5257 0.880 0.967 81.813 84.006 74.947 0.996
8 4957 0.903 0.982 79.445 81.884 72.835 0.995
9 4842 0.868 0.971 78.518 81.400 69.635 0.995
10 4713 0.787 0.959 77.465 82.365 71.000 0.990
11 4352 0.883 0.974 74.439 76.059 67.683 0.996
12 4238 0.877 0.974 73.457 75.604 69.426 0.996
13 2291 0.790 0.965 54.009 70.036 40.706 0.984
14 3593 0.910 0.980 67.637 71.021 62.362 0.998
15 806 0.653 0.896 32.035 44.045 25.456 0.876
16 1235 0.703 0.921 39.654 56.223 29.732 0.946
17 2 0.000 1.000 1.596 1.000 1.000 1.000
18 5 1.000 1.000 2.523 4.000 1.000 1.000
19 4 1.000 1.000 2.257 3.000 1.000 1.000
20 1 0.000 1.000 1.128 1.000 1.000 1.000
21 1 0.000 1.000 1.128 1.000 1.000 1.000
22 4 1.000 1.000 2.257 3.000 1.000 1.000
23 5 1.000 1.000 2.523 4.000 1.000 1.000
24 2 0.000 1.000 1.596 1.000 1.000 1.000
25 5822 0.852 0.966 86.098 89.376 79.981 0.995
26 5058 0.874 0.974 80.250 83.600 74.673 0.997
27 4926 0.917 0.986 79.196 80.281 75.604 0.998
28 4915 0.867 0.973 79.107 81.884 74.007 0.996
29 3019 0.790 0.947 61.999 65.765 53.451 0.973
30 5831 0.898 0.983 86.164 89.185 82.000 0.998
31 5257 0.861 0.973 81.813 85.212 78.090 0.995
32 3954 0.871 0.969 70.953 74.626 65.054 0.997
33 4857 0.895 0.978 78.639 81.216 72.622 0.998
34 5146 0.892 0.975 80.945 84.172 76.059 0.996
35 5515 0.898 0.976 83.797 87.321 79.209 0.997
36 4413 0.898 0.979 74.959 76.276 71.000 0.998
37 5322 0.869 0.966 82.318 86.822 76.557 0.995
38 4868 0.875 0.974 78.728 81.056 74.000 0.995


APPENDIX B
IVEASURED FRUIT PARA1VETERS











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
39 4887 0.883 0.975 78.882 81.541 73.000 0.997
40 4071 0.885 0.966 71.996 77.621 67.720 0.995
41 4494 0.889 0.978 75.643 78.409 71.589 0.998
42 4728 0.878 0.973 77.588 81.413 70.214 0.997
43 5883 0.899 0.982 86.547 87.092 82.219 0.998
44 3743 0.850 0.955 69.034 74.250 63.000 0.990
45 5625 0.846 0.973 84.628 91.608 77.104 0.996
46 6274 0.909 0.983 89.377 92.655 83.935 0.999
47 4411 0.875 0.969 74.942 77.795 69.584 0.997
48 4073 0.850 0.968 72.013 75.961 63.781 0.992
49 5027 0.891 0.983 80.004 83.355 74.545 0.999
50 4437 0.839 0.967 75.162 80.777 69.426 0.994
51 3619 0.828 0.961 67.881 73.007 61.000 0.987
52 4428 0.794 0.956 75.086 79.246 67.209 0.991
53 4163 0.845 0.971 72.805 76.420 67.067 0.996
54 3320 0.734 0.920 65.017 70.880 56.613 0.967
55 129 1.000 0.991 12.816 14.036 9.899 0.976
TOTAL of
TRIAL #1 208915 3488.016 Number of fruit: 55
1 5716 0.890 0.973 85.310 88.730 82.000 0.997
2 3744 0.805 0.957 69.044 73.007 62.241 0.989
3 4638 0.880 0.970 76.846 80.505 69.354 0.996
4 3085 0.867 0.980 62.673 73.007 50.000 0.989
5 1946 0.735 0.965 49.777 67.365 32.016 0.965
6 5460 0.895 0.977 83.378 84.906 78.772 0.998
7 5835 0.839 0.966 86.194 90.139 80.156 0.996
8 5245 0.906 0.981 81.720 84.172 77.389 0.998
9 4919 0.909 0.980 79.140 80.623 73.410 0.996
10 5184 0.910 0.979 81.243 85.726 77.885 0.998
11 5337 0.879 0.976 82.433 85.475 77.782 0.995
12 3255 0.707 0.918 64.377 73.062 49.518 0.966
13 1841 0.661 0.950 48.415 70.292 32.650 0.941
14 2352 0.725 0.954 54.723 70.093 38.210 0.949
15 3993 0.768 0.952 71.302 78.746 59.059 0.972
16 4573 0.893 0.974 76.305 78.600 71.000 0.997
17 3294 0.778 0.961 64.761 80.411 45.000 0.966
18 1840 0.771 0.936 48.402 57.219 38.000 0.978
19 4150 0.898 0.975 72.691 75.538 69.260 0.997
20 2732 0.793 0.965 58.979 70.859 45.000 0.968
21 4333 0.917 0.980 74.276 77.414 68.447 0.997
22 4700 0.913 0.980 77.358 80.957 69.635 0.996
23 4158 0.768 0.943 72.761 79.310 58.000 0.991











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
24 4411 0.871 0.970 74.942 76.844 68.184 0.995
25 3203 0.851 0.963 63.861 65.000 58.181 0.994
26 4977 0.882 0.972 79.605 81.498 74.847 0.997
27 5282 0.850 0.962 82.008 87.321 74.967 0.991
28 3782 0.921 0.979 69.393 70.605 66.483 0.998
29 3728 0.875 0.969 68.896 72.945 64.133 0.995
30 4735 0.887 0.975 77.645 91.220 75.584 0.997
31 5729 0.894 0.977 85.407 87.618 81.000 0.997
32 6002 0.900 0.976 87.418 91.831 84.054 0.998
33 4484 0.901 0.973 75.559 90.686 70.576 0.998
34 4756 0.872 0.973 77.817 80.808 70.264 0.994
35 3855 0.445 0.944 70.060 76.792 63.514 0.981
36 4626 0.862 0.970 76.746 78.549 70.064 0.996
37 2740 0.769 0.956 59.065 70.349 42.579 0.967
38 1922 0.852 0.965 49.469 60.000 36.346 0.992
39 5347 0.899 0.978 82.511 85.866 77.466 0.998
40 5120 0.895 0.982 80.740 82.735 76.322 0.995
41 3349 0.885 0.973 65.300 67.624 60.000 0.996
42 4928 0.879 0.975 79.212 81.609 74.686 0.998
43 2558 0.811 0.971 57.070 71.007 39.319 0.974
44 5606 0.905 0.978 84.485 87.321 81.786 0.998
45 4848 0.872 0.977 78.566 80.753 71.840 0.998
46 3423 0.882 0.973 66.017 67.268 59.540 0.994
47 3337 0.898 0.974 65.183 67.357 60.000 0.997
48 2196 0.787 0.945 52.878 55.227 40.112 0.980
49 1794 0.709 0.959 47.793 67.007 30.000 0.963
50 4120 0.876 0.968 72.428 75.432 67.365 0.994
51 4771 0.894 0.979 77.940 91.241 70.880 0.996
52 3899 0.876 0.973 70.458 72.835 63.820 0.996
53 4415 0.906 0.983 74.976 76.968 70.000 0.998
54 5034 0.915 0.982 80.059 81.345 73.756 0.998
55 3873 0.874 0.964 70.223 72.567 61.814 0.995
56 3471 0.863 0.968 66.479 68.884 62.434 0.993
57 5469 0.880 0.979 83.447 85.440 79.057 0.997
58 5213 0.865 0.970 81.470 84.865 73.348 0.995
59 4422 0.862 0.971 75.035 79.429 65.192 0.996
60 4650 0.870 0.975 76.945 79.322 71.421 0.996
61 4200 0.906 0.979 65.844 75.690 69.462 0.998
62 3972 0.864 0.971 71.115 73.926 65.000 0.995
63 5042 0.934 0.986 80.123 83.630 77.000 0.997
64 3642 0.838 0.959 68.097 70.725 64.070 0.992











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
65 3442 0.861 0.964 66.200 68.622 55.326 0.988
66 1464 0.757 0.934 43.174 58.464 28.636 0.963
67 3762 0.842 0.961 69.209 72.139 62.000 0.991
68 4757 0.904 0.978 77.825 80.957 73.348 0.997
69 4440 0.907 0.981 75.188 77.162 70.000 0.997
70 4156 0.893 0.974 72.743 78.109 65.069 0.996
71 3864 0.880 0.972 70.141 72.719 65.069 0.994
72 4225 0.865 0.969 73.345 76.538 68.593 0.996
73 3826 0.817 0.962 69.796 72.897 66.648 0.992
74 3584 0.782 0.948 67.552 73.546 58.549 0.985
75 4636 0.883 0.977 76.829 80.777 68.593 0.997
76 4626 0.909 0.977 76.746 79.322 70.937 0.997
77 5125 0.889 0.975 80.780 82.970 76.118 0.997
78 5102 0.872 0.972 80.598 82.347 72.691 0.997
79 4869 0.892 0.979 78.736 80.623 74.000 0.998
80 4159 0.891 0.974 72.770 74.953 68.949 0.997
81 4362 0.863 0.973 74.524 80.411 66.000 0.982
82 2952 0.803 0.967 61.307 72.339 47.518 0.965
TOTAL of
TRIAL #2 338612 5881.860 Number of fruit: 82
1 5350 0.852 0.969 82.534 86.954 77.782 0.995
2 2585 0.750 0.923 57.370 65.734 51.088 0.967
3 2900 0.721 0.956 60.765 81.056 41.110 0.964
4 4429 0.886 0.976 75.094 77.878 68.447 0.996
5 4914 0.870 0.964 79.099 81.006 72.897 0.996
6 3537 0.809 0.955 67.108 73.007 60.000 0.992
7 4315 0.873 0.967 74.122 77.524 72.000 0.995
8 5530 0.888 0.975 83.911 84.900 78.447 0.998
9 4627 0.845 0.964 76.755 81.302 70.214 0.994
10 5227 0.861 0.969 81.580 99.725 77.201 0.998
11 4908 0.859 0.971 79.051 83.487 71.847 0.996
12 4595 0.871 0.969 76.489 79.177 72.125 0.996
13 3690 0.864 0.963 68.544 81.345 65.069 0.994
14 4379 0.831 0.962 74.669 77.782 67.469 0.995
15 4568 0.878 0.972 76.264 77.466 72.615 0.996
16 5306 0.886 0.974 82.194 84.504 74.465 0.997
17 6377 0.875 0.973 90.108 94.303 85.000 0.998
18 3047 0.787 0.936 62.286 66.129 53.310 0.984
19 3579 0.838 0.960 67.505 70.725 61.555 0.993
20 4080 0.861 0.971 72.075 76.276 67.067 0.996
21 4751 0.889 0.978 77.776 81.025 73.335 0.997
22 2443 0.770 0.919 55.772 65.788 45.891 0.959











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
23 4030 0.839 0.962 71.632 77.897 63.000 0.986
24 3298 0.702 0.913 64.801 74.733 57.428 0.956
25 3036 0.814 0.962 62.174 70.774 48.662 0.977
26 2775 0.766 0.938 59.441 64.008 50.220 0.982
27 2985 0.826 0.957 61.649 71.197 49.000 0.982
28 4800 0.398 0.806 78.176 127.319 15.000 0.871
29 4261 0.911 0.983 73.656 75.432 69.296 0.997
30 3469 0.889 0.974 66.460 68.000 60.415 0.994
31 4367 0.854 0.959 74.567 78.854 68.731 0.995
32 5049 0.842 0.964 80.178 85.802 71.421 0.996
33 4322 0.813 0.950 74.182 77.833 68.066 0.992
34 5128 0.888 0.978 80.803 84.149 77.000 0.998
35 4616 0.904 0.980 76.663 79.404 70.000 0.997
36 4084 0.902 0.977 72.110 74.813 67.186 0.996
37 4727 0.906 0.980 77.580 82.006 70.093 0.999
38 3338 0.870 0.962 65.193 68.000 59.414 0.995
39 4467 0.884 0.971 75.416 77.782 71.701 0.997
40 2733 0.897 0.969 58.990 62.650 51.971 0.995
41 1 0.000 1.000 1.128 1.000 1.000 1.000
42 6 0.000 1.000 2.764 3.162 0.000 1.000
43 1 0.000 1.000 1.128 1.000 1.000 1.000
44 4729 0.714 0.921 77.596 90.272 65.215 0.968
45 4824 0.890 0.974 78.372 80.430 74.061 0.997
46 2902 0.834 0.957 60.786 65.795 54.083 0.990
47 2164 0.648 0.889 52.491 71.000 38.053 0.933
48 4517 0.848 0.955 75.837 80.281 69.311 0.995
49 282 0.924 0.962 18.949 21.095 16.000 0.958
50 4588 0.874 0.971 76.431 82.006 70.000 0.997
51 3586 0.867 0.968 67.571 69.462 62.642 0.993
52 3787 0.839 0.956 69.439 74.000 65.069 0.993
53 3838 0.762 0.939 69.905 78.390 56.000 0.963
54 194 0.820 0.878 15.717 16.401 10.050 0.941
55 228 0.897 0.926 17.038 19.026 13.000 0.951
56 4172 0.876 0.971 72.883 76.322 68.593 0.995
57 3261 0.914 0.981 64.436 67.535 58.694 0.998
58 4913 0.875 0.966 79.091 81.123 73.000 0.996
59 3548 0.743 0.941 67.212 74.330 59.464 0.987
60 3736 0.814 0.953 68.970 72.567 64.885 0.993
61 3405 0.786 0.945 65.844 73.824 56.223 0.985
62 4473 0.888 0.973 75.467 76.851 70.178 0.998
63 4218 0.874 0.966 73.284 75.664 68.680 0.996











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
64 3295 0.827 0.958 64.771 69.635 56.303 0.993
65 4290 0.752 0.941 73.907 81.541 70.328 0.983
66 5496 0.876 0.974 83.652 85.006 78.492 0.998
67 4957 0.875 0.974 79.445 82.620 71.386 0.996
68 3969 0.858 0.962 71.088 74.330 66.888 0.995
69 4130 0.915 0.981 72.515 75.073 67.469 0.998
70 3694 0.806 0.944 68.581 75.180 58.181 0.983
71 3739 0.870 0.965 68.997 72.560 64.661 0.996
72 3559 0.829 0.959 67.316 71.176 61.351 0.994
73 4346 0.823 0.952 74.387 81.566 60.959 0.984
74 5762 0.893 0.977 85.653 87.664 77.414 0.998
75 3866 0.782 0.944 70.159 74.967 65.856 0.992
76 4634 0.910 0.980 76.813 80.056 72.125 0.998
77 5324 0.874 0.977 82.333 84.012 77.337 0.998
78 5684 0.874 0.973 85.071 89.196 80.056 0.997
79 3603 0.791 0.946 67.731 73.824 61.620 0.992
80 4271 0.863 0.963 73.743 78.892 65.795 0.995
81 3657 0.893 0.970 68.237 72.090 61.612 0.993
82 3762 0.898 0.977 69.209 72.945 63.000 0.992
83 1785 0.727 0.950 47.673 66.189 30.000 0.979
84 4225 0.875 0.969 73.345 76.059 67.476 0.996
85 3709 0.841 0.962 68.720 72.471 65.192 0.995
86 3070 0.845 0.957 62.521 66.068 55.732 0.990
87 3711 0.884 0.973 68.739 70.937 63.820 0.997
88 4249 0.886 0.972 73.553 79.076 67.417 0.996
89 3635 0.902 0.978 68.031 68.884 65.000 0.996
90 4656 0.908 0.977 76.995 78.518 71.589 0.997
91 4659 0.901 0.977 77.020 81.939 70.725 0.997
92 4715 0.915 0.981 77.481 80.056 71.021 0.996
93 5676 0.855 0.964 85.011 89.320 76.577 0.996
94 3556 0.799 0.948 67.288 69.065 63.411 0.991
95 3560 0.792 0.948 67.326 71.589 59.034 0.987
96 4017 0.901 0.978 71.516 72.835 68.154 0.997
97 4072 0.860 0.970 72.004 75.186 64.382 0.991
98 5217 0.900 0.977 81.501 85.288 74.277 0.997
99 3914 0.878 0.967 70.594 72.139 65.765 0.995
100 3587 0.874 0.969 67.580 71.197 61.000 0.994
101 4602 0.839 0.966 76.547 81.688 69.000 0.995
102 4350 0.829 0.957 74.422 76.322 68.096 0.994
103 4707 0.830 0.965 77.415 84.900 66.219 0.990
104 4073 0.907 0.980 72.013 74.000 68.593 0.998











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
105 2951 0.799 0.941 61.297 65.000 55.154 0.981
106 4626 0.881 0.972 76.746 79.630 69.376 0.996
107 3294 0.755 0.920 64.761 71.840 50.636 0.974
108 3403 0.895 0.972 65.824 67.268 61.057 0.995
109 4614 0.891 0.977 76.647 80.306 70.228 0.997
110 3851 0.784 0.947 70.023 75.822 61.522 0.990
111 2584 0.794 0.966 57.359 66.189 44.000 0.972
112 1223 0.779 0.945 39.461 50.160 27.659 0.982
113 4660 0.874 0.974 77.028 90.554 70.612 0.996
114 4746 0.866 0.970 77.735 79.120 71.505 0.997
115 3986 0.856 0.960 71.240 73.164 66.468 0.995
116 3257 0.917 0.986 64.397 66.287 58.000 0.998
117 3695 0.893 0.977 68.590 82.037 63.789 0.997
118 1914 0.647 0.901 49.366 52.240 41.785 0.964
119 1840 0.741 0.953 48.402 63.008 34.132 0.961
120 3696 0.853 0.974 68.600 88.238 58.941 0.984
121 3565 0.882 0.969 67.373 80.610 63.253 0.994
122 4478 0.889 0.973 75.509 89.828 71.386 0.997
123 4883 0.898 0.977 78.849 82.377 71.063 0.997
124 3924 0.895 0.977 70.684 72.719 63.506 0.995
125 5071 0.870 0.971 80.353 83.433 72.069 0.996
126 5384 0.812 0.958 82.796 88.363 75.133 0.987
127 4244 0.862 0.961 73.509 75.584 65.000 0.992
128 4024 0.842 0.961 71.579 74.653 65.307 0.995
129 3593 0.768 0.946 67.637 70.605 59.540 0.989
130 621 0.695 0.904 28.119 38.000 20.000 0.897
131 1932 0.639 0.886 49.597 62.936 36.069 0.921
132 4229 0.859 0.964 73.379 75.690 65.803 0.995
133 3452 0.859 0.960 66.296 69.181 62.000 0.993
134 4031 0.814 0.958 71.641 76.118 66.400 0.994
135 3574 0.889 0.971 67.458 70.000 61.855 0.997
136 3701 0.887 0.972 68.646 70.831 63.600 0.997
137 2680 0.856 0.975 58.415 65.620 45.354 0.974
138 1824 0.638 0.887 48.191 60.216 33.956 0.931
139 1021 0.754 0.911 36.055 46.000 24.000 0.945
140 1271 0.736 0.925 40.228 51.245 30.414 0.947
141 1888 0.719 0.939 49.029 65.620 33.000 0.979
142 4495 0.883 0.972 75.652 81.302 70.725 0.996
143 3656 0.864 0.968 68.227 70.725 62.008 0.995
144 4037 0.885 0.974 71.694 73.246 65.437 0.997
145 3083 0.898 0.972 62.653 65.765 59.008 0.995











Fruit Max. Min.
Area
Number Cruaiy Cneiy daee imtrDaee litct
(pixeP) issiy Cncil indr Daee Daee litct
(pixels) (pixels) (pixels)
146 3094 0.816 0.951 62.765 65.795 55.000 0.991
147 4546 0.799 0.950 76.080 81.320 69.836 0.992
148 4261 0.810 0.956 73.656 78.518 66.491 0.990
149 3577 0.920 0.979 67.486 70.831 63.008 0.998
150 4728 0.825 0.957 77.588 83.006 69.260 0.995
151 3335 0.823 0.956 65.163 68.593 58.310 0.991
152 3267 0.914 0.981 64.496 68.942 59.548 0.997
153 3639 0.909 0.975 68.068 71.281 65.742 0.995
154 4322 0.893 0.975 74.182 74.953 70.292 0.997
155 4139 0.854 0.965 72.594 75.133 68.154 0.995
TOTAL of TRIAL
#3 583687 10471.110 Number of fruit: 155
OVERALL
TOTAL 16837761 291627.063 Total Num. of fruit: 5384










LIST OF REFERENCES

Aleixos, N., J. Blasco, E. Molto, and F. Navarron. 2000. Assessment of citrus fruit quality using
a real-time machine vision system. In Proc. 15th International Conf: on Pattern
Recognition. 15(1): 482 -485.

Annamalai, P., W. S. Lee, and T. Burks. 2004. Color vision system for estimating citrus yield in
real-time. ASAE Paper No. 043054. St. Joseph. Mich.: ASAE.

Balastreire, L. A., J. K. Schueller, J. R. Amaral, J. C. G. Leal, and F. H. R. Baio. 2002. Coffee
yield mapping. ASAE Paper No. 021166. St. Joseph. Mich.: ASAE.

Beucher, S., and C. Lantuejoul. 1979. Use of Watersheds in contour detection. In Proc. Inter.
Workshop on Image Processing. Rennes, France: CCETT/IRISA, Rennes, France, 1979:
12-21.

Beucher, S., and F. Meyer. 1992. The morphological approach to segmentation: the watershed
transformation. In Dougherty, E. (ed.), Mathematical morphology in image processing,
Chap. 12, pp. 433-481. Marcel Dekker, Inc., New York.

Blasco, J., N. Aleixos, and E. Molto. 2003. Machine vision system for automatic quality grading
of fruit. Biosystems Engineering 85(4), 415-423 doi:10. 1016/S1537-51 10(03)00088-6.

Bora, G. C., M. R. Ehsani, R. Goodrich, and G. Michaels. 2006. Field evaluation of a citrus fruit
pick-up machine. ASAE Paper No. 061141. St. Joseph. Mich.: ASAE.

Borgelt, S. C., and K. A. Sudduth. 1992. Grain flow monitoring for in-field yield mapping.
ASABE Paper No. 92-1022. St. Joseph. Mich.: ASAE.

Breunig, M., H. P. Kriegel, R. Ng, and J. Sander. 2000. LOF: Identifying density-based local
outliers. Proc. SIGM~OD.

Brown, G. K. 2002. Mechanical harvesting systems for the Florida citrus juice industry. ASABE
Paper No. 021108. St. Joseph, Mich.: ASABE.

Buie, Jr., A. P. 1965. Fruit picking and transporting device. U.S. Patent No. 3165880.

Buker, R. S., J. P. Syvertsen, J. K. Burns, F. M. Roka, W. M. Miller, M. Salyani, and G. K.
Brown. 2004. Mechanical harvesting and tree health. EDIS. UF/IFAS. FL.

Campbell, R. H., S. L. Rawlins, and S. Han. 1994. Monitoring methods for potato yield mapping.
ASAE Paper No. 94-1584. St. Joseph, Mich.: ASAE.

Casasent, D., A. Talukder, P. Keagy, and T. Schatzki. 2001. Detection and segmentation of items
in X-ray imagery. Trans. ASAE 44(2): 337-345.

Chinchuluun, R., and W. S. Lee. 2006. Citrus yield mapping system in natural outdoor scenes
using the Watershed transform. ASAE Paper No. 063010. St. Joseph, Mich.: ASAE.










Digabel, H., and C. Lantuejoul. 1978. Iterative algorithms. Proc. 2nd European Symp.
Quantitative Analysis of2~icrostructures in Material Science, Biology and Medicine. Caen,
France, Oct. 1977. J. L. Chermant, Ed. Stuttgart, West Germany: Riederer Verlag, 1978:
85-99.

Duda., R. O., P. E. Hart, and D. G. Stork. 2000. Pattern Classification. 2nd ed. New York, N. Y.:
John Wiley and Sons.

Florida Agricultural Statistics Service (FASS). 2005. Washington, D.C.: USDA. Available at:
www.nass.usda. gov/fl. Accessed 29 May 2006.

Flood, S. J., 2006. Design of a robotic citrus harvesting end effector and force control model
using physical properties and harvesting motion tests. PhD diss. Ganesville, FL: University
of Florida, Agricultural and Biological Engineering department.

Forsyth, D. A., and J. Ponce. 2001. Computer Vision: A M~odern Approach. Prentice Hall, 2001.
Upper Saddle River, NJ.

Futch, S. H., and F. M. Roka. 2005. Continuous canopy shake mechanical harvesting systems.
EDIS. UF/IFAS. FL.

Futch, S. H., J. D. Whitney, J. K. Burns, and F. M. Roka. 2005. Harvesting: From manual to
mechanical. EDIS. UF/IFAS. FL.

Grift, T., R. Ehsani, K. Nishiwaki, C. Crespi, and M. Min. 2006. Development of a yield monitor
for citrus fruits. ASABE Paper No. 061192. St. Joseph, Mich.: ASABE.

Hannan, M. W., and T. F. Burks. 2004. Citrus developments in automated citrus harvesting.
ASABE Paper No. 043087. St. Joseph, Mich.: ASABE.

Hodges, A., E. Philippakos, D. Mulkey, T. Spreen, and R. Muraro. 2001. Economic impact of
Florida' s citrus industry. University of Florida, Gainesville, IFAS publications. Economic
information report 01-2.

Juste, F., I. Fornes, F. Pla, and F. Sevila. 1992. An approach to robotics harvesting of citrus in
Spain. In Proc. Intl. Soc. Citriculture : 7th Intl. Citrus Congress, 3:1014-1018. Riverside,
Cal.: International Society of Citriculture.

Kane, K. E., and W. S. Lee. 2006. Spectral sensing of different citrus varieties for precision
agriculture. ASABE Paper No. 061065. St. Joseph, Mich.: ASABE.

Gender, W. J., U. Hartmond, M. Salyani, J. K. Burns, and J. D. Whiney. 1999. Injurious effects
of metsulfuron-methyl when used to stimulate abscission of Hamlin oranges. HortScience
34(5): 904-907.

Zoning, C. T. J., L. Speelman, and H. C. P. Vries. 1994. Size grading of potatoes: Development
of a new characteristic parameter. J. Agric. Eng. Res. 57, 1 19-128.










Lee, W. S., J. K. Schueller, and T. F. Burks. 2005. Wagon-based silage yield mapping system. Dr
Agricultural Engineering hIternational: the CIGR E. Journal. Vol. VII. Manuscript IT 05
003.

Lee, W. S., and D. C. Slaughter. 2004. Recognition of partially occluded plant leaves using a
modified Watershed algorithm. Trans. ASAE 47(4): 1269-1280.

Leemans, V., and M. F. Destain. 2004. A real time grading method of apples based on features
extracted from defects. Dr Journal ofFoodEngineering 61, 83-89.

Leemans, V., H. Magein, and M. F. Destain. 2002. On-line fruit grading according to their
external quality using machine vision. Biosystems Engineering 83(4), 397-404
doi:10. 1006/bioe.2002. 0131.

MacArthur, D., J. K. Schueller, and W. S. Lee. 2006. Remotely-piloted helicopter citrus yield
map estimation. ASABE Paper No. 063096. St. Joseph, Mich.: ASABE.

Macidull, J. C. 1973. Fruit harvesting apparatus. U. S. Patent No. 3756001.

Marchant, J.A., and C. M. Onyango. 2003. Comparison of a Bayesian classifier with a multilayer
feed-forward neural network using the example of plant/weed/soil discrimination. In
Computers and Electronics in Agriculture 39, 3-22.

Mehta, S. S. 2007. Vision-based control for autonomous robotic citrus harvesting. MS thesis.
Ganesville, FL: University of Florida, Agricultural and Biological Engineering department.

Muscato, G., M. Prestifilippo, N. Abbate, and Rizzuto. 2005. A prototype of an orange picking
robot: past history, the new robot and experimental results. hId. Robot. 32(2): 128-138.

Perry, C. D., G. Vellidis N. Wells, R. Hill, A. Knowlton, E. Hart, and D. Dales. 2005.
Instantaneous accuracy of cotton yield monitors, Instantaneous accuracy of cotton yield
monitors. In Proc. Beltwide Cotton Conf 486-504, New Orleans, LA, National Cotton
Council, Memphis, TN.

Regunathan, M., and W. S. Lee. 2005. Citrus fruit identification and size determination using
machine vision and ultrasonic sensors. ASAE Paper No. 053017. St. Joseph. Mich.: ASAE.

Sanders, K. F. 2005. Orange harvesting systems review. Biosystems Eng. 90(2): 115-125.

Schueller, J. K., and Y. H. Bae. 1987. Spatially attributed automated automatic combine data
acquisition. In Computers and Electronics in Agriculture 2: 1 19-127.

Schueller, J. K., J. D. Whitney, T. A. Wheaton, W. M. Miller, and A. E. Tunner. 1999. Low-cost
automatic yield mapping in hand-harvested citrus. In Computers and Electronics in
Agriculture 23(2): 145-153.

Searcy, S. W., J. D. Schueller, Y. H. Bae, S. C. Borgelt, and B. A. Stout. 1989. Mapping of
spatially variable yield during grain combining. Trans. ASAE 32(2): 826-829.










Shahin, M. A., and S. J. Symons. 2001. A machine vision system for grading lentils. Cana~dian
Biosystems Engineering, 43: 7.7-7. 14.

Sui, R. X., and J. A. Thomasson. 2001. Field testing of Mississippi State University cotton yield
monitor. In Proc. Beltwide Cotton Conf D. A. Richter ed., 332-342. Memphis, TN: Nat.
Cotton Council Am.

Vellidis, G., C. D. Perry, J. S. Durrence, D. L. Thomas, R. W. Hill, C. K. Kvien, T. K. Hamrita,
and G. C. Rains. 2001. The peanut yield monitoring system. Trans. ASAE 44(4):775-785.

Vincent, L. 1991. Recent developments in morphological algorithms. Proc. 8t hzter. Congress
for Stereology, Irvine CA, August 1991.

Welte, H. F. 1990. Forecasting harvest fruit size during the growing season. Acta Hort (ISHS)
276:275-282.

Whitney, J. D., Q. Ling, W. M. Miller, and T. A. Wheaton. 2001. A DGPS yield monitoring
systems for Florida citrus. Applied Engineering in Agriculture 17(2): 115-119.

Yoshida, J., S. Okuyama, and H. Suzuki. 1988. Fruit harvesting apparatus with television camera
and monitor. U. S. Patent No. 4519193.









BIOGRAPHICAL SKETCH

Radnaabazar Chinchuluun was born in Bulgan, Mongolia in 1978. He received his

bachelor degree in computer hardware and software engineering from Mongolian University of

Science and Technology in 2000. After his undergraduate degree, he started working at the same

university as a lecturer and computer engineer.

Radnaabazar j oined the Master of Engineering degree program in Agricultural and

Biological Engineering, and Computer and Information Science and Engineering at the

University of Florida on June, 2005. While pursuing his master' s degree, he has been working as

a graduate research assistant at the Precision Agricultural Laboratory under supervision of Dr.

Won Suk Lee. During thi s time, he has undertaken two maj or citrus yield mapping proj ects.





PAGE 1

MACHINE VISION BASED CITRUS YIELD MAPPING SYSTEM ON A CONTINUOUS CANOPY SHAKE AND CATCH HARVESTER By RADNAABAZAR CHINCHULUUN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF T HE REQUIREMENTS FOR THE DEGREE OF MASTER OF ENGINEERING UNIVERSITY OF FLORIDA 2007

PAGE 2

2007 Radnaabazar Chinchuluun

PAGE 3

To my God who always blesses me and t o my father, Chinchuluun Ochirhuyag, who always takes care of me from heaven

PAGE 4

4 ACKNOWLEDGMENTS I want to express my deepest gratitude to Dr. Won Suk Lee for his enormous support and guidance. He is the best advisor I have ever had in my life. He always provided me greatest guidance in my life and study with immense kindness and patie nce. He opened my eyes by introducing me to the precision agriculture and machine vision. I also thank my committee members Dr. Thomas F. Burks and Dr. James J. Ferguson for their kindness and guidance, and thank Dr. Reza Ehsani for his support and allowin g me to collaborate in his research. All that I have learnt during the course of my thesis would not have been possible without their dedications. I especially thank Dr. Kyeong Hwan Lee and Dr. Ganesh Bora for their support in my experiment, and Mr. Michae l Zingaro, Mr. Harmon Pearson for sharing me their hands throughout my research. Mr. Greg L. Pugh shared his programming experience with me. I would like to thank my friends Mr. Sanghoon Han, Vijay Subramanian and Kevin Kane for their encouragements and su pport. Most importantly I would like to express my dearest appreciation to my family for their dedication, support and love.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ........................... 7 LIST OF FIGURES ................................ ................................ ................................ ......................... 8 ABSTRACT ................................ ................................ ................................ ................................ ... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 1 2 Background ................................ ................................ ................................ ............................. 12 Precision Agricultu re for Citrus ................................ ................................ .............................. 13 Citrus Harvesting ................................ ................................ ................................ .................... 13 Motivation for Citrus Yield Mapping System ................................ ................................ ........ 14 Objectives ................................ ................................ ................................ ............................... 15 2 LITERATURE REVIEW ................................ ................................ ................................ ....... 16 Citrus Harvesting ................................ ................................ ................................ .................... 16 Robotic Harvesting ................................ ................................ ................................ .......... 16 Mechanical Harvesting ................................ ................................ ................................ .... 18 Yield Mapping ................................ ................................ ................................ ........................ 19 Machine Vision and Fruit Size Estimating System ................................ ................................ 22 3 IMAGE ANALYSIS ................................ ................................ ................................ .............. 25 Color Models ................................ ................................ ................................ .......................... 25 The RGB Color Model ................................ ................................ ................................ .... 25 The HSI Color Model ................................ ................................ ................................ ...... 25 The YIQ Color Model ................................ ................................ ................................ ..... 27 Image Classification ................................ ................................ ................................ ............... 28 Color Classification ................................ ................................ ................................ ......... 29 Bayesian classifier ................................ ................................ ................................ .... 29 Shape Classification ................................ ................................ ................................ ........ 31 4 MATERIALS AND METHODS ................................ ................................ ........................... 34 Hardware Design of the System ................................ ................................ ............................. 34 System Operations ................................ ................................ ................................ ........... 35 ................................ ................................ ............. 36 Camera Parameters ................................ ................................ ................................ .......... 36 Image Synchronization with Encoder ................................ ................................ ............. 38

PAGE 6

6 System Software Des ign ................................ ................................ ................................ ......... 39 System Experiments ................................ ................................ ................................ ............... 41 Test bench Experiment ................................ ................................ ................................ .... 41 Image acquisition during the te st bench experiment ................................ ................ 42 Citrus Canopy Shake and Catch Harvester Experiment ................................ .................. 42 Image Processing Analysis ................................ ................................ ................................ ..... 43 Fruit Parameters ................................ ................................ ................................ ............... 47 Yield Prediction Model ................................ ................................ ................................ ........... 47 Camera Calibration ................................ ................................ ................................ ................. 50 5 RESULT AND DISCUSSION ................................ ................................ ............................... 52 Color Segmentation Result on Test bench ................................ ................................ ............. 52 Color Segmentation Result on the Harvester ................................ ................................ .......... 56 System Exec ution Time ................................ ................................ ................................ .......... 58 System Limitation ................................ ................................ ................................ ................... 59 6 CONCLUSIONS ................................ ................................ ................................ .................... 60 APPENDIX A PROGRAMMING CODE ................................ ................................ ................................ ...... 62 Encoder Thread Code ................................ ................................ ................................ ............. 62 ImageWarp Thread Code ................................ ................................ ................................ ........ 63 Image Processing ImageWarp Script in Real time Mode ................................ ...................... 64 Image Processing ImageWarp Script in Post mode ................................ ............................... 65 B MEASURED FRUIT PARAMETERS ................................ ................................ .................. 66 LIST OF REFERENCES ................................ ................................ ................................ ............... 74 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ......... 78

PAGE 7

7 LIST OF TABLES Table page 4 1. Conveyor belt specification ................................ ................................ ................................ .... 35 4 2. Camera resolution ................................ ................................ ................................ ................... 38 4 3. Means of hue, saturation and I (YIQ) components of fruit and background classes ............. 45 5 1. Summary of the test bench experiment results ................................ ................................ ....... 54 5 2. System execution time ................................ ................................ ................................ ............ 59 B 1. Fruit parameters ................................ ................................ ................................ ..................... 66

PAGE 8

8 LIST OF FIGURES Figure page 3 1. RGB color cube. ................................ ................................ ................................ ..................... 26 3 2. HSI color model representation. ................................ ................................ ............................. 27 3 3. Image classification design cycle. ................................ ................................ .......................... 28 3 4. Watershed representation. A) Two touching oranges. B) Binary image of the image A. C) Distance image. D ) Topographic representation of the image (Source: Vincent (1991)). E) Separated oranges. ................................ ................................ ........................... 33 4 1. General overview of the system. A) Camera and four halogen lamps. B) System housing. ................................ ................................ ................................ .............................. 35 4 2. Pictorial illustration of system operation. ................................ ................................ ............... 37 4 3. System connection diagram. ................................ ................................ ................................ ... 37 4 4. Image acquisition of moving object at the different shutter speeds. A) Auto shutter speed. B) 1/2000 s. C) 1/2150 s. D) 1/2230 s. E) 1/2300 s. ................................ ............... 38 4 5. The system software overview. ................................ ................................ .............................. 40 4 6. Encoder thread algorithm. ................................ ................................ ................................ ...... 41 4 7. Test bench experiment. A) Feeding conveyor belt. B) Weight measurement of t ested fruit. ................................ ................................ ................................ ................................ .... 42 4 8. Canopy shake and catch harvester experiment. ................................ ................................ ...... 43 4 9. Color image of fruit and background. ................................ ................................ .................... 44 4 10. Images in different color models and their histograms. A) Hue component of HSI color model. B) Hue histogram of fruit and background. C) Saturation component of HSI color model. D) Saturation histogra m of fruit and background. E) I component of YIQ color model. E) I (YIQ) histogram of fruit and background. ................................ 44 4 11. Image processing algorithm. ................................ ................................ ................................ 46 4 12. Discriminant function between fruit and background classes. ................................ ............. 47 4 13. Relationship between actual fruit weight and diameter. ................................ ....................... 48 4 14. Relationship between actual fruit weight and area. ................................ .............................. 49 4.15. Relationship between actual fruit weight and volume. ................................ ......................... 50

PAGE 9

9 4 16. Camera perspective projection. ................................ ................................ ............................ 51 5 1. Image segmentation result. A) Original image. B) Bayesian classifier result. C) Morphological operation result. D) Watershed tra nsform result. ................................ ...... 53 5 2. Actual fruit weight versus the number of fruit by image processing algorithm: regression analysis. ................................ ................................ ................................ ............ 54 5 3 Actual fruit weight versus fruit area: regression analysis. ................................ ...................... 55 5 4. Fruit diameter versus actual fruit weight: regression analysis. ................................ .............. 55 5 5. Color segmentation result for field experiment. A) Original image. B) Bayesian classifier result. C) Morphological operation result. D) Watershed transform result. ....... 56 5 6. Circul ar filtering result for field experiment 2. ................................ ................................ ...... 57 5 7. Regression analysis between human counting and algorithm counting. ................................ 58

PAGE 10

10 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 Engineering MACHINE VISION BASED CITRUS YIELD MAPPING SYSTEM ON A CONTINUOUS CANOPY SHAK E AND CATCH HARVESTER By Radnaabazar Chinchuluun August 2007 Chair: Won Suk Lee Major: Agricultural and Biological Engineering Yield mapping is a well known beneficial tool in precision agriculture. Traditionally citrus growers ignored in field variabilit y of soil properties, environmental conditions and citrus yield because they dealt with comparatively small scale groves. They treated whole grove as a uniform single unit. Citrus growers always seek new production practices to increase profit by maximizin g citrus yield. Because of lack of information of grove and expertise, growers sometimes cannot make optimal decisions. Hence, as a result of inappropriate management, ve growers are able to adopt new technologies such as Global Positioning System (GPS), yield monitoring, remote sensing (RS), variable rate technology (VRT) and sensing systems. With advance of these technologies, farmers can manage large scale grove at sm aller scale or individual tree basis. Consequently growers are able to increase yield, fruit quality and economic return. In this study, m achine vision based citrus yield mapping and fruit size detection system wa s developed, which could be used on a cont inuous canopy shake and catch harvester. The system consisted of a 3CCD camera, four halogen lamps, a DGPS receiver, an encoder and a laptop computer. A total of 3,653 images were taken d uring an experiment on the test bench at

PAGE 11

11 the Citrus Research and Educ ation Center, Lake Alfred, Florida and 709 images were used for analysis Th e system was also tested on a canopy shake and catch harvester at a Lykes grove located in Fort Basinger FL. A total of 773 images were acquired as well. Fruit weight was measured in 14 test trials of image acquisitions during the test bench experiment as well as in two test trials of image acquisitions during a field trial with a commercial canopy shake and catch harvester. A supervised image processing algorithms that could ident ify and inspect fruit qualities were developed. The number of fruit and total fruit areas were measured from the sets of color images using the develope d algorithm Finally number of citrus fruit that was found by the image processing algorithm during the test bench experiment was compared against actual fruit weight. The coefficient of determination, R 2 was 0.962 between them. For a validation purpose of the canopy shake and catch harvester experiment, number of fruit was counted manually from a total of 60 images. Human counting was compared with algorithm counting. The coefficient of determ ination, R 2 was 0.891 between the actual and estimated number of fruit

PAGE 12

12 CHAPTER 1 INTRODUCTION The thesis addresses a machine vision based citrus yield mapping system coupled with a continuous canopy shake and catch harvesting system. Background One of the major agricultural products in Florida is citrus. Citrus production co ntributes enormous value to not only Florida state economy, but also the U.S. economy. Florida citrus industry accounted approximately 67 percent of the total U.S. citrus production. According to the USDA statistics that was published in 2005, Florida deal t with a total of 679 thousand bearing acreages and produced 13,045 thousand tons of citrus in 2003 2004 season (Florida Agricultural Statistics Service [FASS], 2005). A total of $891,500,000 value was added in the season. In Florida most citrus fruits go to processing industry rather than fresh fruit industry. In addition to juice processing, citrus processing industry produces important by products such as citrus pulp and meal, molasses, and D limonene (Hodges et al., 2001). These by products are used to make a variety of products starting from cleaners, disinfectants to livestock feeds. In 2003 2004 crop season, Florida citrus processing industry produced more than 1.1 million tons of citrus pulp and meal, 38,000 tons of molasses, and 16363 6 tons of D li monene. The total value of these by products was $136 million. Managing such huge production in 2003 2004 season, Florida citrus industry created 76,336 jobs, with 61,307 jobs in the processing sectors and 15,029 jobs in fresh fruit. Moreover, citrus indus try impacted to major industrial sectors such as construction, finance and insurance, health and social sciences, retail trade, wholesale trade, professional scientific and technical services, real estate and rentals (Hodges et al., 2001). Based on these f acts, it is apparent that citrus industry becomes a vital part of Florida economy.

PAGE 13

13 Precision Agriculture for Citrus As a result of mechanizing agricultural grove management, most growers started to deal with large scale areas in which in grove variability of nutrients, tree size, soil types, etc. highly likely existed. Traditionally, growers dealt with comparatively small citrus groves. Thus small groves usually have less in field variability. At the same time, they were easily manageable by citrus growers As grove size increases, citrus growers could not deal with it without technologies. However, precision farming, also known as site specific crop management, enables treating small areas of a large scale field as separate management units using new tech nologies. As becoming tools of the precision farming, the Global Positioning System (GPS), Geographic Information Systems (GIS), Remote Sensing (RS), Variable Rate Application (VRA), yield mapping and advanced sensors and information technologies benefit f armers by managing agricultural operations with less human involvements, reducing overall costs, and eventually improving overall effectiveness and productivity by considering in field variability. Moreover, these technologies would help growers collect la rge amount of production information over past seasons of their groves. One of the major goals of all citrus growers is to increase yield, fruit quality and profit by applying advanced management practices to the groves. A yield mapping allows growers to know where a site specific management would be beneficial. Eventually the growers are able to make appropriate management practices on production inputs if necessary. Citrus Harvesting Most citrus groves in Florida are harvested manually. Citrus harvestin g is a labor intensive process involving large number of workers depending upon the size of grove. At the same time, cost of hand harvesting operations has steadily been increasing. Harvesting productivity is

PAGE 14

14 declined depending on grove condition and heigh t of tree since workers use a ladder and picking bags. Recently mechanical harvesting use is increasing. During 2002 2003 season, about 1.7 sq. km of the 58.6 sq. km of oranges were mechanically harvested. Mechanical harvesting systems can increase labor productivity by 5 to 15 times and reduce unit harvesting cost by 50% or more (Brown, 2002). Generally harvesting of citrus fruit takes 35% to 45% of total production cost (Sanders, 2005). It is therefore clear that mechanical harvesting is a major solutio n for growers Motivation for Citrus Yield Mapping System A site specific crop management would give opportunity to growers optimizing crop production based on in field variab ility. By identifying different factors for in grove variability, citrus growers would make cost effective and environmentally friendly decisions on crop production by changing inputs such as soil property, limestone, herbicide, insecticide, etc. At the sa me time, mechanical harvesting would improve citrus production extensively. Significant efforts have been done to improve productivity and effectiveness of mechanical harvesting since mid 1960s (Futch et al., 2005). A numerous research on fruit external q uality inspection systems based on machine vision has been conducted (Leemans et al., 2002; Blasco et al., 2003). Machine vision systems are usually implemented at processing plants, not at grove. They basically examine fruit for color, size, blemishes and diseases, and sort them by given criteria. However, no research has been done to measure citrus size at the grove and simultaneously record citrus yield in real time. Thus, the purpose of research in this thesis was to work towards the development of a c itrus yield mapping and fruit size estimating system on a continuous canopy shake and catch

PAGE 15

15 harvester. This machine vision system allows growers to know not only yield variability, but also fruit size variability. Objectives Because of motivations descr ibed previously, the objectives of this study were to build a machine vision based citrus yield mapping and fruit size measuring real time system coupled with a continuous canopy shake and catch harvester. More specific objectives were: To build hardware components of a real time citrus yield mapping system that is able to count number of fruit, and measure size of fruit for individual tree or unit area on a citrus canopy shake and catch harvester, To develop an image processing algorithm to recognize ind ividual fruit and measure its size, and To test the complete system in a commercial citrus grove.

PAGE 16

16 CHAPTER 2 LITERATURE REVIEW Citrus Harvesting Manual citrus harvesting has been used for many years and still most of groves have been harvested manually in Florida. A hand harvesting requires a crew of people carrying ladders and picking bags. They detac h citrus fruits from stems, put them into the picking bags, which weigh several pounds, and transfer them to a nearby tub. Finally tubs are collected by a goat truck. As citrus tree height increases, harvesting productivity is declined since they use ladde rs to reach fruit on top of the tree canopy. Another concern of using manual harvesting is that cost of manual harvesting operations is high. Because harvesting constitutes about 35% to 45% of a total cost of citrus production (Sanders, 2005), there is a n eed to improve citrus harvesting practice, to be cost effective and productive. To mechanize and automate citrus harvesting practice is the major goal of the citrus growers in Florida. A numerous studies have been conducted to develop and improve such sy stems that are eventually able to eliminate manual labor, increase productivity and deliver cost effective practices to the growers. Robotic Harvesting Robotic harvesting has been studied since 1960s. During this time, a lot of work has been done. Several research groups all over the world tried to develop automatic citrus harvesters. Robotic harvesters were usually designed mimicking human arm movements. It involves a sequence of operations. First robotic arm or manipulator detects fruit from tree canopy. Detection is usually accomplished with help of imaging sensors by recognizing color or shape of fruit. Consequently it positions the end effector to the relative location of fruit to be ready to

PAGE 17

17 grasp it. The final operation is fruit removal. It involves gripping fruit and detaching fruit from stem. The most common design of robot manipulator is a three degree of freedom (DOF). The objective of gripping fruit is to locate its coordination in the three dimensional (3D) coordinate system. The position coordi calculate the three joint positions of the three DOF needed to reach to the desired position. The limitation of the three DOF is that it cannot reach desired position due to its lack o f degree of freedom. To remedy such limitation of robot manipulators, researchers started designing five to six DOF arms (Hannan and Burks, 2004). Mehta (2007) developed a control system of robotic manipulator using cameras. By using robot arm to the referenced coordinate in the Euclidean coordinate system. Fruit detection has been done using color thresholding. However, fruit occlusion was major problem. At the end of the robotic arm, several pneumatically actuated fingers or end effectors are usually mounted to grasp citrus fruits. Some of the robotic systems have pneumatic suction instead of several fingers. The end effectors fall into several types depending on frui t detaching design: cutting, pulling, twisting or twisting/pulling. The most common design is a cutting end effector. Cutting tools include a circular micro saw (Muscato et al., 2005) and a blade (Buie, 1965; Macidull, 1973; Yoshida et al., 1988). Other ty pe of end effector is a pulling end effector. Several pulling based robotic harvesting systems were designed. Pulling force usually applies along the fruit stem axis. However, the pulling force can be reduced if the force is applied at a 90 angle to the a xis of attachment, as is often done in hand harvesting. Pulling actions disturb the surrounding branches and leaves of tree, so it makes harder for consequent picking since it is

PAGE 18

18 not easy to catch moving fruit. Third type of robotic design is a twisting me thod. In the twisting method, researchers usually use a rotating suction cup. This type of automatic harvesting systems was conducted in Juste et al. (1992). They reported that the detachment rate was 64% to 67%. The reasons of failures were caused by obst acles and mechanical failures. Flood (2006) conducted research on an end effector design. He started investigating physical properties to see the safe grasping limits of the end effector. He reported that although end effector did not meet all criteria, i t was able to harvest, and further refinements needed to be done. Although robotic harvesting is involved a lot obstacle, it will be one of the solutions in a citrus production. Mechanical Harvesting Significant efforts have been devoted to improve produc tivity and mechanize the harvesting of the Florida citrus crop (Futch et al., 2005). T he usage of mechanical harvesting system has been comparatively i ncreasing in last several years because of increasing cost of hand harvesting operations. Today, in Flori da, two types of mechanical harvesters are being used (Futch and Roka, 2005). One is a trunk shake harvester that shakes a tree trunk, consequently fruit will drop to the ground and then dropped fruit is picked up manually by a crew. The second one is a ca nopy shake and catch harvester. It shakes tree canopies causing fruit to fall onto catch frame. Then fruit would be carried through the conveyor system to the goat like trucks. A canopy shake and catch harvester is the most effective and productive mechani cal harvester in Florida. It can harvest 200 to 400 trees per hour assuming it travels down the tree row at ground speed of 0.8 to 2.1 km per hour (Futch and Roka, 2005). A numerous research has been devoted to improve productivity of these harvesting sys tems. As reducing the fruit detachment force, harvesting efficiency would improve significantly. Several research has been conducted to understand how variety, harvesting season

PAGE 19

19 and abscission chemicals would affect fruit removal efficiency. Earliest resea rch was started in 1967 to examine reattachment characteristics of oranges for the development of mechanical harvesting and abscission chemicals. Whitney et al. (2001) also studied different abscission chemicals and tested how they could improve the fruit removal efficiency of trunk shakers. In general, fruit removal efficiencies were increased 10% to 15% when orange detachment forces were reduced 50% to 80%, but they reported that abscission chemicals could increase fruit removals up to 14% and 30% on tru nk shake harvesting system. However, higher concentration of chemicals such as prosulfuron and metsulfuron methyl negatively affects the trees (Kender et al., 1999). Another research was to eliminate manual picking of the trunk shake harvester. OXBO Intern ational Corp. developed and evaluated fruit pick up machine (Bora et al., 2006). As a result of evaluation, its picking rate was 116.1 to 195.9 kg per min and its picking efficiency was 80% to 97% depending on grove conditions. that shake harvesting systems may affect negatively to the groves as they make bruises to the tree branches. However, Buker et al. (2004) studied the effect of mechanical harvesting to tree health and yield reduction over ten years. They found that there was no negative effect on overall fruit yield although the harvester made bark injuries to citrus trees. Yield Mapping Yield mapping is one of the crucial precision technologies since in grove variability of nutrients, tree size, soil types, etc. always exists. It allows growers to know where a site specific management would be beneficial. A large number of yield monitoring and mapping systems were studied and developed for various fruit and crops such as grains (Schueller and Bae, 1987, Searcy et al., 19 89; Borgelt and Sudduth, 1992), potatoes (Campbell et al., 1994), peanuts (Vellidis et al., 2001), silage (Lee et al., 2005), cotton ( Sui and Thomasson, 2001; Perry et al.,

PAGE 20

20 2005), coffee (Balastreire et al., 2002) and citrus ( Schueller et al., 1999; Whitne y et al., 2001 ; Annamalai et al., 2004; MacArthur et al., 2006; Grift et al., 2006; Kane and Lee, 2006; Chinchuluun and Lee 2006) Yield monitoring for citrus can be done either early season when fruit is on trees ( Annamalai et al., 2004; MacArthur et al ., 2006; Kane and Lee, 2006; Chinchuluun and Lee 2006) or harvesting season when fruit is harvested (Schueller et al., 1999; Whitney et al., 2001; Grift et al., 2006). In earlier research for citrus yield mapping (Whitney et al., 2001 ), yield was defined by the number of tubs and their locations in a grove. Thus, yield was based on several trees, not a single tree. In addition, manual harvesting was required to collect and put fruit into tubs to create yield maps. However, other researchers ( Annamalai et al., 2004; MacArthur et al., 2006; Grift et al., 2006; Kane and Lee, 2006; Chinchuluun and Lee 2006) developed yield mapping systems that used more advanced technologies such as machine vision and weighing systems. Annamalai et al. (2004) initiated the id ea of acquiring yield information when fruit was on trees. Their machine vision system was required taking non overlapping sequence of images of tree canopies while the system was driven through grove. The goal was to take images of fruit that was visible from outside of tree canopies based on an assumption that number of fruit growing outside canopies was proportional to the total number of fruit growing in the tree. Image location was recorded by a DGPS receiver as well. Yield information was predicted by the number of visible fruit of tree canopy per unit area. After they grabbed images, they counted the number of fruit from each image using their developed image processing algorithm. Eventually they found a yield prediction model to find citrus yield ind irect way. Hand harvested yield obtained from growers was compared against the predicted yield and a coefficient of determination, R 2 was

PAGE 21

21 0.53. The reason that predicted yield had low correlation to hand harvested yield was that one camera sometimes could not cover whole tree height. Thus some fruit that was outside the yield mapping system by adding four ultrasonic sensors to estimate fruit size along with citrus yi eld information. In addition, they developed three different color classification techniques background. They found that the neural network classifier outperform ed for estimating the linear discriminant with root mean square errors of 4.2 and 7.2 fruit, respectively. Chinchuluun and Lee (2006) added one more CCD camera to the system to provide full tree height coverage. In addition to color image classification technique, they implemented marker based Watershed algorithm to separate individual fruit since occlusions were a major problem. During field experiment, images of a total of 42 trees were acquired and these trees were hand harvested as well. The coefficient of determination between the algorithm predicted yield and actual harvested yield was 0.64. MacArthur et al. (2006) conducted preliminary research on yield ma pping using a mini helicopter. The system had latest technologies such as camera and wireless technology to control the mini helicopter remotely. However, most difficult problem was to take clear images. Vibrations of the mini helicopter platform made blur riness on image acquisition. They suggested that the system needed a lot of further improvements. A canopy shake catch harvester gives an opportunity to adopt site specific crop management in the sense that the harvester would travel in every unit area and collect fruit from every tree in real time. In addition to the shake and catch system, the harvester has conveyor system that carries out all fruit on the catch system to load trucks.

PAGE 22

22 Grift et al. (2006) conducted research in a l aboratory condition on p ossibility of developing yield mapping for fruit counting system on the conveyor system of the harvester using a dual photo sensor as well as a laser sensor. Passing fruit through conveyor system was compared with flows of fertilizer particles since many f ruits passing per second formed a mass flow. They tested several approaches. Firstly, a dual photo sensor was placed on conveyor to count individual fruits. Since fruit was not singularized, this method counted several fruit as one fruit. Then they built a gutter system allowing single fruit would pass each line at a time. In each approach among their methods was the one used a singulation. They advised that furt her research was needed to make them reliable. In this thesis, a feasibility of machine vision based citrus yield mapping system that would not only count number of fruits per unit area, but also estimate fruit size was investigated. Machine Vision and F ruit Size Estimating System In a machine vis ion system, recognizing the fruit along with its characteristics such as size, external color, blemishes and presence of defects from images would be very useful A numerous machine vision oriented applications w ere investigated in precision agriculture and remote sensing such as fruit grading by external and internal qualities, determining bruised areas, finding calyx, classifying forest image, etc. (Koning et al., 1994; Shahin and Symons, 2001; Whitney et al., 2 001; Leemans and Destain, 2004) Fruit quality inspection is an important step in fruit production and marketing. Fruit external quality inspection and grading systems are usually based on color, bruises, stem, calyx, blemish, disease and insect damage. A n umber of such systems were developed for apple (Leemans and Destain, 2004), potato, lentil (Shahin and Symons, 2001) and citrus ( Whitney et

PAGE 23

23 al., 2001). Fruit size is also one of the most important factors that customers recognize. Koning et al. (1994) deve loped machine vision systems to grade potato by its size. A variety of classifiers were used to identify fruit from background in image processing. Classification methods include neural network, Bayesian classifier and discriminant analysis based on diffe rent features of fruit surface. Leemans and Destain (2004) used a Bayesian extracted except russets, since russet was sometimes confused with the transition area between ground and blush color. Marchant and Onyango (2003) compared a multilayer feed forward neural network classifier with a Bayesian classifier for classifying color image pixels into plant, weed and soil. They found that the Bayesian classifier outpe rformed the neural network in the sense of total misclassification error. However, if the number of features increased, more than five or so, the Bayesian classification was infeasible. Thus they recommended using Bayesian classifier over feed forward neur al network when the number of features was few enough to require a feasible amount of storage. Another major obstacle for outdoor machine vision and robotic harvesting systems is fruit occlusion. One of the commonly used methods for occlusion is a watersh ed segmentation method developed by Beucher and Lantuejoul (1979). Casasent et al. (2001) segmented nuts from X ray images using the Watershed transform. Lee and Slaughter (2004) used the modified Watershed transform for separating occluded tomato plant le aves. Besides color analysis in the visible range, near infrared (NIR) spectrum plays an important role in machine vision. Aleixos et al. (2000) developed a citrus fruit inspection system using a machine vision in a controlled environment. In the system, C CD and NIR cameras were used to estimate fruit size and shape, detect blemishes and classify them by color. Although they tried to singularize fruit while

PAGE 24

24 passing through an inspection system, individualizing mechanism did not work efficiently. Thus, they allowed traveling several fruit at the same time. Another problem was fruit touching when they tried to extract contours of individual fruit.

PAGE 25

25 CHAPTER 3 IMAGE ANALYSIS Color Models The use of color in image processing is motivated by the fact that color is a powerful descriptor that often simplifies object identification and extraction from scene. Colors that humans perceive in an object are dete rmined by the nature of light reflected from the object. Red, green and blue colors are called primary colors because human eye is more sensitive to these three colors. Variation of colors is composed of these primary colors. To make specification of color standards, several color models are created. A color model is a specification of a coordinate system and a subspace within that system where each color is represented by a single point. The RGB Color Model The red, green and blue (RGB) color model is bas ed on a Cartesian coordinate system. Figure 3 1 shows a cube of RGB color model in which subspace of interest is defined. Each color consists of primary spectral components: (R, G, B). For instance, red, green, blue, black, and white colors are defined by (255, 0, 0), (0, 255, 0), (0, 0, 255), (0, 0, 0) and (255, 255, 255), respectively. The scale that is from 0 to 255 along each axis defines brightness. The diagonal from black to white represents gray levels with different brightness. The HSI Color Model Although RGB is the most suited color model for human perception, it cannot provide all needs in image processing algorithms. The HSI has advantages in the sense that color (Hue) component is separated from intensity (I), so it alleviates the image proces sing complexity. It consists of hue (H), saturation (S) and intensity (I) components. Hue component defines color. It is measured by degrees. Saturation is color purity represented by vector. The length of vector defined purity. The last component is inte nsity that measures lightness. HSI color model is

PAGE 26

26 represented by any shape such as a hexagon, triangle or circle. The choice of shape does not matter since any of the shape can be warped one from another. Figure 3 2 shows the representation of the model. Figure 3 1. RGB color cube. Converting from RGB to HSI is straightforward using the following equations. Hue component of each RGB pixel is obtained using the equation (3 1) where, Saturation can be defined as

PAGE 27

27 (3 2) The intensity component is defined by (3 3) Figure 3 2. HSI color model representation. The YIQ Color Model The YIQ color model is intended to take advantage of human color res ponse characteristics. The eye is more sensitive to changes in the orange blue (I) range rather than in the purple green range (Q). This is the same color model as HSI model in which brightness information (Y) is separated from chrominance information (IQ) To convert from RGB to YIQ,

PAGE 28

28 the following formulas are used. R, G, B and Y components range from 0 to 1, while I component ranges from 0.5957 to 0.5957 and Q ranges from 0.5226 to 0.5226. (3 4) (3 5) (3 6) Image Classification Classification involves recognizing objects based on their certain features from image. The first step of designing classification is to choose features of object. Classification accuracy would be improved if the feature values of an interested object are more differentiated than feature values of the rest of objects. This step requires general knowledge of the object such as its color, shape, texture, size, etc. Majority of conducted studies ha ve used color as a feature for object classifications. Depending on color models, any components of color models can be chosen as features of object. Figure 3 3 shows general steps of classification. Figure 3 3. Image classif ication design cycle. Next step involves designing and choosing model or classifier. There is no universal classification technique. Goodness of model choice depends on a nature of problem. The process of using data to determine discriminants of the classi fier is referred to as training the classifier. In Choose features Choose model Train classifier Evaluate classifier

PAGE 29

29 this step, a set of feature values is collected. Different classifiers are discussed later in this chapter. The final step is to evaluate whether chosen model really works in a given problem. The classific ation can be divided into two types based on chosen features: color and shape classification. In this chapter, different classification methods are discussed. Color Classification Classification can be of two types depending on availability of prior knowle dge of an object: supervised classification, and unsupervised classification. In supervised classification, a prior knowledge of an object is required. This prior knowledge comes from training operations. For instance, in the orange classification problem, all possible color values of object should be sampled before classifier would be designed. However, collecting all possible set of feature values is infeasible. Thus, some approximations are used in pattern recognition systems, so that it estimates the un known values of the model. In this research, the following Bayesian classification technique was implemented as a supervised classification. Bayesian c lassifier formulize a decision based recognition approach. Let x be an instance of n dimensional feature vector. Also let be one of W decision or discriminant functions. For W pattern classes the basic problem of pattern recogni tion is to find W discriminant functions with the property that, if a vector x belongs to class then (3 7)

PAGE 30

30 In other words, an unknown pattern x is said to belong to the i th class i f, upon substitution of x in to all discriminant functions, g i ( x 8. (3 8) where = posterio r probability, = likelihood or probability density function of class w i P ( w i ) = prior probability and Bayesian classification rule requires that prior probability P ( w i ) and likelihood of each class mus t be known before discriminant functions would be designed. The posterior probability can be chosen as discriminant function in the sense that for given x vector, P ( w i | x ) is a probability that x vector belongs to class w i Later decisions can be made by comparing these probabilities. If the summation of Equation 3 8 would be removed, Equation 3 8 becomes the equation shown in Equation 3 9. (3 9) For the sake of simplification, let us multiply both sides of equal sign of Equat ion 3 9 by logarithm, so the product of two terms becomes a sum of two terms. Thus, the final Baysian discriminant function becomes the following. (3 10) As mentioned before, sampling all possible feature values of each object is infeasible, so that unknown distribution of likelihood is approximated by Normal distribution (Duda et al., 2000). If it is assumed that likelihood is a normal distribution, then multivariate normal density in d dimensions is de fined as

PAGE 31

31 (3 11) Where, d = dimension of x vector, = mean vector of w i class and = d by d covariance matrix. Mean and covariance matrix can be found from the sampling set. Thus, the discriminant function is simplified as follows: (3 12) Shape Classification Besides color, shape is an important feature that objects can be recognized. There are two group of techniques used in segmentation: those based on conto ur detection and those involving region growing. These are primarily used in gray tone image segmentations. One of the region growing tools is a watershed transformation, which is used in separating partially occluded objects in gray images. It was first introduced as a morphological tool by Digabel and Lantuejoul (1978). Since then, many variations were introduced. They were reported by Beucher and Lantuejoul (1979) and Beucher and Meyer (1992). Many researchers used Watershed segmentation algorithm in ma ny applications by means separating touching oranges, apples, nuts, plants from weeds, etc. Some of them have already been mentioned in Chapter 2. The transformation named watershed because in the segmentation, a gray level image is considered as topograph ic surface. The value of each pixel in a gray level image represents an 4 A) and see how the Watershed transformation is applied.

PAGE 32

32 Figure 3 4 B shows that two orange reg ions were extracted by using color classification. Orange region is shown by white color and the background is shown by black color. A distance image is calculated by finding shortest distance from any point that is inside of fruit region to the nearest po int of the fruit boundary. The distance image is shown in Figure 3 4 C where white points are located furthest from the boundary. A region which has highest numerical values of the distance image is represented as the deepest part of the topographic surfac e. Those deepest regions are called catch basins. In Figure 3 4 D, two catch basins are shown. The deepest point of each catch basins is called a minima. A boundary of these two basins is called a watershed line or dam. The goal of Watershed transformation is to find these dams. Let us assume that water continuously floods the different catch basins. During the flooding, two or more floods coming from different minima may merge. To avoid these floods merged each other, dams are constructed on the points of surface where the floods would merge. These dams or watershed lines will separate the touching objects as shown in Figure 3 4 E. A B C D

PAGE 33

33 E Figure 3 4. Watershed representation. A) Two touching oranges. B) Binary image of the image A. C) Distance image. D) Topographic representation of the image (Source: Vincent (1991)). E) Separated oranges.

PAGE 34

34 CHAPTER 4 MATERIALS AND METHOD S In this chapter, hardware and software designs of the citrus yield mapping and fruit quality inspection system as well as conducted image analysis will be discussed. Hardware Design of the System A hardware system to acqui re high quality images for a machine vision based yield monitoring and fruit size measurement system was designed. A housing (99.06 cm x 96.52 cm x 43.42 cm, 0.63 cm thick) for the system was constructed using aluminum sheet to acquire high quality images (Figure 4 1 B) because outdoor imaging is much more complex than indoor imaging. The system consisted of a 3CCD progressive scan digital color camera (HV F31 Hitachi Kokusai Electric Inc, Woodbury, New York ) f our halogen lamps (Master Line Plus 50W GU5.3 12V 38D, Phillips Electronics Atlanta, GA ) a laptop (CF 51, Panasonic, Secaucus, NJ), and a data acquisition card (DAQCard 6036E, National Instruments, Austin, TX). In Figure 4.1 A, only two components are shown. Polarizing filters (25CP, Tiffen Co. LLC Hauppauge, NY) we re placed in front of each lamp as well as camera to remove glare from the lamps Camera had a lens (TF4DA 8, Fujinon Inc., Wayne, NJ) with focal length of 4 mm. As discussed in Chapter 1, a citrus canopy shake and catch harvester was us ed in this research. The harvester had conveyor system that could deliver harvested fruit to load trucks. On the conveyor system, the machine vision based yield mapping system was installed. To simulate the citrus canopy r system, a test bench was designed (Figure 4 1 B). The specification of the test bench was exactly the same as the bench of the harvester as shown in Table 4 1.

PAGE 35

35 Table 4 1. Conveyor belt specification Specification Scale Width of the conveyor belt 86.3 cm Speed of the shaft 200 RPM Diameter of gear wheel 16 cm Speed of the conveyor belt 167.6 cm/s System Operations A simple illustration of system operation is shown in Figure 4 2. The conveyor along with fruits moves constantly at high speed (1.67 m/ s). A B C Figure 4 1. Gen eral overview of the system. A) Ca mera and four halogen lamps. B) System housing. Camera takes a sequence of non overlapping and non skipping images of the conveyor belt. Camera parameters are adjusted properly. T he speed of conveyor belt is sensed by a hall effect encoder, so that the system knows when exactly it takes the next image. Consequently C amera Housing Encoder wheel with 30 teeth Test bench

PAGE 36

36 images are analyzed by the developed algorithm to count number of fruit and record size of each fruit. System Componen At the end of the shaft of conveyor system, an encoder wheel with 30 teeth was installed (Figure 4 1 C). A hall effect sensor was placed next to the encoder wheel to acquire the speed of shaft. The Hall Effect sensor would generate sequ ential pulses. The number of pulses is proportional to the number of gear wheel tooth passed per second. The data acquisition card was used as an interface to the laptop computer. Camera Parameters To acquire clear images from the moving conveyor belt, di fferent camera shutter speeds were studied in this research. Shutter speed is the time for which the shutter is held open during the taking of image to allow light to reach the camera image sensor. Although the camera has auto shutter and auto aperture mod e, these modes could not provide high quality images since the conveyor system moves along with fruits at high speed. Thus, the shutter should be opened very shortly to freeze fast moving object. On the other hand, since high shutter speed would allow less light coming to imaging sensor, image would not be bright enough. Therefore, the camera was tested by changing the shutter speed while moving the camera and taking objects in order to trade off sharpness and brightness of images. The shutter speed of the camera can be set in the range of 4 to 1/100,000 second.

PAGE 37

37 Figure 4 2. Pictorial illustration of system operation. Figure 4 4 shows images taken at the different shutter speeds (auto, 1/2000 s, 1/2150 s, 1/2230 s and 1/2300 s) while the camera was movin g. Fruits in the first two images look blurry. Figure 4 3. System connection diagram. However, in the last three images, oranges can be seen clearly. An image in Figure 4 4 E is generally looked fuzzy although oranges can be seen. Thus, as the result of test trials, the optimal shutter speed was determined to be 1/2200 s and aperture was set to auto mode. Encoder Data acquisition board Laptop computer Image acquisition card 3CCD camera DGPS reciever

PAGE 38

38 Image Synchronization with Encoder lled at the proper height to cover the frame width of 68.6 cm. A size of 800 x 600 pixels was chosen as an optimal frame size of the camera (Table 4 2). Table 4 2. Camera resolution Frame size Max FPS Grabbing delay (ms) Actual frame width (cm) 1024x768 24 bit 7.5 133.3 190.5 800x600 24 bit 15 66 68.6 The width of housing was 41.1 cm. It means that the camera must capture every 41.1 cm width of the conveyor since housing was installed on the conveyor system. As a result of several trials, the 41.1 cm distance of conveyor movement was equal to 30 teeth of the encoder wheel. The speed of conveyor belt was 167.6 cm/s, whereas the width of housing was 41.148 cm. So, image had to be acquired at every 41.1 cm /167.6 cm/s = 245 ms. Therefore, image acquisitio n was synchronized with the conveyor belt using 30 teeth of the encoder. The accuracy of conveyor belt movement is defined as 41.1 cm/30 teeth = 1.3 cm. A B Figure 4 4. Image acquisition of moving object at the different shutter speeds. A) Auto sh utter speed. B) 1/2000 s. C) 1/2150 s. D) 1/2230 s. E) 1/2300 s.

PAGE 39

39 C D E Figure 4 4. Continued System Software Design A multithreading system software was developed using Microsoft Visual C++ 6.0 MFC/COM and ImageWarp (BitFlow Inc. Woburn, MA ). Since an image had to be taken in 245 ms assuming that the conveyor speed is 167.64 cm/s, the program had to be run fast enough to read from encoder, acquire an image, read DGPS receiver and analyze images. General overview of the multithread based progra m is shown in Figure 4 5. by operating system. A preemptive scheduler in the Windows operating system kernel divides CPU time among active threads so that they appe ar to run simultaneously. Multiple threads

PAGE 40

40 The system software consisted of three working threads each of which worked independently from each other. Communication between these threads w as accomplished by creating events. An encoder thread would always read signal that comes from the hall effect sensor, and count period of square waves (Figure 4 6). When the counter reached at the encoder counter limit, N the thread created an event tell ing to the ImageWarp thread that it is a right time to acquire next image. The counter limit N was determined as 30, based on camera view and traveling distance of conveyor system. Figure 4 5. The system software overview. Th e system software was able to run in two modes: real time and post processing. In real time mode, the ImageWarp thread would acquire a digital image and directly determine the number of fruits and size of each fruits. Whereas, in post processing mode, the thread would acquire image and save it to the hard drive. The GPS thread would constantly read signal coming from the DGPS receiver. It then parses row of characters to latitude, longitude and time. Finally it saves GPS information in a raw file. GPS thread Read GPS data 1.Grab Image 2.Process Image 3.Save Images to the Hard drive. 1.Read Encoder counter 2.If counter reaches N, it sends a signal to the Thread 2 and Thread3. Encoder thread ImageWarp thread Thread1 Thread2 Thread3 Start End.

PAGE 41

41 Figure 4 6. Encoder thread algorithm. System Experiments The machine vision based citrus yield mapping and fruit size measurement system was examined both in a laboratory on a designed test bench and on a shake and catch harvester. Val encia oranges were used throughout these experiments. Test bench Experiment The machine vision based citrus yield mapping and fruit inspecting system was first tested in a laboratory at the Citrus Research and Education Center (CREC) at the Lake Alfred, F lorida during February and March of 2007. Begin Read period of square wave Increase a counter n = n + 1 Whether the counter reaches N Create an event telling to other two threads End.

PAGE 42

42 Image acquisition during the test bench experiment An algorithm for counting the number of fruit and measuring the size of fruit was developed and tested on the test bench. A total of 14 trials were conducted. In each trial, different number of boxes of oranges was manually fed to the conveyor belt. Total fruit weight was measured with a balance (PS60 Parcel Scale, Mettler Toledo, Columbus, OH) as shown in Figure 4 7 and a sequence of fruit images was acquired for each trial. A total of 3,653 images were acquired. Among these, 703 images were randomly used for classification. A B Figure 4 7. Test bench experiment. A) Feeding conveyor belt. B) Weight measurement of tested fruit. For each image, fruit parame ters such as the number of fruit, fruit diameter and fruit area were extracted during image analysis. Consequently these parameters were added up for images of every test in order to see how they were correlated with actual fruit weights. Citrus Canopy Sh ake and Catch Harvester Experiment The harvester experiment was conducted during March of 2007 at the commercial citrus grove (Lykes Inc.) located in Fort Basinger, FL. A total of 773 images were taken during static operation. Fruit was loaded into the goa t like truck and its weight was measured using a balance ( RW 05S, CAS, Korea). The fruit weight was determined by subtracting only truck weight from the measured total weight.

PAGE 43

43 Figure 4 8. Canopy shake and catch harvester experiment. Image Processing An alysis The basic goal of image processing analysis was to recognize fruit pixels and estimate the number of fruit and their size from images. Although the housing of the machine vision system with artificial lighting was built to improve image acquisition, color similarity between fruit and background, sunlight coming from the bottom, and fruit occlusions were major obstacles (Figure 4 9) to accomplish the objectives stated in Chapter 1. The most visible feature to identify fruit from image was their color Different color components were carefully examined in various color models how fruit pixel values could be isolated from background pixel values. The most isolated components were saturation and hue components of HSI color model and I component of YIQ co lor model. Therefore, hue, saturation and I components were studied further to be chosen as features in color image classification.

PAGE 44

44 Figure 4 9. Color image of fruit and background. A total of 1,118,618 fruit pixels and 2,365,368 background pixels were collected from 60 random images to implement supervised color classification algorithm. These 60 images were taken during the test bench experiment. In color classification, any given pixel has to be assigned to either fruit class or background class. The histograms along with images in different color components are shown in Figure 4 10. Hue and I components provided better separation of two classes (Figure 4 10 B, F). Their means are shown in Table 4 3. A B Figure 4 10. Images in different color mod els and their histograms. A) Hue component of HSI color model. B) Hue histogram of fruit and background. C) Saturation component of HSI color model. D) Saturation histogram of fruit and background. E) I component of YIQ color model. E) I (YIQ) histogram of fruit and background.

PAGE 45

45 C D E F Figure 4 10. Continued Table 4 3. Means of hue, saturation and I (YIQ) components of fruit and background classes Classes Hue Saturation I (YIQ) Fruit class 60.6 153.0 30.8 Backround class 108.6 127.7 4.5 The imag e processing algorithm steps are shown in Figure 4 11. After images were acquired, digital images were converted to HSI and YIQ color models. Then color classification algorithm was followed. Different color classification approaches were studied. Although linear threshold based color classification was implemented for color classification, it did not give good fruit recognition. Therefore, quadratic classification models were investigated rather than linear one to accomplish the objectives stated in Chapte r 1.

PAGE 46

46 Figure 4 11. Image processing algorithm. As formulated in Chapter 3, Bayesian classifier was used in the color classification. According to the Equation 3 10, fruit and background discriminant functions are defined as Equ ation 4 1 and Equation 4 2, respectively. (4 1) (4 2) Thus, the decision is made as (4 3) The prior probabilities depend on the coverage of fruit pixel in an image. The cover age of fruit pixel would change from image to image, hence the priors are not really known for any individual test cases. Marchant and Onyango (2003) advised to choose equal prior probabilities in these situations. However, in randomly sampled images, some images had no fruit, but some of them had full coverage of fruit. Therefore, the prior probabilities in this study are defined as follows: (4 4) Image acquisition RGB to HSI and YIQ conversion Color classification Morphological operation Separating touching fruits Extracting number of fruit and their size

PAGE 47

47 The priors was found as P(fruit) = 0.22 and P(backg round) = 0.88 using the fruit and background samples. The quadratic discriminant function was found on I and Hue as shown in Figure 4 12. Figure 4 12. Discriminant function between fruit and background classes. Fruit Parameters In image analysis, th e following parameters were calculated for every individual fruit: area, circularity, convexity, average diameter, maximum and minimum diameter and ellipticity. The ImageWarp provided all functions to calculate these parameters. Yield Prediction Model To d evelop a citrus yield prediction model, a relationship between fruit weight and fruit diameter was studied. The first assumption was that the correlation between them might be linear. Welte (1990) studied Jonagold apple growth size during growing season, a nd found that apple weight was directly correlated to apple diameter with regression coefficient of 0.9899.

PAGE 48

48 Thus orange samples were collected from the CREC. Fifty three randomly selected ter was referred to an average of minor and major dimension of fruit since most fruits were not complete circle. As the result of measurement (Figure 4 13), coefficient of determination was 0.971. Thus, the relationship between them was assumed linear. Figure 4 13. Relationship between actual fruit weight and diameter. Fruit area can be measured from images in 2 dimension (2D). If fruit is assumed as a complete circle, then area (A) is determined as (4 5) Where R = radius of fruit, d = diameter of fruit.

PAGE 49

49 If fruit weight was compared against with two dimensional fruit area, then the correlation between fruit weight and fruit diameter would not be linear, but it is quadratic. The regression analysis was conducted with them an d the coefficient of determination was found to be 0.972. Figure 4 14. Relationship between actual fruit weight and area. In 3 dimension, fruit shape is sphere. The weight (M) of any object is defined by density (g) and volume (V). (4 6) Thus, if it is assumed that density is uniform for all fruit, then fruit mass (M) is defined as (4 7)

PAGE 50

50 A regression analysis was conducted on fruit weight and fruit cubic diameter. Th e coefficient of determination was 0.969. Figure 4.15. Relationship between actual fruit weight and volume. From above analysis, fruit weight and fruit area comparison showed better performance than others. Camera Calibration To measure actual size o f fruit from images, camera had to be calibrated. Since camera was placed at constant height of 891.8 mm, camera intrinsic parameters were used to find the actual object size. In this calibration, a pinhole camera model was used (Forsyth and Ponse, 2001). 3 mm x 4.6510 3 mm. The perspective projection figure can be represented in Figure 4 16, where

PAGE 51

51 defines an object size in an imaging plane, defines the camera focal length, defines a distance from camera to the object and X defines a object size in a real world coordinate system. Figure 4 16. Camera perspective projection. A ratio of similar trian gles is defined as: (4 8) Each unit of the parameters is written in Equation 4 8. To make parameter units consistent, pixel should be converted into millimeters. Since the camera pixel size is 4.6510 3 mm x 4.6510 3 mm, th e actual object size is defined as: (4 9)

PAGE 52

52 CHAPTER 5 RESULT AND DISCUSSIO N In this chapter, the result summaries of the test bench as well as canopy shake and catch harvester experiments will be discussed. The chapter starts with a summary of image classification and processing. Then yield predicti on results will be discussed later. Color Segmentation Result on Test bench As mentioned in the previous chapter, a total of 14 test trials were conducted on the test bench at CREC, Lake Alfred, FL. The image processing algorithm was applied to the 609 im ages that were taken during the experiment at CREC. The color segmentation algorithm was started with image acquisition (Figure 5 1 A) and followed by Bayesian classification (Figure 5 1 B). After Bayesian classification, morphological operation removed no n fruit pixels that came into fruit regions. Finally only fruit regions were extracted (Figure 5 1 C). Since some neighboring fruit regions joined each other, watershed transform separated them into individual fruit (Figure 5 1 D). Finally, fruit paramete rs were extracted from the segmented digital images taken during 14 trials. The parameters included number of fruit, diameter and area of each fruit were determined by the ImageWarp software. The fruit region was estimated by the number of pixels in a blob However, the fruit diameter is defined as (5 1) Besides these parameters, fruit circularity, ellipticity, convexity and maximum/minimum diameters were measured from each fruit in digital images. Some portion of 207 page re sults is included in Appendix B.

PAGE 53

53 A B C D Figure 5 1 Image segmentation result. A) Original image. B) Bayesian classifier result. C) Morphological operation result. D) Watershed transform result. Table 5 1 summarizes the test bench experiment results by the image analysis. A regression analysis was conducted on fruit parameters which were found by the image analysis. First, a relationship between the actual fruit weight and the number of fruit found by image processing algorithm was studied (Fi gure 5 2). The relationship was almost linear with the coefficient of determination of 0.892. Moreover, a trend between the actual fruit weight and fruit area measured by the algorithm was studied as well (Figure 5 3). The relationship between them was eve n better with R 2 of 0.962. The relationship between the actual fruit weight and fruit diameter measured by the algorithm was also conducted (Figure 5 4). The coefficient of determination was 0.963.

PAGE 54

54 Table 5 1. Summary of the test bench experiment results Test Total images Actual fruit weight (kg) Sum of fruit area (pixel) Algorithm fruit counting Sum of Fruit Diameter (pixel) Sum of Fruit Diameter (mm) 1 37 22.59 208915 55 3424 3549.78 2 27 25.92 338612 82 5889 6105.33 3 28 31.18 583687 155 10471 10855 .7 4 41 46.11 660859 169 11792 12225.2 5 29 49.28 867118 305 15798 16378.3 6 23 48.99 884879 252 15591 16163.7 7 46 75.30 1245102 336 22040 22849.6 8 40 71.03 1151251 331 20043 20779.3 9 51 97.61 1677222 518 29588 30674.9 10 55 93.85 1524808 415 264 76 27448.6 11 51 91.58 1572474 547 28024 29053.5 12 60 114.26 2073615 758 37307 38677.5 13 60 111.22 2202060 836 38888 40316.5 14 61 120.29 1847159 623 33336 34560.6 Figure 5 2. Actual fruit weight versus the number of fruit by image processing algo rithm: regression analysis.

PAGE 55

55 Figure 5 3. Actual fruit weight versus fruit area: regression analysis. Figure 5 4. Fruit diameter versus actual fruit weight: regression analysis.

PAGE 56

56 Color Segmentation Result on the Harvester The system was tested on a comm ercial canopy shake and catch harvester (Freedom Series 3220, Oxbo, Clear Lake, WI) at a commercial citrus grove (Lykes grove, Fort Basinger, FL). Fruit weight was measured using a balance ( RW 05S, CAS, Korea). The developed image processing algorithm was applied to images taken during the experiment. The result is shown in Figure 5 5. A B C D Figure 5 5. Color segmentation result for field experiment. A) Original image. B) Bayesian classifier result. C) Morphological opera tion result. D) Waters hed transform result. The Bayesian classifier worked relatively well on the images. As a result of morphological operations, some fruit regions were incorrectly classified. Fruit regions became smaller than the

PAGE 57

57 original sizes. Consequently, incorrect segme ntation made harder the individual fruit extraction (Figure 5 5 D). Breunig et al. (2000) investigated detecting outliers from large data sets based on a local density. An outlier is defined as an isolated object with respect to the surrounding neighborho od. The decision for detecting outlier objects is the following: (5 2) Using this density clustering approach, fruit pixels from binary image was clustered. The circle radius was chosen as two pixels based on several trials. Th e density threshold value (5 pixels) was chosen too. The result of density cluster or circular filtering is shown in Figure 5 6 from Figure 5 5 C. Figure 5 6. Circular filtering result for field experiment 2. Although fruit weight was measured twice, t wo measurements were not enough for the analysis. Thus, the number fruit from 60 images was counted to compare it with the result of the image processing algorithm. The regression analysis was also conducted on number of fruit that was extracted by human c ounting and algorithm counting. The coefficient of determination was 0.891 (Figure 5 7).

PAGE 58

58 Figure 5 7. Regression analysis between human counting and algorithm counting. System Execution Time The system was tested on a laptop computer with 2.1 GHz CPU and 64 MB RAM specification. Two functions were developed to display actual execution time on a dialog. The average execution time of 50 trials for acquiring an image, reading DGPS output and reading the encoder was measured (Table 5 2). It was constant at 12 0 ms. The average time for color segmentation was 138 ms. The execution time of morphological operation and Watershed transform were not constant. They depended on the number of fruit in an image. The average time of morphological operation and Watershed t ransform were 40 ms and 89 ms, respectively. Thus, total execution time was 367 ms.

PAGE 59

59 Table 5 2. System execution time Execution steps Average execution time (ms) Image, DGPS, encoder acquisition and saving an image 120 Color segmentation 118 Morpholo gical operation 40 Watershed transform 89 Total execution time 367 System Limitation The machine vision system has not been tested at citrus grove when the continuous canopy catch and shake system was used for harvesting. Several obstacles are expecte d to test it at citrus grove. First of all, canopy catch and shake system creates vibration in which the machine vision system components such as wire and interconnections may be malfunctioned. In addition, the canopy shake and catch system creates dust wh ile harvesting, so that it may effect to the image acquisition by reducing image quality. Another issue is that some foreign materials will also be harvested along with fruit, such as leaves or branches, which will pose some difficulty in identifying frui t. Those leaves and branches may affect image segmentation negatively. Some of the pesticide residues such as copper (greenish color) or disease such as rust mite could also interfere with color segmentation. Some other complications can be expected, but these are not known until the machine vision system is tested at actual citrus grove.

PAGE 60

60 CHAPTER 6 CONCLUSIONS In this study, a machine vision based citrus yield mapping and fruit size measurement system on a canopy shake and catch harvester was developed successfully. The advantage of this system from previous citrus yield mapping systems is that it can record fruit size information in real time and without any manual labor involvement. The system was tested on a test bench as well as on a commercial canopy shake and catch harvester. To develop and test the system, a test bench with the same specification as the mechanical harvester was designed at the Citrus Research and Education Center located at Lake Alfred, FL. The system had to acquire high quality images of citrus fruit that was passing through a conveyor system. Multithread based syste m software was developed using Visual C++ and ImageWarp image processing software. A supervised image processing algorithm was developed to count number of fruit and measure the size of individual fruit. The fruit area, the number of fruit and fruit diamet er were measured using an image analysis algorithm from the set of images for the test bench trial. Actual fruit weight was also measured. The coefficients of determination of the sum of areas, the number of fruit and the sum of fruit diameters against act ual fruit weight were 0.962, 0.892 and 0.963, respectively. When relationships between actual fruit weight (W) and actual diameter (D) of a single fruit were studied, a quadratic (W~D 2 ) and linear (W~D) relationships showed the highest linear trend with R 2 of 0.972 and 0.971, respectively. In a test bench trial, these two relationships were still held. Thus, the image processing algorithm performed quite well on images taken during the test bench trial. When the citrus yield mapping system was tested on a canopy shake and catch harvester, the number of fruit was counted manually from 60 images. The coefficient of determination

PAGE 61

61 between the number of fruit counted by image processing algorithm and the human counting was 0.891 for the harvester trial. The reas on for the low performance of the system for the harvester trial was due to the following: 1. Although illumination was controlled by utilizing housing for the system, the sun light came from under the conveyor system, so that image lighting was not always uni form. 2. Due to this non uniform lighting, the color segmentation algorithm could not perform well. To improve the performance of the system, a cover under the conveyor system should be placed to block the sunlight coming into the system. The next step would be to test the system at the citrus grove while the canopy shake and catch harvester is harvesting citrus fruit.

PAGE 62

62 APPENDIX A PROGRAMMING CODE Encoder Thread Code UINT WorkerThreadProc( LPVOID lParam ) { EncoderReading m_oEncoderInst; //InfoStruct *m_info = (InfoStruct*)lParam; CComDemoDlg* m_dlg = (CComDemoDlg*)lParam; unsigned long imNum = 0, LIMIT = 20; unsign ed int lastCount = 0, data=0, counter=0; CString ss, ss2, ss3; LIMIT = ::GetDlgItemInt (m_dlg >m_hWnd, IDC_ENCODER_LIMIT, NULL, true); ss2.Format("LIMIT=%d", LIMIT); ::PostMessage (m_dlg >m_hWnd, WM_SEND_STATUS_MESSAGE, 1, 0); while (!STOPFLA G) { data = m_oEncoderInst.DAQmxRun(); counter = data lastCount; ss.Format ("counter value=%d", counter); ASSERT(m_dlg >m_pImWThread >m_hThread != NULL); ASSERT(m_dlg >hIWEvent != NULL); ::SetDlgItemInt (m_dlg >m_hWnd, IDC_EDIT_COUNTER, (UINT)counter, true); // Read counter. When the counter reaches LIMIT, then it sets the hEncoderEvent event. if (counter > LIMIT) { imNum++; lastCount = data; // Raised if (!SetEvent(m_dlg >hEncoderEvent)) { AfxMessageBox(_T("SetEv ent failed ")); m_dlg >SendStatus(_T("SetEvent failed ")); } else {

PAGE 63

63 m_dlg >SendStatus(_T("SetEvent for Encoder ")); } //::PostMessage (m_info >hWnd, WM_COUNT, counter, imNum); // wParam = counter value, lParam = imNumber } } // 1 = Worker thread exiting... // 2 = Busy. Script running... // 3 = Script thread exiting... ::PostMessage (m_dlg >m_hWnd, WM_SEND_STATUS_MESSAGE, 1, 0); return 0; } ImageWarp Thread Code UINT WorkerIWProc(LPVOID lParam) { DWORD dwWaitRe sult; int flag1, flag2; // InfoIWThread* tempIW = (InfoIWThread*)Param; CComDemoDlg* m_dlg = (CComDemoDlg*)lParam; HANDLE hEvents[2]; flag1 = 0; flag2 = 0; ::PostMessage (m_dlg >m_hWnd, WM_SEND_STATUS_MESSAGE, 3, 0); while (!STOPFLAG) { hEvents[0 ] = m_dlg >hEncoderEvent; hEvents[1] = m_dlg >hIWEvent; //m_dlg >SendStatus(_T(" IMThread ")); dwWaitResult = WaitForSingleObject (m_dlg >hEncoderEvent, INFINITE); if (dwWaitResult == WAIT_OBJECT_0) { m_dlg >SendStatus(_T(" Inside of wait ")); if (!ResetEvent(m_dlg >hEncoderEvent) || !ResetEvent(m_dlg >hIWEvent)) { m_dlg >SendStatus(_T(" Failed on ResetEvent ")); }

PAGE 64

64 ::PostMessage(m_dlg >m_hWnd, WM_IW, 0, 0); } else {// error m_dlg >SendStatus(_T(" Failed r eading WaitForSingleObject ")); } } m_dlg >SendStatus(_T(" IMThreadExit ")); // 1 = Worker thread exiting... // 2 = Busy. Script running... // 3 = Script thread exiting... ::SetDlgItemText(m_dlg >m_hWnd, IDC_FIRST, _T("IW thread exiting...")) ; //::PostMessage (m_dlg >m_hWnd, WM_SEND_STATUS_MESSAGE, 3, 0); //CButton* th2 = (CButton*)::GetDlgItem(m_dlg >m_hWnd,IDC_CHECK_TH2); //th2 >SetCheck(0); return 0; } Image Processing ImageWarp Script in Real time Mode closeAll() setCurDir("C: \ \ RESEARCH \ \ CitrusYieldMapping \ \ 07 0226 TestImages \ \ ") Dim I as Integer I = 1 do resetParam() setShowGrid(FALSE) grabIm(1) grabIm(100) setSelect(100,SH_RECT, SEL_NEW, 124,172,891,622) splitRGB (100, 102,101,103) R=102 splitYIQ(100, 103,105,104) I=105 threshold(102, 106,M_PRESET,48.,236.,TRUE,FALSE) Thresholding using red scrap(106,107,0,8,0) dilate(107,106,1,PR_VBAR) fill(106,107,1,0) erode(107,108,1,PR_CROSS) open(108,109,2,PR_CIRCLE) scrap(109,110,0,1470,0) '110

PAGE 65

65 threshold(105,106,M_PRESET,132.,187.,TRUE,FALSE) Thresh by I scrap(106,107,0,8,0) dilate(107,108,1,PR_VBAR) fill(108,109,1,0) erode(109,111,1,PR_CROSS) open(111,107,2,PR_CIRCLE) scrap(107,108,0,800,0) andim(108,110, 112) 108 AND 110 = 112 separate(108,113,1,1) scrap(113,114,0,400,0) duplicate (114, 2) resetParam() selectParam("Count","Area","Diameters","DMax","DMin") measObjects(114,105) saveIm (100, strFormat ("image%d.tif", I)) saveIm (114, strFormat ( "ProcessedImage%d.tif", I)) I = I + 1 SendMessage(0) loop Image Processing ImageWarp Script in Post mode setCurDir("C: \ \ RESEARCH \ \ CitrusYieldMapping \ \ 07 0226 TestImages \ \ ") Dim I as Integer I = 1 do grabIm(1) grabIm(100) saveIm (100, st rFormat ("image%d.tif", I)) I = I + 1 SendMessage(0) loop End

PAGE 66

66 APPENDIX B MEASURED FRUIT PARAM ETERS Table B 1. Fruit parameters Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 1 6035 0.895 0.984 87.658 88.837 82.680 0.998 2 5222 0.845 0.961 81.541 85.703 75.644 0.995 3 4353 0.857 0.971 74.447 78.294 68.542 0.996 4 3722 0.833 0.960 68.840 75.432 60.407 0.994 5 5048 0.881 0.978 80.171 83.487 73.756 0.997 6 6012 0.910 0.984 87.491 90.449 81.000 0.999 7 5257 0.880 0.967 81.813 84.006 74.947 0.996 8 4957 0.903 0.982 79.445 81.884 72.835 0.995 9 4842 0.868 0.971 78.518 81.400 69.635 0.995 10 4713 0.787 0.959 77.465 82.365 71.000 0.990 11 4352 0.883 0.974 74.439 76.059 67.683 0.996 12 4238 0.877 0.974 73.457 75.604 69.426 0.996 13 2291 0. 790 0.965 54.009 70.036 40.706 0.984 14 3593 0.910 0.980 67.637 71.021 62.362 0.998 15 806 0.653 0.896 32.035 44.045 25.456 0.876 16 1235 0.703 0.921 39.654 56.223 29.732 0.946 17 2 0.000 1.000 1.596 1.000 1.000 1.000 18 5 1.000 1.000 2.523 4.000 1.00 0 1.000 19 4 1.000 1.000 2.257 3.000 1.000 1.000 20 1 0.000 1.000 1.128 1.000 1.000 1.000 21 1 0.000 1.000 1.128 1.000 1.000 1.000 22 4 1.000 1.000 2.257 3.000 1.000 1.000 23 5 1.000 1.000 2.523 4.000 1.000 1.000 24 2 0.000 1.000 1.596 1.000 1.000 1. 000 25 5822 0.852 0.966 86.098 89.376 79.981 0.995 26 5058 0.874 0.974 80.250 83.600 74.673 0.997 27 4926 0.917 0.986 79.196 80.281 75.604 0.998 28 4915 0.867 0.973 79.107 81.884 74.007 0.996 29 3019 0.790 0.947 61.999 65.765 53.451 0.973 30 5831 0.8 98 0.983 86.164 89.185 82.000 0.998 31 5257 0.861 0.973 81.813 85.212 78.090 0.995 32 3954 0.871 0.969 70.953 74.626 65.054 0.997 33 4857 0.895 0.978 78.639 81.216 72.622 0.998 34 5146 0.892 0.975 80.945 84.172 76.059 0.996 35 5515 0.898 0.976 83.797 87.321 79.209 0.997 36 4413 0.898 0.979 74.959 76.276 71.000 0.998 37 5322 0.869 0.966 82.318 86.822 76.557 0.995 38 4868 0.875 0.974 78.728 81.056 74.000 0.995

PAGE 67

67 Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) M in. Diameter (pixels) Ellipticity 39 4887 0.883 0.975 78.882 81.541 73.000 0.997 40 4071 0.885 0.966 71.996 77.621 67.720 0.995 41 4494 0.889 0.978 75.643 78.409 71.589 0.998 42 4728 0.878 0.973 77.588 81.413 70.214 0.997 43 5883 0.899 0.982 86.547 87 .092 82.219 0.998 44 3743 0.850 0.955 69.034 74.250 63.000 0.990 45 5625 0.846 0.973 84.628 91.608 77.104 0.996 46 6274 0.909 0.983 89.377 92.655 83.935 0.999 47 4411 0.875 0.969 74.942 77.795 69.584 0.997 48 4073 0.850 0.968 72.013 75.961 63.781 0.99 2 49 5027 0.891 0.983 80.004 83.355 74.545 0.999 50 4437 0.839 0.967 75.162 80.777 69.426 0.994 51 3619 0.828 0.961 67.881 73.007 61.000 0.987 52 4428 0.794 0.956 75.086 79.246 67.209 0.991 53 4163 0.845 0.971 72.805 76.420 67.067 0.996 54 3320 0.734 0.920 65.017 70.880 56.613 0.967 55 129 1.000 0.991 12.816 14.036 9.899 0.976 TOTAL of TRIAL #1 208915 3488.016 Number of fruit: 55 1 5716 0.890 0.973 85.310 88.730 82.000 0.997 2 3744 0.805 0.957 69.044 73.007 62.241 0.989 3 4638 0.880 0.970 76. 846 80.505 69.354 0.996 4 3085 0.867 0.980 62.673 73.007 50.000 0.989 5 1946 0.735 0.965 49.777 67.365 32.016 0.965 6 5460 0.895 0.977 83.378 84.906 78.772 0.998 7 5835 0.839 0.966 86.194 90.139 80.156 0.996 8 5245 0.906 0.981 81.720 84.172 77.389 0.9 98 9 4919 0.909 0.980 79.140 80.623 73.410 0.996 10 5184 0.910 0.979 81.243 85.726 77.885 0.998 11 5337 0.879 0.976 82.433 85.475 77.782 0.995 12 3255 0.707 0.918 64.377 73.062 49.518 0.966 13 1841 0.661 0.950 48.415 70.292 32.650 0.941 14 2352 0.725 0.954 54.723 70.093 38.210 0.949 15 3993 0.768 0.952 71.302 78.746 59.059 0.972 16 4573 0.893 0.974 76.305 78.600 71.000 0.997 17 3294 0.778 0.961 64.761 80.411 45.000 0.966 18 1840 0.771 0.936 48.402 57.219 38.000 0.978 19 4150 0.898 0.975 72.691 75 .538 69.260 0.997 20 2732 0.793 0.965 58.979 70.859 45.000 0.968 21 4333 0.917 0.980 74.276 77.414 68.447 0.997 22 4700 0.913 0.980 77.358 80.957 69.635 0.996 23 4158 0.768 0.943 72.761 79.310 58.000 0.991

PAGE 68

68 Number Area (pixel 2 ) Circularity Convexity Fr uit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 24 4411 0.871 0.970 74.942 76.844 68.184 0.995 25 3203 0.851 0.963 63.861 65.000 58.181 0.994 26 4977 0.882 0.972 79.605 81.498 74.847 0.997 27 5282 0.850 0.962 82.008 87.3 21 74.967 0.991 28 3782 0.921 0.979 69.393 70.605 66.483 0.998 29 3728 0.875 0.969 68.896 72.945 64.133 0.995 30 4735 0.887 0.975 77.645 91.220 75.584 0.997 31 5729 0.894 0.977 85.407 87.618 81.000 0.997 32 6002 0.900 0.976 87.418 91.831 84.054 0.998 33 4484 0.901 0.973 75.559 90.686 70.576 0.998 34 4756 0.872 0.973 77.817 80.808 70.264 0.994 35 3855 0.445 0.944 70.060 76.792 63.514 0.981 36 4626 0.862 0.970 76.746 78.549 70.064 0.996 37 2740 0.769 0.956 59.065 70.349 42.579 0.967 38 1922 0.852 0 .965 49.469 60.000 36.346 0.992 39 5347 0.899 0.978 82.511 85.866 77.466 0.998 40 5120 0.895 0.982 80.740 82.735 76.322 0.995 41 3349 0.885 0.973 65.300 67.624 60.000 0.996 42 4928 0.879 0.975 79.212 81.609 74.686 0.998 43 2558 0.811 0.971 57.070 71.0 07 39.319 0.974 44 5606 0.905 0.978 84.485 87.321 81.786 0.998 45 4848 0.872 0.977 78.566 80.753 71.840 0.998 46 3423 0.882 0.973 66.017 67.268 59.540 0.994 47 3337 0.898 0.974 65.183 67.357 60.000 0.997 48 2196 0.787 0.945 52.878 55.227 40.112 0.980 49 1794 0.709 0.959 47.793 67.007 30.000 0.963 50 4120 0.876 0.968 72.428 75.432 67.365 0.994 51 4771 0.894 0.979 77.940 91.241 70.880 0.996 52 3899 0.876 0.973 70.458 72.835 63.820 0.996 53 4415 0.906 0.983 74.976 76.968 70.000 0.998 54 5034 0.915 0 .982 80.059 81.345 73.756 0.998 55 3873 0.874 0.964 70.223 72.567 61.814 0.995 56 3471 0.863 0.968 66.479 68.884 62.434 0.993 57 5469 0.880 0.979 83.447 85.440 79.057 0.997 58 5213 0.865 0.970 81.470 84.865 73.348 0.995 59 4422 0.862 0.971 75.035 79.4 29 65.192 0.996 60 4650 0.870 0.975 76.945 79.322 71.421 0.996 61 4200 0.906 0.979 65.844 75.690 69.462 0.998 62 3972 0.864 0.971 71.115 73.926 65.000 0.995 63 5042 0.934 0.986 80.123 83.630 77.000 0.997 64 3642 0.838 0.959 68.097 70.725 64.070 0.992

PAGE 69

69 Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 65 3442 0.861 0.964 66.200 68.622 55.326 0.988 66 1464 0.757 0.934 43.174 58.464 28.636 0.963 67 3762 0.842 0.961 69.209 72.139 62.000 0.991 68 4757 0.904 0.978 77.825 80.957 73.348 0.997 69 4440 0.907 0.981 75.188 77.162 70.000 0.997 70 4156 0.893 0.974 72.743 78.109 65.069 0.996 71 3864 0.880 0.972 70.141 72.719 65.069 0.994 72 4225 0.865 0.969 73.345 76.538 68.593 0.996 7 3 3826 0.817 0.962 69.796 72.897 66.648 0.992 74 3584 0.782 0.948 67.552 73.546 58.549 0.985 75 4636 0.883 0.977 76.829 80.777 68.593 0.997 76 4626 0.909 0.977 76.746 79.322 70.937 0.997 77 5125 0.889 0.975 80.780 82.970 76.118 0.997 78 5102 0.872 0.9 72 80.598 82.347 72.691 0.997 79 4869 0.892 0.979 78.736 80.623 74.000 0.998 80 4159 0.891 0.974 72.770 74.953 68.949 0.997 81 4362 0.863 0.973 74.524 80.411 66.000 0.982 82 2952 0.803 0.967 61.307 72.339 47.518 0.965 TOTAL of TRIAL #2 338612 5881 .860 Number of fruit: 82 1 5350 0.852 0.969 82.534 86.954 77.782 0.995 2 2585 0.750 0.923 57.370 65.734 51.088 0.967 3 2900 0.721 0.956 60.765 81.056 41.110 0.964 4 4429 0.886 0.976 75.094 77.878 68.447 0.996 5 4914 0.870 0.964 79.099 81.006 72.897 0. 996 6 3537 0.809 0.955 67.108 73.007 60.000 0.992 7 4315 0.873 0.967 74.122 77.524 72.000 0.995 8 5530 0.888 0.975 83.911 84.900 78.447 0.998 9 4627 0.845 0.964 76.755 81.302 70.214 0.994 10 5227 0.861 0.969 81.580 99.725 77.201 0.998 11 4908 0.859 0 .971 79.051 83.487 71.847 0.996 12 4595 0.871 0.969 76.489 79.177 72.125 0.996 13 3690 0.864 0.963 68.544 81.345 65.069 0.994 14 4379 0.831 0.962 74.669 77.782 67.469 0.995 15 4568 0.878 0.972 76.264 77.466 72.615 0.996 16 5306 0.886 0.974 82.194 84.5 04 74.465 0.997 17 6377 0.875 0.973 90.108 94.303 85.000 0.998 18 3047 0.787 0.936 62.286 66.129 53.310 0.984 19 3579 0.838 0.960 67.505 70.725 61.555 0.993 20 4080 0.861 0.971 72.075 76.276 67.067 0.996 21 4751 0.889 0.978 77.776 81.025 73.335 0.997 22 2443 0.770 0.919 55.772 65.788 45.891 0.959

PAGE 70

70 Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 23 4030 0.839 0.962 71.632 77.897 63.000 0.986 24 3298 0.702 0.913 64.801 74.733 57.428 0.956 25 3036 0.814 0.962 62.174 70.774 48.662 0.977 26 2775 0.766 0.938 59.441 64.008 50.220 0.982 27 2985 0.826 0.957 61.649 71.197 49.000 0.982 28 4800 0.398 0.806 78.176 127.319 15.000 0.871 29 4261 0.911 0.983 73.656 75.432 69.296 0.997 30 3469 0.889 0.974 66.460 68.000 60.415 0.994 31 4367 0.854 0.959 74.567 78.854 68.731 0.995 32 5049 0.842 0.964 80.178 85.802 71.421 0.996 33 4322 0.813 0.950 74.182 77.833 68.066 0.992 34 5128 0.888 0.978 80.803 84.149 77.000 0.998 35 4616 0.904 0. 980 76.663 79.404 70.000 0.997 36 4084 0.902 0.977 72.110 74.813 67.186 0.996 37 4727 0.906 0.980 77.580 82.006 70.093 0.999 38 3338 0.870 0.962 65.193 68.000 59.414 0.995 39 4467 0.884 0.971 75.416 77.782 71.701 0.997 40 2733 0.897 0.969 58.990 62.65 0 51.971 0.995 41 1 0.000 1.000 1.128 1.000 1.000 1.000 42 6 0.000 1.000 2.764 3.162 0.000 1.000 43 1 0.000 1.000 1.128 1.000 1.000 1.000 44 4729 0.714 0.921 77.596 90.272 65.215 0.968 45 4824 0.890 0.974 78.372 80.430 74.061 0.997 46 2902 0.834 0.95 7 60.786 65.795 54.083 0.990 47 2164 0.648 0.889 52.491 71.000 38.053 0.933 48 4517 0.848 0.955 75.837 80.281 69.311 0.995 49 282 0.924 0.962 18.949 21.095 16.000 0.958 50 4588 0.874 0.971 76.431 82.006 70.000 0.997 51 3586 0.867 0.968 67.571 69.462 6 2.642 0.993 52 3787 0.839 0.956 69.439 74.000 65.069 0.993 53 3838 0.762 0.939 69.905 78.390 56.000 0.963 54 194 0.820 0.878 15.717 16.401 10.050 0.941 55 228 0.897 0.926 17.038 19.026 13.000 0.951 56 4172 0.876 0.971 72.883 76.322 68.593 0.995 57 32 61 0.914 0.981 64.436 67.535 58.694 0.998 58 4913 0.875 0.966 79.091 81.123 73.000 0.996 59 3548 0.743 0.941 67.212 74.330 59.464 0.987 60 3736 0.814 0.953 68.970 72.567 64.885 0.993 61 3405 0.786 0.945 65.844 73.824 56.223 0.985 62 4473 0.888 0.973 7 5.467 76.851 70.178 0.998 63 4218 0.874 0.966 73.284 75.664 68.680 0.996

PAGE 71

71 Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 64 3295 0.827 0.958 64.771 69.635 56.303 0.993 65 4290 0.752 0.941 73.907 81.541 70.328 0.983 66 5496 0.876 0.974 83.652 85.006 78.492 0.998 67 4957 0.875 0.974 79.445 82.620 71.386 0.996 68 3969 0.858 0.962 71.088 74.330 66.888 0.995 69 4130 0.915 0.981 72.515 75.073 67.469 0.998 70 3694 0.806 0.944 68. 581 75.180 58.181 0.983 71 3739 0.870 0.965 68.997 72.560 64.661 0.996 72 3559 0.829 0.959 67.316 71.176 61.351 0.994 73 4346 0.823 0.952 74.387 81.566 60.959 0.984 74 5762 0.893 0.977 85.653 87.664 77.414 0.998 75 3866 0.782 0.944 70.159 74.967 65.85 6 0.992 76 4634 0.910 0.980 76.813 80.056 72.125 0.998 77 5324 0.874 0.977 82.333 84.012 77.337 0.998 78 5684 0.874 0.973 85.071 89.196 80.056 0.997 79 3603 0.791 0.946 67.731 73.824 61.620 0.992 80 4271 0.863 0.963 73.743 78.892 65.795 0.995 81 3657 0.893 0.970 68.237 72.090 61.612 0.993 82 3762 0.898 0.977 69.209 72.945 63.000 0.992 83 1785 0.727 0.950 47.673 66.189 30.000 0.979 84 4225 0.875 0.969 73.345 76.059 67.476 0.996 85 3709 0.841 0.962 68.720 72.471 65.192 0.995 86 3070 0.845 0.957 62. 521 66.068 55.732 0.990 87 3711 0.884 0.973 68.739 70.937 63.820 0.997 88 4249 0.886 0.972 73.553 79.076 67.417 0.996 89 3635 0.902 0.978 68.031 68.884 65.000 0.996 90 4656 0.908 0.977 76.995 78.518 71.589 0.997 91 4659 0.901 0.977 77.020 81.939 70.72 5 0.997 92 4715 0.915 0.981 77.481 80.056 71.021 0.996 93 5676 0.855 0.964 85.011 89.320 76.577 0.996 94 3556 0.799 0.948 67.288 69.065 63.411 0.991 95 3560 0.792 0.948 67.326 71.589 59.034 0.987 96 4017 0.901 0.978 71.516 72.835 68.154 0.997 97 4072 0.860 0.970 72.004 75.186 64.382 0.991 98 5217 0.900 0.977 81.501 85.288 74.277 0.997 99 3914 0.878 0.967 70.594 72.139 65.765 0.995 100 3587 0.874 0.969 67.580 71.197 61.000 0.994 101 4602 0.839 0.966 76.547 81.688 69.000 0.995 102 4350 0.829 0.957 74.422 76.322 68.096 0.994 103 4707 0.830 0.965 77.415 84.900 66.219 0.990 104 4073 0.907 0.980 72.013 74.000 68.593 0.998

PAGE 72

72 Number Area (pixel 2 ) Circularity Convexity Fruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 105 2951 0.799 0.941 61.297 65.000 55.154 0.981 106 4626 0.881 0.972 76.746 79.630 69.376 0.996 107 3294 0.755 0.920 64.761 71.840 50.636 0.974 108 3403 0.895 0.972 65.824 67.268 61.057 0.995 109 4614 0.891 0.977 76.647 80.306 70.228 0.997 110 3851 0.784 0.947 70.023 75.822 61.522 0.990 111 2584 0.794 0.966 57.359 66.189 44.000 0.972 112 1223 0.779 0.945 39.461 50.160 27.659 0.982 113 4660 0.874 0.974 77.028 90.554 70.612 0.996 114 4746 0.866 0.970 77.735 79.120 71.505 0.997 115 3986 0.856 0.960 71.24 0 73.164 66.468 0.995 116 3257 0.917 0.986 64.397 66.287 58.000 0.998 117 3695 0.893 0.977 68.590 82.037 63.789 0.997 118 1914 0.647 0.901 49.366 52.240 41.785 0.964 119 1840 0.741 0.953 48.402 63.008 34.132 0.961 120 3696 0.853 0.974 68.600 88.238 58 .941 0.984 121 3565 0.882 0.969 67.373 80.610 63.253 0.994 122 4478 0.889 0.973 75.509 89.828 71.386 0.997 123 4883 0.898 0.977 78.849 82.377 71.063 0.997 124 3924 0.895 0.977 70.684 72.719 63.506 0.995 125 5071 0.870 0.971 80.353 83.433 72.069 0.996 126 5384 0.812 0.958 82.796 88.363 75.133 0.987 127 4244 0.862 0.961 73.509 75.584 65.000 0.992 128 4024 0.842 0.961 71.579 74.653 65.307 0.995 129 3593 0.768 0.946 67.637 70.605 59.540 0.989 130 621 0.695 0.904 28.119 38.000 20.000 0.897 131 1932 0. 639 0.886 49.597 62.936 36.069 0.921 132 4229 0.859 0.964 73.379 75.690 65.803 0.995 133 3452 0.859 0.960 66.296 69.181 62.000 0.993 134 4031 0.814 0.958 71.641 76.118 66.400 0.994 135 3574 0.889 0.971 67.458 70.000 61.855 0.997 136 3701 0.887 0.972 6 8.646 70.831 63.600 0.997 137 2680 0.856 0.975 58.415 65.620 45.354 0.974 138 1824 0.638 0.887 48.191 60.216 33.956 0.931 139 1021 0.754 0.911 36.055 46.000 24.000 0.945 140 1271 0.736 0.925 40.228 51.245 30.414 0.947 141 1888 0.719 0.939 49.029 65.62 0 33.000 0.979 142 4495 0.883 0.972 75.652 81.302 70.725 0.996 143 3656 0.864 0.968 68.227 70.725 62.008 0.995 144 4037 0.885 0.974 71.694 73.246 65.437 0.997 145 3083 0.898 0.972 62.653 65.765 59.008 0.995

PAGE 73

73 Number Area (pixel 2 ) Circularity Convexity F ruit diameter (pixels) Max. Diameter (pixels) Min. Diameter (pixels) Ellipticity 146 3094 0.816 0.951 62.765 65.795 55.000 0.991 147 4546 0.799 0.950 76.080 81.320 69.836 0.992 148 4261 0.810 0.956 73.656 78.518 66.491 0.990 149 3577 0.920 0.979 67.486 70.831 63.008 0.998 150 4728 0.825 0.957 77.588 83.006 69.260 0.995 151 3335 0.823 0.956 65.163 68.593 58.310 0.991 152 3267 0.914 0.981 64.496 68.942 59.548 0.997 153 3639 0.909 0.975 68.068 71.281 65.742 0.995 154 4322 0.893 0.975 74.182 74.953 70. 292 0.997 155 4139 0.854 0.965 72.594 75.133 68.154 0.995 TOTAL of TRIAL #3 583687 10471.110 Number of fruit: 155 OVERALL TOTAL 16837761 291627.063 Total Num. of fruit: 5384

PAGE 74

74 LIST OF REFERENCES Aleixos, N., J. Blasco, E. Molto, and F. Navarron. 2000. Assessment of citrus fruit quality using a real time machine vision system. In Proc. 15th International Conf. on Pattern Recognition 15(1): 482 485 Annamalai, P., W. S. Lee, an d T. Burks. 2004. Color vision system for estimating citrus yield in real time. ASAE Paper No. 043054. St. Joseph. Mich.: ASAE. Balastreire, L. A., J. K. Schueller, J. R. Amaral, J. C. G. Leal, and F. H. R. Baio. 2002. Coffee yield mapping. ASAE Paper No. 021166. St. Joseph. Mich.: ASAE. Beucher, S., and C. Lantuejoul. 1979. Use of Watersheds in contour detection. In Proc. Inter. Workshop on Image Processing. Rennes, France: CCETT/IRISA, Rennes, France 1979: 12 21. Beucher S., and F. Meyer. 1992. The morp hological approach to segmentation: the watershed transformation. In Dougherty, E. (ed.), Mathematical morphology in image processing Chap. 12, pp. 433 481. Marcel Dekker, Inc., New York. Blasco, J., N. Aleixos, and E. Molto. 2003. Machine vision system f or automatic quality grading of fruit. Biosystems Engineering 85(4), 415 423 doi:10.1016/S1537 5110(03)00088 6. Bora, G. C., M. R. Ehsani, R. Goodrich, and G. Michaels. 2006. Field evaluation of a citrus fruit pick up machine. ASAE Paper No. 061141. St. Jo seph. Mich.: ASAE. Borgelt, S. C., and K. A. Sudduth. 1992. Grain flow monitoring for in field yield mapping. ASABE Paper No. 92 1022. St. Joseph. Mich.: ASAE. Breunig, M., H. P. Kriegel, R. Ng, and J. Sander. 2000. LOF: Identifying density based local out liers. Proc. SIGMOD Brown, G. K. 2002. Mechanical harvesting systems for the Florida citrus juice industry. ASABE Paper No. 021108. St. Joseph, Mich.: ASABE. Buie, Jr., A. P. 1965. Fruit picking and transporting device. U.S. Patent No. 3165880. Buker, R. S., J. P. Syvertsen, J. K. Burns, F. M. Roka, W. M. Miller, M. Salyani, and G. K. Brown. 2004. Mechanical harvesting and tree health. EDIS. UF/IFAS. FL. Campbell, R. H., S. L. Rawlins, and S. Han. 1994. Monitoring methods for potato yield mapping. ASAE Pa per No. 94 1584. St. Joseph, Mich.: ASAE. Casasent, D., A. Talukder, P. Keagy, and T. Schatzki. 2001 Detection and segmenta tion of items in X ray imagery. Trans. ASAE 44(2): 337 345. Chinchuluun, R., and W. S. Lee. 2006. Citrus yield mapping system in nat ural outdoor scenes using the Watershed transform. ASAE Paper No. 063010. St. Joseph, Mich.: ASAE.

PAGE 75

75 Digabel, H., and C. Lantuejoul. 1978. Iterative algorithms. Proc. 2 nd European Symp. Quantitative Analysis of Microstructures in Material Science, Biology an d Medicine Caen, France, Oct. 1977. J. L. Chermant, Ed. Stuttgart, West Germany: Riederer Verlag, 1978: 85 99. Duda., R. O., P. E. Hart, and D. G. Stork. 2000. Pattern Classification 2 nd ed. New York, N. Y.: John Wiley and Sons. Florida Agricultural Stat istics Service (FASS). 2005. Washington, D.C.: USDA. Available at: www.nas s.usda.gov/fl. Accessed 29 May 2006 Flood, S. J., 2006. Design of a robotic citrus harvesting end effector and force control model using physical properties and harvesting motion te sts. PhD diss. Ganesville, FL: University of Florida, Agricultural and Biological Engineering department. Forsyth, D. A., and J. Ponce. 2001. Computer Vision: A Modern Approach Prentice Hall, 2001. Upper Saddle River, NJ. Futch, S. H., and F. M. Roka. 200 5. Continuous canopy shake mechanical harvesting systems. EDIS. UF/IFAS. FL. Futch, S. H., J. D. Whitney, J. K. Burns, and F. M. Roka. 2005. Harvesting: From manual to mechanical. EDIS. UF/IFAS. FL. Grift, T., R. Ehsani, K. Nishiwaki, C. Crespi, and M. Min 2006. Development of a yield monitor for citrus fruits. ASABE Paper No. 061192. St. Joseph, Mich.: ASABE. Hannan, M. W., and T. F. Burks. 2004. Citrus developments in automated citrus harvesting. ASABE Paper No. 043087. St. Joseph, Mich.: ASABE. Hodges, A., E. Philippakos, D. Mulkey, T. Spreen, and R. Muraro. 2001. Economic impact of information report 01 2. Juste, F., I. Fornes, F. Pla, and F. Sevila. 1992. An appr oach to robotics harvesting of citrus in Spain. In Proc. Intl. Soc. Citriculture : 7 th Intl. Citrus Congress, 3:1014 1018. Riverside, Cal.: International Society of Citriculture. Kane, K. E., and W. S. Lee. 2006. Spectral sensing of different citrus variet ies for precision agriculture. ASABE Paper No. 061065. St. Joseph, Mich.: ASABE. Kender, W. J., U. Hartmond, M. Salyani, J. K. Burns, and J. D. Whiney. 1999. Injurious effects of metsulfuron methyl when used to stimulate abscission of Hamlin oranges. HortS cience 34(5): 904 907. Koning, C. T. J., L. Speelman, and H. C. P. Vries. 1994. Size grading of potatoes: Development of a new characteristic parameter. J. Agric. Eng Res. 57, 119 128.

PAGE 76

76 Lee, W. S., J. K. Schueller, and T. F. Burks. 2005. Wagon based silag e yield mapping system. In Agricultural Engineering International: the CIGR E. Journal Vol. VII. Manuscript IT 05 003. Lee, W. S., and D. C. Slaughter. 2004. Recognition of partially occluded plant leaves using a modified Watershed algorithm. Trans. ASAE 47(4): 1269 1280. Leemans, V., and M. F. Destain. 2004. A real time grading method of apples based on features extracted from defects. In Journal of Food Engineering 61, 83 89. Leemans, V., H. Magein, and M. F. Destain. 2002. On line fruit grading accordin g to their external quality using machine vision. Biosystems Engineering 83(4), 397 404 doi:10.1006/bioe.2002.0131. MacArthur, D., J. K. Schueller, and W. S. Lee. 2006. Remotely piloted helicopter citrus yield map estimation. ASABE Paper No. 063096. St. Jo seph, Mich.: ASABE. Macidull, J. C. 1973. Fruit harvesting apparatus. U.S. Patent No. 3756001. Marchant, J.A., and C. M. Onyango. 2003. Comparison of a Bayesian classifier with a multilayer feed forward neural network using the example of plant/weed/soil d iscrimination. In Computers and Electronics in Agriculture 39, 3 22. Mehta, S. S. 2007. Vision based control for autonomous robotic citrus harvesting. MS thesis. Ganesville, FL: University of Florida, Agricultural and Biological Engineering department. Mus cato, G., M. Prestifilippo, N. Abbate, and Rizzuto. 2005. A prototype of an orange picking robot: past history, the new robot and experimental results. Ind. Robot 32(2): 128 138. Perry, C. D., G. Vellidis N. Wells, R. Hill, A. Knowlton, E. Hart, and D. Da les. 2005. Instantaneous accuracy of cotton yield monitors, Instantaneous accuracy of cotton yield monitors. In Proc. Beltwide Cotton Conf ., 486 504, New Orleans, LA, National Cotton Council, Memphis, TN. Regunathan, M., and W. S. Lee. 2005. Citrus fruit i dentification and size determination using machine vision and ultrasonic sensors. ASAE Paper No. 053017. St. Joseph. Mich.: ASAE. Sanders, K. F. 2005. Orange harvesting systems review. Biosystems Eng 90(2): 115 125. Schueller, J. K., and Y. H. Bae. 1987. Spatially attributed automated automatic combine data acquisition. In Computers and Electronics in Agriculture 2: 119 127. Schueller, J. K., J. D. Whitney, T. A. Wheaton, W. M. Miller, and A. E. Tunner. 1999. Low cost automatic yield mapping in hand harve sted citrus. In Computers and Electronics in Agriculture 23(2): 145 153. Searcy, S. W., J. D. Schueller, Y. H. Bae, S. C. Borgelt, and B. A. Stout. 1989. Mapping of spatially variable yield during grain combining. Trans. ASAE 32(2): 826 829.

PAGE 77

77 Shahin, M. A., and S. J. Symons. 2001. A machine vision system for grading lentils. Canadian Biosystems Engineering 43: 7.7 7.14. Sui, R. X., and J. A. Thomasson. 2001. Field testing of Mississippi State University cotton yield monitor. In Proc. Beltwide Cotton Conf ., D. A. Richter ed., 332 342. Memphis, TN: Nat. Cotton Council Am. Vellidis, G., C. D. Perry, J. S. Durrence, D. L. Thomas, R. W. Hill, C. K. Kvien, T. K. Hamrita, and G. C. Rains. 2001. The peanut yield monitoring system. Trans. ASAE 44(4):775 785. Vincent, L. 1991. Recent developments in morphological algorithms. Proc. 8 th Inter. Congress for Stereology Irvine CA, August 1991. Welte, H. F. 1990. Forecasting harvest fruit size during the growing season. Acta Hort (ISHS) 276:275 282. Whitney, J. D., Q. Lin g, W. M. Miller, and T. A. Wheaton. 2001. A DGPS yield monitoring systems for Florida citrus. Applied Engineering in Agriculture 17(2): 115 119. Yoshida, J., S. Okuyama, and H. Suzuki. 1988. Fruit harvesting apparatus with television camera and monitor. U. S. Patent No. 4519193.

PAGE 78

78 BIOGRAPHICAL SKETCH Radnaabazar Chinchuluun was born in Bulgan, Mongolia in 1978. He received his bachelor degree in computer hardware and software engineering from Mongolian University of Science and Technology in 2000. After his undergraduate degree, he started working at the same university as a lecturer and computer engineer. Radnaabazar joined the Master of Engineering degree program in Agricultural and Biological Engineering, and Computer and Information Science and Engineering at the University of F a graduate research assistant at the Precision Agricultural Laboratory under supervision of Dr. Won Suk Lee. During this time, he has undertaken two major citrus yield mapping projects.


xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID E20101114_AAAAEK INGEST_TIME 2010-11-14T13:37:36Z PACKAGE UFE0021209_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
FILES
FILE SIZE 1053954 DFID F20101114_AABMQN ORIGIN DEPOSITOR PATH chinchuluun_r_Page_40.tif GLOBAL false PRESERVATION BIT MESSAGE_DIGEST ALGORITHM MD5
4482a34a03536658580f18940218e0af
SHA-1
bbd97414b1db20794a14d4c5577dddc161c2a939
32075 F20101114_AABMLQ chinchuluun_r_Page_55.jpg
11c58eea9050473fa39484072325f6e8
9ac955635f9866d3b64727b25ac6bf8169142f68
1782 F20101114_AABMVL chinchuluun_r_Page_31.txt
c8d1c6f6b4fe102e93d71509d6aafd61
912d16ad9f0323e86dcc878446899112f04ecea7
F20101114_AABMQO chinchuluun_r_Page_41.tif
7eb867a37c15a64b44786c86dfe2a4eb
8397dd649d1ef57bd4d75ff1a74b74cb36005438
88110 F20101114_AABMLR chinchuluun_r_Page_56.jpg
03a16254afa79a63a45ac67ec97cd467
747fce6b3eaa6e488e8a898d8982c6b93b27bf22
1456 F20101114_AABMVM chinchuluun_r_Page_32.txt
557de27142fdf93ebcef4201fe2cce51
a1e8976d58a48de2eab30084f41f1a035f49fbfd
25271604 F20101114_AABMQP chinchuluun_r_Page_42.tif
7fd1f9968301a1fdc3e36d5106a6b3db
e10b907d4362b347a9fbac9360681abdf43d8a62
63645 F20101114_AABMLS chinchuluun_r_Page_57.jpg
f9ea020ab496772421b6cdac3ba82fd5
0a5e130a6d3400346d0c2d997168e03d072ba872
1878 F20101114_AABMVN chinchuluun_r_Page_34.txt
1136d981e74ae693555aba625805f704
324d1c2be178ff4c40cda05299c4b0e1f263da5f
F20101114_AABMQQ chinchuluun_r_Page_43.tif
1802bcd24115b671d9402f8b350aaa4c
c6355707d3c6b99a7cc20169ec5fbf8eaeec2b63
56814 F20101114_AABMLT chinchuluun_r_Page_59.jpg
1e3941a6fa6a95f30483152728a889bd
edd3d99137c01eda481f55c8cbde77e847d21218
833 F20101114_AABMVO chinchuluun_r_Page_35.txt
98ae0c64fb63e09c4e8ed3cd6b8476fe
404d54a862a5a4f65aa57409f13c14e99baff532
F20101114_AABMQR chinchuluun_r_Page_44.tif
9f87785cdd6d58f82de0e90d32c9a1f3
cccf2bd5fe4e50524073c69f487838c6f11d129f
84631 F20101114_AABMLU chinchuluun_r_Page_60.jpg
81e66cefa5628bff0e0400d7c44b9977
8cc7f09a80711eff24721e4ebeb4861d3adb37dc
1538 F20101114_AABMVP chinchuluun_r_Page_36.txt
d4cc647941152ddf7a118551fa5bfebb
36f126e170b57da16ed40149d69a0e0385afc286
F20101114_AABMQS chinchuluun_r_Page_45.tif
eb765f3d28f012aa1ed0ef44f201ef99
e01d85803e99c81c46574425b5fae09d192237f8
36363 F20101114_AABMLV chinchuluun_r_Page_61.jpg
97dffd1c807b5a87e1fcc01ac56910ad
5b80bacb3ab1e1e570164993b1ddf53efb7c1b38
1331 F20101114_AABMVQ chinchuluun_r_Page_38.txt
4a8d2259f9cce8770085779d784f0823
d25bb4b631173f1aa22fb38cbff7faf6f1647ce8
F20101114_AABMQT chinchuluun_r_Page_47.tif
1f079961dac385928cd7213cfb3f67c6
3c10ce9b5f9a8afe264cf36f0e6c34f3f97a7d7e
55273 F20101114_AABMLW chinchuluun_r_Page_62.jpg
1334467653f25294af9fd02a30c666ed
7f9a28aa5c4e5b1479f77fb6680bf4aa6d2de838
F20101114_AABMQU chinchuluun_r_Page_48.tif
ce0992c6c77ca48f511ff9806015af9b
a47c37249855fdbac65e765c9a53856f5b88c2b4
51164 F20101114_AABMLX chinchuluun_r_Page_63.jpg
07d006f1714cf7e95e7321732ab67ed9
07f20052debb688674ff33abfa608c08f873ef61
1792 F20101114_AABMVR chinchuluun_r_Page_40.txt
a18adb7164a49b24a83cc5300e4275e2
fcc16299ecda856935857ea8f921f1413f4b0d0c
F20101114_AABMQV chinchuluun_r_Page_49.tif
f3bbdc8cb54cc97f24f8d14b1acd9022
3481e12e8eccde2edaca6ae81027bb14a38dd141
4211 F20101114_AABMJA chinchuluun_r_Page_33.QC.jpg
270f836e515495782b66ef3a3555ee9a
2cc9f993fe0e41d3308228673cc3ee7fe2af38ec
53899 F20101114_AABMLY chinchuluun_r_Page_64.jpg
e6c79172a8f775446eaa4b031aa6f5e1
2b32d56cde934a5b6113affab86731b3c1d53f94
680 F20101114_AABMVS chinchuluun_r_Page_41.txt
4a77e39be4f35a76c35a945e8e9fa015
1ba4144eb098ab697b5f5bd732e8c626a5d0845f
F20101114_AABMQW chinchuluun_r_Page_50.tif
f56426c2f8c134a2e4f80b91d5852d73
d590435c1db05173d5654494a68aa1604b9af94a
18173 F20101114_AABMJB chinchuluun_r_Page_05.QC.jpg
591f521702825e62c0822212e8f13e23
fc2b02a9e9a70ed40da69ea72100cf0fc4b1ee1e
39462 F20101114_AABMLZ chinchuluun_r_Page_65.jpg
fef14cb4922da26801d85a25ec9390ce
6f4529a48349c9ccafaf59618e332d7e642ff809
1595 F20101114_AABMVT chinchuluun_r_Page_42.txt
2c92b3e67859595c6b7c5358dcaeb46a
3110abb1b61ea010e9d48b53723d3baab393edd7
F20101114_AABMQX chinchuluun_r_Page_51.tif
8354db0b0e7c6d83bad1fca14a82b597
e1fbaf365f9f053605195ec192b6949eb36a0b1b
28289 F20101114_AABMJC chinchuluun_r_Page_56.QC.jpg
eafa5b35cd3b793004ef25fe5369e9e0
da2318dc0418f71e5f2992fb508dd2f4eb8e464e
1114 F20101114_AABMVU chinchuluun_r_Page_43.txt
da4a559a7d511df1037fcae9d93d0cfc
a449da55bce78ad62e363473abf004dc597edbd0
F20101114_AABMQY chinchuluun_r_Page_52.tif
9c383463a7fcc92b2dc967067da472e1
29364432a546841128842ec6908b3e5127448e2d
54996 F20101114_AABMJD chinchuluun_r_Page_58.jpg
c98a582b35eb78272cd653f99d380834
cf4c74ae932e1c46500f144dcfafa140447df709
1225 F20101114_AABMVV chinchuluun_r_Page_44.txt
ca3721ae1d1d34e8d51e6029081858ec
5df3af55222c422c4d4562aa3a1e40cc50e78292
50413 F20101114_AABMOA chinchuluun_r_Page_41.jp2
0062c0de9b4f11c00dd5421501c66f01
d57a41f4f970490014bc709fd16bb6c8cfbaade4
F20101114_AABMQZ chinchuluun_r_Page_53.tif
08a9f9cb29ccbcbfe16042241d140d23
5411e292032af182f73508f606bbed986c75b04a
F20101114_AABMJE chinchuluun_r_Page_11.tif
f11db17dcb5080891e11d35f4f45e0a4
01e78c6e802d2782b0e967a3eb06298d656c6717
1458 F20101114_AABMVW chinchuluun_r_Page_45.txt
2979680573d74b2cbabcbce0ac2dea3f
258724952691c19d502664a7125b08e7a6081761
1051972 F20101114_AABMOB chinchuluun_r_Page_42.jp2
13a722f038491cabd7aba24929151d24
2420818971046f91e8ec2576d14d842f916c99ff
1266 F20101114_AABMVX chinchuluun_r_Page_46.txt
b01353c9891975f398abe6f2e566f13d
3e37d90e99f908a51d1e39fd7f70db40418d13e2
1051976 F20101114_AABMOC chinchuluun_r_Page_43.jp2
2a1bf66286a36d891fc1ca9e3957eca4
f51471de54211c9bacc0f292738e12cc60ecb04b
271 F20101114_AABMJF chinchuluun_r_Page_33.txt
04ad641f4ccafc3e3d11984f636951ad
98ef9289a3c9b36009438e2a2c237046c3bae9b4
31842 F20101114_AABMTA chinchuluun_r_Page_38.pro
eb0fa801a99477060eb76e1dc6f05746
4f9accfa22bfe2cd190ae2e7af340670ece244a1
1257 F20101114_AABMVY chinchuluun_r_Page_47.txt
25db170550452b9ba59432871becd0a5
5e2d5dc1e419b5cb80f1f970b983ca2fc80aed95
1051920 F20101114_AABMOD chinchuluun_r_Page_44.jp2
25e8e104107b7481f0bb5fd86b5e5198
c785ca2471fec248cb1dd75c2a1df5425e243767
6589 F20101114_AABMJG chinchuluun_r_Page_67thm.jpg
d9899db56f04d4c9a3b825e26a564ca4
3a2c63345e15c78fe8d20ef635493191fee75adb
21828 F20101114_AABMTB chinchuluun_r_Page_39.pro
fc052b5f23c553ded829ce1159e869cb
e20a258798ce4a5e6b0413d7757e12e5898715d6
873 F20101114_AABMVZ chinchuluun_r_Page_48.txt
2d468ba773524cc495794ad6f2bed36c
5447700da68d69be712e310fafde22b9c39ff391
1051966 F20101114_AABMOE chinchuluun_r_Page_45.jp2
54c689dc9aeb873cabb17cff041f6abf
a3847085105c37f0f72d44d406d14743e64efab0
1051979 F20101114_AABMJH chinchuluun_r_Page_56.jp2
50207dd6a00abf7c93666ecdf487aa42
3719d28156a0d1f2867b0bf4d952c55a1322fe8b
37173 F20101114_AABMTC chinchuluun_r_Page_40.pro
21981fe769fae4d484db1ef72ee13b71
c3d0c669156e2427a2e696a4ce6507c2d10fe3ba
77142 F20101114_AABMOF chinchuluun_r_Page_46.jp2
71baa3009c482e68479a709e724e0e87
53aa479e14889c7051d789220851f1225ebd9dd3
23599 F20101114_AABNAA chinchuluun_r_Page_44.QC.jpg
f5d2d3345e968dd84d192461ebc90163
eb5ac9ee435a07425a6d0f804da5d89bbaa384be
26598 F20101114_AABMYA chinchuluun_r_Page_14.QC.jpg
d891b3c5fcef0ef45f3d18dc86518ba2
f1f92ab797f40dab3effc71df97965265381e392
6671 F20101114_AABMJI chinchuluun_r_Page_68thm.jpg
adc5f330818d444318b2baef3c0a399d
0962877a688840ba2965322d6288d815f2b2186b
14676 F20101114_AABMTD chinchuluun_r_Page_41.pro
bd01bc0b0da4df9c471bbc808b36dbfb
c20d54b80f14a578ad55858a16e1f330aee3b9e4
64423 F20101114_AABMOG chinchuluun_r_Page_47.jp2
7a0c2106d5b04f6ae7ceb1338bfd2ae5
107c4d6bdb40240d5eb015dabbcd4432c8e30ecd
5521 F20101114_AABNAB chinchuluun_r_Page_45thm.jpg
a16432bf3d7f6ea92400f715ddc4ff6c
1655d90915d570c64019a4713535de59c0cc8603
2871 F20101114_AABMYB chinchuluun_r_Page_15thm.jpg
41862a797d6f989998fd828d52ef9936
f085d5ae689375ae5db071efe650d20ebe6825c2
117161 F20101114_AABMJJ chinchuluun_r_Page_66.jp2
881828a084197a7aa3ce11738deb1b1b
24b1db9251df523371619ad8be9d9b55bb4e6d98
26201 F20101114_AABMTE chinchuluun_r_Page_43.pro
81eb7d6697deed73b4186afcaee014b1
642db5f1814267071a1188cf934f92f91f8f5c83
601478 F20101114_AABMOH chinchuluun_r_Page_49.jp2
70329a3d817a5e02d2a4c57aad39c2fe
099808a50516fb723d11a0f692fd65918b1ee7d7
19692 F20101114_AABNAC chinchuluun_r_Page_45.QC.jpg
47267916b9ab525f3990437aa6cbfa34
aff4ec1e2926cb83d9221677a9766b8e3798b2d3
11510 F20101114_AABMYC chinchuluun_r_Page_15.QC.jpg
9f3dd2cf8f038cac83150ef18bb4f136
4503205455cbb434f5a445d43db16932e7843327
F20101114_AABMJK chinchuluun_r_Page_03.tif
4b3e087a848161c1bef20e76c43d8180
e90d77d72a91ee8c35acc3ba029d74f0960d2be9
26059 F20101114_AABMTF chinchuluun_r_Page_44.pro
b77e45979c4c41ed01ff04e7a631db1e
b8097c0da53b37a7af7a4efa410d4b25387512dc
719973 F20101114_AABMOI chinchuluun_r_Page_50.jp2
2ed8a99d6cbe971b17c00a0daced270c
c4b6ee635689c6ce7fcaf990be5335877bf53696
5476 F20101114_AABNAD chinchuluun_r_Page_46thm.jpg
feeae7d3887235148439401d55e3b889
7acb241969738c47fefc6d40427581123d020f77
5692 F20101114_AABMYD chinchuluun_r_Page_16thm.jpg
63591295568eed5a6e000ac049e055ca
6434cde94b477d29fa78e8af4c27362dcca7cf09
71192 F20101114_AABMJL chinchuluun_r_Page_11.jp2
2d2975bdc14d29fc2d2ccc4ceef0521a
e8866d6d540f77d7ff3731109c317af74409e899
22510 F20101114_AABMTG chinchuluun_r_Page_45.pro
e74409977fce3cd818416b5734a612d7
25a0d2b4c49f3d65c4ce5a5bf724c8e50f70a082
417632 F20101114_AABMOJ chinchuluun_r_Page_51.jp2
55fa177d62c84e00a7843c3ad189dd57
63728113e868a6a834fb1b52cbb99f67c1829a6b
19337 F20101114_AABNAE chinchuluun_r_Page_46.QC.jpg
dfc7811ca313facee8609d5ea6d29b93
b8bf6e90aee018cec2b492acb02f347a2089ef9b
25027 F20101114_AABMYE chinchuluun_r_Page_16.QC.jpg
19d0bdc3f54b70462cd17dbced4ca76f
7c50b647e939aa7c5801976f484cdd8eff785c72
6682 F20101114_AABMJM chinchuluun_r_Page_19thm.jpg
9e884edc65b3c686ecf1beb327c39102
d11ac2a38872d3b2c6ade7d82c8f296b176a76fa
28078 F20101114_AABMTH chinchuluun_r_Page_46.pro
7851f02b8050e2e80976faf36a7dc876
c8d9638efc5e07f8abe74885bb0d194692b3bdfb
4177 F20101114_AABNAF chinchuluun_r_Page_47thm.jpg
7affdf5fc0a470cce33afc72dcbf6bb8
8f54b8997bbfd3e89894757e275f9c64bf68daaf
6648 F20101114_AABMYF chinchuluun_r_Page_17thm.jpg
ff438e14b628f85ae1e477ac186de555
a18b19ce8af446a3359e3a0f5d071ab05f007efa
28447 F20101114_AABMJN chinchuluun_r_Page_21.QC.jpg
974cd75aa667b9d953375eb39b19d578
8adeaa69bab8c433d2fc8a4f54f92b6b6f9f9cd3
26828 F20101114_AABMTI chinchuluun_r_Page_47.pro
267d42a679a14b94908c9ec8ab18d9a5
5a072bf78943e7753043736c76ec27617d6e478d
92174 F20101114_AABMOK chinchuluun_r_Page_52.jp2
d41a800b1fa88aea1a865e2b32f3eee0
061f307f2198ee3409e401bb6e3ca9a2dddf736e
15466 F20101114_AABNAG chinchuluun_r_Page_47.QC.jpg
8bd950c7c23bc49324f71d6546254452
2257822fc9ccdd54a1fdeb4a2f378e439bc61518
6615 F20101114_AABMYG chinchuluun_r_Page_18thm.jpg
c8abc04e076d1508603e1d3d2c1a2a1d
ceef8626ed25a7a69f54d046c365712e08b85b96
1142 F20101114_AABMJO chinchuluun_r_Page_65.txt
8a49aac9e00cb70e89a10fefca2cab77
012876ab532cdfb06069bea00be69fe2efff0846
20722 F20101114_AABMTJ chinchuluun_r_Page_48.pro
cf9208f85c2bd581b98b4c0aafdb44a1
9dab7719672b1556126899ff53a13e25df73a19b
1051962 F20101114_AABMOL chinchuluun_r_Page_53.jp2
81c4c13ef03096180c1d54b846c575f0
8dbe957c8dbf9f779e8fa8858619080deace376c
14260 F20101114_AABNAH chinchuluun_r_Page_48.QC.jpg
6123780100bff5ac35f8cea60591cfd5
4379511e1a7d68a1b07bb3e762fd6f562c22052d
28373 F20101114_AABMYH chinchuluun_r_Page_18.QC.jpg
b037b95d318202ba9158cb01fee54f52
1d3d3804d7fe653ea9e3ce04ac0515044486433f
20659 F20101114_AABMTK chinchuluun_r_Page_49.pro
e1e2dd0b1ae24d07ac8f3ea1d29498f4
a83bab73fb089de126f3facdd6ec17c8d27dea6c
662747 F20101114_AABMOM chinchuluun_r_Page_54.jp2
ef4b7ad2777535a8a9cd44a7ead42bd4
8d66471429dcc621f71cce354c93f6b934f00ff7
4126 F20101114_AABMJP chinchuluun_r_Page_48thm.jpg
c966ede196e0a15a321c215235072a0f
0a9f19f5ba2a82280a978cafdf83626c69f9e208
4318 F20101114_AABNAI chinchuluun_r_Page_49thm.jpg
6dfcd79778f54da85eb0b2612d0facf9
254f8d66e856fd68f1a1fa6fd40f90320a3b74ff
29024 F20101114_AABMYI chinchuluun_r_Page_19.QC.jpg
75a2511aa5bd4e6bc39a4ad210e494d4
cce8330664f61706250cc81d04d1de2baed5330a
25175 F20101114_AABMTL chinchuluun_r_Page_50.pro
037ce69219ed4f7df9399c34004a0696
501d8f01707afd52caa2832ae6574337603fa7ba
447535 F20101114_AABMON chinchuluun_r_Page_55.jp2
2c59913737ef65ba4d8a4cae15dd4fea
b4e6acc4b500a6a4c360173320e1e60e24162ac7
77848 F20101114_AABMJQ chinchuluun_r_Page_34.jpg
1f13403d2b6db892b99a155b5d583424
2eb0d96e62c5fdb56b271c413117abc4a0747607
14311 F20101114_AABNAJ chinchuluun_r_Page_49.QC.jpg
404e8f732299e4254354493820b175ab
deafdcd17cfd0bd3f1b255201e58d118d44686b1
6835 F20101114_AABMYJ chinchuluun_r_Page_20thm.jpg
5f4f0d23f6a9ef634f905852ff9f8f26
fbf61208db8e6bfd2c82460457938daf9e337e44
17189 F20101114_AABMTM chinchuluun_r_Page_51.pro
d25f77754609373a75b3010fe35b03ac
f2f99c13f69f213089384d3fd7166e3d6af55238
760310 F20101114_AABMOO chinchuluun_r_Page_58.jp2
3749ebef36f2508410e0de7951000df0
47b42143c5fae49662aa6969b15bf61fb3c87d96
F20101114_AABMJR chinchuluun_r_Page_61.tif
d8d5cac2e625333198870df5ea1bde7e
4b5579dfb5dcd802cbd68234fc780e1227163ba8
4872 F20101114_AABNAK chinchuluun_r_Page_50thm.jpg
4bd7869e519a4065df11bc940496b7c2
46764c0b9e62214c5b978926b85bc6803b4b1aa9
27725 F20101114_AABMYK chinchuluun_r_Page_20.QC.jpg
5345c5ce8319a603aca34c8774a0c698
5bd2310481ffeca9233737b55c7b2332c1e41989
42020 F20101114_AABMTN chinchuluun_r_Page_52.pro
f277f86aa807ee39174768dff23e9d57
d4d088f5d16ebf8c70ee9c6a241399b3e6c93334
74562 F20101114_AABMOP chinchuluun_r_Page_59.jp2
80debb6c0919ae7d0db4afe3385e152f
7b17d57aac3d0d04fcd7b17bc79e0ee393b404fa
2088 F20101114_AABMJS chinchuluun_r_Page_12.txt
538487470c51a71cbd1ef1dbdd74bc0b
651c18f415638623eb870dcbba85211a44ea0b14
17572 F20101114_AABNAL chinchuluun_r_Page_50.QC.jpg
3e735d154d8e7a7b1edc56e5b060633b
5e31f0937a8ef685b1956dc9be4404d3b55c5b51
6705 F20101114_AABMYL chinchuluun_r_Page_21thm.jpg
3cd08321997a01b5397439230399f315
cbd779372734b098255a0c2d3e8f08b5aa5a1b41
34442 F20101114_AABMTO chinchuluun_r_Page_53.pro
f62d0eb808b40e97bfb1ce538458549e
0c56c3b2aa9cd0cadcd455cf2d31c94b1970c390
111068 F20101114_AABMOQ chinchuluun_r_Page_60.jp2
9d2047c617068f4426b5f65df47557a5
fc0dab50b1372b67772f7cd38530b2a88a3292fb
121065 F20101114_AABMJT UFE0021209_00001.xml FULL
527b48e55b9cba33a6f05ccb645a7fde
1e1444889267ed539f20a0012cab8304884fd534
3345 F20101114_AABNAM chinchuluun_r_Page_51thm.jpg
cb789d9a8605c33771dcef169eb487ac
a336dabca61c791cf481e93b92ac11e52622691f
6620 F20101114_AABMYM chinchuluun_r_Page_22thm.jpg
29d7689fd289032e834e8da5bcb99346
7279a3606cebc244c71507ed721938cf9d82cf71
46013 F20101114_AABMOR chinchuluun_r_Page_61.jp2
b21af1bf9ff3684e54a2d635c5b91dda
4aa0d923ea3ee17502e3c25c1b8191555690097a
11938 F20101114_AABNAN chinchuluun_r_Page_51.QC.jpg
1c30f122eb3951746f03cc501e391992
490602da6d632967cfdfe9ebcc34f8f68549a2ab
27333 F20101114_AABMYN chinchuluun_r_Page_22.QC.jpg
cbd4de27e3f3a0058183b1853bbd0b98
de3c1cbdbdf61498812f8fa4cdca0ad09db05241
24468 F20101114_AABMTP chinchuluun_r_Page_54.pro
8f0e2531cd0ca31160deff2dd0599e46
e5ef74fe2cd332271600a359aed86285798fdae5
65475 F20101114_AABMOS chinchuluun_r_Page_62.jp2
1e8f695d99a395a1d58162e88dac352a
2cde6f2cb192482c735029cc3d4ac1a0c816eed2
5633 F20101114_AABNAO chinchuluun_r_Page_52thm.jpg
bcce530e33ae2a682c71d30d7c13b4de
9b157908aff340bab849bb7312b2e1a0f717f40e
6572 F20101114_AABMYO chinchuluun_r_Page_23thm.jpg
b229127a96b7985f691c044131f3e3bf
013479828cab6e193846664335293c9f8220e170
12895 F20101114_AABMTQ chinchuluun_r_Page_55.pro
a44aa52d12fc08c2dd21269ef96f0578
802a8a2f60e110eaaf6584089edd619aff545b8b
66206 F20101114_AABMOT chinchuluun_r_Page_63.jp2
58623eddeb32575564bb0460da1f775e
9a3740dc81988c3e808ab4a583d2fd6919514f1d
23241 F20101114_AABMJW chinchuluun_r_Page_01.jpg
f4bc8c64ce18e736896afc6050deb666
eabd5c65d659bd86d39f7e41405d3ee3eb41d579
7815 F20101114_AABNAP chinchuluun_r_Page_53thm.jpg
db5159b676142ffcf5d0b0525afbda59
a121a8576f9a571728e56af89d167805c21d9513
28353 F20101114_AABMYP chinchuluun_r_Page_23.QC.jpg
b4b766245c9c30d4fc639d428f5298eb
4f36b5e16d3538bb0f79e8e371551b381ed566ce
21780 F20101114_AABMTR chinchuluun_r_Page_56.pro
865df819544d5a08e2347059cadc11da
c36a19118296fe1ea786431a862b3af811da871c
67830 F20101114_AABMOU chinchuluun_r_Page_64.jp2
0c97593aa283d28f291808ed85371fdb
8b908c6281677d301a7382a7e355bf1a71758b56
4050 F20101114_AABMJX chinchuluun_r_Page_02.jpg
9ce7b869f8e8e977dd6fb9e35c498d82
47f1e688220a6898558cbfe47accbf5db7e06b63
25831 F20101114_AABNAQ chinchuluun_r_Page_53.QC.jpg
c28bc5c373b5c823be97119e88e60024
66523b237c2022d5c3afe146b0036d70c6ef0785
1272 F20101114_AABMYQ chinchuluun_r_Page_24thm.jpg
390d7179b35eb1db77a60b383f17d66f
3c8f9d44331afe1ca37c23c33d601048180145f8
32334 F20101114_AABMTS chinchuluun_r_Page_57.pro
233d4c10d909fb97da1dbf6857ea3e61
48606a82271eda98df0f0d9488b2e380751be713
124906 F20101114_AABMOV chinchuluun_r_Page_67.jp2
cd77447ffca5265d1510e7d55a5b8674
ce41187d321bfcdba6450e24082f5a17fdf0d9f9
84488 F20101114_AABMJY chinchuluun_r_Page_05.jpg
7ee8ddbcce7d0f41bfad6ae8f697d960
385f2a7312ebab6634e90b2d6bac0887c7d4d78b
4874 F20101114_AABNAR chinchuluun_r_Page_54thm.jpg
6075b48ec76017a08599bf29ae032543
b8c2490a1ea7b1022388d491936788b3f8777afd
4825 F20101114_AABMYR chinchuluun_r_Page_24.QC.jpg
db3bb5e514046dd64cdc02c3f2df5ca0
27251ab57cda56d642346913fd89876ca4030b09
25433 F20101114_AABMTT chinchuluun_r_Page_58.pro
5d9bba160fbf3dd45d24bec3a375e045
8c18db1a9e8de8c1ce07433ff2346a5b070ecb42
126193 F20101114_AABMOW chinchuluun_r_Page_68.jp2
e3a43eeaf13f78bea67f61d88d748ea4
7d88ec6273f30f9248bfbc750f49151cbcc91bad
53240 F20101114_AABMHB chinchuluun_r_Page_22.pro
b14f5be23820d1b6f6e48c0c60221b99
e48f28c03ea40d40b6c5aea64fb375dba6b20110
26020 F20101114_AABMJZ chinchuluun_r_Page_07.jpg
159687440c7a99a574da45080aec7250
5e0c961fe77ef5239be4ac9719e91fe6aa200b3f
3766 F20101114_AABNAS chinchuluun_r_Page_55thm.jpg
457d12671487f529091e6be7a4ffb601
bfa983d1673a2e4fda9b03df90ad69e22d19adb3
6222 F20101114_AABMYS chinchuluun_r_Page_25thm.jpg
72ce3a1994000bba958104335e1ba1ed
91a6ea14e3d8822dbc28b80595a32443b353c359
33558 F20101114_AABMTU chinchuluun_r_Page_59.pro
1075092454e59fd10f91161842c3e2cd
5afac417636b2953aa293a3ea39ff11323c18b8d
125590 F20101114_AABMOX chinchuluun_r_Page_69.jp2
90cb0b58d74c677456ff2bc7d277a70a
9c28eed317af4c30aeef69eede51aaa49d18a3c6
666 F20101114_AABMHC chinchuluun_r_Page_07.txt
d728ebf2a411d5726fae65ad3f487507
328fbc28ef4a3f1e30f6ef8761a3f1c473b321d7
11448 F20101114_AABNAT chinchuluun_r_Page_55.QC.jpg
343f3038ee2d6e7c20ddbc829e665f03
2d6b062428c1056d69f8134814686db9ae0e6a1f
25686 F20101114_AABMYT chinchuluun_r_Page_25.QC.jpg
5d7ea238fb910124ff87c233a52419aa
e9b3b9af2792428e69d9076201d5ee1a5d59a036
50995 F20101114_AABMTV chinchuluun_r_Page_60.pro
dc326a7894673768a26fb5ea2ed4c672
44d379472c87d8276381ec925f5718662b6c2418
110593 F20101114_AABMMA chinchuluun_r_Page_66.jpg
dcf41ce93b74e00281db920797263c91
79c739342f606c84079a1e04ea83ddef19655f64
126072 F20101114_AABMOY chinchuluun_r_Page_71.jp2
3dc62dd204095f7ef3e3c94fdd693dd3
e639e1d3f1c74fef4399a001b41f672744ca6f16
8453 F20101114_AABNAU chinchuluun_r_Page_56thm.jpg
494e34f29dce482a1c3c3d39c13cd693
3c633ceed686d722efd3e610709f9d5602ff8528
20250 F20101114_AABMTW chinchuluun_r_Page_61.pro
67c4176054bfe27f697f0e661590f648
b29de9a6368c444d158e3f4b4993b6b2859e9393
116275 F20101114_AABMMB chinchuluun_r_Page_67.jpg
045ef2bf58aea9bb2eefe2c0600b8679
664afb5bc7ab52a4aed11e76d46d7cbe7e16671b
45912 F20101114_AABMOZ chinchuluun_r_Page_73.jp2
d6981f5f77d67b69605ec10d2acc259e
580fad2638961b3e79133e9962aaa388dfc8b511
1503 F20101114_AABMHD chinchuluun_r_Page_59.txt
540b710532ccb319cefcabe7987c0c8a
6f739c4d2b7169d8bd44ea7c6c395317c625a582
5823 F20101114_AABNAV chinchuluun_r_Page_57thm.jpg
ac142c38b554533e8138e220b73d270b
7f7afcca344f3ca16d8bbff269a8d0a81ed34852
3666 F20101114_AABMYU chinchuluun_r_Page_26thm.jpg
049585ff663a021e1745f05b246f7d48
a823778396b5c02e8fd9ae80580a0ed7cdc8dc27
27628 F20101114_AABMTX chinchuluun_r_Page_62.pro
f5a8f1c48fbef52ea8eda95a517037cf
7e9709c3f50679efd4914d12b1c5b2224e34afd8
117284 F20101114_AABMMC chinchuluun_r_Page_68.jpg
301c5c8d5f67b2f782cefc44e371c82c
49adb770ee0ad583db59f910fb49a7782875aba7
F20101114_AABMHE chinchuluun_r_Page_46.tif
e59b05ad11e9f40b829446640167efad
463e45d765c8ed124c7b00e6f81b3cf1a6a7ce28
22134 F20101114_AABNAW chinchuluun_r_Page_57.QC.jpg
a06491d54555ca391c23083141cd919b
51f45d45e135f23068c6eb8167611d0b5e5351db
12874 F20101114_AABMYV chinchuluun_r_Page_26.QC.jpg
851417018f3c65ff0af26fa0c9651ef4
547e278d382db7a12edd3e18d9c70120558953f9
117298 F20101114_AABMMD chinchuluun_r_Page_69.jpg
95f51e5545e2ecaa5d377303415de905
2fed1482e5a12d9a297d702cafd1d38b8246b0dd
50146 F20101114_AABMHF chinchuluun_r_Page_65.jp2
591dadb4560eada734e071bd189158e8
332d4d329a1a0914ac3611fdda76912e6848d309
F20101114_AABMRA chinchuluun_r_Page_54.tif
1caf01cfaf95d1de7e4f8065ada3e9b3
6942d56ba1dc9f8c1bf90dffa86750107230dfd5
27842 F20101114_AABMTY chinchuluun_r_Page_63.pro
37ecc46224438a23ebc90b8aa13c0726
e16a9d858e56e00e85b17998813d8eb0f2b99d4a
5026 F20101114_AABNAX chinchuluun_r_Page_58thm.jpg
2f46fe263757221eba10386ab6e50d0c
c9841d43b95e051dd314845e85e04e9f4eacbf75
3338 F20101114_AABMYW chinchuluun_r_Page_27thm.jpg
345a23ae62cd231cc0f3eeb964021f29
333d299a81ff482eb457970391c8d78fa9bdbef7
116015 F20101114_AABMME chinchuluun_r_Page_70.jpg
ddf4e2bb965d4c0329689965e81a5300
db8ddbf4e66565c7876d072fa5c6f0607898ae53
585211 F20101114_AABMHG chinchuluun_r_Page_48.jp2
b19978908ae3cf3efc6f5b9a93f45736
6ae58754b4af213e90bb6938d28997892ff0af86
F20101114_AABMRB chinchuluun_r_Page_55.tif
d00e4655896c95f029d6a8d5919377fa
ac3d8a4e328d90d7aecc04101efb02408ab1f192
28069 F20101114_AABMTZ chinchuluun_r_Page_64.pro
91ed8fc55532f51efe8dc0fdc9a4d193
b0a2f8ecdb05054d3c793ee5207fa9b00dc42138
17219 F20101114_AABNAY chinchuluun_r_Page_58.QC.jpg
8bc9729497fddd271abffb289fd28fc4
4ca8fda674bf9543f24ddfb890d44309a982693f
11417 F20101114_AABMYX chinchuluun_r_Page_27.QC.jpg
97ef814c66d8859f2902e0d56c307e87
e851add65ff67a5da29f5db4deca4d215f335549
117160 F20101114_AABMMF chinchuluun_r_Page_71.jpg
00645da9d4018bf9d45eb03315b10a29
0aed428db804dd58b13374763772bae65e41d91e
27460 F20101114_AABMHH chinchuluun_r_Page_09.pro
be1a65b85ca3aeae2c0dbd3317cc9b44
d56cf13472f12e1219d8765909661bf75bc1934e
F20101114_AABMRC chinchuluun_r_Page_57.tif
1bccf3f858ba785c5dd92e26ff31e967
f1c9cd663201465658833eaaa883148c4b24d83f
4444 F20101114_AABNAZ chinchuluun_r_Page_59thm.jpg
0a20edabf36a2c8fdb8f4b0fae7a4ea8
88f5e4b4f87245f0e9417c81c45c4e34d69eca31
4745 F20101114_AABMYY chinchuluun_r_Page_28thm.jpg
fcca60257836e6f15fa24728563a1c88
6dc20d0fa1fb92ed51cf8ffb20e5f9b420acbf31
117242 F20101114_AABMMG chinchuluun_r_Page_72.jpg
f01657acc6d480b6db7a268eec55f32d
c9724124ec39c81e81142d2717b920b6d0b9fd3a
948 F20101114_AABMWA chinchuluun_r_Page_49.txt
70fcaaab6286c7a6b2c63308dd892bed
e8a9ba26d2e26c00afdfc4f7c803d9ce64adf6db
48790 F20101114_AABMHI chinchuluun_r_Page_06.pro
89ef2cb7fe12ec44cfd2bb33687a636a
ed934a7a1b9669b2a899c2466fce371d958dd8f0
F20101114_AABMRD chinchuluun_r_Page_58.tif
a2382c24486831b7862cc54bc65e9877
713feb086a3f6f77c50c9ee5e04f58b3381b38b5
20540 F20101114_AABMYZ chinchuluun_r_Page_28.QC.jpg
8f4a5430c236178490df5587ea7516fb
26ff90647276cc07fd2acd3412b25fdc324ec6dd
43068 F20101114_AABMMH chinchuluun_r_Page_73.jpg
cdb69f7e1dd6e1fb46405137d876725d
c302f81239cca392a4eb6577b44f57b4d8392b86
1107 F20101114_AABMWB chinchuluun_r_Page_50.txt
aa8fcd641d0303fe7b8b5fe59c503b2d
16a1a957f61e3ca4c01e8ef33636555b11d2fcfc
F20101114_AABMHJ chinchuluun_r_Page_10.tif
750c921c396aa8a62f363e94c9ab94f9
fd30de5714eb29f61e805a9a108f4a7b43619dd3
F20101114_AABMRE chinchuluun_r_Page_59.tif
c7f57dcc1d8aa26e563469e74706bc72
928bae3cee3187cac7d7cb7a02314f4a373000fa
863 F20101114_AABMWC chinchuluun_r_Page_51.txt
0d0d2971b2769dd8479ea70d8bd414e5
f77c7fd7ab2a1656a13fc96c28ab2b0cd763d879
102 F20101114_AABMHK chinchuluun_r_Page_02.txt
a58f8f5381b062e7e6b8faa1ce11bab5
aa5bf14bef8595f0b6e8a7996f12a7a220fed2ba
F20101114_AABMRF chinchuluun_r_Page_60.tif
7cb9c24daac21669188a88835a960e40
53571fe4ed6f4c655504d258271a08ee1be82c8c
101163 F20101114_AABMMI chinchuluun_r_Page_74.jpg
b5ac53a764dd31f8ead2277602533030
6bde756a5a084a72aa2345a14ed0b0e1c623591f
1812 F20101114_AABMWD chinchuluun_r_Page_52.txt
0e42724b2a27b1c578a51055630d75ff
270251a9f22ac3b81bfb3b060b5f206c4534546e
16968 F20101114_AABMHL chinchuluun_r_Page_54.QC.jpg
836069ec1322468832915244a8ad0f65
1a2d8574e0dd700b72b28e0f48cc28a6fd7c3279
F20101114_AABMRG chinchuluun_r_Page_62.tif
e869bd3a285f19af59fe1f0e9a232161
e5dc11b529458b4dea73eecd07de2b5289f83cdc
108877 F20101114_AABMMJ chinchuluun_r_Page_75.jpg
0abd207c9bf1a8a2a2d4d9c1cf1a94ba
6f3881aeedaffb3a8c98ab07a60cf3a4e2e93e28
1810 F20101114_AABMWE chinchuluun_r_Page_53.txt
8b581b46b07a24d439ef0753d02649e1
0c3c597859a133c8c6f8e40695dedfdd3eebf13e
5474 F20101114_AABMHM chinchuluun_r_Page_30thm.jpg
eb8956ea2282952900898486e4120c73
8b75270c8d8c570bf7073098e5622e12ac611931
F20101114_AABMRH chinchuluun_r_Page_63.tif
c7ebdf606c60c7e7ca543e5775fc40fe
79f77cc028ca3b16c82bc40b384f084703140bef
109543 F20101114_AABMMK chinchuluun_r_Page_76.jpg
3944c499276f43378274a556ce6571e0
a538bfd68b80b30cbac164f355cff5cf308165de
1089 F20101114_AABMWF chinchuluun_r_Page_54.txt
e17d7cc7cebe2bd7015ec55ba656d9a0
ee4759ce0dab4ad40e153ee0071219e6eb3e7ea3
2136 F20101114_AABMHN chinchuluun_r_Page_17.txt
3374351627fb590033bd10bf35f885ef
a800a63a5c4451d47a7e34daa04c6f659d1240ce
F20101114_AABMRI chinchuluun_r_Page_64.tif
08dca7506709c84fa4e0ba446ce1134b
4ed2d44390332127f02254b05f9ece416b90d8c8
693 F20101114_AABMWG chinchuluun_r_Page_55.txt
b10cd7c7726565110ce8e767d195723e
b058b4496c9e7e10714285c390ccf1f6d3cdf414
65418 F20101114_AABMHO chinchuluun_r_Page_45.jpg
d77034f5d6f53f871e99a6b3f0d9d55b
91e95d03a3c0d0c93ab9b4a50c8d3ad5f3c7bc58
F20101114_AABMRJ chinchuluun_r_Page_65.tif
964a4d990114b4a940e141a2cbce34b9
518453c1edb19d533b5dd66a910eb6896ff30c71
48092 F20101114_AABMML chinchuluun_r_Page_77.jpg
a6ca7f588be989fcc027cd11e67260e6
0fe992adc67fc8d41a75518841ebf3d29c6b2350
908 F20101114_AABMWH chinchuluun_r_Page_56.txt
f872418021583be90c602f22dbfc5434
5a447e3dffd083b9167ada37e28f3434731236b8
123502 F20101114_AABMHP chinchuluun_r_Page_70.jp2
1d1f046c7d4ad557e0752a753af3d3ff
0a41d2fe29a6d5a5afbf58ba18c76189a71a65dc
F20101114_AABMRK chinchuluun_r_Page_66.tif
1863cfd6d53d04187e2867cefd37ec61
ce50f58b20ec6b8cde012b31bc99d1253ce4f3d3
38448 F20101114_AABMMM chinchuluun_r_Page_78.jpg
e363530dcf493c94fb86c9ea5e49928b
4bdbcb720d032cb7b7e63a033e5ac092c9d642bc
1322 F20101114_AABMWI chinchuluun_r_Page_57.txt
e6973e8a50a9ddec034a09de0d46edc8
76bf2c3d1772477bd29ae94ddf8f17d3f39cf180
27252 F20101114_AABMHQ chinchuluun_r_Page_17.QC.jpg
e90c413aff777420f4eb8959134c35f0
2442b70b9508cb2a7865fdb766e453c1f45cf68e
F20101114_AABMRL chinchuluun_r_Page_67.tif
ba8856875e1a5f5cbedfc80c98c4033d
c51c27b2fd7637e48ddce6e111e65e7db5f04b89
26341 F20101114_AABMMN chinchuluun_r_Page_01.jp2
cd9f441ded68bd7dbd79757f1e3becfd
c7a01010ec2761e1b72897277d73e6b2bc98ae0d
1202 F20101114_AABMWJ chinchuluun_r_Page_58.txt
37cd71dc269e8b12f68b40fe5e63869a
f4728ef4ba2225ef67d35489ac86bf9308f0d0d4
65388 F20101114_AABMHR chinchuluun_r_Page_36.jpg
1540243408bbc0e3992aa4d2ab922f13
4ee038b62046d6cb8a38aab88ad9e2ff7ad0d541
F20101114_AABMRM chinchuluun_r_Page_68.tif
8c6a151d6e0c57ac8ce2a310a6e24797
4f69604cdbc9af0018700db3914f4d296be03d7b
5686 F20101114_AABMMO chinchuluun_r_Page_02.jp2
6bba6f5b58fe5485b94e0cebeb2d66f8
e6117345787d4880362af7e0da2ca4e15bc148d1
2095 F20101114_AABMWK chinchuluun_r_Page_60.txt
b1c18f59a222ea3a9f37cbdd2314a815
f3237024b5ef5f19d8b3f3b9c5f0edef0bd69db8
21689 F20101114_AABMHS chinchuluun_r_Page_52.QC.jpg
6477f5c08137c95451185b65429b8d11
93904cd2c598e978274f6560dccb734401d7a3c9
11146 F20101114_AABMMP chinchuluun_r_Page_03.jp2
458c9586caa6d7f914d4679a16f4b882
f8a7fc92ac32349df77772394c2207ef32fb5cec
855 F20101114_AABMWL chinchuluun_r_Page_61.txt
b0f637ef7b5620fb9cb38439ab6c6a5b
889b2225b28e0e6f78c248dadb1b681c5aa294dc
F20101114_AABMHT chinchuluun_r_Page_23.tif
36c2585f536f11e1fcd086f5b949e515
a593da10aff97c38f8010fe173e0ee7f095dbc7b
F20101114_AABMRN chinchuluun_r_Page_69.tif
8b8e1ed3c71f96e3586a1e68581038b0
03330259169b287b04fca56df71881bdd293afaa
69325 F20101114_AABMMQ chinchuluun_r_Page_04.jp2
f39df4f43d984b08d4074fc1296960ef
feacb31135cec45ee394f7579f131eabb0bc3ce2
1475 F20101114_AABMWM chinchuluun_r_Page_62.txt
589cf9f8538bca666d7450e4b2e49c5e
05f9398f34d7ce417b021c9511f71ea487e944a1
15632 F20101114_AABMHU chinchuluun_r_Page_27.pro
ab841251ff66eeef8374ade2d923da15
e3ee0f4e77931ab3ad762a44ef70fe352bee1b1b
F20101114_AABMRO chinchuluun_r_Page_70.tif
d9268637de7f82e43f82c616c73d4f47
a59c7d0d3dba7cff3cafe021ddba9e8dbb93a7ee
1051983 F20101114_AABMMR chinchuluun_r_Page_05.jp2
4f3e995f2a9be987e3161bc4ec813a7e
37ef108d2fee9bdc1c592d64c5dc113bb7e38596
1222 F20101114_AABMWN chinchuluun_r_Page_63.txt
eefbd1ad7b459659bf272ab85d747bd2
497f330bc7a18645de883323bcb46d1e1c0a1225
20756 F20101114_AABMHV chinchuluun_r_Page_78.pro
cd989b3dbe8448f677d55311f4dfc2fe
b6dd55971a8110a63c54aabb19e9d0625aac9f1e
F20101114_AABMRP chinchuluun_r_Page_71.tif
609a973d947307d610cd90842a8537fb
b693163d51f9395a9a0b63b14f6b47c10b259560
1051984 F20101114_AABMMS chinchuluun_r_Page_06.jp2
086dea1e843b0bd0da4fcca5737d3a1a
b3002670d6cb4cdc6e5f9f442dcb0bb1e577c18a
1388 F20101114_AABMWO chinchuluun_r_Page_64.txt
46b4be62b551f846a977307142abedd5
031e0a33bf1a8fa552b4d7e67dd51abc8bf008e0
70342 F20101114_AABMHW chinchuluun_r_Page_29.jpg
8f51e7b6e3871bf438fce764039f6736
8bc4f837105d0b88342122eb9ac6240db39142c9
F20101114_AABMRQ chinchuluun_r_Page_72.tif
f94207a7f58f8caa5d5bd4e0b5c5729d
0e3cf4c504850499bbfc991f49a872a6b9d016a2
535310 F20101114_AABMMT chinchuluun_r_Page_07.jp2
9832e6ae54d8d5dfb57c6d216a9bf020
ff6d974b0f9e1be34b5e4f654263fc1a5647fb97
2707 F20101114_AABMWP chinchuluun_r_Page_66.txt
84668bdf7651bcc0083338f8c49d8ab7
8507bf1d9f24e67286c9097f8ebb6411e524dfa7
23920 F20101114_AABMHX chinchuluun_r_Page_34.QC.jpg
477591b36f63c2dd5b7a5077597ef59d
9d2f3b66cc2c8eb03f043c5a823431c9b295baee
F20101114_AABMRR chinchuluun_r_Page_73.tif
9e1f26c27a4add787535c8b069fc6cbe
c938ee8ae645a4cc7bd3e31a5ae54a417b3e7aaf
1051975 F20101114_AABMMU chinchuluun_r_Page_08.jp2
f5bf111e2a754ed6d181b14f41371496
918ca3388248aed1c86e5e859408f9c8f8c4fd85
2824 F20101114_AABMWQ chinchuluun_r_Page_67.txt
d11bf86a93d1df51516cbe7c64af5354
3bfbed7c786b587fc2fc64c20f82449da23a8762
4498 F20101114_AABMHY chinchuluun_r_Page_05thm.jpg
92ac6c9a0f07b51a56a2d289953eba5f
d2d7c0fbccd90db13ecd5fdbf2a617a5c6ec050c
F20101114_AABMRS chinchuluun_r_Page_76.tif
6cac230cd3a20e1e815cf964f120ffd7
bd279bca97358bf83ec97af8e14200f74a366c6d
1051977 F20101114_AABMMV chinchuluun_r_Page_09.jp2
4203aee694580d6857f6eba4cb4cf272
17bf4309cb88b5f799699bfd3f7526c8b7a859ff
2635 F20101114_AABMWR chinchuluun_r_Page_68.txt
5e04bbd5a85ba57e281dd93ec60bc7d9
d303a10a5bbf5a880ab223af39e38ea09c37c23e
F20101114_AABMHZ chinchuluun_r_Page_56.tif
07f1be531112c1fef71f543710929119
cf70515a4f81037457cc5c4e7bf5c5f6a7d9926c
F20101114_AABMRT chinchuluun_r_Page_77.tif
83d06ccf77c30fd56f11d5b3688ad92f
ed2f741c1c986fc892bddd2cae159e73661f1ef8
104101 F20101114_AABMMW chinchuluun_r_Page_10.jp2
97e550c1e70b2c9130a4fe59ac3db3ff
63cf6b27907657622b7215378ab3f8514a6d9010
F20101114_AABMRU chinchuluun_r_Page_78.tif
8be0cf1d4f27eb90fb781532e7081e33
c1239bbee338336b1d2907c7401fdd8f7acb7ff7
107704 F20101114_AABMMX chinchuluun_r_Page_12.jp2
633feaae2edf846f8015edebe8e65ca6
a0ad9975f4d44262940bfd8ab6d6cc7aebeaa587
2825 F20101114_AABMWS chinchuluun_r_Page_69.txt
ed12b89392a78f6b249a56d1b64a99b9
4f57dae6f01bae403fd488857bd8968f58f10fa4
8705 F20101114_AABMRV chinchuluun_r_Page_01.pro
327a59458fc1b96e8c84408371023f5f
5da239eb0a0efbf3b18762b6bf56a47241e8b9f8
98073 F20101114_AABMKA chinchuluun_r_Page_08.jpg
a87601d93eaabc9ba1864839485e9f6c
6b99a3b2033791a6f571af377c61ae9f3e04c0d1
112593 F20101114_AABMMY chinchuluun_r_Page_13.jp2
6cc9a8a56ce1c7ff5b4056bf8a667132
09b505c19f0e22ba7a8914413f94bf71363bc6b6
2615 F20101114_AABMWT chinchuluun_r_Page_70.txt
74c8d6e2b0f4eaff62c9b9b7946e516c
e4ece3072c04cf6d9970b6fd7675007cd35c54c6
1042 F20101114_AABMRW chinchuluun_r_Page_02.pro
90035c3d83c369ecb66ea7762e9343b1
e07607dc2cd6ac33a3ff1b93ce35fe172a3d3e9c
45112 F20101114_AABMKB chinchuluun_r_Page_09.jpg
3c9cc4870198f83400c9a7170c6932e7
d710e8744cc797545d5c0853e1eb83fa70c089eb
112554 F20101114_AABMMZ chinchuluun_r_Page_14.jp2
c9af32c6b1729f1220c66df6f3030a36
d79e0359ff54d8e0c645f96e12cae1a8a99576df
2640 F20101114_AABMWU chinchuluun_r_Page_71.txt
8b011d7746793a52426e10f996ff432e
0cd2874df9e34f18934ecae815b37e7896747b4d
3943 F20101114_AABMRX chinchuluun_r_Page_03.pro
f46aacd1f9364a351e495fc1ca72e2d8
98f16c38fbe1bfc4080ac8df53bfb07ef8e6f2a5
82416 F20101114_AABMKC chinchuluun_r_Page_10.jpg
8bae4dec00d25e17afb1c259e534ebbd
b497121613fb228a6ed80ebad86f55ac7ad1b0c6
2630 F20101114_AABMWV chinchuluun_r_Page_72.txt
a0cfbde027da9d9e5a51ecafa83cfca8
5d16a9bb4f724cf676788fa135728bafe82f5b63
132744 F20101114_AABMPA chinchuluun_r_Page_74.jp2
cc0f386c0b1be977af96a4f0bf2827c1
2c89bc96f36c48517faf8b0225fe3d2270407305
30384 F20101114_AABMRY chinchuluun_r_Page_04.pro
abecd73aa3a502b7cf37ef00cca088c2
e8cbb037362827a70bf9feabb65320be203fa968
82991 F20101114_AABMKD chinchuluun_r_Page_12.jpg
98e7ed7269484eeff92112e7ef7625a6
22f60042af9d2ebc9b0eb39634d572047c878790
1057 F20101114_AABMWW chinchuluun_r_Page_73.txt
ee28df07e49fb7982d79a7565985aa34
2a407c11b2ae7fa54ad1077e31ff6d1b6653dedc
131656 F20101114_AABMPB chinchuluun_r_Page_75.jp2
6f0b98806e1cec8b87428f926f0a9213
3e5ebcddb84e384e3ce691c8b99c74a68dd8c9dd
70269 F20101114_AABMRZ chinchuluun_r_Page_05.pro
f0a88b776271b603e875de447c3514ed
69f9be428ad326edb1c4fe6ba2ee5446bf4751dd
84716 F20101114_AABMKE chinchuluun_r_Page_13.jpg
0aa321a57538e83ef7945d962bab29be
2b84ba6b2dcdb393cc55ef00862f0faf5556bd31
2600 F20101114_AABMWX chinchuluun_r_Page_74.txt
54e7a33c6dfaf8e8c68b2a6f16c3f99c
db58904bc5e77c160feed56398c1e4ac74a21e9a
144703 F20101114_AABMPC chinchuluun_r_Page_76.jp2
4a2b95d30ab91622bb478509e85d7d59
b4d9fbbd06241ee818578b05f668b494b583cc0c
85064 F20101114_AABMKF chinchuluun_r_Page_14.jpg
21d53998b121c659efa6fe0acfd8e718
37e153e98e0dc4100453ee96765e961424bb9928
2574 F20101114_AABMWY chinchuluun_r_Page_75.txt
6c09884de77efb4b654896be2a0b7449
9b21c622ba0bd3af2e3eb5aa65ec402b010840ad
61246 F20101114_AABMPD chinchuluun_r_Page_77.jp2
97a814918b8c279eeeb336734240581d
bdc2d4ca03356a5ecdc2fb702d89aa5211a1a130
20342 F20101114_AABMUA chinchuluun_r_Page_65.pro
c5c8c2051362232cdb108f96fae5933f
1dddc06168fe0fab4f4a80182bff749e006f130a
2793 F20101114_AABMWZ chinchuluun_r_Page_76.txt
f4c14502f9125c5c0ae449c270dfd268
539f0dc7185daeaaed353b25b9e5a855041bb45d
49902 F20101114_AABMPE chinchuluun_r_Page_78.jp2
789eaf7f121e060506982add04eb42bc
3dcf5346f1105dc106c09fd3698895499505364f
35503 F20101114_AABMKG chinchuluun_r_Page_15.jpg
156125c1ef5c9ed5bb221c6ac2eafee8
752e8447267dbbd7f7d98730ac86a0a29c6aec87
52897 F20101114_AABMUB chinchuluun_r_Page_66.pro
d02b8c27d083cd6dc60ea1031c7eb11f
efa67df67331330d2d31d5869c6b1fb6e07dccbd
F20101114_AABMPF chinchuluun_r_Page_01.tif
4b06e2046f91078a380198ff1e4f7e96
41c001c40891d91ae7005834372191c6ae3e3aeb
77454 F20101114_AABMKH chinchuluun_r_Page_16.jpg
974d4154b1c5f7df7f03fffc009c8906
50311c2d6f40ea8318a80b4e8063d3291873aee4
56217 F20101114_AABMUC chinchuluun_r_Page_67.pro
a73220a91edc3db6305d93555b196776
57928e497d583ac7bc7bec025b54f8b511964d30
17783 F20101114_AABNBA chinchuluun_r_Page_59.QC.jpg
d43f576e4f54f3a93d5bf5a77df8d98b
373068414a8390e5695ec5ebd4b10d76641a9e6c
5531 F20101114_AABMZA chinchuluun_r_Page_29thm.jpg
810f9e2586a6efba3a6d522408a4167f
8785ee1114004971af95a9998d643144da2d1e69
F20101114_AABMPG chinchuluun_r_Page_02.tif
523a83b8e33fa02fd2bdda505d7c7f96
48a39f8cabe18123b9522ce97d76393a6e0fd5f1
87505 F20101114_AABMKI chinchuluun_r_Page_17.jpg
ba0623e28a5b2b245fe0eb941748b936
ea162ee92f215a2083ff981f6e444c9835c02dce
56260 F20101114_AABMUD chinchuluun_r_Page_68.pro
1f11e46346b9c6c8df969ab36cf94502
965629c70c6e1ca6265bb198ba8d7dd2ff461915
6439 F20101114_AABNBB chinchuluun_r_Page_60thm.jpg
5cd59d74cf2eac8673d445b51ade87e0
f7eec5950325b7394ec7152d97ed9c1224f9d51b
22163 F20101114_AABMZB chinchuluun_r_Page_29.QC.jpg
f9dba3059dbb29c173ed4ae0468ace9c
963689c70e080a63ce68cb17be97abd6cc523476
F20101114_AABMPH chinchuluun_r_Page_05.tif
44fabc166137bcbab6a1ee31f0354283
6d1da347747bfc3dfd6a32597f59a3042daeea6e
88648 F20101114_AABMKJ chinchuluun_r_Page_18.jpg
5213126ecf4e87a21724b46e981d262a
892901f6b15e91f2a02aedeebf42bf03dd738bdc
56269 F20101114_AABMUE chinchuluun_r_Page_69.pro
c62ed79d1beb83ddfc191b465eee4028
e5f37dd6566267f53051a7659224fa08ace870ce
26505 F20101114_AABNBC chinchuluun_r_Page_60.QC.jpg
d3240326fcda384666b77655cf47d364
ffc9d2bb61fcc44d8609cc1f859bb5dd826e1850
21569 F20101114_AABMZC chinchuluun_r_Page_30.QC.jpg
d9ccff67336e78ef826e15c3477691c3
b35e96ba7a0eca85e094c0bee4fa06790172c04e
89069 F20101114_AABMKK chinchuluun_r_Page_19.jpg
585ef2441ce65faa5de5df43bc20291f
f5635879de9e1adb2cca1a36c557d68a5dd38194
55740 F20101114_AABMUF chinchuluun_r_Page_70.pro
36be8f57070590a04e4c30f1e1792c70
b3f0f5ee8da7b7800228de598ecfa868e256ab8a
F20101114_AABMPI chinchuluun_r_Page_06.tif
d8a5aeb87c27d3c9c9b34b2da00bfaf3
f29b3b2e76c7b7b999c2e2e6935dd32252de624d
4296 F20101114_AABNBD chinchuluun_r_Page_62thm.jpg
66c15108848e55be598f96d545c62c28
37095eef8d8b8279b15b10882db49cddd155d2d4
5522 F20101114_AABMZD chinchuluun_r_Page_31thm.jpg
78d20fc463a3c32fd17ebb572db7468b
51bfd5aa9b67c96fe877e6e9a33ec9c657a0f35d
88915 F20101114_AABMKL chinchuluun_r_Page_20.jpg
cbb18ac7e1f6341c5bdd059e27887af3
deae433d69c44a4f8c4777b9502060f4aaa96590
56400 F20101114_AABMUG chinchuluun_r_Page_71.pro
553730969d314879d79269bc04540acf
a0c33d41888f418a08184e32be458bcaeccdcec3
F20101114_AABMPJ chinchuluun_r_Page_07.tif
aab2c8a51c79d98774dec6e44b3d50d5
0e2b9cd2efad48f081db134eef6acef51ce32d99
16145 F20101114_AABNBE chinchuluun_r_Page_62.QC.jpg
e7234fcad42826511054b1fd2a7f8ae0
31d6a9d0e16434915dcb31b56bcf82d8845d2cb7
6824 F20101114_AABMZE chinchuluun_r_Page_32thm.jpg
4a7b859df1d30ac186780d93d33bc5cf
d401acf0d5db1f99b6141d00bbf665b5b65687d8
91597 F20101114_AABMKM chinchuluun_r_Page_21.jpg
694be85ab9a87483f2fd39abc50d06ce
98b6592f2c030c58700dd2f486182db5a296a9a4
57273 F20101114_AABMUH chinchuluun_r_Page_72.pro
9973a11814f9850e1780c3a29789e9f0
bf628fcbbe464ed344c212fe28ba89a9133182d3
F20101114_AABMPK chinchuluun_r_Page_08.tif
061073c7cf56879a65e154bac74ebded
10ca48869bd5bd4f0cbeb11da1d40f792e0dbda3
4218 F20101114_AABNBF chinchuluun_r_Page_63thm.jpg
38a7b46bba1c7c50f77c839743064728
5ea5dce8d236a2ace6a469a7df5b508913979ce6
24168 F20101114_AABMZF chinchuluun_r_Page_32.QC.jpg
ffa38826774086664c021792dcac7c9d
dd71580cdc8ebb5e1941eed3dc2d8585c5146b83
87495 F20101114_AABMKN chinchuluun_r_Page_22.jpg
19cabce6e4a028e20b0164cc6c4f491f
558bdfd02958e1d2419c82d3f70551957f2982e4
20689 F20101114_AABMUI chinchuluun_r_Page_73.pro
652902412e21be13919e89ba67e7b90e
7fcc160f9b3d151ee51e23a14bb8cd5a715fdc2a
15661 F20101114_AABNBG chinchuluun_r_Page_63.QC.jpg
d3e23b4548095e1c2ab0715acb320bea
ac21ff2546ee8731501f95666f99484f90763f6d
1328 F20101114_AABMZG chinchuluun_r_Page_33thm.jpg
3a47e94dc94267d3ea8f8183a9d59846
61472e2c25e5f80cc52ec014ee97dd917309b3a6
90803 F20101114_AABMKO chinchuluun_r_Page_23.jpg
0f8793dff2291425716a9cb00aa8688e
61da02de56456378277449daf203f4c5236c8b5e
64160 F20101114_AABMUJ chinchuluun_r_Page_74.pro
8730e53352a9350512f4d0a6e9ed72d4
083363a863bc33a1169f9daa78f13e43b6327d2a
F20101114_AABMPL chinchuluun_r_Page_09.tif
af9f1528273f680a52fc46d5f900abc2
67e5b6996a10c30f7b6ba31c70b466b65ced627f
3823 F20101114_AABNBH chinchuluun_r_Page_64thm.jpg
d32e233cde96bc36c5d313a979fce25b
3ff2b73956770a25fe2d84ca1d80975371f739f0
5707 F20101114_AABMZH chinchuluun_r_Page_34thm.jpg
58652d0ec8a5be2b95f4530edcf1117f
e6dc26ad50d7cd7fceec058101bbf49ceec00fe6
13914 F20101114_AABMKP chinchuluun_r_Page_24.jpg
48bce4ff7956feb9832648bcc608977f
501b51ff4fe44ff62bbfdffb77ca08a8f997b1fa
63816 F20101114_AABMUK chinchuluun_r_Page_75.pro
e47cfaf204310c710782958d774b7fe7
1c457e106652e603b58b266e387fc05c80625dfd
F20101114_AABMPM chinchuluun_r_Page_12.tif
236b707144ca6f344078635b51541b7d
94b1f62886fc82da08761b2153dc459798d76b6e
14889 F20101114_AABNBI chinchuluun_r_Page_64.QC.jpg
b28a89d4e0847f8b1f58ab6e291951f8
c48a877d2dbd069d0dc5be5580a6f1680aea96e7
6449 F20101114_AABMZI chinchuluun_r_Page_35thm.jpg
115d160c9ea15753a8aa576a2a6e908d
438a51ef6c7121910ecf37d2c17182914d0aac12
81898 F20101114_AABMKQ chinchuluun_r_Page_25.jpg
ac35652bc4d41a35b5b1e1ab397c5c2c
10adf210602c9786362c82738e4f902a9b421a1f
69744 F20101114_AABMUL chinchuluun_r_Page_76.pro
d58fc77d2174bff6e4dcc2a194b08dd2
9a056f548702449fc58abb61b25b93553835e6b9
F20101114_AABMPN chinchuluun_r_Page_13.tif
dc5544ceb7e336581006d6ec0085613a
204fc358448cd20df70910addf38ece052217679
3000 F20101114_AABNBJ chinchuluun_r_Page_65thm.jpg
cbdbc4d0eb24275006ae9a25b99eff10
e55c503d927c590f873b6e13986b974bb3e7d75d
23816 F20101114_AABMZJ chinchuluun_r_Page_35.QC.jpg
61955e8ed2f8967abae3428975556987
359072b488cc4729c0c57be60348698f163ad59a
35669 F20101114_AABMKR chinchuluun_r_Page_26.jpg
c1de20e9908bcc558a06fd9f8455a60e
02051f8447b5257dc2368694817ab257503bac55
28059 F20101114_AABMUM chinchuluun_r_Page_77.pro
d6b5fbcb5b090c7ebe2681792cd1d7bf
234d9a7e6d476784498fcea73921e0a445fe0c7b
F20101114_AABMPO chinchuluun_r_Page_14.tif
b9cd3be24dc10210cbf65024510ce90d
5949cd27b16e7c8aea8d6fef78faaafd1c9a828f
11705 F20101114_AABNBK chinchuluun_r_Page_65.QC.jpg
20d029d699d363ddd60e4e9a9dcaf0a3
fdd7ef534113aef4d93059b6bce4fb648111bf54
5116 F20101114_AABMZK chinchuluun_r_Page_36thm.jpg
e4cc74e2f01d85a2f2755f4cb2f3ac37
711dec1f55efc3d9a4ee14862d41a2206c314816
35581 F20101114_AABMKS chinchuluun_r_Page_27.jpg
c8b0d7946829b92966a49ea47e91e1d9
b3f5c82118c6eae02ceeaf5b5eec43a831e3587e
480 F20101114_AABMUN chinchuluun_r_Page_01.txt
69edd5abc2b691c57d333e275d5063c2
f6793cf352b7e352bfc4f7a03a9bd69380427d4e
F20101114_AABMPP chinchuluun_r_Page_15.tif
f93746fc073ef31193d18d43a24dbf7e
5ec5dd890e7d0bba1c15ba9711c676c454cc1ad2
6346 F20101114_AABNBL chinchuluun_r_Page_66thm.jpg
f2182314561363b215280bb61e0fd75b
cb17895b14b8246534a7b10e48c2c2311281ea04
5186 F20101114_AABMZL chinchuluun_r_Page_37thm.jpg
c12ee2a655b6cc54633ff1430a44c524
4d81da1ef8679bbc7c33c197712c403bbbccc5ba
62546 F20101114_AABMKT chinchuluun_r_Page_28.jpg
2b0ea180db88089241a897cc786d0025
618c0e63cc1aac5d3dff10167f27ae55455d55fc
244 F20101114_AABMUO chinchuluun_r_Page_03.txt
8249bcd323eb29139785ba8541b8eda3
9d0d4851b13a149466fd66804fa2563eab3180f4
F20101114_AABMPQ chinchuluun_r_Page_16.tif
bfb0b5e71d601c1ab79dd915a9ae4561
5533f92098d2c553793507a82507251697b5bc23
30938 F20101114_AABNBM chinchuluun_r_Page_66.QC.jpg
d623e8f402c634da6bdd505081243fc3
0b0fe3b99f89ae749940ee31ae7eed667f1a4fe1
18435 F20101114_AABMZM chinchuluun_r_Page_37.QC.jpg
52675d1d37c35fa3740c519d436be0f6
356917362a51060b672a88fcb3bd906bf4f35804
70279 F20101114_AABMKU chinchuluun_r_Page_30.jpg
4e595d82bfc1c9635693d064b5ae9181
79d15403dd16616e7bad8861884d484f5f1433b6
F20101114_AABMUP chinchuluun_r_Page_04.txt
4731908be577492ac2f8933cd62f9e03
b6a9d1d3a812df9398515bfe39caaec1448b7216
F20101114_AABMPR chinchuluun_r_Page_17.tif
474b1b5815429b1637cd09036fee33e6
a96a416e407162adba230a9c514ed7738b9ab942
32855 F20101114_AABNBN chinchuluun_r_Page_68.QC.jpg
5b7083508409050676492222ab2ea188
d0b4cf7815294eb25520b8b8bf8b04a3a7c444e9
5751 F20101114_AABMZN chinchuluun_r_Page_38thm.jpg
6468277c1922cfeb8279b93cd36749e9
dbdf5702a340797482af8f220c9117f27bfc0725
69874 F20101114_AABMKV chinchuluun_r_Page_31.jpg
6e710800cc122c0262c2d92f28183515
79be6863753e6abbea1a22ebd9d6cd5585e5c12f
F20101114_AABMPS chinchuluun_r_Page_18.tif
44ea76ad64ecff178bf8800a253812cb
8cfeb907f77585da9ee7ce14b527ba6eb46c6682
6640 F20101114_AABNBO chinchuluun_r_Page_69thm.jpg
edc99d6ca3acefa2120bb4d337282410
5c8a09e22a1f83b25556025186589f6d04aa373b
21182 F20101114_AABMZO chinchuluun_r_Page_38.QC.jpg
b4a2fff33e0669234bff288b45fdeaf1
c78343b074ebf391d20089fc077967c281be41c9
71474 F20101114_AABMKW chinchuluun_r_Page_32.jpg
d922f75c809d841e59c5bd12e9f97794
036fe2eb2574dd4ac03fd890b158c17e4c077683
3045 F20101114_AABMUQ chinchuluun_r_Page_05.txt
1773c74522783498a68cf136b91083f0
4b8a6bd51fbac50d3896196892aea5e97859104c
F20101114_AABMPT chinchuluun_r_Page_19.tif
aaf5308466edf9492408d17feac4f58d
a2675ddaa4a9daf3b6c165f85559f8e7e38b8f67
33072 F20101114_AABNBP chinchuluun_r_Page_69.QC.jpg
b673fb947bc22e2b8c3f70a093731ba0
a674917530091ccb5862900f9efb375f6b828530
5010 F20101114_AABMZP chinchuluun_r_Page_39thm.jpg
d7bd086b79d16be113208b7ec4706825
f2e8b9e35b307076dc5a2b1913bfd10aece4badd
16004 F20101114_AABMKX chinchuluun_r_Page_33.jpg
e364a4daefb9005ae7203907126901d5
b0bf9224a4872099db4b47015efab6aca688fdaf
2098 F20101114_AABMUR chinchuluun_r_Page_06.txt
2b0c909ebbebc30a849e1c0c278be633
b0e6d6c9ebbce6630275e2976f7e947413fcd932
F20101114_AABMPU chinchuluun_r_Page_20.tif
cf3e1e9ae568970974e8b10a30489f73
05c97863be7852e0687b3091a4d4396fe8d797c2
F20101114_AABNBQ chinchuluun_r_Page_70thm.jpg
a2c3a4631c035fa3f07c37cc7c15adb0
787aa0a9669c775c795ca265c92184b61cf0abe0
17867 F20101114_AABMZQ chinchuluun_r_Page_39.QC.jpg
8d0272c0eebe78cceb43b06072e80209
6b4165c35363fc1cd447757d27b641e327bb660f
72937 F20101114_AABMKY chinchuluun_r_Page_35.jpg
bffd6a9298c49ad7a1b81425b4cf266e
688a5a7be100f005904a8fdd5bcc550e4e15f6ba
2709 F20101114_AABMUS chinchuluun_r_Page_08.txt
d9c72d9c26b89e929d3b04f5b7889d3b
1e4d9c8bae99327764db09e0fca4739e445aca76
F20101114_AABMPV chinchuluun_r_Page_21.tif
44367f42a95d97400d85cfb64d4c1dd0
243fa7ea0f9ab6c479387e491a4253bd4289084e
940 F20101114_AABMIA chinchuluun_r_Page_39.txt
6c52763c82e7c084c7bc31dc09054769
fb7e8591e2dd3b060072bbac056b850e21f94117
32700 F20101114_AABNBR chinchuluun_r_Page_70.QC.jpg
6d91619d731a1a5d0db966d563366de2
e9f5b8fd5d43d67a82ad1074abcf7d14563e8a64
6002 F20101114_AABMZR chinchuluun_r_Page_40thm.jpg
4cb8ad701b11a7fc148e85e23f7252b5
700d2345bccf36efeb82dc3d4daf911bd263c525
57155 F20101114_AABMKZ chinchuluun_r_Page_37.jpg
fd4552da22eb448ad9b9aa1593621b08
ce849529b836c6c0655804cb455c65eb81d0843a
1135 F20101114_AABMUT chinchuluun_r_Page_09.txt
c2ebf0b34a83bbf1d8df89886faf8d46
f4f754fb25528e0aec97ec187f4d732e07df2121
F20101114_AABMPW chinchuluun_r_Page_22.tif
c2d0e0b52b70638a267b79a8f814d8ed
57175c2fc5ea90906d46ac51a7dd915f41925a2f
38640 F20101114_AABMIB chinchuluun_r_Page_42.pro
67f51d72936a7cc866c2ffc707cef6db
2080c4ecd67032668aadb9b62bb9e9dc48102a67
6643 F20101114_AABNBS chinchuluun_r_Page_71thm.jpg
30c064cc54e8cc14daa015572a804411
78c8752cce8861c0dc76bb2eac39d2234b24ac2a
23754 F20101114_AABMZS chinchuluun_r_Page_40.QC.jpg
5cef584b3d91924eee13289447264fb0
09f3c8f9d45f3edfe4a9a21e698ff8cfaccaf2aa
1270 F20101114_AABMUU chinchuluun_r_Page_11.txt
c0a57c036623a89eecc19444fc8831b0
8ffac2228d2d85b376f3f61138175ccba41354d3
F20101114_AABMPX chinchuluun_r_Page_24.tif
ad65f1690f566058eda1e82af824deb4
c54adfdbf9b45a809dd472b054145b99a93f331a
65524 F20101114_AABMIC chinchuluun_r_Page_08.pro
1511e2947fb396d8957c58b4e611b834
555e758fdde771d3daa1eabd25d120bb1a7bf67a
32868 F20101114_AABNBT chinchuluun_r_Page_71.QC.jpg
a9d122627d8b818ce610976e1fe2ce66
c40449ab3ea16e41cf91e2ddcc81f0de18eff2f7
3698 F20101114_AABMZT chinchuluun_r_Page_41thm.jpg
076b8275fb6e5471d65ca36e51dd1a79
354f010897c549a2cbe0f1d6cf90c80ba9020887
2080 F20101114_AABMUV chinchuluun_r_Page_13.txt
2c21ff84e0b64b67740d1e8ca38aa229
7efbb7b1a41e0a4e18b2df4174ac34cccdff3a85
47257 F20101114_AABMNA chinchuluun_r_Page_15.jp2
b8bac54359339eb24f41e911f2685365
85192e4ccbf725c07f2a471aa2dea6fef4272e6b
F20101114_AABMPY chinchuluun_r_Page_25.tif
ca8de6a261f1ea92bd0927e39153516f
0a7d05ec20df101239bf26ef82fc0d7cb65ceb37
F20101114_AABMID chinchuluun_r_Page_75.tif
bd1fafc7bc975abb6c25246f2cf17c9f
56f385985a4554850561d99b522fdf713684ce9d
F20101114_AABNBU chinchuluun_r_Page_72thm.jpg
37e9737d96e9456ff451e0cd48310a29
80ad511687ae01fce2b032c6bf83059705b85726
14160 F20101114_AABMZU chinchuluun_r_Page_41.QC.jpg
6e3439fe7e3167e6b4a5022fd1a0a5ed
6af677fb85fabff381bbf85acaf5634a1792dab4
2057 F20101114_AABMUW chinchuluun_r_Page_14.txt
31dd5732e19e86f73c8f7a4d2c3b3d9b
879e33c13cdefb7d8f2e3d2a30e42797b10a8e22
99920 F20101114_AABMNB chinchuluun_r_Page_16.jp2
54eef4424b7abbcdc2b9689d30a74367
b4b3233c1d6aa4d528ccb6a0a3f8297a2cc04c0c
F20101114_AABMPZ chinchuluun_r_Page_26.tif
79eaba82cd3f6c93677e14ecca1eede4
73dd807590b6aa84d7a543fd16094002ddb842c1
32954 F20101114_AABNBV chinchuluun_r_Page_72.QC.jpg
598aff95d874e77850fe09c5a823f4c3
3f6f697f026ed1a509da1b04b086eb8d09c6162b
884 F20101114_AABMUX chinchuluun_r_Page_15.txt
17f21fe0b633caa43c9f5a27076ee9ef
3fb465e855ed02dd11acf9bc868d99e55f38558a
115201 F20101114_AABMNC chinchuluun_r_Page_17.jp2
cb1e75c3069a520283ef445facb07610
fdbc65ae1dfbf56f6cbb1335c8b74b3126832159
F20101114_AABMIE chinchuluun_r_Page_74.tif
ffcc1fc0402df903dbb7c0e5ef4e1d62
33dcf4113c997a3ca79175cbba43508c9ee4a9b6
12958 F20101114_AABNBW chinchuluun_r_Page_73.QC.jpg
9aa1ebeb9b57a7b788b135c96a4dda3b
dfbd1ec5d0d3e988915c3c9959a4656cb559a4ca
7057 F20101114_AABMZV chinchuluun_r_Page_42thm.jpg
28e9e9abaa07607beea969c22127fbd1
eb1a18ac9ef436772ec088a296ef1e278927b00c
14945 F20101114_AABMSA chinchuluun_r_Page_07.pro
4475dc676e2ca158c4ccb61f61556624
80682cea84e9ec9aaf9b2b4ce0504cdd4645f3de
1937 F20101114_AABMUY chinchuluun_r_Page_16.txt
cee36bc70e28822bf51267ac6f15d860
c800cf4cd1c08e45138069a99243f325c3c6a290
116927 F20101114_AABMND chinchuluun_r_Page_18.jp2
c4f57d5df8b4d8d01117e203ce4c5c1c
a873c028d36f89e15426ef52b6271902531eea9b
22130 F20101114_AABMIF chinchuluun_r_Page_31.QC.jpg
87c897e27e91dfdb9e7fbbd2f79751b0
b02c8e70bc30644586681cbd309b0fb6d6fe6ba1
7161 F20101114_AABNBX chinchuluun_r_Page_74thm.jpg
9182ebdfd07be739851d2725b0fed96e
eab1e6828a9c2da9c99ad1a5250825622d756c36
26982 F20101114_AABMZW chinchuluun_r_Page_42.QC.jpg
42ae631807cf0cd17e6c1f2eb259a25e
51b2f52cc85cc3db0c9c9d295920012832048e6d
46752 F20101114_AABMSB chinchuluun_r_Page_10.pro
b323a9406a3a385a8e302d56046cddba
0294e153b1b48c21b8ce62e9868f188131fe4fe3
2148 F20101114_AABMUZ chinchuluun_r_Page_18.txt
052de317026367cf0872a24e714cabe9
237ef85ac1fceebd4112c3999f947542275ad111
118539 F20101114_AABMNE chinchuluun_r_Page_19.jp2
4766d88e000c6a8dee7fdab834efacec
af3f5d1bb7024c53e745f4867405a60351e3bcf2
16897 F20101114_AABMIG chinchuluun_r_Page_04.QC.jpg
d556ae6dec911e2adc3bdc392aae2b3d
93628d3bf7a533691afea120fea6eca535d51c7e
29897 F20101114_AABNBY chinchuluun_r_Page_74.QC.jpg
d425982eecc7fe78359e6a167c734fa9
3948c61d06503288de5dde0bc42cd282d62e3662
5620 F20101114_AABMZX chinchuluun_r_Page_43thm.jpg
dfe9e2516326cbd20baeb5a9c4f2db1c
8b415f51b7cdbc847b17eecad42f8098577bf37e
32039 F20101114_AABMSC chinchuluun_r_Page_11.pro
9301918874269e1bf53afe50fc7aabaf
0b39993a4b45bfba7befe644c197b8e044866ec8
115332 F20101114_AABMNF chinchuluun_r_Page_20.jp2
a94b38190df704b811aaa2fd4ec2d78d
53ff49ca79e093428e366d731a8baa6fb8f57887
633 F20101114_AABMIH chinchuluun_r_Page_26.txt
2ef0ae7c72a188863596eb0b94dfd3a3
c27463d16bc122d3975dd3709b7d776537853ed4
29463 F20101114_AABNBZ chinchuluun_r_Page_75.QC.jpg
fbbd96a477d6ceeae206371f6fabd247
bea12b71a06b0d9c3c8d765d28cdffd0a44f34c3
20823 F20101114_AABMZY chinchuluun_r_Page_43.QC.jpg
98c8699b059c8fd24fbd4b53ddfaeefe
6a8ae6e293bb39a97212083ea374723f41056c01
1145 F20101114_AABMXA chinchuluun_r_Page_77.txt
4b850a6a40b813b748ca4d484591f4fd
27b30afc173349b1eeb064070e07a9be47ea48c0
50009 F20101114_AABMSD chinchuluun_r_Page_12.pro
6703616d98b8e7f3ec8c0047ec7a2dd5
dc87aeabdeaf2a46f12a4cf1ffd62c8219cdcea4
118252 F20101114_AABMNG chinchuluun_r_Page_21.jp2
90e55b55e4fb1cc69725e96abff89f85
4b41d1af98eb5dcc28504245a3e15836f622fa7c
54500 F20101114_AABMII chinchuluun_r_Page_04.jpg
f98fd51205004fa7fb17f9ab000a53a3
a55a2e52e8f692a8f81900b295ea0d39ca250e5e
6561 F20101114_AABMZZ chinchuluun_r_Page_44thm.jpg
6d96942915ce02ec35ad0b2a76b7d49f
7a79a87252165b06da0d1d80d1a6e88a2ac1cfd2
2093144 F20101114_AABMXB chinchuluun_r.pdf
6a4dfb9ea104df05d45ccae836709c86
827db299e8133caa06c73dd2c46f58a3c3956303
114442 F20101114_AABMNH chinchuluun_r_Page_22.jp2
6d124e1f76540acdeabfc26660e034d3
f78f8da0c5cef14c45233ebed2065dc5762345e3
55593 F20101114_AABMIJ chinchuluun_r_Page_11.jpg
82018447fa94f0074b8180badb9c3340
9564ea4b9aee1d04ba358fdd2e3d8088f0df8d26
51373 F20101114_AABMSE chinchuluun_r_Page_14.pro
f8fb532ee479f831a28002810533e819
4b2e68600364853efd6e4af5e39fc22358017dc2
1705 F20101114_AABMXC chinchuluun_r_Page_01thm.jpg
9a17b7f0c73819c34c2dd658a00bf211
3403acfbd0529c88b7637e852fb828953da5c047
118988 F20101114_AABMNI chinchuluun_r_Page_23.jp2
6b8bfe53f5c54bbb5d6e4e335b049db2
62202985efa894e166fcc408928a8dd5353ac6d8
69088 F20101114_AABMIK chinchuluun_r_Page_06.jpg
4d8afcfa4af49c46f5c4ed2ed53929ff
995061de8e74d137a3e0a985aff97aba1440f6f6
20630 F20101114_AABMSF chinchuluun_r_Page_15.pro
e3011c3573991b807f3586c1624e4032
05f183cd21f8c936fb9fbb00d71a497c87c02480
7053 F20101114_AABMXD chinchuluun_r_Page_01.QC.jpg
8a77b5485c24af4ba618423c10241b4e
e350d08cd654fab63e7bcbf1146c3f776b48b19a
939 F20101114_AABMIL chinchuluun_r_Page_37.txt
592910b73cb21e9143397b08ff3aaa93
2ae3eb7706976aa550f52ec8c53025c81e9a15ca
46142 F20101114_AABMSG chinchuluun_r_Page_16.pro
155fe859af3c06ec64b904f3d1371b0c
c20a72dfc4cd718bde4766d06eb7d0f0afe8c2cf
457 F20101114_AABMXE chinchuluun_r_Page_02thm.jpg
c4ce6aa61fe15eadc3b7d0dbd1211af9
9a8e570622f4fdb1975d35d32a88d9da3270244b
17418 F20101114_AABMNJ chinchuluun_r_Page_24.jp2
1c46e38be134ad77226a2bacad67a4e8
fd32018992e8b99213024d85a0fccc226db449d1
F20101114_AABMIM chinchuluun_r_Page_04.tif
c0639fffb6c8c0663ed0d5c64334b145
2662ef201fb3127743f18cae2dc8f5c78e84eb52
54223 F20101114_AABMSH chinchuluun_r_Page_17.pro
6ae25875c8c3d1f23335b7c6f7cdeec9
88a6406cd623ee473e739fddef18554b79b3a60c
1087 F20101114_AABMXF chinchuluun_r_Page_02.QC.jpg
05a724437cf69377f97beecb53243043
6e3a026033cdb1afabd9de878f13f8bee331ce09
105374 F20101114_AABMNK chinchuluun_r_Page_25.jp2
79596fd36a2a3c2c548cbdf67141a771
85fa7e389a7879c105d45c5c4267d461270a146e
866 F20101114_AABMIN chinchuluun_r_Page_78.txt
126c58a5482ffc95936ff216b28485e4
f4496f9a3031b8a0b93f57800a5910c95a147e5b
54529 F20101114_AABMSI chinchuluun_r_Page_18.pro
cb299dc7eaf6c836ca1e0d639b2259fd
de601622fd727fc2de73a7013654088641f491c4
800 F20101114_AABMXG chinchuluun_r_Page_03thm.jpg
3bfb832c5048212f75585982302e4d91
bee9d7410662bc6c98cc4a5c704fae95c1af39aa
412924 F20101114_AABMNL chinchuluun_r_Page_26.jp2
04156a610c1e83f58cbd6cd7d5ff20e1
2f5a37cfb375bf53322874dc86b79e72f87bee63
809898 F20101114_AABMIO chinchuluun_r_Page_57.jp2
063b6fd402d11e6c4242fe5ba6232968
38f1ce964de8b5bd6d404c0b59ca6ea6c37d1c33
54539 F20101114_AABMSJ chinchuluun_r_Page_19.pro
58fb70e98c2ef03392f25761883b7e6c
b27f90d4e4556b865975ff80a2a7396013700e27
2118 F20101114_AABMXH chinchuluun_r_Page_03.QC.jpg
dba91ff7a07affeafa1bb67e56f0110b
4d7aaa87826e8ff606e597829ec2ed3882f15a6c
428757 F20101114_AABMNM chinchuluun_r_Page_27.jp2
88de69a574c8781b9630b9866071a079
1830183bb895884654e7ddacc6f70da4102f7977
2034 F20101114_AABMIP chinchuluun_r_Page_10.txt
0776802a12a4b1dcd121a7df7c56ec10
3f4805a0313c5a21d37859fd9c6d90ad59d1a774
53616 F20101114_AABMSK chinchuluun_r_Page_20.pro
5a5bdf79c9025b5b81e1ad3d377e6196
a752af13900a440c24d942a680560b7e60c8cf75
4332 F20101114_AABMXI chinchuluun_r_Page_04thm.jpg
28f8cb8685fcccccfbee8daed6e5567f
14dfbb3a8cf6bc3d5eaffd9af65729d1a96c80ae
81045 F20101114_AABMNN chinchuluun_r_Page_28.jp2
2aacf78d6fe56ca7eddec07e439812c8
3bdaa6fc1e48ec6fab959747de5c1bbc0896a76b
2996 F20101114_AABMIQ chinchuluun_r_Page_61thm.jpg
3a1e986bb53138a3ec3b082c18706daf
5e43ba75b629ec9eacab55c9f5be3374715ace59
55779 F20101114_AABMSL chinchuluun_r_Page_21.pro
7ec1020b8987745b638d340ee6c18550
bacda9af09deffde35e542a560e2e759e01e3693
3866 F20101114_AABMXJ chinchuluun_r_Page_06thm.jpg
d3932316a2a94b0bf2a7637f082e102e
6b1323162640a834fc1ecf6797f08f3f4e0b4925
95192 F20101114_AABMNO chinchuluun_r_Page_29.jp2
a786851c6b315fe0ba5d9d7b817efa69
d1da5322464974dfc0496e4f185d418db88bb554
50961 F20101114_AABMIR chinchuluun_r_Page_13.pro
060ae621114571b20ad6f5789476a930
d1e3f2870fc4046ef4ed4dc26a1b563091bfd493
54896 F20101114_AABMSM chinchuluun_r_Page_23.pro
9d6b951e161ce08178d4be07f80e08c4
47609b6b220f8fee49a205e5969570eb9098b7ac
15520 F20101114_AABMXK chinchuluun_r_Page_06.QC.jpg
392951d791c3e86fac1ceedd28959306
b9b65240b18dfde3be6f2fa800e33d8125ea03bd
90301 F20101114_AABMNP chinchuluun_r_Page_30.jp2
65c382d6599539568f3ccfd9ebccec50
2c6507783126faaf7b7ab32aa33059680f9202d2
20746 F20101114_AABMIS chinchuluun_r_Page_36.QC.jpg
ff16e466c18752af386479bc3457be73
b20d02986bafdd54951dde0a5a10981c567fba18
6608 F20101114_AABMSN chinchuluun_r_Page_24.pro
60055d6a719da5eb22d780a4a9fabc9f
c12f5ea6ae4c24d29e0740c390e1e63a2a920fe3
2027 F20101114_AABMXL chinchuluun_r_Page_07thm.jpg
9085a00e43bf197c247f9a29cdedb8e5
4e8ab0bda6201ad990272bccc68b1e74acb7731a
90872 F20101114_AABMNQ chinchuluun_r_Page_31.jp2
ed9f7c7c481deb11225020c356a5f555
c0f8ac4beff1110142611bd6c603e9cab20f6236
32899 F20101114_AABMIT chinchuluun_r_Page_67.QC.jpg
6307bfd17186151a51894625fff8e4f4
bdb66df4d6e5b1bf652387e730ae900891eb9bd3
6779 F20101114_AABMXM chinchuluun_r_Page_07.QC.jpg
ba39ee4cf2b56070f9db7b506a449748
dfdbb0406add208c01f6c7478b055ada90112b97
884223 F20101114_AABMNR chinchuluun_r_Page_32.jp2
4ddd4269fe7d6aa99992ce8977508dff
ebe0517aff02ea185ca44dff028b8103cac07a89
3539 F20101114_AABMIU chinchuluun_r_Page_77thm.jpg
01b548fb6d089ab4b43d887e60f012a8
52359dda85d1e9fe1a944f5407f010dece1e54ea
49234 F20101114_AABMSO chinchuluun_r_Page_25.pro
18ccb45e5144e8f4fbb954836de941e3
0f7e34197551778a07059e2092c27213512987bd
5943 F20101114_AABMXN chinchuluun_r_Page_08thm.jpg
fbfb9ff9d8f40228f24f377b9b6c7c56
9a25ab97cfe7aaef172ece416907ff1035bb96d6
168325 F20101114_AABMNS chinchuluun_r_Page_33.jp2
8116df8b9de41c7707d7bdad8ba501e8
b89e05b03871ec22bbcfa47518c70e175c5d73bb
10879 F20101114_AABMIV chinchuluun_r_Page_61.QC.jpg
d7201b6711aa523bc25fdc6583ac21e2
b287d27c40dbd2633a4a3aa9fb035e21f37ec48b
13432 F20101114_AABMSP chinchuluun_r_Page_26.pro
ed3d6ce8f814d9266bc0c53a5c33409c
18fbc50e5819feb3857eea3b0572ef7acb37610e
24915 F20101114_AABMXO chinchuluun_r_Page_08.QC.jpg
a80d3e8151df8e4d7ed3de8ab62a730a
f7bc04eb2452e54345f9de6f922979ffe613c1d6
100674 F20101114_AABMNT chinchuluun_r_Page_34.jp2
07a69c574e3f76a7a5639e44797cffb9
9e72bdb0a1adcc93180012c72e8863488c451f89
2909 F20101114_AABMIW chinchuluun_r_Page_73thm.jpg
680b2a8c04913d660a80520a5517189e
9e1794553d248c51325ed648d69074da39773cc6
36974 F20101114_AABMSQ chinchuluun_r_Page_28.pro
8151d681d45241e475e8b63fab77d9e5
453936c7f378c1eb1a7df14dff62dac07a637dd9
3088 F20101114_AABMXP chinchuluun_r_Page_09thm.jpg
e5da63ea0fc7025f1b082a2a80a10da6
63dd1a9ebffe4cdf714794a4922c9f7ba474555d
1051958 F20101114_AABMNU chinchuluun_r_Page_35.jp2
da904189bc6c2a03ec1fdbad1a5a2861
00b836f8d34c40318e1310b615c61b4be7c67917
8547 F20101114_AABMIX chinchuluun_r_Page_03.jpg
e8ab4bc18638193ef12ac5a3d6947666
dbb1cbd930451c75ab491a487b3c990b4c0d1809
44177 F20101114_AABMSR chinchuluun_r_Page_29.pro
fd141ac8648a9eb08d9367a0d894efc9
df15e7d1d9bfc05eac6ade25945f6b28c9148970
11985 F20101114_AABMXQ chinchuluun_r_Page_09.QC.jpg
3a110e51ba14665e9f8562d86fa93d41
4f28281f274876038520c861c05a0c1583241ad3
85744 F20101114_AABMNV chinchuluun_r_Page_36.jp2
e016eb6ffb2ede2621f07547e7e3afd1
0a5f775dfde4304ea70e1b6405c66776b612779e
126581 F20101114_AABMIY chinchuluun_r_Page_72.jp2
a510534f7c4410ed309bea796fec1c09
b2227db72f679095bf8fe04a80ca12b3e2a7edd7
42478 F20101114_AABMSS chinchuluun_r_Page_30.pro
2448012443d04d1154fc5fe2cdc0f29e
178a8a1d73b870f519c40c82f04903ebacd75202
5974 F20101114_AABMXR chinchuluun_r_Page_10thm.jpg
7a86e2694d2d2e5017aba765db1208c2
6ecb53061eb133fb62a3fb733916fcbee964e137
723677 F20101114_AABMNW chinchuluun_r_Page_37.jp2
25204c8c3d9da4459beb960b6d82ad11
eef55413224e2c60f4d15b6f516ba6b0eccf70a9
7105 F20101114_AABMIZ chinchuluun_r_Page_75thm.jpg
a9d234b749ce99dcb4ab54beafe3c98e
53311d03857aff34840973dc5b750a39482de835
41053 F20101114_AABMST chinchuluun_r_Page_31.pro
3a95d21451607d973b19822ea0cd69b3
56b715651e8cee01a4ce1e00af9caca8629dc123
24155 F20101114_AABMXS chinchuluun_r_Page_10.QC.jpg
e0e531a63494766280f1a8dc5369e4bf
ba66c57388dc7a35e121711a68c7b8cd8dc02bb4
921137 F20101114_AABMNX chinchuluun_r_Page_38.jp2
2eacd1636a219e780c664416a3923f2c
6860f06e205d2bf9f45431310a08292e85166740
32538 F20101114_AABMSU chinchuluun_r_Page_32.pro
3efb1b172970c2df32995ef2ccc0e947
2a7ff5e0ea7c99926985aad2c37c133c7b3ca58f
1051930 F20101114_AABMNY chinchuluun_r_Page_39.jp2
e61572a840047cc28f0ef2a01076e970
56205a2787683fc710dd96d53328ae9500e51855
5622 F20101114_AABMSV chinchuluun_r_Page_33.pro
1b482b88c6a026bd926a2294282dc63c
76d1fba6f2006b81c0166bab64312fbaaab9da87
67099 F20101114_AABMLA chinchuluun_r_Page_38.jpg
801fdca97c706a9a6b9c6a2daa325f96
81791ffbbb6f70d6be1f2fb0622f2fdf433d6fc8
4078 F20101114_AABMXT chinchuluun_r_Page_11thm.jpg
7135a2e5b04905cda9e08bb2ee773076
9025652e191c904e3ad6e726f04c63f42e3157a8
88301 F20101114_AABMNZ chinchuluun_r_Page_40.jp2
931e320d872895b1e6a290d40f42f21c
3c84c59e73e4753899ffe038e734a3287e91c6fd
44883 F20101114_AABMSW chinchuluun_r_Page_34.pro
d09e5573908c3c51999c3c8359374a6f
e08082fa3a687491ab00a45bba5ec072cc0d88b0
55276 F20101114_AABMLB chinchuluun_r_Page_39.jpg
b6161214113091bde1effac435460d96
843b26f70f0b322d6ebc4b1ec51d3c1e0ee5dbf9
17247 F20101114_AABMXU chinchuluun_r_Page_11.QC.jpg
15b510cbe6c57ad26206508f9523ada7
1436aae56550f83ffb1a53d8c2b216b379edbe33
19729 F20101114_AABMSX chinchuluun_r_Page_35.pro
d645ac8396c8a2f5ead701f7a5dfa046
04db6b57de006edf72d773ccfb2455f250c69b27
71147 F20101114_AABMLC chinchuluun_r_Page_40.jpg
b276360188f4373a75a33f3ffde69a7f
be4ea8aba95ddc2f84273a8abb96a3adaa3bfa0c
5977 F20101114_AABMXV chinchuluun_r_Page_12thm.jpg
a60b7ed70c88b9623031fa734c8e1c0a
4b67cfb13082fb091e73ebecdc18353ef5b15d13
F20101114_AABMQA chinchuluun_r_Page_27.tif
d9f1433817086b5b04abc97f6318ad38
887534eba9ca56a8118d77bbba83478dd5ee9241
38388 F20101114_AABMSY chinchuluun_r_Page_36.pro
2bc934336a45f37efbda73c22a442736
b667d3caefeb18e78ec2ec783c710c147cf4b7c3
41740 F20101114_AABMLD chinchuluun_r_Page_41.jpg
7454b967cc3905d9c0c5295d7064f35e
253b6b79d39a505d076b93043012e51c5be167e7
25441 F20101114_AABMXW chinchuluun_r_Page_12.QC.jpg
b42874627ca977bd0b6f6e0127f14a10
c19f254fb2f82e8bcb83939b5b1a567e40123efe
F20101114_AABMQB chinchuluun_r_Page_28.tif
c72048813f74b862671ce7371f151454
674209a47903d9261790c742750c1a217bd55ab5
18647 F20101114_AABMSZ chinchuluun_r_Page_37.pro
de11b1a025fcd531cf98395fd235d5c5
9b16ab53eff7729413d472ba50dab9d1a61848ed
86707 F20101114_AABMLE chinchuluun_r_Page_42.jpg
ce1e47a63623904af2bfca357124ee43
f31540081f45dc78392d276766bea6b95e9ebb74
6081 F20101114_AABMXX chinchuluun_r_Page_13thm.jpg
3d5b1e6506e7fd1ccbb890bb5aceaadd
ff5b9c16cf8f98b3a3c80c07bed603052e24e0e7
F20101114_AABMQC chinchuluun_r_Page_29.tif
c743298cd7d19f19e9c80fccdd626484
a74983db59117e25fe00aa0c49d3c4af205b83c3
67734 F20101114_AABMLF chinchuluun_r_Page_43.jpg
da90d6c121e98c42c266b7e5440e559e
5406164db8b173fe14e6cc19624ad3007a7f021d
26318 F20101114_AABMXY chinchuluun_r_Page_13.QC.jpg
5dbd358769e6e9f596e03f220f0ccc2a
964b8cee58d099e7ecde73304f29e3a03e3c80ff
F20101114_AABMQD chinchuluun_r_Page_30.tif
88845aadaf11040a781467d76b44b5bc
613feed3b12a6e9f495b5da1290ec5055f32abaf
77913 F20101114_AABMLG chinchuluun_r_Page_44.jpg
58df7f661a5726847c6ce06da3fb3a41
5167eb13e23cd27e205a51b939f7a537343d7223
2175 F20101114_AABMVA chinchuluun_r_Page_19.txt
7112856c05079962155973837da1ef68
4e49e5d1b6d601c13c9154aeb81e0bdf817dfa05
6483 F20101114_AABMXZ chinchuluun_r_Page_14thm.jpg
d8dbee777af2327095e66afe9b6e6554
37ed3adbbf34a4a7c683d3f32f54c95b072910f2
F20101114_AABMQE chinchuluun_r_Page_31.tif
2066735df420eddc1f484ab2014eb968
087a0d95be3ff5a8918eab9d3242c5e506d01c48
2104 F20101114_AABMVB chinchuluun_r_Page_20.txt
5416b8c98405e573d49c9bd7c77bfa7b
386759a65a094aa9b9e3bf29048aaec09807c455
F20101114_AABMQF chinchuluun_r_Page_32.tif
beba25b14f59c3d820fed46592a25eac
be3a14df0c3b6a0dedc46a963432897074753e9e
61879 F20101114_AABMLH chinchuluun_r_Page_46.jpg
79cb9775c1253d5432dbd313740f65da
2796c9190cb5437c9048664d822029194df29e45
2185 F20101114_AABMVC chinchuluun_r_Page_21.txt
da1662d28d0b82a5884ced4522f3a84e
5bb3df742476558dd1f824cbacb0ac3cd3eb8fb9
7082 F20101114_AABNCA chinchuluun_r_Page_76thm.jpg
55a430ba1162dfa39ba4ed5bbb8af816
ee3d72198c31d664fbc7c00e0bf64af56b201ab7
8423998 F20101114_AABMQG chinchuluun_r_Page_33.tif
a09a55609eaba8552b7e0e0a959aa76f
01b8841d29652f6267681769d8290b7ed59a62ca
49260 F20101114_AABMLI chinchuluun_r_Page_47.jpg
99fa22d29e600f3a2c8605b1b8ad614a
872b36b027bf8cdc0354c65dbd7d69cd744a0da4
2127 F20101114_AABMVD chinchuluun_r_Page_22.txt
60fb8dabf2431c16462ec353140d54cd
d779f805fc23f6924c0bd1451de052187b3d73d9
31021 F20101114_AABNCB chinchuluun_r_Page_76.QC.jpg
393c76a539ddf2f50bab41262f764835
463bbcdd4931a90e550ba51bbffe8c665f33325c
F20101114_AABMQH chinchuluun_r_Page_34.tif
ad4b7cd8ae672a6e0c79264488f464ff
5f001622f27a10ff29396569c6e17df1fe262586
42894 F20101114_AABMLJ chinchuluun_r_Page_48.jpg
856b35718ae8891478e015dc8eecfb4e
7f74241c98ab7bf0ca0d084ddfb161048ff129ac
2157 F20101114_AABMVE chinchuluun_r_Page_23.txt
35c1e41488f0631e38474b828515a261
17674476da380f028c2bf267fdfbb3ba8f5b21d9
14137 F20101114_AABNCC chinchuluun_r_Page_77.QC.jpg
606b521ec82096b8bacf69654884a7e4
9eabaaf1f667815273598a90b28717d3ce199d62
F20101114_AABMQI chinchuluun_r_Page_35.tif
9e978a942e2a7fd5c087d8125860d7cf
9b296f098a620f230d0e7b8ca076f81b01fff9c3
43803 F20101114_AABMLK chinchuluun_r_Page_49.jpg
0702e7bf0c50762346cb3f152ed039ae
edfb942da62d2619603ee8a6ffc4ca30af20e601
265 F20101114_AABMVF chinchuluun_r_Page_24.txt
b1da0550fbdd75300ee7b7e06f1dc18c
8096bbe794a74b3f9b2734121fc80e92d558c02e
3087 F20101114_AABNCD chinchuluun_r_Page_78thm.jpg
9e73079f000cf8f313cc36fee20f145c
e1fb5d943bdc0be6f67b004f6c44190bcad61dcf
F20101114_AABMQJ chinchuluun_r_Page_36.tif
030e9800d38a76d37e7f6ea22821cbbf
4b5f6e68a6e9073469115f942533bd5ba4d7cdfa
52717 F20101114_AABMLL chinchuluun_r_Page_50.jpg
4137225e9a724a71de1b1c9dbb285d2a
8f62af020f049e7f7494f98367d53193b0e2061d
2053 F20101114_AABMVG chinchuluun_r_Page_25.txt
48ad41f260c10d3e94586c8a482e734b
71eab7dc2c4cf583e5e9775c06588515eb8219c1
11931 F20101114_AABNCE chinchuluun_r_Page_78.QC.jpg
291e0adcc5790ced82176b949b08af91
1d9b72ee787dc3199185c3a2e4fb90ab8dad154d
F20101114_AABMQK chinchuluun_r_Page_37.tif
1de944f1947014d2e54c91f790dccfa5
d1895d9207f4ce33e368c5b7dd90526b10791537
36159 F20101114_AABMLM chinchuluun_r_Page_51.jpg
193e0e4b091ae67e64cc26deec860c94
871cd62a60c8929121565fe15cd2c7a94c45b404
651 F20101114_AABMVH chinchuluun_r_Page_27.txt
7ab659a32508e19f6cff66dce8900580
c7d52e93117aa62e60da35967f7a037bc800ef0c
93645 F20101114_AABNCF UFE0021209_00001.mets
6efac94339a82f1cb6f2b2f9da2545e7
e543d12f0cc9c3e7a9b541527062f143a0f37757
F20101114_AABMQL chinchuluun_r_Page_38.tif
b13bdfed1c5fc9a8e7b8e4cf3155ed78
916ba6f7997ac5084fd24a8f734a24de8c365505
70155 F20101114_AABMLN chinchuluun_r_Page_52.jpg
770b100ea2ef6a3392f477a876bdc9df
f5f1c9b96e3a75681b5293594d607e25a7edc10a
1569 F20101114_AABMVI chinchuluun_r_Page_28.txt
4e8ea3a8218685a418146ec8f0a79e11
e24ecd6d84ca07e3af55df50a6b0f50ce51374d3
82820 F20101114_AABMLO chinchuluun_r_Page_53.jpg
81774a236e0beaae155c3bcdb02bab24
f8b929e380683eece1396269251b1f6e54426f65
1795 F20101114_AABMVJ chinchuluun_r_Page_29.txt
bb496849bc62e750f1ff5080b8887e59
5e49c273b4aba9c2a579387de5eb23cab8036ba6
F20101114_AABMQM chinchuluun_r_Page_39.tif
3651a3d760fb8aa44997c07235fed409
db13a035458734e6baad3dcbe3512fa45365eca9
52698 F20101114_AABMLP chinchuluun_r_Page_54.jpg
1fe8fb1263dede6ca32ee1e71eb04822
186e050da7911f4e95af97d45a18cfdc9ff8d730
1793 F20101114_AABMVK chinchuluun_r_Page_30.txt
449f566803ff353ebb9989895f348254
98a2cb82feea08de5815cd76ff7fb6e882033fcb