Available online at www.sciencedirect.

com
Available online at www.sciencedirect.com

Procedia
Engineering

Procedia
Engineering
0023
(2011)
000000
Procedia
Engineering
(2011)
351 366
www.elsevier.com/locate/procedia

Review on fruit harvesting method for potential use of

automatic fruit harvesting systems
Peilin Li, Sang-heon Lee, Hung-Yao Hsua*
Student, Division of ITEE,School of AMME,University fo South Australia,Adelaide, South Australia
PHD,Division of ITEE,School of AMME,University of South Australia,Adelaide,South Australia

Abstract
In horticultural industry, conventional harvesting is done by handpicking methods to remove hundreds of fruits such
as citrus fruits in random spatial locations on the individual fruit trees. It is well known that harvesting fruits in a
large scale is still inefficient and not cost effective. To solve this challenging task, mechanical harvesting systems
have been investigated and practiced to enhance profitability and efficiency of horticultural businesses. However they
often damage fruits in the harvesting process. Development of efficient fruit removal methods are required to
maintain the fruits quality. This paper reviews fruit harvesting systems from purely mechanical based systems in
which operator involvement is still required, to automatic robotic harvesting systems which require minimal or no
human intervention in their operation. The researches on machine vision system methodologies used in the automatic
detection, inspection and the location of fruits for harvesting are also included. The review is focused on the citrus
fruits due to the fact that the research on citrus fruit harvesting mechanism is a bit more advanced than others. Major
issues are addressed in the camera sensor and filter designs and image segmentation methods used to identify the
fruits within the image. From this review, the major research issues are addressed as future research directions.

2011 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.

Keywords: Fruit Harvest; Mechanical Harvest; Automatic Harvest; Physical Sensor; Machine Vision; Image Processing

1. Introduction
In recent years the horticultural industry has funded the research and the development of novel fruit
harvesting systems. Motivation for the research is to decrease harvesting cost and increase the value of
their product to the consumer. Conventional harvesting method is highly labour intensive and inefficient
in terms of both economy and time. Machine harvesting systems are a partial solution to overcome these

* Corresponding author. Tel.: +61-8-83023826; fax: +61-8-83025292.

E-mail address: liypl001@mymail.unisa.edu.au.

1877-7058 2011 Published by Elsevier Ltd. Open access under CC BY-NC-ND license.
doi:10.1016/j.proeng.2011.11.2514

2352

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

issues by removing fruits from the trees efficiently thus to reduce the harvesting cost to about 35-45% of
total production cost[1]. Two streams of harvesting systems have been researched and attempted through
years since early 60s. They are mechanical harvester and automatic harvester.
Mechanical harvesting systems are designed to achieve mass removal of the fruit during the harvesting
season. This method has been practiced using such as shaker or air blast with chemical mechanics of
abscission as pre-harvest agents to loosen the mature fruits. There are some issues in mechanical
harvesting system such as the quality and size selection, the damage to the fruit and trees in some cases
and the layout of the grove design for mechanical harvester. A post selection process can be appended to
maintain the aesthetic appearance and value of product for the consumer. However, the mechanical
system operates blind when it come to removing quality ripe fruit.
In the early 1960s, the concept of an automatic harvester was proposed and investigated by Schertz
and Brown[2] using automatic robotic picking device. They proposed a system which uses a robotic arm
to position a manipulator within the picking range of target fruit before detaching the fruit from the tree.
The guidance for the manipulator is achieved by a machine vision system to detect the fruit. However the
nature of the horticultural environment makes the fruit detection a challenging task. It is well known that
the robust solutions are still largely underdeveloped. The issues are multifactor of such as the
unstructured environment, limitations of the sensors, and a robust methodology. Hence this review paper
is focused toward the novel sensor designs and image processing methods which aim to present the reader
with an up-to-date account of useful methods found in literature to overcome such issues.
The remaining of the paper is organized as follow. An overview of the mechanical harvester is
outlined in section two. Relevant research articles regarding the development of automatic harvesting
machines are reviewed in section three. In an automatic harvester, the image sensor for data acquisition
and computer aided processing are the main components used to successfully detect the fruit. Those are
detailed in section four and five. The major issues are discussed in section six. Last section presents the
summary on this review paper with the future directions of the research.
2. Mechanical harvester
Mechanical harvesting methods have been investigated and practiced since early 1960s[3]. Coppock
stated that citrus fruit can be harvested by mechanical means after proving that a tree could be
mechanically shaken to remove the fruit from the branches without destroying the whole tree. To reduce
the physical damage to the tree, a pre-harvest abscission spray was also proposed to loosen the fruits on
the tree. In order to improve the design of mechanical harvester the biological and physical properties of
the fruit were also studied by Coppock et al[4]. This section will review those mechanical harvesting
methods and others very briefly. For a more complete review, see Sanders[1]. The mechanical harvesting
methods reviewed here are limb shaking, air blasting, canopy shaking, trunk shaking, and the use of an
abscission chemical agent to loosen the fruits.
2.1. Limb Shaker
An early limb shaker was represented by Coppock and Jutras[5] using inertia developed by Adrain and
Fridley. An eccentric weight about 85 pounds was rotated in the mechanism to produce the shacking
action after the shaker was attached to the tree limb. Notably some damage was made to the bark of the
tree by the clamping mechanism. An alternative tree shaker was represented using fixed stoke, inertia,
and direct impact on trees limbs[6]. The issues from this practice included such as fruit damage due to fall
foliage, lower removal rate in earlier and mid of harvesting season, and large or small immature fruit
removal. Another tree shaker with two catching frames each with an inertia type limb shaker was

Peilin
et al.//Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

developed[7]. Still, immature fruits were removed with damages to the fruits. A self-propelled limb
shaker was tested on Valencia orange[8]. A self-propelling full powered positioning limb shaker was also
evaluated with abscission aid by Summer and Hedden[9].
2.2. Air Blast
The application of force generated by airblast to remove the fruit started in 1961. An oscillating air
blast machine was tested and practiced by Jutras and Patterson[10]. Fruit removal was maximized by the
oscillation rate. The air blast model and all the subsequence models were made and named after FMC
(Food Machinery Corporation)[11]. The performance of FMC series was dependent on factors such as
structure of tree, size and weight of fruits. Later, an air shaker was designed and constructed to alleviate
issues such as the high power requirement[12]. However still damages to the fruits and leaves were the
major issues addressed in the project.
2.3. Canopy Shaker
A canopy shaker was designed to clamp secondary limbs to shake vertically[13]. The shaker was
extended into tree with a pantograph lift unit and shake always vertically. An excessive immature oranges
were removed during tests conducted by Summer. Two continuing canopy shakers were reported by
Futch and Roka[14], one was self-propelled unit and another was tractor-drawn unit. These two units
were used for juice processing plants. Manual workers were needed to collect the fruits after the harvest.
Shaking frequency and stroke are important factors in the performance in this type of harvester and it
requires more tests to determine the optimal values.
2.4. Trunk Shaker
Trunk shaker was used to remove deciduous fruits and nuts but difficult to apply this technique on
citrus fruit removal[15, 16]. However, Hedden and Whitney designed the experiment to evaluate the trunk
shaker for earlier season Hamlin orange and late season Valencia orange using different unbalanced mass
and multidirectional shakers for years. The linear low frequency shaker with a larger displacement
performed better than the canopy shaker machine. The bark was more or less damaged during the
experiment. Later the trunk shakers were tested along with other canopy shakers by Whitney[17]. The
efficiency or removal was from 67% on large trees to 98% on small trees. More recently, a tractor
mounted trunk shaker was tested on varieties of oranges and mandarins in Spain by Torregrosa et al[18]
in comparison to a hand-held shaker. Overall the tractor mounted shaker was more effective with 72%
detachment than the hand-held shaker with 57% detachment. In test, the fruits picked up from ground had
high percentage of bruise. Defoliation was high at high shaking frequency and the bark was damaged in
season of May and June.
2.5. Abscission Chemical
Abscission chemical agent was designed to loosen the mature fruit and improve the rate of removal of
fruit in harvesting season. There are many kinds of abscission agent such as Ethephon and 2-chloroethyl
phosphoric acid[19]. The use of abscission agent was applied as pre-harvest process and constituted part
of harvesting such as air shaker[20]. Air shaker was tested with applying abscission agent in advance on
FMC-3 by Wilson et al[20]. Limb shaker used abscission to loosen the fruit on stem[9]. It was noted that
abscission chemical was inconsistent in practical use[8]. The abscission use was subjective to many

3353

4354

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

factors such as weather factors, tree injury, and cost of using chemical such as equipment. The
Prosulfuron, an abscission chemical agent which was used on Hamlin and Valencia oranges loosening,
were studied by Kender et al[21]. This abscission was more effective in Hamlin than others. However, the
immature Valencia was loosen before harvest. The CMN-P abscission chemical was tested on Hamlin
orange before the harvesting by trunk shaker[22]. In the study, the fruit rind burn and pre-harvest drop of
fruit were observed. The orange removal with various chemicals was discussed in[23]. Three abscission
chemicals and four types of trunk shaker with scissors type clamps were used in the test. Usually the
chemicals helped to loosen both Hamlin and Valencia oranges and improved the removal rate by 30%.
The main issue from the test was the pre-harvest drop of fruits. The chemical Methyl Jasmonate (MJ) was
tested on whole tree or canopy sectors of Valencia oranges[24]. High MJ caused great loosening of fruit
but excessive defoliation as well. The combined use with other compounds was proposed for later
experiment especially on Valencia orange harvest. A CMNP abscission chemical was tested on Hamlin
and Valencia oranges by Kostenyuk and Burns[25] before the mechanically harvesting. For all of these
methods mentioned, the consistency of the performance is the major issue.
2.6. Post-Harvest
Peripheral mechanical harvesting systems such as catching frames have been designed to reduce the
damage on falling fruits[26]. These frames also help to collect the fruit efficiently. In addition, a tractor
pulled rake was made to windrow the fruits to the drip-line and the windrowed machine was used to pick
up windrowed fruits[27]. Both methods were not adaptable to all grove conditions. The grove conditions
should be considered to maximize the efficiency of all mechanical harvesters based on the comparison
study on different mechanical harvesters[28]. The quality of fruits picked up by a harvesting machine was
evaluated. Also the potential for microbiological contamination to the process was also presented by
Goodrich et al[29].
3. Automatic harvester
The mechanical harvesting system cannot maintain the quality and size selection that the human vision
can. The automatic individual harvester was considered as an alternative method to the mechanical
harvester by Schertz and Brown[2]. The two detachment devices, a vacuum twist device and a rotating
cutoff device, were used in the research. The photometric comparisons showed the potential use of the
light reflectance for the fruit detection. The concepts were further developed by Parrish and Goskel from
University of Virgia in 1976[30]. Much concrete works began at around 1983 at Kyoto University at
Japan, and at University of Florida. Then CEMAGREF Montpellier France[31] extended it on the
subsequent projects.The state of arts of individual robotic harvesting systems has been reviewed in this
section. The study focused on the sensors and the vision system in mechanical manipulator. Even though
the review is majorly focused on the citrus fruit harvester, some other significant robotic harvesting
systems for different kinds of fruits harvesting are selected and reviewed as a complementary reference.
3.1. MAGALI Project
The basic idea of automatic harvesting system was early developed at Montpellier France[32]. A
camera was fixed in the pathway to detect the fruits and a grasping tool was sent to the fruit within
straight line trajectory. This basic concept was tested on a labouratory prototype in May 1984. A simple
black and white(B/W) camera was positioned above the telescope arm aiming fruit horizontally in this
model. Years later, the video colour camera was applied to replace the B/W camera with a proximity

Peilin
et al.//Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

sensor to sense the fruit touch in grasper. This first prototype of self-propelled robot was built to harvest
the golden apple[33, 34]. The prototype used three degrees of freedom and hydraulically powered
spherical coordinate with joint position compensation and an end effecter shaped as suction cup. The
camera was positioned at the centre of rotary joints at the base of the manipulator. The vision coordinates
coincided with the robot coordinates making the end effecter in line of sight. More than 50% fruits
were detected and harvested with 75% fruits with the stem on. In average, each fruit was harvested in
around 4 seconds.
3.2. Florida Citrus Picking Robot
Issue for the line of sight from MAGALI project was the obscure between the vision and the target
during the later picking stage. Inaccuracy of guidance was accounted when the fruit was moved by wind.
Harrel et al overcame this issue by attaching the colour video at the end of the effecter on a spherical
manipulator with an ultrasonic transducer to compensate the distance to the fruit[35, 36]. This eye-inhand approach used the contrast colour between the fruit and the background to track the fruit. The cycle
for harvesting was about 3 seconds to 7 seconds with 75% fruit detected.
3.3. Eureka Projects
Spanish and French research team cooperated on a project to harvest citrus fruits[37, 38]. The gripper
was a spherical manipulator with hydraulic power, and the camera positioned in the centre of the
manipulator. In vision detection, a single B/W camera of gray level scheme was tested using 635nm
wavelength filter and 560nm wavelength filter with a supported flashing light. Then the colour image for
the mature fruit detection was used with Bayesian classifier as discrimination function. The colour
scheme was superior to the monochromatic scheme with over 90% fruits detected. The occlusion by
leaves was one issue found in the project. The most failures were the distance problem in the approaching
stage when it is close to the fruits. Hence the need of knowing distance from manipulator to the fruit
target is of all importance in tests.
3.4. Agribot Project
Agribot project from Spain developed a semi-automatic harvester by combining both human and
machine functions. Two jointed harvesting arms were built and were mounted on the human guided
vehicle[39]. The manipulator was three degree of freedom(DOF) design with one vertical rotation axes
and two horizontal rotation axes and the rotation on the end effecter with gripper. The gripper was
specially designed with the pneumatic suction cup and proximity IR sensor to sense the attached fruit
followed by the saw cutter to cut the stem. Detection was done by the human operator when the vehicle
was placed opposite the fruit tree using a joystick to control the pan and tilt mechanism pointing to the
target. Fruits were localized in spherical coordinates with distance detected by laser telemetry and the
angular position by two rotational axes. The laser ranger finder was sensitive to the outdoor lighting
conditions. Hence the special goggles were used to block and enhance the red laser spot. The harvesting
cycle time for this machine was 2 second.
3.5. CRAM Citrus Picking Robot
In Italy, researchers and collabourators from the University of Catania developed a citrus picking
robot[40]. A Cartesian based manipulator was housed on a vehicle with a caterpillar to maximize the

5355

6356

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

reachability. The top arm was equipped with a pneumatic piston to extend the gripper far inside the
canopy while the lower arm was not equipped the pneumatic extension. In the vision system, the index RG from RGB colour model was adopted to segment fruits from the background. The fruit was located by
estimating the diameter. The distance was estimated based on the dynamic incremental movement of
manipulator. Kalman filter was used to estimate the real dimension of the fruit in the reaching stage in
sequential estimation. In simulation test, the picking time per fruit is about 5.93 second.
3.6. Other Fruit and Vegetable Robotic Harvester
Some other robotic harvesters have been reviewed and listed. Buemi et al[41] from Italy presented a
robotic harvesting system on AGROBOT project to harvest the tomato in green house. The system was
based on a vehicle carrying a six DOF of picking arm consisting of gripper and hand, two micro cameras,
and the control system with the image frame grabber and the image processer. The components of hue
and saturation from HSI colour space were used to perform threshold to segment the image. A prototype
of robotic melon harvester was developed and practiced by joint researchers from Israel and USA[42, 43].
In Japan, Kondo et al[44] has investigated harvesting method with a robot for tomato, cucumber, and
grape. In Netherland, Henten et al[45] presented a greenhouse robot to harvest the cucumber using an
industrial manipulator with 7DOF to position the end effecter. Hayashi et al[46] presented an eggplants
robotic harvester. In Belgium, Baeten et al[47] integrated the industrial manipulator with the newly
designed and patented flexible gripper which consisted of the silicon funnel with camera mounted inside
the center of the funnel to harvest apples.
4. Machine vision system in automatic harvesting
The vision system used in the automatic harvester aims to detect the fruits and provides the
information of the location and the distance to the fruits to the robotic controller. In vision recognition
system, vision cameras are mainly the solution to communicate with the environment. The major
achievements of vision systems and the performance of the various sensors in the harvesting have been
reviewed by Jimenez et al[48]. The sensor techniques applied in agricultural applications have also been
reviewed by Chen et al[49]. Normally, the digital image data has been categorized in three types;
intensity, spectral, and laser range finder[48]. On top of the basic colour processing, new technologies
have also been investigated especially in the measurement and spectral analysis such as the hyperspectral
imaging technology(HIS)[50] and the thermal imaging technology[51]. The visibility of the fruit citrus on
canopy has been studied using multiple perspective viewing method analysis as well for a harvesting
robot[52]. This section will review those approaches.
4.1. Monocular Scheme
In applications, a single gray scale camera or a colour camera has been widely applied on citrus or
orange detection[48]. The Italian AID robot vision system adopted a colour camera with artificial lighting
to detect oranges. Approximately 70% of the visible fruit were recognized. Slaughter and Harrell[53]
used a digital colour camera with a filter of 675nm wavelength to enhance the contrast between orange
fruits and others. Over 75% orange were detected with some non-fruit part misclassified. There are three
separate vision system development projects found in European Eureka Project[37, 38]. A single
grayscale camera was tested using 635nm filter for citrus and 560nm filter for leaves. Then two cameras
with a red filter and a green filter were tested to segment the fruit respectively. Finally a colour image for
mature fruits was used for fruit identification with over 90% fruits detected. Regunathan and Lee[54]

Peilin
et al.//Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

used a colour camera to capture the citrus fruit with an ultrasonic sensor to obtain the distance between
fruit and camera. Issues such as partial occlusion, variable illumination, and clustering of fruit caused
errors in both fruit count and size estimation were addressed.
A single camera has been applied also in apple detection. Parrish et al[30] used a single gray scale
camera with a colour filter attached to detect the fruit apples. In MAGALI project, both B/W and colour
camera were used to detected the apples[33, 34]. Later to overcome the obscure issue of this scheme,
Harrell et al[35, 36] designed the vision system by positioning the colour video at the end of effecter on a
spherical manipulator with an ultrasonic transducer to compensate the distance to the fruit. The success
rate was about 75% for picking attempts and about 50% from all fruits[48]. Sites et al[55] used a single
CCD camera to catch gray level fruit images of peach and apple with filters. Up to 92% fruits were
detected. Bulanon et al[56] used a colour CCD camera to acquire images and 80% apples were detected
under natural lighting condition. Zhao et al[57] used a single colour image to segment the fruit apples
using both colour and texture. In tests, 18 apples were recognized out of 20 visible apples. One of
interesting practices with video sequences was presented by Tabb et al[58] to segment the apples fruit in
labouratory environment. Up to 95% of apples were identified. In Belgium, Baeten et al[47] used the
colour camera positioned inside the center of the gripper in a funnel shape. About 80% apples were
detected and harvested in the experiment.
Some other fruits and vegetables have also been practiced using this scheme. Whittacker et al[59] used
a greyscale video camera to capture the tomato images. Average 68% tomato was detected with a proper
threshold. In melon harvesting system, two single B/W cameras were positioned as far vision and near
vision sensors for planning and controlling level respectively[42]. About 80% melons were detected in
experiment. The author proposed the improvement by combining infrared and visible image or by
merging multiple sensor fusion methods. In AGROBOT project, two colour cameras were positioned on
each of two DOF manipulator head mechanism. Later it was also found that the use of stereoscope
matching scheme could extract the 3D information of object from both segmented results from two colour
images[41].
4.2. Binocular Stereoscope Scheme
An early approach using stereo vision system was presented by Kassay[60] from Hungary AUFO
robot project. Two colour cameras were designed and positioned separately in a distance with an angle
converging to the same scene. The photogrammetric principle was applied based on a triangulation
algorithm. In the experiments about 41% of visible apples were detected. Similar to this scheme, Grasso
et al[61] developed a stereo matching algorithm to recognize oranges using two images captured from
different view of angles. Another stereo matching method was presented by Plebe and Grasso[62] in
which two cameras were carried in the center of the gripper on each of two robotic arms. The 3D match
was done based on disparity constraints. In Italian AGROBOT project, two colour cameras was designed
and positioned on two DOF manipulator heads[41]. The 3-dimensional information was then extracted
from stereoscope matching of two tomato images on the same scene. Takahashi et al[63] presented
another binocular stereo vision system. Some issues were found commonly. The measured disparity was
difficult to guarantee the estimation of the dynamic shape of the natural fruits. The intensive computation
of transforming the disparity between the objects in two images into depth information was costly. The
lighting conditions and occlusion problems affected the detection efficiency.

7357

8358

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

4.3. Structured Light Scheme

Jimenez et al[64] presented a laser range finder sensor for AGRIBOT robot on orange detection. This
sensor gave the spherical coordinates of the scanned scene points. The range and attenuation data values
of scanned surface were identified by the laser energy depending on the distance, surface type, and the
orientation of the sensed points. Totally four types of images were used in the system. In the image
processing, the shape analysis could detect both mature and green fruits if the contour of the fruit was
visual more than 70% of the whole. Overall 87% visible fruits were detected without any false detection.
Benady and Miles[65] presented the use of the structure light with a laser profiling system to locate the
melon. Saint-Marc et al[66] placed two cameras on both sides of the sheet light plane with an angle to
overcome the occlusion problem. Two scanned images from both cameras were matched based on the
calibration. In Japan, a custom built laser range finder was applied by Kondo[44] for tomato, cucumber,
and grape harvesting. The custom-built active range finder projected and received two lasers light in the
same optical axis and the distance from the sensor to the target was calculated by the demodulated signal
from both beams. A similar idea was found in a cherry robotic harvesting system from Japan by
Tanigaki[67]. The 3D sensor was custom built and equipped with red and infrared laser diodes to detect
the ripe cherry. In the experiment, 10 out of 12 fruits were recognized. The use of structured light had
faced some issues such as the sparseness of measurements, occlusion, and the reading of the orientation
of the surface. Scanning could be improved by parallelizing the scanning and the analysis process[64].
4.4. Hyperspectral Scheme
Hyperspectral technique has been applied in quality inspection analysis and measurement applications.
Generally, the hyperspectral sensor is designed to disperse the whole sensible spectral area into discrete
wavebands of spectra. In the output image, each pixel collects stacks of the waveband segments
continually across the whole sensible spectrum. A hyperspectral system was developed by Mehl et al[68]
to detect the defects and contaminations on apples. The wavebands between 542nm to 752nm gave a
great statistical disparity between defect surface and normal surface of apples. Ye et al[69] used an
airborne hyperspectral remote sensing to predict the citrus yield. Ariana et al[70] used a near infrared
hyperspectral imaging to detect the bruises on pickling cucumbers. In the experiment, the waveband ratio
and difference were preferable since the classification between the normal and defection areas was more
consistent in those two. Okamoto et al[71] present a green citrus fruit detection using a hyperspectral
imaging camera. The camera employed could sense from 369nm to 1042nm waveband areas. Three
different types of green citrus fruits were captured by this sensor; Tangelo, Valencia, and Hamlin. The
time to acquire a hyperspectral image varied from 22s to 65s depending on the scanning area and the
image resolution. The main cause of incorrect identification was the similar spectral characteristics
between the green leaves and the green fruits.
4.5. Thermal Imaging Scheme
The infrared thermograph has been applied in application using sensors responding to longer
wavelength. Stajnko et al[72] developed a method for estimating the number and the diameter of apples in
an orchard using a thermal camera. The thermal sensing was based on the heat emission depending on the
exposure time from the sun heat. A recent project using thermal infrared imaging in fruit citrus canopy
was presented by Bulanon et al[51, 73]. The thermal radiation which was based on the Stephan-Boltzman
law was used as an indicator of the temperature of the emitting object. It was found that the large
temperature difference between the fruit and the canopy occurred around 4:00 pm. The histogram based

Peilin
et al.//Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

algorithm performed well during the selected time. Later the author developed a fusion technique to
improve the fruit detection using both colour image and thermal image. The registration was required to
align two different images from thermal sensor and the visible camera by a half-meter PVC pipe frame. In
the experiment, the success rate was high when the temperature of fruit was warmer than the canopy. The
fusion technique using thermal imaging and colour imaging could be a practical alternative during an
appropriate time period.
4.6. Multispectral Scheme
Instead of dispersing spectrum into the discrete numerous wavebands, multispectral image capture the
specific across wavelength area in the spectrum. The early multispectral scheme was presented by
Rabatel[74]. The custom built camera consisted of three CCD micro cameras side by side with three
different optical waveband filters in terms of 550nm, 650nm, and 950nm wavelength respectively[48].
Three images were calculated and combined to generate a binary image. The detection of fruit was about
75% when the sky was overcast. This scheme initially gave the implication to possibly detect immature
green fruits by combining different waveband spectra. Kane and Lee[75] used a monochromatic near
infrared camera equipped with multi waveband pass filters to capture citrus fruit tree images. Averagely
84.5% correct citrus pixels were identified. This work was extended after the measurements on the green
leaves and the green citrus fruits through types of citrus and seasons[76]. Three waveband filters were
attached to the camera respectively to catch waveband spectral area images. However the resultant
multispectral images were not well synchronized and matched in dynamic scenes. Another issue was the
saturation of sensor value and the dark area on the image. The reason that the number of leaves caused the
diffuse reflectance was theoretically studied[76]. It was proposed that the multispectral images should be
captured at the same time with capability of acquiring more wavebands and a smarter image processing
technique. Dobrusin et al[77] carried out their preliminary investigation on the detection of melon using
infrared image. They proposed the method by combining the infrared and visible image to improve the
detection result.
The multispectral scheme has been more designed in inspection analysis applications. In Spain,
Aleixos et al[78] presented a machine vision scheme on citrus online inspection using two CCD cameras
with one camera for visible area detection and the second for infrared area detection. Lu et al[79]
presented a multispectral scheme to predict the firmness and the soluble solids content of the fruit apples.
More recently, Kise et al[80, 81] has developed two types of portable multispectral imaging system, a
dual-band spectral imaging system and three-band spectral imaging system for contaminant detection in
food inspection industry.
5. Methods in image data analysis on fruit harvesting
Image processing schemes applied to the images are basically a recognition problem to recognize the
fruits. In[48] the methods have been categorized into two groups as local based and shape based. In
general a procedure may be accomplished by several steps: image acquisition followed by the feature
construction, preprocessing to improve the quality of images by such as noise removal, segmentation to
cluster the image into different classes, and classification or interpretation by assigning meaning to
classes. Regarding the nature of the task in the fruit image analysis, the segmentation part seems to be the
centrepiece of all other tasks in analysis. Some of the analyses have been further discussed in this section.

9359

360
10

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

5.1. Color Based Analysis

In colour based analysis, the use of threshold or index is considered as a fast way of segmentation. The
physical optical filters have been widely adopted to enhance the specific area for the threshold purpose
with improvement on the non-uniform distribution. A large handful work has been applied on citrus fruit
images. For example, Slaughter et al[53] used the hue and saturation components as chrominance
information to segment the oranges from the background. From European Eureka project, the first
approach used a fast threshold algorithm on the data acquired from a B/W camera with 630nm
wavelength filter and achieved 80% visible fruits detected. In the second approach, two cameras were
equipped with 630nm and 560nm wavelength filters on each of them. The ratio between both images was
calculated and used. About 80% visible fruit were detected finally[38]. Similar approaches on citrus
images have been reviewed and detailed in[78, 82, 83]. A similar method has also been applied on
hyperspectral images[71] to detected the green citrus and on a thermal images[73] to detect the citrus
fruits. Later, a method by fusing the visible image and the thermal image was developed to improve the
fruit detection in[51].
The physical method has also been applied on fruit apples and other vegetables. For example, Parrish
et al[30] used a No.29 red filter to enhance the apples in image. A logical approach was used to find the
region of the apples by comparing to the defined region of the apple templates. Some similar approaches
on other fruits and vegetables are reviewed and detailed in works[41, 43, 55, 56, 74, 84].
Some colour indices have been applied to segment the image data. Zhao et al[57] approached an
algorithm on apple images using the texture properties. The index r = 3R (G + B ) gave high contrast
values for red apples. An index R B was used on citrus fruit image segmentation by Xu et al[85]. A
dynamic threshold value was calculated based on the image maximum and minimum intensity to classify
the citrus fruits and the background after segmentation. Stajnko[72] used a thermal imaging camera to
estimate the number and diameter of the apple. A normalized difference index NDI ( g r ) / ( g + r ) was
also adopted to contrast the image[86]. The global threshold was then applied to segment the fruits form
the background. Based on the study on the spectral measurement using spectrophotometer[76], Kane and
Lee[75] applied a normalized difference index similar to NDI on three waveband images instead of RGB.

5.2. Shape Based Analysis

Whittaker[59] applied a circular Hough transform on tomato image segmentation. This circular Hough
transform (CHT) uses the shape information which is independent of the colour difference between the
fruits and foliages. The detection rate was 68% with respect to percentile classification of 99%.
Benady[65] applied a modified CHT on laser range image data of melon. The interference of background,
leaves, or curvature from the other non-fruit features would be misinterpreted. Another issue was the
intensive computation by CHT[87]. Pla et al[88] presented a study on the characterization of spherical
objects on monochrome images based on Lambert law of diffuse reflector. The ellipsoidal mathematical
model on image was based on the illumination function received relative to the position of object and
intensity. Totally 75% visible fruits were detected and false detection was also caused by occlusion issue.
Pla[89] developed a method for a partial circle recognition using the segment contours information. This
method could be considered in the case that the contour could be extracted on for example the green
mature fruits. In the algorithm, the quantification of parameters like circle parameters was based on a
minimization criterion using the tangent at each point. The shape analysis could be integrated with the
advanced technique for the acquisition of multi channel images on the same scene. For example,
Jimenez[64] used a range sensor to acquire the multi images on the same scene and a dual colour and
shape methods were employed on different images from the same scene to locate the fruits.

Peilin
et al.//Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

5.3. Machine Learning Based Analysis

The difficulty underlying in the fruit image data is essentially the unpredictable distribution. Based on
the empirical data, the machine learning has been researched to predict the fruit image data distribution
using computational learning theory such as statistical tool method.
5.3.1. Statistical Based Analysis
One of the multivariate statistical pattern classification techniques, Bayes parametric optimal classifier,
has been widely applied on the classification based on posterior probability theory. Slaughter et al[90]
presented the Bayes classifier on the orange images. The Bayesian discriminant correctly classified 75%
of oranges. In European Eureka project, Juste et al[38] used a pattern classification method of Bayes
rules which was similar to the method by Slaughter. The success rate was reported up to 90% fruits
detected. Ferri et al[91] presented a method based on the nearest neighbor classification and multieditcondensing technique in a vision system for citrus harvesting. The multiedit algorithm allowed to select
prototypes which belong to Bayes accepted region while the condensing algorithm attempted to eliminate
those multiedit references which embedded in the internal regions of Voronoi polygons of each class. A
modified K-means algorithm approach was presented by Weeks et al[92] on orange image segmentation.
About 67% orange were detected with 54% accurate detection in the results. Mehl[68] used a
hyperspectral camera on apple surface defects analysis. The principle component analysis(PCA) was
employed in this scheme. Kane and Lee[76] used a spectrometer to measure the different citrus fruits
through seasons. PCA was applied to discriminate the classes between the fruits and the background. The
Fisher linear discriminant was also used to find the direction of the major eigenvector for the projection of
the low dimension space.
5.3.2. Neural Network Approach
In A.I.D. project from Italy, Grasso[61] transferred RGB to a modified HSV colour space. A
multilayer perceptron trained by backpropagation error algorithm was employed on the segmentation of
orange tree images. In the experiment, about 88% fruits were detected. Bulanon[93] reported that both
artificial neural network (ANN) and decision theoretic classifier were used on the segmentation of apples
in image for the harvesting robot. Both methods achieved 80% fruits detected. Regunathan[54] employed
a colour camera and an ultrasonic sensor with a neural network to compare with Bayesian and Fisher
linear discriminant methods. In the experiments, the neural network performed better in fruit size
estimation. Ye[69] approached a neural network via backpropagation learning algorithm on airborne
hyperspectral images to estimate the citrus yield. The study indicated that the hyperspectral data in May
had high correlation with the yield. Jan-jun et al[94] studied and compared a linear vector quantization
(LVQ) network based segmentation method with other three methods: dynamic threshold segmentation,
extended Otsu method, and improved Otsu method combined with genetic algorithm on fruit image. The
result showed that the traditional Otsu method was faster than other two Otsu methods. The segmentation
based on LVQ generated a poor result due to the nonuniform reflection and the high bright area. In
addition, the time cost of LVQ was higher than Otsu methods. The author concluded that the LVQ was
not adaptable for real time applications. However, the parametric adjustment for estimation functions in
learning algorithm barrier the use of such advanced method in real time harvesting application. On top of
that, more literature can be found for analytical or inspection purpose applications using artificial neural
network[95].

11361

362
12

Peilin
al. / Procedia
Engineering
23 (2011)
351 366
Author
nameLi/ et
Procedia
Engineering
00 (2011)
000000

6. Research issues and discussion

Although it is difficult to cover all materials on top of the projects undergoing to the data, from this
review, there is a need for the much further improvement to develop more practical and efficient fruit
harvesters. Even though mechanical harvesters, especially for citrus fruits, can provide an efficient and
economical way to replace the human labour, the damage sustained to the fruits during the harvesting is a
major issue that needs to be solved[96]. The premature dropping of underdeveloped fruit will affect the
subsequent seasons harvest[97]. Hence this is also another major issue for the mechanical harvester. In
addition, due to the space requirements by the mechanical harvester, the grove design and the tree
locations should be considered in advance to better suit for the harvesting. The abscission chemical
applied to help to increase the efficiency of the mechanical harvesters might also be a critical issue due to
the health regulation[1]. Despite undergoing improvement on mechanical shakers and the post selection,
the issue of the size and quality selection still limits the capability of the mechanical harvesters.
The recent prominent achievements in research have provided a vision for the potential use of
automatic harvesters. However, the operations under the open dynamic natural environment with
potential multi-factor make practical applications of automatic harvesters challenging and even
unacceptable. Obviously the final fully autonomous fruit picking robot would be based on the integration
of developments of those components under the principles for agricultural robotics[98]. A mobile
locomotive system would be self-navigated with the capability of automatic path planning, mapping or
position in the grove. The degree of freedom of manipulator is still limited in reachability on positioning
and extending further inside of the fruit tree canopy since the natural pose of fruits on trees requires a
wide degree of the orientation of the end-effecter. With the consideration of the multi-purpose application
in mind, the end-effecter design could be customized to be interchangeable grippers. In addition, the costbenefit issue will make the automatic fruit picking robot impracticable if the fruit identification and
picking speed is slower than those of picking by the conventional human labour. This is actually a major
challenging issue of the fruit picking robot.
To make the automatic harvesting robot be used in real application, the fruits must be identified
accurately and rapidly. In most literatures, the successful fruit identification rate is averagely around 70%
and up to 90% with variations in some cases. The inaccuracy of identification usually comes from
inefficient illumination, occlusion, or inaccurate distance estimation. Hence those issues should be solved
before any automatic harvester could be used in the practice. On top of it, the speed of response is critical
and challenging for the vision system in dynamic such as in a rapid motion. An efficient method needs to
be developed to extract the information of the fruit object accurately from the image for the manipulator.
On top of all, the ripeness study[76] on specific fruit spectral could be quantified as well to cooperate
with the vision system and methodology development.
7. Conclusion

The fruit harvesting systems have been reviewed in this paper. The main applications from both
mechanical harvesting systems and automatic harvesting systems have been collected. In addition, the
machine vision system has been focused covering the sensor schemes and methods behind them. From
the literature, mechanical harvesting systems show the advantage in mass production. The automatic
harvesting systems have been practiced to tradeoff the selection capability between the conventional
labour harvester and mechanical harvester. Still none of those systems match the capability of the human
labour faultlessly, especially in the vision and the recognition capability. The real time cost-effective and
fully automatic robotic harvester might have a long way from the final prototype yet. However some
interests for future research to develop the potential use of this automatic harvester is possible. The

Peilin
et al./ /Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname

navigation is yet developed for the base of the manipulator. The manipulator and gripper could be
innovated for multipurpose flexibility. The robust sensor development should include the integration of
types of data. The measurement on special fruits for ripeness study could be helpful in cooperation of the
recognition vision system development.
Acknowledgements

The author is indebted to all researchers for their achievements and for all the materials that have
been reviewed in this paper and beyond. This literature review commenced and based on the valuable
project report for robotic citrus harvesting from the centre of Advanced Manufacturing Research at
University of South Australia.
References
[1] Sanders K F, Orange Harvesting Systems Review, Biosystems Engineering, vol.90, no.2, 2005, pp.115-125.
[2] Schertz C E ; Brown G K, Basic considerations in mechanizing citrus harvest, Transaction of the ASAE, 1968, pp.343-346.
[3] Coppock G E, Picking citrus fruit by mechanical means, Florida State Horticultural Society, vol.1362, 1961.
[4] Coppock G E; Hedden S L; Lenker D H, Biophysical properties of citrus fruit related to mechanical harvesting, Transaction
of the ASAE, vol.12, no.4, 1969, pp.561-563.
[5] Coppock G E ; Jutras P J, Harvesting citrus fruit with an inertia shaker, Floriad Agricultural Experiment Stations Journal
Series, vol.1538, 1962.
[6] Coppock G E, Harvesting early and midseason citrus fruit with tree shaker harvest systems, Floriad Agricultural Experiment
Stations Journal Series, vol.2824, 1967.
[7] Coppock G E ; Hedden S L, Design and development of a tree-shaker harvest system for citrus fruit, Transaction of the
ASAE, vol.11, no.3, 1968, pp.339-342.
[8] Coppock G E, Harvesting 'Valencia' oranges with a limb shaker, Floriad Agricultural Experiment Stations Journal Series,
vol.4217, 1971.
[9] Sumer Harold R ; Hedden Scott L, Harvesting oranges with a full-powered positioning limb shaker, Floriad State
Horticultural Society, 1975, pp.117-120.
[10] Jutras P J; Coppock G E; Patterson J M, Harvesting citrus fruit with an oscillating air blast, in American Society of
Agricultural Engineers, no.62-155, 1962
[11] Whitney J D ; Patterson J M, Development of citrus removal device using oscillating forced air, Transaction of the ASAE,
vol.15, no.5, 1972, pp.849-860.
[12] Whitney J D, Design and performance of an air shaker for citrus fruit removal, Transaction of the ASAE, vol.20, no.1, 1977,
pp.52-56.
[13] Summer H R, Selective harvesting of Valencia oranges with a vertical canopy shaker, Transaction of the ASAE, vol.16,
no.6, 1973, pp.1024-1026.
[14] Futch S H ; Roka F M, Continuous canopy shake mechanical harvesting systems, in Horticultural Sciences Department,
Florida Cooperative Extension Service, Institute of Food and Agricultural Sciences, University of Florida, no.HS1006, 2005
[15] Hedden S L; Whitney J D; Churchill D B, Trunk shaker removal of oranges, Transaction of the ASAE, vol.27, no.2, 1984,
pp.372-374.
[16] Hedden S L; Churchill D B; Whitney J D, Trunk shakers for citrus harvesting part II: tree growth, fruit yield and removal,
Applied Engineering in Agriculture, vol.4, no.2, 1989, pp.102-106.
[17] Whitney Jodie D, Field test results with mechanical citrus fruit removal devices, Proc. Fla. State Hort. Soc., vol.110, 1997,
pp.89-92.
[18] Torregrosa A, et al., Mechanical harvesting of oranges and mandarins in Spain, Biosystems Engineering, vol.104, 2009,
pp.18-24.
[19] Wilson W C, Abscission chemicals aid to citrus fruit removal, Floriad Agricultural Experiment Stations Journal Series,
vol.522, 1977.
[20] Wilson W C; Donhaiser J R; Coppock G E, Chemical and air shaker orange removal in South Florida (Labelle), Proc. Int.
Soc. Citriculture, vol.92, 1979, pp.56-58.

13363

364
14

et al. / Procedia
Engineering
23 (2011)
351 366
AuthorPeilin
nameLi/ Procedia
Engineering
00 (2011)
000000

[21] Kender W J, et al., Effects of concentration and application time of 'Prosulfuron' on the abscission of 'Hamlin' and
'Valencia' oranges, Proc. Fla. State Hort. Soc., vol.111, 1998, pp.108-112.
[22] Koo Young M; Salyani Masoud; Whitney Jodie D, Effects of abscission chemical spray deposition on mechanical harvest
efficacy of 'Hamlin' orange, Proc. Fla. State Hort. Soc., vol.112, 1999, pp.28-33.
[23] Whitney J D, et al., Abscission chemicals affect trunk shaker orange removal, Proc. Fla. State Hort. Soc., vol.113, 2000,
pp.93-96.
[24] Hartmond Ulrich, et al., Citrus fruit abscission induced by Methyl-jasmonate, J. Amer. Soc. Hort. Sci, vol.125, no.5, 2000,
pp.547-552.
[25] Kostenyuk Igor A ; Burns Jacqueline K, Mechanical wounding and abscission in citrus, Physiologia Plantarum, vol.122,
2004, pp.354-361.
[26] Sumner H R ; Churchill D B, Collecting and handling mechanically removed citrus fruit, Proc. Int. Soc. Citriculture, vol.2,
1977, pp.413-418.
[27] Hedden Scott L; Sumner H R; Churchill D B, Collecting and handling mechanically harvested oranges in South Florida
(Labelle), Proc. Int. Soc. Citriculture, vol.92, 1979, pp.59-61.
[28] Whitney J D, Field test results with mechanical harvesting equipment in Florida oranges, Applied Engineering in
Agriculture, vol.15, no.3, 1999, pp.205-210.
[29] Goodrich Renee M; Ehsani Reza; Friedrich Loretta M, Effect of fruit pick up systems on the microbial quality of citrus fruit
surfaces, Proc. Fla. State Hort. Soc., vol.119, 2006, pp.195-197.
[30] Parrish E A; Jr; Goksel A K, Pictorial Pattern Recognition Applied to Fruit Harvesting, Transactions of the ASAE, vol.20,
no.5, 1977, pp.822-827.
[31] Sistler Fred E, Robotics and Intelligent Machines in Agriculture, Ieee Journal of Robotics and Automation, vol.RA-3, no.1,
1987, pp.3-6.
[32] D'Esnon A Grand, Robotic Harvesting of Apples, in ASAE; SME, no.01-85, 1985, 210-214.
[33] D'Esnon A Grand, et al., Magali: a Self-Propelled Robot to Pick Apples, American Society of Agricultural Engineers,
vol.87, no.1037, 1987.
[34] Tillett N D, Robotic Manipulators in Horticulture: A Review, J.agric. Engng Res., vol.55, 1993, pp.89-105.
[35] Harrell R C; Adsit P D; Munilla R D; Slaughter D C, Robotic picking of citrus, Robotica, vol.8, 1990, pp.269-278.
[36] Harrell R C; Slaughter D C; Adsit P D, A Fruit-Tacking System for Robotic Harvesting, Machine Vision and Applications,
vol.2, 1989, pp.69-80.
[37] Rabatel G; Bourely A; Sevila F; Juste F, Robotic Harvesting of Citrus: State-Of-Art and Development of the French Spansh
EUREKA Project, 1995.
[38] Juste F ; Sevila F, Citrus: An European Project to Study the Robotic Harvesting of Oranges, in Proceedings of the 3.
International Symposium on: Fruit, Nut, and Vegetable Harvesting Mechanization Jordbrugsteknik Agricultural Engineering, no.67,
1992, 331-338.
[39] Ceres R, et al., Design and implementation of an aided fruit-harvesting robot(Agribot), Industrial Robot, vol.25, no.5, 1998,
pp.337-346.
[40] Muscato G ; Prestifilippo M, A prototype of an orange picking robot: past history, the new robot and experimental results,
Industrial Robot: An International Journal, vol.32, no.2, 2005, pp.128-138.
[41] Buemi F; Massa M; Sandini G, Agrobot: A Robotic System for Greenhouse Operations, Robotics in Agriculture & The
Food Industry, vol.4, 1995, pp.172-184.
[42] Edan Yael, Design of an autonomous agricultural robot, Applied Intelligence, vol.5, 1995, pp.41-50.
[43] Edan Yael; Rogozin Dima; Flash Tamar; Miles Gaines E, Robotic Melon Harvesting, IEEE Transactions on Robotics and
Automation, vol.16, no.6, 2000, pp.831-834.
[44] Kondo N; Monta M; Fujiura T, Fruit harvesting robots in Japan, Adv. Space Res., vol.18, no.1/2, 1996, pp.181-184.
[45] Henten E J Van, et al., An autonomous robot for harvesting cucumbers in greenhouses, Autonomous Robots, vol.13, 2002,
pp.241-258.
[46] Hayashi Shigehiko; Ganno Katsunobu; Ishii Yukitsugu; Tanaka Itsuo, Robotic harvesting system for eggplants, JARQ,
vol.36, no.3, 2002, pp.163-168.
[47] Baeten Johan, et al., Autonomous fruit picking machine: a robotic apple harvester, Laugier and R. Siegwart (Eds.): Field
and Service Robotics, vol.STAR 42, 2008, pp.531-539.
[48] Jimenez A R; Ceres R; Pons J L, A Survey of Computer Vision Methods for Locating Fruit on Trees, Transactions of the
ASAE, vol.43, no.6, 2000, pp.1911-1920.
[49] Chen Yud-Ren; Chao Kuanglin; Kim Moon S, Machine Vision Technology for Agricultural Applications, Computers and
Electronics in Agriculture, vol.36, 2002, pp.173-191.
[50] Suzuki Y, et al., Spatial Distribution Mapping of Pasture Herbage Mass by Hyperspectral Imaging, in Biennial Conference
of the Australian Society for Engineering in Agriculture (SEAg). Brisbane, Australia, 2009

Peilin
et al./ /Procedia
ProcediaEngineering
Engineering00
23(2011)
(2011)000000
351 366
AuthorLiname
[51] Bulanon D M; Burks T F; Alchanatis V, Image fusion of visible and thermal images for fruit detection, Biosystems
Engineering, vol.103, 2009, pp.12-22.
[52] Bulanon D M; Burks T F; Alchanatis V, Fruit Visibility Analysis for Robotic Citrus Harvesting, American Society of
Agricultural and Biological Engineers, vol.52, no.1, 2009, pp.277-283.
[53] Slaughter David C ; Harrell Roy C, Color vision in robotic fruit harvesting, American Society of Agricultural Engineers,
vol.30, no.4, 1987, pp.1144-1148.
[54] Regunathan Murali ; Lee Won Suk, Citrus Fruit Identification and Size Determination Using Machine Vision and
Ultrasonic Sensors, in American Society of Agrigultural Engineers, no.053017, 2005
[55] Sites Peter W ; Delwiche Michael J, Computer Vision to Locate Fruit on a Tree, American Society of Agricultural
Engineers, vol.31, no.1, 1988, pp.257-263.
[56] Bulanon D M; Kataoka T; Ota Y; Hiroma T, A Machine Vision System for the Apple Harvesting Robot, Agricultural
Engineering International: the CIGR Journal of Scientific Research and Development, vol.III, 2001, pp.1-11.
[57] Zhao Jun; Tow Joel; Katupitiya Jayantha, On-tree Fruit Recognition Using Texture Properties and Color Data, in 2005
IEEE/RSJ International Conference on Intelligent Robots and Systems, 2005
[58] Tabb Amy L; Peterson Donald L; Park Johnny, Segmentation of Apple Fruit from Video via Background Modeling, in
American Society of Agricultural and Biological Engineers, no.063060, 2006
[59] Whittaker A Dale; Miles G E; Mitchell O R; Gaultney L D, Fruit Location in a Partially Occluded Image, American Society
of Agricultural Engineers, vol.30, no.3, 1987, pp.591-597.
[60] Kassay L, Hungarian robotic apple harvester, in American Society of Agricultural Engineers, no.92-7042, 1992, 1-14.
[61] Grasso Giorgio M ; Recce Michael, Scene Analysis for an Orange Picking Robot, in International Congress for Computer
Technology in Agriculture. Wageningen, Netherlands, VIAS Wageningen, Netherlands, 1996
[62] Plebe Alessio ; Grasso Giorgio, Localization of spherical fruits for robotic harvesting, Machine Vision and Applications,
vol.13, 2001, pp.70-79.
[63] Takahashi Teruo; Zhang Shuhuai; Fukuchi Hiroshi, Measurement of 3-D Locations of Fruit by Binocular Stereo Vision for
Apple Harvesting in an Orchard, in American Society of Agricultural Engineers, no.021102, 2002
[64] Jimenez A R; Jain A K; Ceres R; Pons J L, Automatic fruit recognition: a survey and new results using Range/Attenuation
images, Pattern Recognition, vol.32, 1999, pp.1719-1736.
[65] Benady Meny ; Miles Gaines E, Locating melons for robotic harvesting using structured light, in American Society of
Agricultural Engineers.; Society for Engineering in Agricultural, Food, and Biolo, no.92-7021, 1992
[66] Saint-Marc P; Jezouin J-L; Medioni G, A Versatile PC-Based Range Finding System, IEEE Transactions on Robotics and
Automation, vol.7, no.2, 1991, pp.250-256.
[67] Tanigaki Kanae; Fujiura Tateshi; Akase Akira; Imagawa Junichi, Cherry-harvesting robot, Computers and Electronics in
Agriculture, vol.COMPAG-2106, 2008, pp.1-8.
[68] Mehl Patrick M; Chen Yud-Ren; Kim Moon S; Chan Diane E, Development of Hyperspectral Imaging Technique for the
Detection of Apple Surface Defects and Contaminations, Journal of Food Engineering, vol.61, 2004, pp.67-81.
[69] Ye Xujun, et al., Estimation of citrus yield from airborne hypersepctral images using a neural network model, Ecological
Modelling, vol.198, 2006, pp.426-432.
[70] Ariana Diwan P; Lu Renfu; Guyer Daniel E, Near infrared hyperspectral reflectance imaging for detection of bruises on
pickling cucumbers, Computers and Electronics in Agriculture, vol.53, 2006, pp.60-70.
[71] Okamoto Hiroshi ; Lee Won Suk, Green citrus detection using hyperspectral imaging, Computer and Electronics in
Agriculture, vol.66, 2009, pp.201-208.
[72] Stajnko D; Lakota M; Hocevar M, Estimation of number and diameter of apple fruits in an orchard during the growing
season by thermal imaging, Computers and Electronics in Agriculture, vol.42, 2004, pp.31-42.
[73] Bulanon D M; Burks T F; Alchanatis V, Study on temporal variation in citrus canopy using thermal imaging for citrus fruit
detection, Biosystems Engineering, vol.101, 2008, pp.161-171.
[74] Rabatel G, A vision system for Magali, the fruit picking robot, in Agricultural Engineering. 1988: Paris.
[75] Kane Kevin E ; Lee Won Suk, Multispectral Imaging for In-field Green Citrus Identification, in American Society of
Agricultural and Biological Engineers, no.073025, 2007
[76] Kane Kevin E ; Lee Won Suk, Spectral Sensing of Different Citrus Varieties for Precision Agriculture, in American Society
of Agricultural and Biological Engineers, no.061065, 2006
[77] Dobrusin Yuri, et al., Real-time image processing for robotic melon harvesting, in American Society of Agricultural
Engineers, no.92-3515, 1992
[78] Aleixos N; Blasco J; Navarron F; Molto E, Multispectral Inspection of Citrus in Real-time Using Machine Vision and
Digital Signal, Computer and Electronics in Agriculture, vol.33, 2002, pp.121-137.
[79] Lu Renfu, Multispectral imaging for predicting firmness and soluble solids content of apple fruit, Rostharvest Biology and
Technology, vol.31, 2004, pp.147-157.

15365

366
16

et al. / Procedia
Engineering
23 (2011)
351 366
AuthorPeilin
nameLi/ Procedia
Engineering
00 (2011)
000000

[80] Kise Michio; Park Bosoon; Lawrence Kurt c; Windham William R, Design and calibration of a dual-band imaging system,
Sens. & Instrumen. Food Qual, vol.1, 2007, pp.113-121.
[81] Kise Michio, et al., Multispectral imaging system with interchangeable filter design, Computers and Electronics in
Agriculture, vol.72, 2010, pp.61-68.
[82] Bulanon D M; Burks T F; Alchanatis V, Study on Fruit Visibility for Robotic Harvesting, in American Society of
Agricultural and Biological Engineers, no.073124, 2007
[83] Hannan M V; Burks T F; Bulanon D M, A Real-time Machine Vision Algorithm for Robotic Citrus, in American Society of
Agricultural and Biological Engineers, no.073125, 2007
[84] Cardenas-Weber Martha; Hezroni Amots; Miles Gaines E, Machine vision to locate melons and guide robotic harvesting,
American Society of Agricultural Engineers, no.91-7006, 1991
[85] Xu Huirong ; Ying Yibin, Citrus fruit recognition using color image analysis, Intelligent Robots and Computer Vision XXII:
Algorithms, Techniques, and Active Vision., no.5608, 2004, pp.321-328.
[86] Woebbecke David M; Meyer George E; Bargen Kenneth Von; Mortensen David A, Plant species identification, size, and
enumeration using machine vision techniques on near-binary images, Proc. SPIE, vol.1836, no.208, 1993.
[87] Illingworth J ; Kittler J, A Survey of the Hough Transform, Computer Vision, Graphics and Image Processing, vol.44, no.1,
1988, pp.87-116.
[88] Pla F; Juste F; Ferri F, Feature extraction of spherical objects in image analysis: an application to robotic citrus harvesting,
Elsevier Science Publishers B.V. Amsterdam, vol.8, 1993, pp.57-72.
[89] Pla Filiberto, Recognition of Partial Circular Shapes from Segmented Contours, Computer Vision and Image Understanding,
vol.63, no.2, 1996, pp.334-343.
[90] Slaughter David C ; Harrell Roy C, Discriminating Fruit for Robotic Harvest Using Color in Natural Outdoor Scenes,
American Society of Agricultural Engineers, vol.32, no.2, 1989, pp.757-763.
[91] Ferri Francisco ; Vidal Enrique, Colour image segmentation and labeling through multiedit-condensing, Pattern
Recognition, vol.13, 1992, pp.561-568.
[92] Weeks A R; Gallagher A; Eriksson J. Detection of Oranges from a Color Image of an Orange Tree, in Proceedings of SPIE
- the International Society for Optical Engineering. 1999.
[93] Bulanon D M; Kataoka T; Okamoto H; Hata S, Determining the 3-D location of the apple fruit during harvest, in American
Society of Agricultural Engineers, no.701P1004, 2004
[94] Jian-jun Yin; Han-ping Mao; Su-yu Zhong, Segmentation methods of fruit image based on color difference, Journal of
Communication and Computer, vol.6, no.7, 2009, pp.40-45.
[95] Egmont-Petersen M; Ridder D de; Handels H, Image Processing with Neural Networks - A Review, Pattern Recognition
vol.35, 2002, pp.2279-2301.
[96] Whitney J D, Mechanical removal of fruit from citrus trees, Proc. Int. Soc. Citriculture, vol.2, 1977.
[97] Li Kuo-Tan; Syvertsen Jim; Burns Jackie, Mechanical harvesting of Florida citrus trees has little effect on leaf water
relations or return bloom, Proc. Fla. State Hort. Soc., vol.118, 2005, pp.22-24.
[98] Edan Yael, Food and Agriculture Robotics, in Handbook of Industrial Robotics, S.Y. Nof, Editor. 1999, John Wiley & Sons,
Inc.