International Journal of Advanced Engineering Research and

Science (IJAERS)
Peer-Reviewed Journal
ISSN: 2349-6495(P) | 2456-1908(O)
Vol-10, Issue-7; Jul, 2023
Journal Home Page Available: https://ijaers.com/
Article DOI:https://dx.doi.org/10.22161/ijaers.107.8

Fruits and Vegetables Detection using YOLO Algorithm

S. Kanakaprabha1, Dr. Gaddam Venu Gopal2, D. Kaleeswaran3, D. Hemamalini3,
Dr. G. Ganeshkumar5
1Department of Computer Engineering, Rathinam Technical Campus, Coimbatore, Tamilnadu, India
2Department of Computer Science and Engineering, Koneru Lakshmaiah Education Foundation,Vaddeswaram,Guntu.
3Department of Computer Engineering, Rathinam Technical Campus, Coimbatore, Tamilnadu, India
4Department of Computer Science and Engineering, Karpagam Academy of Higher Education, Coimbatore, Tamilnadu, India
5Department of Information Technology, Hindusthan Institute of Technology, Coimbatore, Tamilnadu, India

Received: 08 Jun 2023, Abstract—The robotic harvesting platform's fruit and vegetable
Receive in revised form: 10 Jul 2023, detection system is crucial. Due to uneven environmental factors such
branch and leaf shifting sunshine, fruit and vegetable clusters, shadow,
Accepted: 18 Jul 2023,
and so on, the fruit recognition has become more difficult in nowadays.
Available online: 31 Jul 2023 The current method in this work is used to detect different types of fruits
©2023 The Author(s). Published by AI and vegetables in different size and shape. This method makes the use of
Publication. This is an open access article OpenCV, Dark Flow, a TensorFlow variant of the YOLO technique. To
under the CC BY license train the necessary of network, a range of fruits and vegetable pictures
(https://creativecommons.org/licenses/by/4.0/). were input into the network. The photos were pre-processed using
OpenCV to create manual bounding boxes around the fruits and
Keywords— Dark Flow, Fruit, OpenCV,
vegetables before into the training. YOLO detection algorithm is used.
Vegetable, YOLO
In, this method more accurately and rapidly recognizes of an item in an
image. After the network has been trained, the test input is sent into the
bounding boxes surrounding the recognized fruits and vegetables will
be displayed as a consequence.

I. INTRODUCTION grasping system, building a reliable fruit identification is a

The main factor in the agriculture sector with the crucial first step toward fully automated harvesting robots.
highest cost demands. This is brought on by rising supply Suppose the fruits cannot be detected or seen, it cannot be
costs for items like electricity, irrigation water, and gathered. This level is challenging because of a number of
agrochemicals, among the others. Because of this, the factors, such as changing illumination, occlusions, and
horticulture sector and farm enterprises are suffering from situations in which the fruit seems visually similar to the
thin profit margins. Under these circumstances, food background. To deal with these problems, as well as we
production will need to increase to meet the rising require a highly discriminative feature representations and
demands of a growing world population, which will be a a generalised model that is robust to changes in brightness
major problem in this future. Due to its greater endurance and perspective. Fruits and vegetables are essential for
and repeatability, robotic harvesting has the potential to human diet as well as animals and other living things. The
save labor costs while simultaneously enhancing the fruit requirement for food is two times more than it was
quality. These factors have led to a rise in interest in previously due to the ever-increasing population of all
deploying agricultural robots to harvest fruits and living creatures. Farmers must work extremely hard and
vegetables during the past three decades. It takes a lot of long hours to meet such a large demand, and farms must
challenging tasks, including choosing and manipulating, be monitored at all hours of the day and night. The
to build these platforms. Although it is the first perception product is affected by climate change in addition to the
of the system that comes under later manipulation and expanding population. Untimely rain and sweltering heat

www.ijaers.com Page | 53
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

hinder the farmers' arduous task. position of the fruit goal and this accuracy can be achieved
Convolutional neural networks (CNN), recurrent by the fruit axes of orientation. In this survey, a novel
neural networks (R-NN), fast R-NN, YOLO, and other strawberry result of robot for the ridge-planted berries is
techniques are available in this field and may be used to introduced, along with a rotating yolo r- yolo fruit pose
identify and recognize fruits and vegetables. Based on estimator that increases the efficiency of picking point
training data given to the network, You Only Look Once localization for the lightweight network.
(YOLO) is an efficient object identification method. Each Convolution neural network was replaced with
frame updates into the input are examined, and the mobilenet-v1 is backbone of the network for feature
necessary items are frame into quickly. Various extraction. Alternative, rotation of the angle guideline was
techniques have been to identify into fruits and vegetables. used to design the training set and establish the anchors,
In techniques such as CNN, RNN, and Fast RNN, only and then logistic regression and rotated anchors were used
certain sections of interest are applied to identify objects to predict the spin of the target fruits bounding boxes in a
within an image. The networks mentioned above do not batch of 100 strawberry photos. This considerably boosted
employ a holistic approach. These regions of interest are operating speed. The average identification rate and recall
sent into the Needed model's network, and only those rate for the suggested model were 9443 percent and 9346
things are found there are trained. When compared to the percent, respectively. The integrated controller of the robot
region-based algorithms, YOLO is extremely different. processed eighteen frames per second, which showed
One convolution network is sufficient to retrieve the strong real-time achievement in terms of actual
YOLO class probability and bounding box information. identification and localization efficiency of choosing
places. This study presents technical advice for enhancing
the fixed controller of fruit picking robots target
recognition, showing that the recommended design
outperformed a variety of different target identification
methodologies.
Osman, Y., et al. [3] A two-stage approach
includes recognizing the fruits and then tracking them
framework by framework. The principle of "You Only
Fig.1(a) Fig.1(b) Look Once" is applied to identify threats (YOLO).
Bounding boxes are collected from the finding and Non-
Max Suppression (NMS) is utilized to produce the
The photos used for YOLO identification are
concluded detection. The tracking system is then supplied
shown in Figures 1(a) and 1. (b). Yolo-based real-time
with the boxes. We use a Deep SORT algorithm that was
application detection systems may be used to a variety of
especially developed to deal with fruits for tracking. Using
applications with good generalization. With a little
box coordinates, the original image is cropped to eliminate
amount of example picture data, it is easily adaptable to
each recognized object. ResNet, a convolutional neural
any sort of fruit and vegetable. methods that use a variety
network (CNN), then extracts features from the cropped
of data in both early and late fusion.
image to build the feature map. By comparing the
attributes of new and old detections using a distance
II. LITERATURE SURVEY metric, which links the two things with the smallest
distance, new detections are connected to previous
A non-destructive approach based on thermal
detections.
imaging is suggested by S. Raka, et al. [1] for evaluating
the interior and exterior quality of fruits. Analyzing the Input items with no associations are studied as
fruit surfaces of the heated properties allows for the branded different objects to be monitored. We maintain
precise determination of ripening conditions. The creation path of the fruits through-out the video frames to assure
of an automated system can decrease the time-consuming that we are conditional accurate they are initially
human inspection tasks involved in fruit sorting. observed. We determine the method using videos taken in
an apple garden to demonstrate this approach's very
According to Y. Yu et al. [2], the primary
effectiveness in the natural light. The decision show that
technical challenge to the implementation of robots for
fruit counting on real-time video grain can be performed
strawberry harvesting is the need for improved real-time
with great precision. The new method works with all types
performance in the localization algorithms. To locate the
of fruits and vegetables and doesn't require any
plucking point on the strawberry stem. By estimating the
modifications to the algorithms.

www.ijaers.com Page | 54
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

In this study, a prototype of an autonomous fruit filter, altered histogram of directed gradient is to
harvesting robot built around a mobile chassis and a combined with a pair of the depth comparison of texture
robotic arm is proposed by S. M. Mangaonkar et al. Our caption with a random forest classifier. The effectiveness
suggested architecture can recognize fruits using an object of the fruit recognition algorithms on the fruit’s datasets
identification method and an image pre-processing using the human-labeled images on the ground truth. This
module (YOLO v3). This study proposes a prototype of an methodology is to identify the size invariant, resistant to
autonomous fruit harvesting robot based on a robotic arm partial occlusions, to be precise than existing method for
mounted on a mobile chassis, developed by S. M. identifying potential fruit locations.
Mangaonkar et al. [4]. Our suggested architecture is In order to address issues with human health, K.
capable of recognizing fruits thanks to an image pre- Roy, et al. [9] offer a method for segmenting rotting
processing module and an object detection algorithm. vegetables. Edge Detection, Color Based Segmentation,
K. R. B. Legaspi and colleagues [5] Whiteflies and Marker Based Segmentation were three segmentation
and fruit flies were identified and classified using techniques that delivered effective and beneficial
YOLOV3. The analyst used a Raspberry Pi camera to outcomes. The segmentation techniques outlined above
acquire images, and also set up both desktop and online successfully distinguish between rotting and healthy parts
applications for viewing the images captured by the of a vegetable, allowing the diseased veggies to be
Raspberry Pi camera. The confusion matrix showed that distinguished from the healthy ones. Using an automated
the miniature had the overall accuracy of 83.07 percent in system to sort vegetables can save money on labor and
recognizing and recognizing fruit flies and whiteflies. increase accuracy for any company that manufactures
According to S. K et al. [6], The Regional Built food goods. On numerous levels, the ways to spot rotting
Convolutional Neural Network (RCNN), Fast RCNN, and veggies are examined.
Faster RCNN are examples of pre-trained Deep Neural An image-based technique is to identify the grade
(DNN) representations. To detect fruits in an input image, fruit size is presented by H. Dang and colleagues [10].
the You Only Look Once (YOLO) V3 and the Single Shot Following the acquisition of the fruits image, of several
Multibox Detector (SSD) were implemented on the RISC- fruit characteristics are extracted into detection
V architecture. COCO datasets are used for pre-training to techniques. These characteristics are used to grade
ensure uniformity across all DNN models. In terms of students. This integrated into grading system has to the
accuracy and inference efficiency, experimental results benefits of high grading getting better accuracy, quick
demonstrate that YOLO and SSD-Mobile Net outperform speed, and low cost, according to experiments. It is likely
all existing DNN models for object recognition on the to be applied to yield-related detection and grading.
RISC-V architecture. According to colour and form data, T. Gayathri
The team of Yogesh [7] The fruit quality detection Devi et al. [11] provide an image processing technique for
technique described in this study was built on the completely independent separation and production
basis of the form, size, and colour of the fruits' forecast of fruits. The pre-processing procedure is started
external features. Manual fruit monitoring is using the supplied fruits images. The picture is then
ineffective in the agricultural industry due to growing determined to transform from RGB to HSV Color
demand. Therefore, the agriculture sector needs a information to analyze the berry from the roots. The
capable approach to support it in meeting customer required colors may be hidden using colour edge
demand. The recommended method makes advantage detection. To diminish noise, of Gaussian filter is used.
of a sturdy feature that is speeded up. The approach The picture outline is measured. The photographs are then
discusses object detection by eliminating the local processed using an image analysis technique. Fruit
feature from the segmented picture. Creating a flaw counting based on colour and shape is displayed in the
detection method that can be utilized to quickly result. The fruit and vegetables in the image are
extract features and descriptions is the goal. automatically segmented and counted using feature
A fruit identification technique is suggested by Z. extraction and a circular fitting approach. Various fruits
S. Pothen et al. [8] that makes use of the fruit's surface's such as (orange/tangerine, pomegranate, apple, lemon,
slow change in intensity and gradient orientation. For mango, and cherry) are used for automated conditional.
potential fruit sites, gradient orientation profiles and Using the Open CV, the necessary image processing
monotonically falling intensity profiles are both examined operations are completed.
also named as means by either "seed spots" To Orange fruit pictures taken in natural illumination
categorized into potential fruit spots that pass the first were segmented using edge-based and color-based

www.ijaers.com Page | 55
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

detection techniques done by R. Thendral et al. [12]. The probability refers to the possibility of each class that the
objective of this study was to locate and identify an object. The class possibility is the likelihood that the
orange in each of the twenty digitized fruit images that images in the case that fits to session. As a consequence,
were casually selected from the Net. Edge-based N x N x C possibilities, where C is the number of classes,
segmentation is consistently outperformed by color-based are generated, with each cell forecasting one class
segmentation. The computation is carried out using the probability.
MATLAB image-processing toolbox, and the computed Pre-processing is done on the pictures to get rid
outcomes are exposed in the segmented image results. of noise and outliers, improve contrast enhancement, and
speed up the algorithm. Although additional pre-
III. METHODOLOGY processing techniques may be used in this approach, Non-
Max Suppression is the major emphasis. The network
Fruits and vegetables must be divided in order to
topology resembles that of a typical CNN with 24
be seen clearly against a backdrop of leaves and stems.
convolutional layers and two completely connected layers
Due to the variations in color and lighting, significant
at the end. The Google Net idea is used to construct the
quantities of the occlusion, and other considerations, this
YOLO network architecture. Fast YOLO is a quicker
test is difficult. Yolo is a real-time object tracking system
variation of YOLO that uses nine convolutional layers
that is offered as a technique. Yolo's primary premise is
slightly than 24 and maintains all other limitations
that you only look once when configuring a model for
constant with exception of the system size.
training. This approach then requires that you test the
model with the necessary versions since the model The spatial arrangement of the grid cells that go
versions change. Yolo has overtaken the market leader, into creating the bounding box makes YOLO less
CNN, in terms of popularity. Yolo and CNN are successful at recognizing little objects in big groupings.
equivalent, although CNN does real-time object tracking Since YOLO learns mostly from data, the system cannot
less well. Both boundary boxes with a different CNN are recognize in advanced or changed shape of aspect ratios.
analyzed by YOLO. YOLO is favored due of its speed.
Furthermore, unlike CNN's moving window and
area proposed bill algorithms, it generates predictions
while maintaining a global perspective. The secondary
cause is YOLO's fast learning of generalizable
representations of objects. One of the distinctive
characteristics that the network discovers for each border
box is the size of the boundary box and the many class
choices. Only item classification with a quality greater
than the edge is utilized to identify the images inside the
box when a threshold has been specified. It is crucial to
consider the output encryption technique YOLO employs.
On the basis of the supplied picture segments, a N x N
matrix is created. Even when there are numerous images Fig.2. Yolo Architecture for Detecting Fruit and
are just one square of the grid, cell in the object's centre Vegetables
aids in predicting its existence. Each cell is surrounded by
five bounding boxes, each of which has five distinct
There are several methods to implement the YOLO
characteristics denoted by a letter (x, y, w, h, c).
approach, and our system uses Darknet, an open-source
The coordinates of the box's core cell are: (x, y). neural network framework. An open-source real-time
the bounding box's dimensions are (w, h). The confidence computer vision library is called OpenCV. Since its
score is the last element that determines whether or not an creation in C++, it has been translated into a number of
item is in the box (c). If this is the case, the item is not programming languages. We will use the OpenCV cv2
included within the box, the score will be 0. Ideally, the Python library. Drawing a bounding box that represents the
element should be zero, but if it is present, it should be upper left edge and lowest of the correct coordinates is one
one. The formula used to determine the confidence factor of cv2's four main features. Rectangles are drawn using the
favor’s the intersection of the box and the accuracy over provided coordinates using this function. The expected
the union of the prediction box. Additionally, YOLO class labels and confidence ratings are combined on each
determines the probability for each category. Class bounding box using a second algorithm. The final task is to

www.ijaers.com Page | 56
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

scan the picture and look for class labels. The last approach
is employed to read a movie from regional cache and give
it class names. Additionally, it may be used to access the
real-time video from a webcam or new computer hardware.
Tensor flow operates on both the CPU and the GPU
equally; however, the Yolo GPU version runs quicker than
the Yolo CPU version. According to GPU specs, Yolo
operating on a GPU can analyze video at a rate of 40100
frames per second, whereas Yolo consecutively on a CPU
Fig.3(a) and Fig.3(b) boundary box for the practice
can only manage 38 frames per second.
pictures.
A. Training Image
The first phase in training is to get relevant
B. Testing Images
photographs from the web, which are mostly pictures of
distinct kinds of cucumber, apples, and capsicum. For a To accurately identify and categorize in presence of
quick and precise categorization, train as many images as vegetable images, and that can be loads into the trained
you can. From various web sources, a total of 100 YOLO model and then lots into the dataset that has to be
photographs of each vegetable were collected, with 60 classified. The location and class of the veggie are also
images being utilized for train and the 40 for testing. The determined by the writing. A bounding box with the class
network's distinctive qualities are determined by a number label and confidence score is produced using OpenCV. The
of variables in the YOLO configuration file; this variable classical only identifies bounding boxes with a confidence
quantity must be altered to match our contributions and score of 0.15 or above because a confidence threshold of
productivities. 0.15 has been established. The same method is
demonstrated in a movie to identify fruits and vegetables.
An xml document is created in which the top-left
and bottom-right coordinates of all picture in the train Every frame of the film is taken out, and using a
dataset are listed. After that, a rectangle selector runs a threshold of 0.15, each picture is categorized like every
Python script to complete this task. Moreover, the data is other image. A lot of computing power is required for
supplied to the system, a pre-trained dataset like yolo-tiny movie processing; else, object detection and classification
should be amount. The epochs are set to 300, overwriting on a film would be quite sluggish. In this algorithm is
the default of 1000, and the learning rate is set to 0.001. classified into the 3 modules into actual period, a script is
Each training phase will run the full batch of 16 photos printed into load the classical was trained into capture the
done the unseen layers and adjust the masses accordingly. video. The bounding boxes for all picture in individually
edge by passing the edges to the algorithm. Here, too, the
The complete dataset has been divided into batches.
0.15 cutoff is employed. For a fast comparison, the
All the epoch consists of 11 batches and 11 step due to the
microchip type of Tensor Flow processes webcam footage
180 pictures in the dataset. We have the same number of
at 4-6 frames per second.
batches as there are epochs in our system. The mean error
would have been calculated after each step, the weights
would have been back-propagated, and all of the pictures IV. RESULT AND ANALYSIS
in that all the batch subjected to all hidden layers. The dataset utilized comprises of 180 test photos of
As a consequence, all of the pictures have all three vegetables in different arrangements, such as
disappeared through the hidden layer once at the end of horizontally, vertically, or in challenging lighting as a
each epoch, allowing us to calculate the mean error. The group or as a backdrop. Decreasing the learning rate to
regular error does not alteration training the times, the train 0.001 will accelerate the training process. The program
is still or the learning rate is change. The average error is was set to run for 300 epochs, with 11 stages per epoch, as
discovered to be between 4 and 6 after 137 training the batch size was set to 16. Following 100 epochs, the
iterations and did not drop any other. At the conclusion of average error rate is 4-6. When the dataset was further
every 125 steps, Yolo will save the masses file in the trained, into 137th epoch, the average error did not
resident manual. These mass files test our miniature significantly vary across successive stages.
utilizing it. The training remained halted and the weights that
were obtained were saved in order to evaluate their
accuracy. Then, this classical is tested into a range of

www.ijaers.com Page | 57
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

sample photos. 70% of the veggies were correctly

identified and classified more often than 70% of the time
when a video was used as an input to the algorithm.
Vegetables in a range of situations and orientations,
including vertical, in bunches, and against complex
backgrounds, could all be successfully identified and
classified by the model. The model also produces a
confidence score for every prediction, which was more
than 50% for almost all of the photos.
Table 1. Fruits and vegetables' effectiveness
The quantity of pictures examined in 50
fruits
Average degree of assurance 67.6%
Number of vegetables photographs 50
that underwent testing
Average degree of assurance 67.6%
Number of pictures where different 75%
fruits may be seen
Number of pictures in which different 75%
vegetables are visible
Percentage of photos with false 50%
positives
The proportion of photos with a 65%
confidence rating of at least 50%.
Number of photos with a greater than 20%
80% degree of confidence
Fig.4. Images of several fruits that YOLO detected

The YOLO algorithm works by dividing the

image into a grid and predicting bounding boxes and class Green mango was likewise confidently and
probabilities for each grid cell. This approach allows effectively identified and identified. In the bare minimum
YOLO to detect multiple objects simultaneously and with number of photos, the model accurately classifies fruits
high speed and accuracy. Keep in mind that the quality and and vegetables with a 65% accuracy rate. The model has a
accuracy of YOLO's detections depend on various factors, high degree of confidence in its ability to identify several
including the quality and diversity of the training data it green apples in a picture. The model correctly identifies
was exposed to during its training phase. green capsicum and categorises it, as well as various other
capsicums in complicated backdrops.
As a result, the model was able to correctly identify
and classify the the fruits and vegetables. With a high
confidence score of over 50%, the model accurately detects V. CONCLUSION
cucumber, and it also correctly detects numerous A model for identifying fruits and vegetables
cucumbers. When there are a lot of cucumbers, the model has been developed, along with the recommended
sometimes provides false positives by identifying half of a approach, which has been built, trained, and tested.
cucumber as a full cucumber. There are a few instances Our algorithm can identify and classify 60–70% of the
where the model fails to detect a vegetable due to the crop and can identify different vegetables in a single
unknown orientation of a vegetable. picture under a variety of limitations. 70 of the
photographs were accurately categorized when the
threshold was set to 015 since the bulk of the images
were downloaded from the internet. The more
effective training set is greater than the accuracy in

www.ijaers.com Page | 58
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

categorizing each of the shots if we had utilized off- thermal image processing," 2019 IEEE Pune Section
field images as our training examples. For this model International Conference (PuneCon), 2019, pp.1-4,doi:
to discriminate between foreground and background 10.1109/PuneCon46936.2019.9105726.
[2] Y. Yu, K. Zhang, H. Liu, L. Yang and D. Zhang, "Real-
produce in an automated harvesting system, depth
Time Visual Localization of the Picking Points for a
information is crucial. This may be done by utilizing
Ridge-Planting Strawberry Harvesting Robot," in IEEE
3D photos and altering the system so that it no longer Access, vol. 8, pp. 116556-
uses 2D images for training characteristics like size, 116568,2020,doi:10.1109/ACCESS.2020.3003034.
colour and texture. [3] Y. Osman, R. Dennis and K. Elgazzar, "Yield Estimation
using Deep Learning for Precision Agriculture," 2021
IEEE 7th World Forum on Internet of Things (WF-IoT),
2021, pp. 542-550, doi: 10.1109/WF-
IoT51360.2021.9595143.
[4] S. M. Mangaonkar, R. Khandelwal, S. Shaikh, S.
Chandaliya and S. Ganguli, "Fruit Harvesting Robot
Using Computer Vision," 2022 International Conference
for Advancement in Technology (ICONAT),2022,pp.1-
6,doi: 10.1109/ICONAT53423.2022.9726126.
[5] K. R. B. Legaspi, N. W. S. Sison and J. F. Villaverde,
"Detection and Classification of Whiteflies and Fruit
Flies Using YOLO," 2021 13th International
Conference on Computer and Automation Engineering
(ICCAE), 2021, pp. 1-4, doi:
10.1109/ICCAE51876.2021.9426129.
[6] S. K, A. K. Srivastava, S. Allam and D. Lilaramani,
"Comparative analysis on Deep Convolution Neural
Network models using Pytorch and OpenCV DNN
frameworks for identifying optimum fruit detection
solution on RISC-V architecture," 2021 IEEE Mysore
Sub Section International Conference (MysuruCon),
2021, pp. 738-743, doi:
10.1109/MysuruCon52639.2021.9641594.
[7] Yogesh and A. K. Dubey, "Fruit defect detection based
on speeded up robust feature technique," 2016 5th
International Conference on Reliability, Infocom
Technologies and Optimization (Trends and Future
Directions) (ICRITO), 2016, pp. 590-594, doi:
10.1109/ICRITO.2016.7785023
[8] Z. S. Pothen and S. Nuske, "Texture-based fruit
detection via images using the smooth patterns on the
fruit," 2016 IEEE International Conference on Robotics
and Automation (ICRA), 2016, pp. 5171-5176, doi:
10.1109/ICRA.2016.7487722.
[9] K. Roy, S. S. Chaudhuri, S. Bhattacharjee, S. Manna and
T. Chakraborty, "Segmentation Techniques for Rotten
Fruit detection," 2019 International Conference on Opto-
Electronics and Applied Optics (Optronix),2019,pp.1-
4,doi: 10.1109/OPTRONIX.2019.8862367.
[10] H. Dang, J. Song and Q. Guo, "A Fruit Size Detecting
and Grading System Based on Image Processing," 2010
Second International Conference on Intelligent Human-
Machine Systems and Cybernetics, 2010, pp. 83-86, doi:
Fig.5. YOLO detection of various vegetable images 10.1109/IHMSC.2010.120
[11] T. Gayathri Devi, P. Neelamegam and S. Sudha, "Image
Processing System for Automatic Segmentation and
REFERENCES Yield Prediction of Fruits using Open CV," 2017
[1] S. Raka, A. Kamat, S. Chavan, A. Tyagi and P. International Conference on Current Trends in
Soygaonkar, "Taste- wise fruit sorting system using Computer, Electrical, Electronics and Communication

www.ijaers.com Page | 59
Kanakaprabha et al. International Journal of Advanced Engineering Research and Science, 10(7)-2023

(CTCEEC), 2017, pp.758-762, doi:

10.1109/CTCEEC.2017.8455137.
[12] R. Thendral, A. Suhasini and N. Senthil, "A comparative
analysis of edge and color based segmentation for
orange fruit recognition," 2014 International
Conference on Communication and Signal Processing,
2014, pp. 463-466, doi: 10.1109/ICCSP.2014.6949884.

www.ijaers.com Page | 60