Fruit Classification and Grading

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Fruit Classification and Grading

Akshay Haridas, Ananthakrishnan K., Adwaitha Shyam, Ann Maria Joy
B.Tech Computer Science,
Sahrdaya College of Engineering and Technology
Abstract:- This project aims at classifying fruits production of roughly 15.03 million tonnes, accounting for
according to their quality and then grading them 40.48 percent of global mango production. India sells mango
according the same. At present we focus on classifying to more than 40 countries around the world. Its outward
mango taking its outer physical structure into appearance has an impact on its market value, so careful
consideration. Mango is a commercial fruit that is fruit management after harvesting is critical.
grown and enjoyed all over the world.For business
purposes, mangoes must be categorised according to The colour of the fruit, in general, denotes its ripeness
their quality. This classification is currently done and the existence of faults. In this paper, a CNN is
manually, which is inefficient and prone to human offered as a method for automatically detecting mango fruit
mistake.It also increases personnel, lowering the overall defects using image processing. This framework can be used
cost and effectiveness of the mango processing sectors. in a variety of settings, including manufacturing enterprises
We presented a classification system for mangoes based that manufacture mango juice and supermarkets. The
on changes in their visual characteristics in this research. physical properties of mangoes are used in the grading
The methods proposed can be used on any other mango procedure. This technique is currently carried out by hand
species that changes colour during the ripening phase. and is heavily reliant on the human visual system.Over
Here we use a CNN model for classification. The dataset the last 20 years, agricultural fruit categorization has shifted
was acquired from “The Ministry of Education Artificial from human grading to automatic classification.
Intelligence Competition and Annotated Data Collection
Project (MOE AI competition and labeled data II. LITERATURE SURVEY
acquisition project)”. There were three sets of data for Here we look at some of the related works in the field
training, testing and validation.
of agriculture and defect detection:
I. INTRODUCTION R.Meena Prakash proposed Detection of Leaf
Image processing is a method of converting a physical Diseases and Classification using Digital Image Processing
image to a digital representation and then performing [1].The K-Means technique is used to segment leaves in this
operations on it.It is used to extract enhanced visuals as well paper. GLCM is used to extract texture features.SVM is then
as relevant data from photographs. It is a signaling process used to classify the data. To begin, photos of various leaves
distribution. In this process, if you take the input as an are obtained utilising for higher quality, a digital camera
image and apply an efficient algorithm, you will get the with the needed resolution After that, the input image is
following results: The image, data, or feature associated downsized to 256x256 pixels. To improve the quality of the
with that image.The processing stage begins with image image, preprocessing is performed.It consists of Image
segmentation. There is a certain demand for image improvement and colour space conversion.The RGB L*a*b*
segmentation algorithms. The first of them is speed. In colour space is used to convert leaf images. The colour
process Image segmentation, it does not want to spend a lot space conversion is employed to improve visual analysis.In
of time. The second is object. This will improve the results SVM, the kernel function converts data from the input space
of image verification. If the result of the shape is to a high- dimensional feature space.The computational
incomplete,then many properties must be taken for complexity of kernel Hilbert space is reduced (RKHS).The
recording the resulting edges of intersections.In the agricul- feature vector is fed into the classifier. The feature vectors in
tural field, image processing and computer vision systems the database images are divided into training and testing
have been widely employed for identification, classification, vectors. The classifier is trained on the training set before
grading, and quality assessment. being used to categorise the testing set.The performance of
the classifier is assessed by comparing predicted labels to
Mango fruit defect detection and maturity detection are actual values.
difficult tasks for computer vision to reach near human
levels of recognition. The proposed framework is beneficial Rahul Pralhad Salunkhe [2] proposed a Mango
in supermarkets and may be used in computer vision to Ripening Stage Detection using RGB and HSV method.The
sort fruits automatically from a set of distinct types of mango processing business is one of the largest in the
fruits. The goal of this project is to create an automated world.The image’s red (R), green (G), and blue (B)
programme that can identify defects and detect mango fruit components are separated in the second stage. The average
ripeness based on form, size, and colour features using values of the R, G, and B components are computed
digital image analysis.The proposed method can be utilised to next.Three sets of 24 samples were taken for the testing,
detect obvious faults, stems, size, and form of mangoes, as each set having mangoes at various stages of ripening. The
well as grade them quickly and accurately. The mango is RGB and HSV tech- niques were found to have 90.4 and
grown over the largest area, 4312 thousand hectares, with a 84.2 percent accuracy, respectively, where the result is
IJISRT22JUN781 www.ijisrt.com 608

ISSN No:-2456-2165
only regarded accurate if the detected stage matches the used to start the model. The decoder is a device that
manually selected stage. A false positive rate of 2.57 decodes data. 3 convolution operations make up the basic
percent was recorded for RGB and 7.29 percent for convolution layer. Batch normalisation and ReLU activation
HSV. The RGB and HSV methods had true positive rates to improve segmentation The decoder section has been
of 89.77 percent and 86.74 percent, respectively. redesigned to provide improved performance. a more
extensive multi-scale data collection and analysis
Edwin R Arboleda published a paper on classification Mechanism for fusing . Stride1 is utilised with maximum
of Coffee Bean Species Using Image Processing, Artificial pooling. to minimise the size of the feature map in the
Neural Network and K-Nearest Neighbor. There were 255 encoder To acquire the final pixel-perfect result In the
photos utilised in this project. These images have a res- decoder part, the nearest-neighbor interpolation up-sampling
olution of 5152 x 3864 pixels and were saved in JPEG with multi-times is used.To build feature maps with 512
(Joint Photographic Expert Group) format. To preprocess channels, the input image travels through five encoder
and extract characteristics from coffee sample photographs, stages. There are two paths for the encoder’s final stage and
a computer routine technique was created using the im- the decoder’s stages. The suggested encoder-decoder
age processing toolbox and vislabels function in MATLAB network is used to segment an ore image. as shown
(version R2012a). Morphological features were retrieved below.The result is based on a combination of multi- scale
for categorization after the coffee bean images were pre- contextual features.
processed. The coffee bean features taken from the pho-
tos were utilised to categorise using an Artificial Neural III. PROPOSED SYSTEM
Network (ANN).Multilayer feed forward neural networks
were employed to solve challenges requiring classification Here we use convolutional neural network for classi-
and function approximation.Morphology is the geometric fication. This is much simpler model compared to other
property (shape and size) of objects. It was extracted using image classification algorithm. CNN consists of mainly four
the following methods: Surface area (As): Counting the components: Convolution Maxpooling Padding and Striding
number of pixels with a value of one was used to compute Fully connected layer In this the first three ie, convolution,
the area of a coffee bean. The perimeter (P): was determined maxpooling, padding and striding is for feature extraction or
by counting the number of pixels on the coffee bean’s feature learning. The classification takes place in the fully
border. Deq (diameter equivalent): Deq is the diameter of a connected layer. In fully connected layer each neuron in one
circle with the same area of the item, also known as Ferret level is fully connected to each layers at the next level. The
diameter, and roundness is the diameter of a non-circular number of convolutional layers depends upon the
object (Rs). [3]. number of features that we consider. Here we consider
features that deals with the outer appearance of the fruit.
Han Yang et al proposed an improved encoder-decoder Before convolution we perform
network for ore image segmentation[4]. Three convolution downsampling.Downsampling is a technique for reducing
operations make up the encoder’s fundamental convolution the number of training samples that fall into the majority
layer. The encoder is then followed by a ReLU activation category. As it aids in the balancing of target category
function. VGG16 was pre-trained on ImageNet and was counts. We tend to lose a lot of essential information.
Fig. 1: Block Diagram

ISSN No:-2456-2165
when we remove the acquired data.After [4.] An Improved Encoder–Decoder Network for Ore
downsampling the output is given to the model. The Image Segmentation Han Yang, Chao Huang ,
convolutional neural network is composed of three regions Member, IEEE, Long Wang , Member, IEEE,and
namely backbone, neck and head. Among this regions Xiong Luo , Senior Member, IEEE
backbone and neck deals with feature extraction and the [5.] Farah and T. Ferreira, “The Coffee Plant and Beans:
head region consists of fully connected layers that deals An Introduc- tion,” in Coffee in Health and Disease
with classification. This network was given a fixed size (224 Prevention by Victor R. Preedy, 2015 pp. 5–10.
* 224) RGB image as input, implying that the matrix was of [6.] S. J. G. A. Barbedo, “A new automatic method for
shape (224,224,3).The only preprocessing done here was disease symptom segmentation in digital photographs
subtracting the mean RGB value from each pixel, which was of plant
computed throughout the entire training set. Kernels of size [7.] leaves,” European Journal of Plant Pathology, vol. 147,
(3 * 3) with a stride size of 1 pixel were used. They were no. 2, pp. 349–364, 2016.
able to cover the entire concept of the image as a result of [8.] J. G. A. Barbedo, “A novel algorithm for semi-
this. To keep the image’s spatial resolution, spatial padding automatic segmentation of plant leaf disease symptoms
was applied. Max pooling was done with stride 2 over a 2 * using digital image processing,” Tropical Plant
2 pixel window. Following that, the Rectified linear unit Pathology, vol. 41, no. 4, pp. 210– 224, 2016.
(ReLu) was used to inject non- linearity into the model in
order to enhance classification and computational time.
Previous models had employed tanh or sigmoid functions,
but this proved to be much better.
IV. CONCLUSIONS
Mango is one of the most popular tropical fruits in the

world, with annual production trends increasing. Mango
is India’s most popular fruit and is known as the ”King
of Fruits.” The mango is grown over the biggest area,
4312 thousand hectares, and produces roughly 15.03 mil-
lion tonnes, accounting for 40.48 percent of global mango
production. India sells mango to more than 40 countries
around the world. Its outward appearance has an impact on
its market value, so careful fruit management after
harvesting is critical. The colour of the fruit, in general,
denotes its ripeness and the existence of faults. Using image
processing,a method is suggested to automatically determine
flaw and ripeness in mango fruit.This framework can be
used in a variety of settings, including manufacturing
enterprises that manufacture mango juice and
supermarkets.
REFERENCES
[1.] R.Meena Prakash,G.P.Saraswathy,

G.Ramalakshmi, K.H.Mangaleswari, T.Kaviya,
”Detection of Leaf Diseases and Classification using
Digital Image Processing”, 2017 International
Conference on Innovations in Information, Embedded
and Communication Systems (ICIIECS).
[2.] Rahul Pralhad Salunkhel, Aniket Anil PatiF, ” Image
Processing for Mango Ripening Stage Detection: RGB
and HSV method” , 2015 Third International
Conference on Image Infonnation Processing.
[3.] Edwin R. Arboleda, Arnel C. Fajardo, Ruji P. Medina,
”Classification of Coffee Bean Species Using Image
Processing, Artificial Neural Network and K Nearest
Neighbors”

Fruit Classification and Grading

Uploaded by

Copyright:

Available Formats

Fruit Classification and Grading

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fruit Classification and Grading

Uploaded by

Copyright:

Available Formats

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology

Fruit Classification and Grading

IJISRT22JUN781 www.ijisrt.com 608

Fig. 1: Block Diagram

IJISRT22JUN781 www.ijisrt.com 609

Mango is one of the most popular tropical fruits in the

[1.] R.Meena Prakash,G.P.Saraswathy,

IJISRT22JUN781 www.ijisrt.com 610

You might also like