Proceedings of the Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV 2021).

IEEE Xplore Part Number: CFP21ONG-ART; 978-0-7381-1183-4

Potato Disease Detection Using Machine Learning

2021 Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV) | 978-1-6654-1960-4/20/$31.00 ©2021 IEEE | DOI: 10.1109/ICICV50876.2021.9388606

Md. Marjanul Islam Tarik Sadia Akter Abdullah Al Mamun

Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Daffodil International Engineering Daffodil International Engineering Daffodil International
University Dhaka, Bangladesh University Dhaka, Bangladesh University Dhaka, Bangladesh
[email protected] [email protected] [email protected]

Abdus Sattar
Department of Computer Science and
Engineering
Daffodil International University
Dhaka, Bangladesh
[email protected]

Abstract— In Bangladesh potato is one of the major crops. Nations. According to a report of this organization,
Potato cultivation has been very popular in Bangladesh for Bangladesh is in the eighth place with a production of 722 lakh
the last few decades. But potato production is being hampered 10 thousand tons. Not only is this a wonderful achievement in
due to some diseases which are increasing the cost of farmers production, potato is now one of the most profitable crops in
in potato production. However, some potato diseases are the country. It has also become a means of earning foreign
hampering potato production that is increasing the cost of currency. Potatoes worth 33 million were exported last year.
farmers. Which is disrupting the life of the farmer. An But before that Bangladesh had to import potatoes from 20
automated and rapid disease detection process to increase countries of the world. We believe that, if we detect the
potato production and digitize the system. Our main goal is disease of potato properly and provide the proper treatment,
the production growth will increase to our expectation.
to diagnose potato disease using leaf pictures that we are
going to do through advanced machine learning technology. We have taken the help of image processing to diagnose
This paper offers a picture that is processing and machine potato disease and potato leaf disease. Here we have used
learning based automated systems potato leaf diseases will be these parameters to diagnose the disease which can be
identified and classified. Image processing is the best solution identified by looking at the characteristics of the disease and
for detecting and analyzing these diseases. In this analysis, the type of disease and we have tried to give more antidotes
picture division is done more than 2034 pictures of unhealthy here. Accuracy is very good with training data in the project
potato and potato's leaf, which is taken from openly and accuracy with sample data is above 99%. So our project
accessible plant town information base and a few pre- will be able to accurately diagnose potato disease and leaf
prepared models are utilized for acknowledgment and disease.
characterization of sick and sound leaves. Among them, the
program predicts with an accuracy of 99.23% in testing with II. LITERATURE REVIEW
25% test data and 75% train data. Our output has shown that Recently several works completed within the identification of
machine learning exceeds all existing tasks in potato disease leaf diseases. Here totally completely different strategies that
detection. are analyzed by different researcher’s disease detection are
Keywords— Potato Disease, Image processing, Machine reviewed.
Learning, Disease Detection, Agriculture
This paper planned the magnitude relation of the wavelength
I. INTRODUCTION of sunshine across that's absorbed the spectrum, that counting
Potato is one of the major harvests in our country. The on the leaf structure that detect the potato mold victimization
production of potatoes in Bangladesh is very lower than Segmentation of Images and Machine Learning. Once the
compared to the other developed countries in the world. The right spectrum up from greenhouse research with associate
production of potatoes is hampered by Many kinds of Pests accuracy of 84.6% diseases were analyzed severally, once
and diseases. So we can’t export potatoes to our expectations models area unit healthy and might distinguish between bays
in the other countries. Among them early blight, leaf roll virus, and spectra. Leaves, additionally as 3 categories lately blight
scab, Hollow heart etc are the most terrible disease of potatoes wound progression with 92% of accuracy. For trickster, the
at previous present and times in Bangladesh. In Bangladesh, model distinguish between pre-marked leaves, healthy &
the major area's farmers faced many hampers on this disease black-crowned trees fictions 74.6% classification accuracy
every year. The farmers and businessmen of our country are [1].
facing many problems with those diseases particularly in the
case of export to other countries as Russia, Indonesia, In [2] this work, they gift a strategies for act image process
Malaysia, Sri Lanka, Thailand, Hong Kong, Turkey, Vietnam, and machine learning to consent to diagnose from leaf
Maldives etc. Due to COVID-19, the price of potatoes illustrations. Trendy phenotyping and disease detection offer
increases day by day. Even a decade ago, production was committed steps towards food security and property
below half a million tons. Now it is moving towards billions agriculture. This is often automatic methodology that classify
of tons. This success has brought Bangladesh to the top ten diseases in potato from the publicly accessible plant picture
potato producing countries in Qatar. The recognition was
information mentioned as ‘Plant’ Village '. Use our
given by the Food and Agriculture Organization of the United

978-1-6654-1960-4/21/$31.00 ©2021 IEEE 800

Authorized licensed use limited to: University of Canberra. Downloaded on May 22,2021 at 19:53:37 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV 2021).
IEEE Xplore Part Number: CFP21ONG-ART; 978-0-7381-1183-4

segmentation approach and support Vector machines show a applied as a classifier to distinguish and arrange sicknesses
classification of 300 diseases pictures with the accuracy of on the leaves of two plants for example Tomato and Cotton.
95%. Utilizing this methodology, they had the option to accomplish
This paper [3] offers a picture that area unit process and a classification accuracy of 92.5%.
machine learning-based automatic systems Potato leaf The method proposed in this paper [11] was aimed at
diseases are becoming to be known and classified. During this identifying and discriminating deficiencies of four classes of
paper, figure the division completed over 450 pictures of minerals (nitrogen, phosphorus, potassium and magnesium).
healthy and sick Potato leaves, that area unit taken from Before the resolution, the photos are changed to the HSI and
publicly accessible plant villages Databases and 7 classified L*a*b* shading spaces. Those differentiations are assessed
algorithms area unit utilized for acknowledgment and also the by Euclidean distances decided in both concealing spaces.
classification of sick and healthy leaves, Irregular Forest
Classifier provides 97¬curacy.
This paper,[4] represent a Uses models that like pre-trained III. PROPOSED METHODOLOGY
model VG19 for fine-tuning for lifting pertinent properties All through the proposed model, the CNN algorithm is
from the dataset. That’s the results of many classifications utilized to recognize various kinds of potato infections, having
were complete supplying regression has surpassed others 7 classes of potato sicknesses and accomplished 99% accuracy
significantly the margin of classification accuracy got rate.
over97.8% check dataset.
During this analysis paper they demonstrate Transfer
Learning Technology which may be used for early detection
of potato mold once it's troublesome to gather pictures of
thousands of latest pages. Transfer learning uses already
learned deep learning models to resolve new issues. The
experiments enclosed pictures 152 healthy leaves, one
thousand blight leaves and one thousand early The Blight
Page program predicts with 99.43% accuracy. Tested with
two hundredth check knowledge and eightieth train
knowledge [5].
P.Badar [6] has used how of segmentation
victimization K-Man agglomeration on totally different
characteristics of potato leaves like color, texture, region, etc.
& has applied the back propagation neural network formula
for identification and classification of the diseases within the
image of leaf with a classification of 92% accuracy.

M.R [7] depicted a straightforward infection discovery

framework for the cotton utilizing the leaf picture. A picture
of the sickness influenced leaf is taken. At that point utilizing
picture handling and Artificial Neural Network distinction is
altered over solid and unhealthy examples. What's more,
ANN order of 80% precision.
In [8] dealt with the order of three basic paddy leaf sicknesses
( Leaf impact, Brown spot, and Bacterial curse) and taught
manures or pesticides k-implies grouping is utilized for the
Figure 1. methodology
division of the illness affected part and Visual substance are
utilized as highlights during this framework. At that point A. Dataset
Support, vector machine classifier does the characterization
We collected this information by taking pictures
cycle. The general precision is 92.06%.
directly from the potato fields of the village. We were able to
collect data on about seven types of diseases of potatoes and
The system proposed by Yao [9] is aimed at
potato leaves. We collect total 2034 potato and potato leaves
identifying and classifying three types of rice crop damage.
images. We basically split the data between potato leaves and
In his figure the diseased regions are isolated using the Otsu’s
potato disease. While collecting data, we noticed that diseases
method, then divided. The first picks up the tone, shape and
is more contagious in potato leaves. The class that divides our
surface features, and the second came from HSV shading
data is mentioned below:
space. And finally, the properties are submitted to the vector
x potato leaf roll virus
machine, which classifies in last word.
x Hollow heart of potato
U. Kumari [10] has utilized a methodology of picture x scab of potato
Segmentation during which they have extricated different x Soft rot of potato
highlights of a picture like difference, relationship, energy, x Sutali poka rog
homogeneity, mean, fluctuation and change, and so on. x Virus jonito rog
Subsequent to extricating highlights, Neural Network is

978-1-6654-1960-4/21/$31.00 ©2021 IEEE 801

Authorized licensed use limited to: University of Canberra. Downloaded on May 22,2021 at 19:53:37 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV 2021).
IEEE Xplore Part Number: CFP21ONG-ART; 978-0-7381-1183-4

C. Model Architect
CNNs thinks about piece by piece of picture. The pieces
that CNN looks for are called highlights. It finding the harsh
element matches in two pictures in similar positions, CNNs
improve at seeing closeness than entire picture coordinating
plans. Each component resembles a smaller than normal
picture, a little two-dimensional cluster of qualities.

Figure 2. Dataset example

Figure-4: The architecture of Plant Disease recognition
Table I: Dataset Features match common aspects of the images.
x Give input photograph into convolution
layer
x Choose boundaries, apply channels with
steps, cushioning if requires. Perform
x Convolution on the picture and apply
ReLU enactment to the grid.
x Execute pooling to decrease Dimension
size.
x Add as numerous convolutional layers
until satisfied.
x Flatten the yield and feed into a
completely associated layer.
x Output the class utilizing enactment and
order pictures.
B. Preprocessing
Table: Sequential Model
In this project we have to take four types of image processing
steps to normalize the image, change the color of the image,
and identify the properties, Image processing such as filtering
and transformation of the image. We have used Python's
Opencv Library for this purpose. The features of the OpenCV
library are:
ƒ Read & write images
ƒ Capture and save the videos
ƒ Image_processing such as filtering and
transformation
ƒ Detection the feature
ƒ Video or picture object detection
Total params: 120, 387
The picture document is perused with the OpenCV work the Trainable params: 120, 387
request for colors is BGR. Then again, in Pillow, the request Non-trainable prams: 0
for colors is thought to be RGB.
IV. MODEL PERFORMANCE
A. Training, Testing and the Validation
We first create a sequel model of CNN with 7 levels. We used
Adam Optimizer to measure performance error and tune cross
entropy. We then use transfer learning to create the model.
To do that we only use built in module keras, keras
applications that provide pre-trained weights. We modified
Figure 3. Data Processing the Deep Learning model so that the pre-trained weights
matched our desired output dimension by dropping the last
few layers and adding few Junne lavens.

978-1-6654-1960-4/21/$31.00 ©2021 IEEE 802

Authorized licensed use limited to: University of Canberra. Downloaded on May 22,2021 at 19:53:37 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Third International Conference on Intelligent Communication Technologies and Virtual Mobile Networks (ICICV 2021).
IEEE Xplore Part Number: CFP21ONG-ART; 978-0-7381-1183-4

V. CONCLUSION
This thesis mainly focuses on disease detection of potatoes
from any surface by using machine learning (CNN). We
found that CNN is the best way to perform this type of
detection object. However, this model gains 99% of
validation accuracy. We have a large amount of data set and
to get the best accuracy, we have tried our best. We think this
type of project will play a vital role in our agriculture sector.
Most of the farmers of the village in Bangladesh are not
literate and they can't know about the disease properly. They
Figure 5. Validation data can't know the method of detecting disease. That's why the
insect is destroying the potato and our farmers get to suffer
from it. We think that, this work can change the situation of
the potato grower in Bangladesh. In future, our aim is to
create an android application that can detect the disease of
every type of crop and can provide the proper solution for
those diseases of the crop. In future by increasing our
database, we will able to get better accuracy. By building an
android app we will continue the development process. And
we will create a system where the farmers of Bangladesh can
easily get instant service and advice on their problem by
detecting the disease.
REFERENCES
[1]. Damian Bienkowskia, Matt J. Aitkenheadb, Alison K.
Leesc, Christopher Gallaghera, Roy Neilsona “Detection and
Figure 6: Training model differentiation between potato diseases using calibration models
trained with non-imaging spectrometry data” Computers and
B. Model performance Electronics in Agriculture 167 (2019) 105056.
[2]. Monzurul Islam, Anh Dinh, Khan Wahid ,Pankaj Bhowmik
We have done 15 epochs, This model has made the progress “Detection of Potato Diseases Using Image Segmentation and
rate is 98.36% in train Dataset and 98.81% in preparing Multiclass Support Vector Machine” , 2017 IEEE 30th Canadian
approval set that we make. Trial with arbitrary pictures the Conference on Electrical and Computer Engineering (CCECE).
instructional meeting went easily after the end. It was best [3]. Md. Asif Iqbal and Kamrul Hasan Talukder,”Detection of Potato
aftereffect of precision Overall. Subsequent to dissecting the Disease Using Image Segmentation and Machine Learning”
[4]. Divyansh Tiwari, Mritunjay Ashish,Nitish Gangwar,Abhishek
result and disarray grid, it is perceptible that the exhibition of Sharma, Suhanshu Patel ,Dr. Suyash Bhardwaj.“Potato Leaf
our model is satisfactory. The exhibition of our model is given Diseases Detection Using Deep Learning”. Proceedings of the
beneath : International Conference on Intelligent Computing and Control
Systems (ICICCS 2020) IEEE Xplore Part Number: CFP20K74-
ART; ISBN: 978-1-7281-4876-2.
[5]. Farabee Islam,Md Nazmul Hoq,Professor Dr. Chowdhury
Mofizur Rahman.” Application of Transfer Learning to Detect
Potato Disease from Leaf Image”.
[6]. Athanikar, Girish and Ms. Priti Badar. “Potato Leaf Diseases
Detection and Classification System Mr.” (2016).
[7]. Malvika Ranjan, Manasi Rajiv Weginwar, NehaJoshi, A.B.
Ingole,“Detection and Classification of Leaf Disease Using
Artificial Neural Network,” International Journal of Technical
Research and Applications, 2015, pp. 331–333.
[8]. F. T. Pinki, N. Khatun and S. M. M. Islam, “Content based paddy
leaf disease recognition and remedy prediction using support
Figure-7. Training and validation accuracy vector machine,” 2017 20th International Conference of
Computer and Information Technology (ICCIT), Dhaka, 2017,
pp. 1–5.
[9]. Yao Q, Guan Z, Zhou Y, Tang J, Hu Y, Yang B, “Application of
support vector machine for detecting rice diseases using shape
and color texture features,” 2009 International Conference on
Engineering Computation, IEEE, Hong Kong, 2009, pp. 79–83.
[10]. C. U. Kumari, S. Jeevan Prasad and G. Mounika, "Leaf Disease
Detection: Feature Extraction with K-means clustering and
Classification with ANN," 2019 3rd International Conference on
Computing Methodologies and Communication (ICCMC),
Erode, India, 2019, pp. 1095-1098.
[11]. Wiwart M, Fordonski G, Zuk-Golaszewska K, Suchowilska E
“Early diagnostics of macronutrient deficiencies in three legume
Figure-8: Training and validation loss species by color image analysis,” Compute Electron Agric 65, 20

978-1-6654-1960-4/21/$31.00 ©2021 IEEE 803

Authorized licensed use limited to: University of Canberra. Downloaded on May 22,2021 at 19:53:37 UTC from IEEE Xplore. Restrictions apply.