See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/341025012

Plant Disease Detection using Deep Learning

Article  in  International Journal of Recent Technology and Engineering · April 2020

DOI: 10.35940/ijrte.A2139.059120

CITATIONS READS

0 1,549

5 authors, including:

Murk Chohan Adil Khan

Sukkur Institute of Business Administration Sukkur Institute of Business Administration
6 PUBLICATIONS   3 CITATIONS    6 PUBLICATIONS   3 CITATIONS   

SEE PROFILE SEE PROFILE

Saif Hassan Muhammad Saleem Mahar

Sukkur Institute of Business Administration Sukkur Institute of Business Administration
4 PUBLICATIONS   0 CITATIONS    4 PUBLICATIONS   3 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Ontology Development View project

Plant Disease Detection View project

All content following this page was uploaded by Saif Hassan on 01 May 2020.

The user has requested enhancement of the downloaded file.

 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-9 Issue-1, May 2020

Plant Disease Detection using Deep Learning

Murk Chohan, Adil Khan, Rozina Chohan, Saif Hassan Katpar, Muhammad Saleem Mahar
 dollars [4]. These diseases are caused by fungi or fungal like
Abstract: [Context] Plants play an essential role in climate orgasms. However, other serious diseases of food and feed
change, agriculture industry and a country’s economy. Thereby crops are caused by viral and bacterial organisms [5]. Some of
taking care of plants is very crucial. Just like humans, plants are the diseases may be of spreadable nature, means they may
effected by several disease caused by bacteria, fungi and virus.
spread from one plant to other hence needed to be identified
Identification of these disease timely and curing them is essential
to prevent whole plant from destruction. [Objective] This paper and taken care timely. It’s very challenging job to detect
proposes a deep learning based model named plant disease disease in plants in very early stages [3].
detector. The model is able to detect several diseases from plants Some of the common symptoms of disease in plants disease
using pictures of their leaves. [Methodology] Plant disease are Leaf rust (common leaf rust in corn), Stem rust (wheat
detection model is developed using neural network. First of all stem rust), Sclerotinia (white mold), Powdery mildew,
augmentation is applied on dataset to increase the sample size. Birds-eye spot on berries (anthracnose), Damping off of
Later Convolution Neural Network (CNN) is used with multiple seedlings (phytophthora), Leaf spot (septoria brown spot),
convolution and pooling layers. PlantVillage dataset is used to
Chlorosis (yellowing of leaves). These diseases can be
train the model. After training the model, it is tested properly to
validate the results. [Results] We have performed different identified by physical condition of plants leaves. The experts
experiments using this model. 15% of data from PlantVillage data can identify either the plant is defected or not by looking at
is used for testing purpose that contains images of healthy as well leaves stems or fruit. This approach requires having lots
as diseased plants. Proposed model has achieved 98.3% testing human resources for this particular job. In this era of
accuracy. [Conclusion] This study is focused on deep learning technology and automation it is not very efficient approach, it
model to detect disease in plant leave. But, in future model can be would be much better if we have an automated system which
integrated with drone or any other system to live detect diseases detects disease in plants automatically. There are many
from plants and report the diseased plants location to people so
researches already done to fill this purpose most of them
that they can be cured accordingly.
utilize traditional machine learning approaches [6]. The
Keywords: Plant Disease, Convolution Neural Network purpose of this study is to create such automated system for
(CNN), Deep Learning, Agriculture, and PlantVillage detecting diseases in plants by using deep learning technique.
Deep learning is subset of machine learning. The advantage of
I. INTRODUCTION deep learning over machine learning is that one does not need
to worry about domain expertise as no feature engineering is
Plants play a vital role in economy and climate change. Since required in this, unlike traditional machine learning
climate change has become a global issue also addressed in approaches [7]. Our system just like other previous researches
UN general assembly 2019, various countries including utilizes images of plants leaves to detect disease in plants.
Pakistan are on mission to plant more and more trees and Plant disease detector is computer vision based automated
plants to keep climate balance. Many studies have proved that plant disease diagnostic system which utilizes machine
extinction of plants due to industry use have caused damage to learning techniques to correctly identify disease and healthy
ozone layer and thus resulting global warming. The rate of plants also the type of disease. For achieving so deep learning
climate change forecast for the future is 10-100 times faster network for images like Convolution neural network (CNN)
than the rate of DE glacial warming [1]. These plants also pay can be utilized. CNN is used to extract features from images
a major role in food industry as well. Balance of global food i.e. horizontal edges, vertical edges, RGB values etc. CNN is
production is also a major issue [2]. Apart from this in health best deep learning neural network for visual feature extraction
care also plants play a vital role [3]. Overall plants are very [8]. The CNN based network can be trained for detecting
essential for human survival therefore it’s also a worldwide disease in plants by providing huge amount of images of
concern to take care of them. healthy and sick plants and trained model in future can be
Just like human health plants health can also be affected by used to predict the disease in plants by images of plants
several diseases. In economic terms, annual losses in food, leaves.
fiber and ornamental production systems caused by plant
pests and diseases are estimated in the hundreds of billions of
II. METHODOLOGY
Deep learning is powerful machine learning approach which
Revised Manuscript Received on April 21, 2020.
* Correspondence Author
have mitigated the traditional machine learning headache of
Murk Chohan*, Department of Computer Science, Sukkur IBA feature engineering. It doesn’t need any domain expertise now
University, Pakistan. Email: [email protected] and all credit goes to deep learning. The core of deep learning
Adil Khan, Department of Computer Science, Sukkur IBA University, is artificial neural network (ANN). Artificial neural networks
Pakistan. Email: [email protected]
are mathematical models that replicate with their neurons and
Rozina Chohan, Department of Computer Science, Shah Abdul Latif
University, Khairpur, Pakistan. Email: [email protected] synapses interconnecting them the general principles of brain
Saif Hassan Katpar, Department of Computer Science, Sukkur IBA function [9].To implement
University, Pakistan. Email: [email protected] neural network one of the most
Muhammad Saleem Mahar, Department of Computer Science, Sukkur standard library is Tensorflow
IBA University, Pakistan. Email: [email protected]

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 909 & Sciences Publication
 Plant Disease Detection using Deep Learning

[10]. It provides all libraries related to artificial neural is tensor flow’s high level API for building and training deep
network. With the help of Tensorflow one can perform learning models.
classification tasks on text as well as images.
B. Dataset Discussion
A. Convolution Neural Network (CNN) Two datasets are used to perform plant disease detection.
Convolution Neural Networks (CNNs) [11] are used First dataset consists of 15 classes and second one consists of
to detect the disease in plant’s leaves. CNN is an evolution of 38 classes. Both databases have number of images of each
simple ANN that gives better result on images. Because plant. First dataset have total 2952 images. Final findings of
images contains repeating patterns of particular thing (any this work is on PlantVillage dataset which contains 38 classes
image). Two important functions of CNN are convolution and of different plants. It is also openly available on internet.
pooling. Convolution is used to detect edges of patterns in an Description of these classes and dataset is given in following
image and pooling is used to reduce the size of an image. Table- I (a) and (b).
CNN architectures that were applied on a problem are
following: (a) Simple CNN, (b) VGG [12], and (iii)
InceptionV3. Moreover training of these models are done
using Jupyter notebook and Keras API of Tensorflow. Keras
Table- I (a): Dataset Description
Class Plant Name Healthy or Disease Name Images (Number)
Diseased
C_0 Apple Diseased Apple_scab 2016
C_1 Apple Diseased Black_rot 1987
C_2 Apple Diseased Cedar_apple_rust 1760
C_3 Apple Healthy - 2008
C_4 Blueberry Diseased - 1816
C_5 Cherry_(including_sour) Diseased Powdery_mildew 1683
C_6 Cherry_(including_sour) Healthy - 1826
C_7 Corn_(maize) Diseased Cercospora_leaf_spotGray_leaf_spot 1642
C_8 Corn_(maize) Diseased Common_rust 1907
C_9 Corn_(maize) Diseased Northern_Leaf_Blight 1908
C_10 Corn_(maize) Healthy - 1859
C_11 Grape Diseased Black_rot 1888
C_12 Grape Diseased Esca_(Black_Measles) 1920
C_13 Grape Diseased Leaf_blight (Isariopsis_Leaf_Spot) 1722
C_14 Grape Healthy - 1692
C_15 Orange Diseased Haunglongbing (Citrus_greening) 2010
C_16 Peach Diseased Bacterial_spot 1838
C_17 Peach Healthy - 1728
C_18 Pepper_bell Diseased Bacterial_spot 1913
C_19 Pepper_bell Healthy - 1988
C_20 Potato Diseased Early_blight 1939
C_21 Potato Diseased Late_blight 1939
C_22 Potato Healthy - 1824
C_23 Raspberry Healthy - 1781
C_24 Soybean Healthy - 2022
C_25 Squash Diseased Powdery_mildew 1736
C_26 Strawberry Diseased Leaf_scorch 1774
C_27 Strawberry Healthy - 1824
C_28 Tomato Diseased Bacterial_spot 1702

By using this table you can come to know number of images in plants. Least images are from Raspberry, Soybean, and
each class. Each class contains approximately 2000 images. Squash class. Below image show some images of different
Fourteen different plants are available in this dataset. For leaves which are available in dataset.
every plant healthy as well as diseased images of leaves are
available. Most of the images belongs to Tomato and Apple

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 910 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-9 Issue-1, May 2020

Table- I (b): Dataset Description

Class Plant Name Healthy or Disease Name Images (Number)
Diseased
C_29 Tomato Diseased Early_blight 1920
C_30 Tomato Diseased Late_blight 1851
C_31 Tomato Diseased Leaf_Mold 1882
C_32 Tomato Diseased Septoria_leaf_spot 1745
C_33 Tomato Diseased Spider_mites Two-spotted_spider_mite 1741
C_34 Tomato Diseased Target_Spot 1827
C_35 Tomato Diseased Tomato_Yellow_Leaf_Curl_Virus 1961
C_36 Tomato Diseased Tomato_mosaic_virus 1790
C_37 Tomato Healthy - 1926
Total 70295

Fig. 1: Sample of Images from PlantVillage Dataset

filter. Batch Normalization is also used in this model. Batch
Dataset is divided into two parts one for training and other for normalization is used to scale data on particular scale but the
Testing. Splitting of dataset is 80/20 ratio randomly. 80% for difference is that it not just does it on input layer but it also do
the training dataset and rest 20% for testing dataset. Training it at other hidden layers. At last model is trained on
dataset consists 56,236 image and testing consists 14,059 PlantVillage dataset.
images. Training of model is done using 56,236 images and
14,056 images were kept unseen by model so that accuracy of Table- II: CNN Training Parameters
model can be checked. Parameter Value
Epochs 25
C. Model Description
Batch Size 32
First some preprocessing is applied on dataset in form of
augmentation to increase size of dataset in order to achieve Learning Rate 1e-1
better accuracy. Then images size are reduced by 256x256 Activation in middle layers Relu
pixels. After that a convolution neural network based model Activation in Final layer Softmax
will be created with multiple pooling and convolution layers
and a dense layer for prediction. Five convolution layers with
3x3 filter are used and five MaxPooling2D layers with 2x2

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 911 & Sciences Publication
 Plant Disease Detection using Deep Learning

Fig. 2: Applied Methodology

Fig. 3: Testing of an Image

III. RESULT AND DISCUSSION

This study shows the importance of plant disease detection in
these days. This model were developed using Deep Learning
in python. 20% (14,059) images from PlantVillage dataset
were used to test the accuracy of this model. These images are
from 38 different classes. 20% of each class randomly
selected for testing. Some real time images were also used.
Those images were captured from local environment. They do
not belong to any class which are present in dataset. But
model give us more than 95% accuracy on those images as
well by telling either leaf is healthy of unhealthy. Total 100 Unhealthy Unhealthy
images were used and 96 were classified correctly. Some Fig. 4: Locally Captured Images
images were captures at night with the help of flash light and
some images have dirt upon it so that they were misclassified. Testing dataset gives accuracy more than 98%. It means 1379
Some of the images which we captured from local images from 14,059 images were classified correctly by
environment. model. Below is the Training and Validation accuracy graph
generated by our model on testing dataset.

Unhealthy Healthy

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 912 & Sciences Publication
 International Journal of Recent Technology and Engineering (IJRTE)
ISSN: 2277-3878, Volume-9 Issue-1, May 2020

research was carried out using the publically accessible

collection of 70295 images, and 100 images from
experimental conditions and actual environment. The system
has achieved an overall 98% testing accuracy on publically
accessible dataset, and performed well on images of Sukkur
IBA University plants. It is concluded from accuracy that
CNN is highly suitable for automatic detection and diagnosis
of plants. This system can be integrated into mini-drones to
live detection of diseases from plants in cultivated areas.
Though this system is trained on Plant Village dataset with
only 38 classes it could tell if the plant has a disease or not as
somehow symptoms are same in all kinds of plants. In
addition, more actual environment images can be added to the
dataset to improve the accuracy on real-condition images of
leaves and classify more plant types as well as disease types.
In the future, this system can also adopt 3 layer approach
where the first layer detects if there’s any plant in an image or
not, second layer tells the plant type and the third layer tells if
there is any disease or not and what type of disease is there if
any.
Fig. 5: Training and Validation Accuracy on Testing
Dataset REFERENCES
1. Huntley, B. (1991). How plants respond to climate change: migration
Classes like Corn_(maize) healthy, Tomato rates, individualism and the consequences for plant
Tomato_mosaic_virus, Strawberry healthy, and communities. Annals of Botany, 15-22.
Corn_(maize) Common_rust give approximately 100% 2. Fang, Y., & Ramasamy, R. P. (2015). Current and prospective
methods for plant disease detection. Biosensors, 5(3), 537-561.
accuracy. Just 1 to 5 images misclassified from them. Apple 3. Mustafa, M. S., Husin, Z., Tan, W. K., Mavi, M. F., & Farook, R. S.
Cedar_apple_rust,Cherry_(including_sour) healthy, Grape M. Development of automated hybrid intelligent system for herbs plant
Black_rot, and Raspberry healthy classes gives less accuracy classification and early herbs plant disease detection. Neural
than other classes. Below is the table of image classes which Computing and Applications, 1-23.
4. Williams, S. D. (2017, February 1). Plants Get Sick Too! Retrieved
gives better accuracy while testing on our model. from https://ohioline.osu.edu/factsheet/plpath-gen-1
5. Isleib, J., & Michigan State University. (2018, October 2). Signs and
Table- III: Image Classes with Better Accuracy symptoms of plant disease: Is if fungal, viral or bacterial? Retrieved
Class Tota Correct Mis- from
https://www.canr.msu.edu/news/signs_and_symptoms_of_plant_dise
l Classifie classified ase_is_it_fungal_viral_or_bacterial
d 6. Shruthi, U., Nagaveni, V., & Raghavendra, B. K. (2019, March). A
Review on Machine Learning Classification Techniques for Plant
Corn_(maize)___healthy 186 186 0 Disease Detection. In 2019 5th International Conference on Advanced
Tomato___Tomato_mosaic_virus 179 176 3 Computing & Communication Systems (ICACCS) (pp. 281-284).
IEEE.
Strawberry___healthy 182 175 7 7. LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep
Corn_(maize)___Common_rust_ 191 188 3 learning. nature, 521(7553), 436-444.
8. Sharif Razavian, A., Azizpour, H., Sullivan, J., & Carlsson, S. (2014).
Corn_(maize)___Northern_Leaf_Bl 191 175 16 CNN features off-the-shelf: an astounding baseline for recognition.
In Proceedings of the IEEE conference on computer vision and
ight
pattern recognition workshops (pp. 806-813).
9. Ferentinos, K. P. (2018). Deep learning models for plant disease
detection and diagnosis. Computers and Electronics in
Table- IV: Final Model Performance
Agriculture, 145, 311-318.
Model Dataset for Dataset for Training Testing
10. Abadi, M., Barham, P., Chen, J., Chen, Z., Davis, A., Dean, J., ... &
Training Testing Accuracy Accuracy Kudlur, M. (2016). Tensorflow: A system for large-scale machine
learning. In 12th {USENIX} Symposium on Operating Systems Design
CNN PlantVillage PlantVillage 99% 98%+ and Implementation ({OSDI} 16) (pp. 265-283).
(80%) (20%) 11. LeCun, Y., & Bengio, Y. (1995). Convolutional networks for images,
speech, and time series. The handbook of brain theory and neural
CNN PlantVillage Actual 99% 95%+ networks, 3361(10), 1995
(80%) Environment 12. Simonyan, K., & Zisserman, A. (2014). Very deep convolutional
networks for large-scale image recognition. arXiv preprint
(100 Images) arXiv:1409.1556.

IV. CONCLUSION AND FUTURE WORK

This study has utilized deep learning capabilities to achieve
automatic plant disease detection system. This system is
based on a simple classification mechanism which exploits
the feature extraction functionalities of CNN. For prediction
finally, the model utilizes the fully connected layers. The

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 913 & Sciences Publication
 Plant Disease Detection using Deep Learning

AUTHORS PROFILE
Murk Chohan Completed her BE Computer System Engineering from
Quaid e Awam University of Engineering & Technology Nawabshah in 207
and received her MS Computer Science degree specialized in Data
Knowledge Engineering in 2019 from Sukkur IBA University. He has been
researching in the field of Deep Learning, Software Engineering, and Cloud
Computing. Currently she is a faculty member of Computer Science
department at IBA-IET Khairpur. Pakistan.

Adil Khan is a faculty member of Computer

Science department at IBA Community College
Ubauro, Pakistan. He completed his BS Computer
Science in 2017 and MS Computer Science
specialized in Data Knowledge Engineering from
Sukkur IBA University in 2019, both with
distinction (Gold Medalist). He has been
researching in the field of Deep Learning,
Software Engineering, and Cloud Computing.

Rozina Chohan is a faculty member of Computer Science Department at

Shah Abdul Latif University, Khairpur, Pakistan. She has earned Doctor of
Philosophy (Ph.D.) in Information Security Identity Theft from UCLan
University, United Kingdom. She obtained her MS degree from University
of Northampton, United Kingdom. She has been mainly researching in the
field of Cyber Security, Information Security, and Deep Learning.

Saif Hassan Katper received the B.S degree in

Computer Science from Sukkur IBA University,
Pakistan in 2016 and M.S degree in Computer
Science from Mohammad Ali Jinnah University,
Karachi, Pakistan (Gold Medalist in MS). He is
currently teaching at Sukkur IBA University. He
has teaching experience of more than 4 years. His
specialization areas include Deep Learning,
Computer Vision. Mr. Saif's awards include PM
ICT (R&D) Fund Scholarship for Four Years B.S Program, Gold Medalist in
Masters, online certifications and many appreciation awards and certificates
for conducting workshops/seminars.

Muhammad Saleem obtained BS Computer

Science degree from University of Sindh,
Jamshoro. Currently he is enrolled in MS
Computer Science in Sukkur IBA University. He
is faculty member of Computer Science
Department at IBA Community College,
Jacobabad, Pakistan with over 6 years of
experience.

Published By:
Retrieval Number: A2139059120/2020©BEIESP Blue Eyes Intelligence Engineering
DOI:10.35940/ijrte.A2139.059120 914 & Sciences Publication
View publication stats