Phytopathology and Diagnosis Using Deep Learning
Phytopathology and Diagnosis Using Deep Learning
Volume 5, Issue 4, April – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Phytopathology and Diagnosis using Deep Learning
Mayur S. Shinde Tejas H. Kenjale
Dept. of Information Technology Dept. of Information Technology
Bharati Vidyapeeth College of Engineering, Bharati Vidyapeeth College of Engineering,
Belpada, Navi Mumbai, India Belpada, Navi Mumbai, India
Abstract:- Phytopathology is the scientific study of plant extracts the features while training. The applications such
diseases caused by environmental conditions. Plant as speech recognition, image processing, image restoration,
diseases must be detected in the early stages of harvest. natural language processing makes use of CNN.
For detection of plant diseases as accurately and quickly
as possible with the lower costs, analysis of patterns and CNN easily identifies and classifies the various
symptoms of plant disease are required. Hence, we objects without/less processing; CNN can be used to
present a system which will detect whether the plant is analyze the visual images and also be used to extract the
affected with any disease. If the plant is affected with features with multi-layered structure. Figure 1 represents
any disease, the system will provide the causes for the architecture of CNN.
disease along with solution for the same. This system is
using CNN, that comes under Deep Learning which is
primarily used for classification of images based on the
characteristics.
Keywords:- Phytopathology, CNN (Convolution Neural
Network), Harvest, Plant Disease Detection
I. INTRODUCTION
India is the mostly agriculture dependent country. Fig 1:- General Architecture of CNN
Most of the sectors in India are agriculture based. Indian
agriculture sector provides half of the employment. India is CNN consists of four main layers:
the country which provides the spicy products, rice, wheat,
spices to the globe. Farmers are completely dependent on A. Convolutional layer
agriculture and the crops which they produce. Therefore, in In convolution layer, sequences of mathematical
agriculture sector plant diseases must be detected and operations are performed so that features can be extracted
identified with accurate solutions. from input image. The filter is used to reduce the size of an
image. The filter is moved step by step from left corner of
Due to the plant diseases, farmers cannot produce the an image. At every step, the values present in image are
yield as they are expecting and therefore it affects the multiplied with the filter values and the summation of result
economy as well. So to identify and detect plant diseases as is performed. A new matrix of a smaller size is created
early as possible, we should make use of disease detection from the input image. Figure 2 represents the convolution
techniques. Various parts of the plants such as stem and functionalities in convolution layer of 7x7 image input and
leaves gets affected due to diseases. A person/farmer 3x3 filter
cannot identify the accurate disease. In most of the
countries, farmers are unable to make a contact with
agriculture experts or technicians. Therefore, it is necessary
to adopt new technologies which will perform the plant
disease detection as well as classification.
II. CONVOLUTIONAL NEURAL NETWORK
CNN can be used in various applications such as
natural language processing, speech recognition. It can be
used to identify the satellite images which contain trees,
roads, buildings, etc. CNN algorithm such as neural Fig 2:- Convolution functionalities of 7x7 image with 3x3
networks, where it consist of neurons which takes it’s input filter
and multiplies with its weight and then it is passed through
the activation function layer to another neurons. In CNN,
each layer makes use of output of previous layer as an
input. In CNN there is no need to perform feature
extraction and classification because the model itself
IJISRT20APR731 www.ijisrt.com 1119
Volume 5, Issue 4, April – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
B. Pooling layer
The pooling layer gets implemented after the
convolution layer is performed. In this layer the resulted
output matrix size is get reduced. In pooling layer filters
with different sizes can be used, By default 2x2 size filter is
used. Figure 3 represents an example of pooling
functionalities.
Fig 3:- Pooling functionalities
C. Activation function layer
Activation function layer is simply output of a
function. This layer performs functions such as
identification, tan h, Rectified linear unit, Arc Tan.
D. Fully connected layer.
After performing convolution, pooling and activation
layers, resulted matrix is given as input to the fully
connected layer. In this layer Recognition and classification
both functions are performed. Figure 4 represents fully
connected layer.
Fig 4:- Fully connected layer
III. PROPOSED METHOD
Fig 5
In this study, images from PlantVillage dataset are
used. Dataset is divided into 80-20 ration. 80% of dataset is For disease detection, image is captured or selected.
used for training Neural Network and 20% of dataset is The selected or captured image is then processed by Neural
used for testing neural network. Once the neural network is Network. If the disease is detected by neural network then
trained for disease classification necessary files are it will give a diagnosis report consisting reasons for disease
generated for deploying the neural network. This file is along with how it can be cured.
used for identifying disease.
IJISRT20APR731 www.ijisrt.com 1120
Volume 5, Issue 4, April – 2020 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
IV. EXPERIMENTAL RESULTS
To carry out functionality of this system, we have
performed a demonstration on few leaf image databases
and have performed classification. The challenge occurred
while performing demonstration is that, the more than 2
leaves having different diseases are somehow similar to
each other. Therefore, this challenge can be resulted in
getting the leaves classified in different classes.
V. CONCLUSION
Currently the Convolutional Neural Network (CNN)
system can detect healthy tomato leaves and tomato leaves
affected by bacterial spots with the accuracy of 97%. By
providing more training data to the CNN the system can
predict different diseases accurately. CNN doesn’t require
pre-processing for the input. Which means that the system
does not require to do pre-processing and classification in
two separate parts, CNN does both the pre-processing and
classification. CNN can work with both supervised,
unsupervised and semi-supervised data, Because of this the
prediction model is easy to train.
REFERENCES
[1]. Leninisha Shanmugam, Agasta Adline A. L,
Aishwarya N Krithika G, “Disease Detection in Crops
using Remote Sensing Images”. 2017 IEEE
International Conference on Technological
Innovations in ICT for Agriculture and Rural
Development.
[2]. Zulkifli Bin Husin, Abdul Hallis Bin Abdul Aziz, Ali
Yeon Bin Md Shakaff, Rohani Binti S Mohamed
Farook, “Feasibility Study on Plant Chili Disease
Detection Using Image Processing Techniques”. 2012
Third International Conference on Intelligent Systems
Modelling and Simulation.
[3]. K. Indumathi, R. Hemalatha, S. Aasha Nandhini and
S. Radha, “Intelligent Plant Disease Detection System
Using Wireless Multimedia Sensor Networks”. IEEE
WiSPNET 2017 conference.
[4]. Melike Sardogan, Adem Tuncer, Yunus Ozen, “Plant
Leaf Disease Detection and Classification based on
CNN with LVQ Algorithm”. IEEE 2018 3rd
International Conference on Computer Science and
Engineering.
