2018 International Conference On Advances in Communication and Computing Technology (ICACCT)

Amrutvahini College of Engineering, Sangamner, Ahmednagar, India. Feb 8-9, 2018

PLANT DISEASE DETECTOR

Jagadish Kashinath Kamble
Pune Institute of Computer Technology (PICT)
Pune, INDIA
jagdish750@gmail.com

Abstract— nowadays agriculture crops face many problems x For the adoption of modern agricultural technology
like traits or diseases. Each year plant diseases, viruses and
fungal attacks lead to crop losses up to 30% of total production. and modern equipment in India is still not as much as
Clearly control of plant disease depends on accurate and rapid
you want.
detection and identification of the disease. Detection is the
process of identifying the presence of symptoms of disease. x Plant diseases, insects and pests.
Farmers in India have minimal access to agricultural experts,
who can inspect crop for identifying disease and provide advice Agriculture Plays very important role in economic and
about disease. Delayed expert responses or advice to queries often social affairs of people in India. Indian agriculture basically
reach farmers too late which are useless. Automatic detection of characterized as a means of subsistence is changing fast as per
plant disease is important as it beneficial to monitor large field of market demands both domestically and international. The crop
crops and identifies the presence of disease symptom on the plant yield losses, on field and during post-harvest caused by pests,
leaf or stem. In this project we develop Mobile app for disease and weeds are of paramount importance. The
automatically detecting plant disease through image processing responsibility of protecting food crops from diseases and pests
technique with the objective of providing fast, accurate, ease of in the challenging environment is rising with increase in
use and inexpensive solutions to farmers. human population and its needs. The crop losses due to pest
are assessed to be ranging approximately between 20 to 30%
Keywords— crop; plant disease; detection; mobile app of crop productions. The commercial farmers in developed
countries are able to handle difficult situations up in the best
I. INTRODUCTION way with the high input cost due to the increasing area of
India is an Agricultural country. Agriculture business in production, manpower and others by the application of proper
India has been always a priority. About 70% of India’s knowledge base developed with the latest technology, whereas
population is dependent on the business of farming for it is not so in developing countries like India. This is targeted
employment. Two-thirds of the people who work in to fill the gap between commercial farmers in developed and
agriculture or agriculture related industry. Around 30% of the developing countries and to provide with the latest
population of their indirect food, clothing, shelter depends on technological assistance to farmers. Information Technology
the income of farming. India is the second largest in the world has a key role to play in all facets of Indian agriculture.
for producing and exporting farm output. Agriculture sector Nowadays there are lots of technologies like remote sensing,
share one-third of the country's GDP (Gross Domestic geoinformatics, wireless sensor networks and digital image
Product) and is its single largest contributor. Agricultural processing etc. which are providing the support to Indian
export contributes one-fifth of the total exports of the nation. agriculture and farming sector. The latest technological
In Indian economy, Agriculture is the Leading Contributor and application in this most deprived sector can help the Indian
also major component in the development of socio-economic farmers for improving the efficiency and productivity of
Conformation of India. The Agriculture Sector cannot be agriculture. The technology may provide the E-powering tool
overlooked because more than 50% of the manpower in India for Indian farmers in decision making, who lives in the rural
is employed only in this from last 30-40 years as per the areas. Mobile Technology with image processing techniques
United Nations Food and Agriculture Organization (UNFAO). plays a major role in Indian agriculture sector. The
Therefore, agriculture has always been a priority in India’s requirement of mobile phone based application for targeting
Five Year Plans. The agriculture yield is not very much Indian farmer specific requirements and needs in rural India
distinctive for developing countries like India and other and which depends on agricultural sector. This requirement
thus giving a scope for research in this area. Following are provides a prospect chance to research community to study for
some of the causes of low productivity of agriculture in India. agro-economic development of Indian rural areas. The mobile
application solutions based on Indian farmer’s needs and
x Illiteracy, lack of socio-economic progress, demand are a major challenge for technical and social research
community with rural agro-economic constraints. Image
inadequate or inefficient finance for farm production
processing techniques can be used to detect the plant disease
and Lack of its sales and marketing facilities. at an early stage with the help of captured images by any
imaging device. An expert system can be designed to provide
the best care at the early stages with the help of the disease
present and its security level. In developing countries most of
the farmers identify the disease on the basis of their

978-1-5386-0926-2/18/$31.00 ©2018 IEEE 97

experience, but some time, even expert farmers and plant image is taken directly from the android device back camera
pathologist are not able to recognize the exact disease present [More than 8 mega pixel]. In this step, the captured image of
in the agriculture product. This problem can be solved using the plant leave is chosen as a input from developed app, with
mobile based expert system in which the farmer only needs to field information, plant information and send to the server for
capture the image of the disease leaf and on the basis of processing [Fig.3]. The acquired images can be stored in
texture features the developed system will identify its type and different format such as PNG, JPEG, and BMP etc. This
severity. Various image processing techniques were used in acquired image then undergo through various different image
agriculture like fruits and vegetable recognition, weed and soil processing techniques at the server where feature are
segmentation, disease quantification, disease classification, extraction for further analysis.
disease identification. The need of the rural farmers is the
exact information about disease, seeds, fertilizer, insects and
cures. The dissemination of Information and Communication
Technologies (ICT) in developing countries provides much
opportunity to transfer knowledge and information. Mobile
phones are easily available to across the nation because of
price, usage in day today work. Mobile phone uses in
developing countries play a vital role for the enhancement of
agriculture business. From recent years, the modern mobile
phones become more powerful and handy and also support
wide variety of services.

II. METHODOLOGY
The proposed system designed as mobile application. The
work commence with capturing image (plant leaf) with require
information such as plant information, field information (state
information) using Android Mobile phone with camera (more
than 8 mp)and through internet connectivity send to the server.
Server side image processing is done on input image to find
the infected part on leaf. Generally, the proposed stages
divided into four modules viz. Fig.2. Flow structure of proposed system

x Client Module (Front End) B. Feature Extraction Module

1) Client Module (Mobile Application)
Feature extraction is done on leaf image just to remove
x Server Module (Back-End) background and green pixels as to enhances the infected part
2) Feature Extraction Module and predict the appropriate cure on it.
3) Training Database
4) Decision Making Module

Fig. 1 Proposed systems Architecture

The different image processing steps like image enhancement,
image segmentation are applying on input image. Then
different texture feature values are calculated from processed
image. At last extracted feature values entered into the pre
trained Artificial Neural network (ANN) as input to classify
the test image. Finally predicted result (disease) send to user.
Fig. 3 UI of proposed system
[Fig.2]
a) Image Enhancement (Background Subtraction): In this
A. Image Acquisition (Mobile Application program): process, Image enhance is done based on removing
The application program is developed with Android background, shadow and light reflections. Shadow or light
programming language using Eclipse framework. The reflection detection and removal are an important task when
application program provides user interface to Farmers. The dealing with color outdoor images. It is still a challenging

98
problem especially from the images with complicated more than 750 images of different plants with 7 different
background such as with some interference and overlaps features of each. Dataset was designed on the basis of disease
between two adjacent leaves. For the Background Subtraction classes of different plants. Some images were captured,
EM algorithm for Gaussian Mixture [EMGM] was used. First generated by different research centers of Maharashtra and
background from acquired image is removed and then non- some are collected from Google images. The efficiency and
green pixels as shown in Fig.4 accuracy of classifier is totally dependent on image database.

Fig.4. Flow Diagram of Infected Part Detection

Fig.5. Flow Diagram of Texture Feature Extraction Module
b) Masking and Removal of green-pixels (Disease
Segmentation) setting Threshold: Masking and removing the D. Decision Making Module (Classification)
green pixels is important in the disease segmentation step Decision making module predict the disease using the pre
because it describes the healthy part of the leaf image. For trained classifier and test data. The decision making module
masking image pixel value are set to zero. To mask and perform two steps, training and classification.
remove green pixels from leaf image, need to set some a) Image classification: It is the method where first we
specified or varying threshold value [t]. Assign zero value to train any classifier with the features of images with known
RGB (Red, Green and Blue) component of pixel of an image, category. Any new image is classified using its feature vector
if the pixel intensity of green factor of pixel component is less and trained classifier. The proposed approach classification
than computed threshold value. Because these pixels are not not only describes whether the leaf is diseased or not but also
useful for disease identification and classification step as it identify the disease that affect to the leaf. ANN was used as
represent healthy part of leaf as shown in fig 4. Image classifier to generated or trained using the feature database.
segmentation is simply done to separate foreground and
background of image, by setting background as 0 and x Step 1: Define the classification class: Different
foreground as 1 on gray level filtering of image. classification classes can be defined on the basis of
features and characteristics of image.
c) Texture Feature extraction: Texture is important feature
for classification and retrieval of an Image. Texture feature is x Step 2: Feature selection: Unique features should be
characterizing the surface and structure property of an image select or establish to discriminate between the
which can be utilize in image processing. Image is generally different classification classes etc.
made from pixels and texture property can be said as group of x Step 3: Sampling of training data: The appropriate
pixels. In this approach CCM[5] (color co-occurrence decision rules should be determined by sampling the
method/Harlick method) method used for texture feature training data. The classification techniques such as
extraction which used for classification of plant disease. As Supervised or unsupervised should be selected using
shown in fig 5, with help of texture extraction method, we had training data sets.
combined the black pixel part as one group and brown pixel x Step 4: Train the classifier: From the appropriate
part as another group. training data, train the corresponding classifier for
C. Feature Database (Feature Dataset depending on plant classification.
disease images x Step 5: Classification: Using the pre-trained classifier
For the detection and classification of plant disease or decision rules classification will be takes place.
require large number of image database. The images are used
for testing and training purpose. The image library consist

99
 validity of trained network. At last phase trained
neural network applied on testing samples.

Fig.7. Classifier Sets

Some functions used by ANN for training in MatLab are:
x= input matrix, t= target matrix;
Fig.6. Haralick Texture Features (Trained Data) net = patternnet (10);
b) Training Data for classification: To train, two net= train (net,x,t);
matrices are considered, the input vector matrix and the target E. Predicted Diseases:
vector matrix. The input matrix is the P x M matrix where P is The input or captured image entered by farmer, undergo
feature vector size and M is image database size. The this phases, tested against the classifier to predicted the
corresponding target matrix is the Rx M matrix where R is infected part and to get appropriate remedial action, which is
disease class size and M is image database size. Each then send back to farmer. [fig.8]
corresponding target matrix column consist 0 or 1. For first
disease class in corresponding Target matrix column element
consists 1 (other elements are zeros).
x Artificial Neural Network (ANN): An ANN is the
information processing model based on the human
brain (nervous system). It consist large number of
interconnected components (neurons) that work for
solving specific problems. ANN is highly parallel
and distributed processing paradigm. ANN is train
using training feature data and trained classifier used
to classify the testing feature data. Validation is used
for verify the accuracy of trained classifier. Before
start the training, Proper Network should be set up
which include the type of network (Feed forward
networks or radial basis network) and method of
training. Designing ANN model models follows the
following steps:
1) Build the Network
2) Create Classifier: Train the Classifier and Fig.8. Predicted Disease
Testing the data with trained classifier
x ANN Network: Used feed forward neural networks
of two layers with one hidden layer, in which number III. CONCLUSION
of neurons for hidden layer was 10. Complex
problem require more neurons and more layers. The Detecting the disease is mainly the purpose of this
input to the neural network is feature vector matrix system. There is main characteristics of disease detection are
and its corresponding target vector matrix. The help to farmers with speed and accuracy using Information
output is disease class representing these diseases ( and Communication Technology (ICT). The method reported
for example Alternaria blight (early blight),Downy in the thesis can be used to design a plant disease detector for
mildew, bacterial pustule, frogeye leaf spot, powdery farmers for the early detection of plant disease infection and
mildew, purple seed stain, septoria leaf spot, healthy getting appropriate cure remotely.
leaf etc.). The disease class is different for each crop. In proposed system we define the application of texture
x Classifier: The images samples are divided into analysis for detecting plant diseases with the help of different
different sets like training, testing and validation. The image processing technique. There is require to working on
training set used for NN training (train the development of automatic, efficient, fast and accurate system
classifier).The validation set used for check the which is use for detection of disease and provide the solutions.

100
 REFERENCES [7] Smita Naikwadi, Niket Amoda, Advances In Image Processing For
Detection Of Plant Diseases at International Journal of Application or
Innovation in Engineering & Management vol.3,nov.2013.
[1] Crop Pest Surveillance and Advisory Project (CROPSAP), Maharashtra. [8] M. C. Burl et al. Mining for Image Content. In Systemic, Cybernetics,
[2] P.Revathi, M.Hemalatha, Classification of Cotton Leaf Spot Diseases and Informatics / Information Systems: Analysis and Synthesis,
Using Image Processing Edge Detection Techniques ISBN, pp 169-173, (Orlando,FL), July 1999.
IEEE-2012. [9] Aldrich, B. and M. Desai, 1994. Application of spatial grey level
[3] H. Al-Hiary, S. Bani-Ahmad, M. Reyalat, M. Braik and Z. dependence methods to digitized mammograms, Image Analysis and
ALRahamneh, Fast and Accurate Detection and Classification of Plant Interpretation, 1994. Proceedings of the IEEE Southwest Symposium on,
Diseases IJCA, vol. 17(1), pp. 31-38, March 2011, IEEE-2010. April 21-24, UK., pp: 100-105.
[4] S. Arivazhagan, R. Newlin Shebiah, S. Ananthi, S. Vishnu Varthini, [10] Stone, M.C., 2001. A survey of color for computer graphics. Course at
Detection of unhealthy region of plant leaves and classification of plant SIGGRAPH 2001, (August 2001).
leaf diseases using texture features, CIGR, vol. 15(1), pp. 211-217, http://graphics.stanford.edu/courses/cs448b-02-spring/04cdrom.pdf.
March 2013.
[11] Ford, A. and A. Roberts, 2010. Color space conversions. August 11,
[5] http://in.mathworks.com/help/images/gray-level-co-occurrence- 1998.http://www.poynton.com/PDFs/coloureq.pdf.
matrixglcm.html.
[12] Weizheng, S., Yachun, W., Zhanliang, C., and Hongda, W. (2008).
[6] P.landge, S. Patil, D. Khot, O. Otari, U. Malavkar, Automatic Detection Grading Method of Leaf Spot Disease Based onImage Processing. In
and Classification of Plant Disease through Image Processing at Proceedings of the 2008 international Conference on Computer Science
International Journal of Advanced Research in Computer Science and and Software Engineering - Volume 06 (December 12 - 14, 2008).
Software Engineering Vol.3,issue-7 july-2013. CSSE. IEEE Computer Society, Washington, DC, 491-494. DOI=
http://dx.doi.org/10.1109/CSSE. 2008.1649.

101