You are on page 1of 13

SRI SAIRAM COLLEGE OF ENGINEERING

CONSULTANCY REPORT

The project ’Cotton Crop disease prediction using Deep Learning’ aims at. Identifying the
symptom of the disease of the cotton plant by the naked eye is really a hard task for the farmers
so here we come up with a Plant disease detection model, a model that is capable of detecting
the leaves that are affected andthe cure for them using deep learning

September 2020 – May 2021

By

Mrs. Kalamani P
CONTENTS

1. INTRODUCTION
1.1 Motivation
1.2 Challenges
1.3 Software Requirements Specification
2. SYSTEM ARCHITECTURE AND DESIGN
2.1 Architecture Used
2.1.1 Resnet 50
2.1.2 Inception
2.1.3 VGG 16
3. METHODOLOGY
3.1 Procedure of disease Prediction
3.2 Design of module
4. RESULTS AND DISCUSSIONS
4.1 Results Achieved
4.2 Discussion
5. CONCLUSION AND FUTURE ENHANCEMENT
5.1 Conclusion
5.2 Future Enhancement
1. INTRODUCTION
1.1MOTIVATION

As we all know India is the second-largest in population and along with it many of the crops are grown in
India out of which cotton is one of the main cash crops in India.

The first mill of cotton in India was established in 1818’s near Kolkata and the cottonmill was started in 1854
by KGN Daber is its name was Spinning and Weaving Company Cotton crops play an important role in the
agricultural economy of India.There are nine states in India that grow cotton in which Gujarat is the leading
state following Maharashtra. Maharashtra has the largest area for cotton cultivation. Nearly 60% area of
cotton cultivation is rain fed in India. Along with being the most common crop, it comes with the con that its
leaves get affected by bacteria or funguseasily.

Many of the farmers get affected by these bacteria and it also leads to a reductionin their production as it
not only restricts the growth of the plant but also destroys its crop and also losses of money. General
observation by farmers is time-consumingand at times, it is not very accurate so here we come up with a Plant
disease detection model, a model that is capable of detecting the leaves that are affected and the cure for the
musing deep learning. We will be detecting spot disease in leaves and will be classifying the leaves based
on disease, leaf types using machine learning (deep learning) concepts . The most common cotton disease
are ‘Verticillium wilt’, ‘Boll Rot’, ‘Fusarium wilt’, ‘Alternaria leaf spot’,‘Tobacco Streak virus’. These
diseasescause harm to the agricultural economy of India in more than one way.

1.2CHALLENGES

 Finding the best model with highest accuracy.


 Dataset collection in huge number for training the model.
 Time taken in training the model.
 Deciding the model which is not over fitted or under fitted.
 Removing bias from the model (some affected leaves getting predicted to behealthy).
 Making the web application of the module more user friendly

1
1.3 Software Requirement Specification

1.3.1 Python

Python is a worldwide largely used programming language and it is an extremely amazing language.
Python upholds numerous ideal models including object-situated and utilitarian programming. The
framework is adaptable and incorporates programmed memory, the executives and has large and up to date
standard libraries. It is a verypowerful high-level language. Python utilizes dynamic composing just as
reference computation to oversee memory. A significant element of Python is the adaptability of the words
switch, which incorporates the technique also, the variablewords during framework execution. Python has a
huge standard library, regularly referred to as one of Python’s incredible powers, giving the right devices to
many tasks.
The Python Package Index, Python’s outsider programming bundle, contains north of 72,000 bundles that
offer a wide scope of capacities, including client graphical, interfaces, web system, interactive media,
information, organization and correspondence test structures, mechanization and web scratching, text
handling.
1.3.2 Spyder

The abbreviation of spyder is “Scientific Python Development Environment” and it is a free IDE. Which
can be accessed using Anaconda Navigator. It is mainly used for scientific programming in the Python
programming language.

1.3.3 HTML and CSS

The abbreviation of HTML is “Hyper Text Markup Language” it a language mainlyused when a user want
to design and implement webpages and the CSS stands for “Cascading Style Sheets” and it issued to design
the web page. In our project, we have used the language HTML and CSS to create a User Interface for our
web application.

1.3.4 System SpecificationHardware Equipment:

RAM: 2GB or more


Hard Disk Space: 20 GB or more
Processor: 64 bits or 32 bits Compatible with Clock Speed of 1.7GHz

Software Requirements:

Operating System: Windows (7/8/10/11), Linux, macOS Software Used: Anaconda Navigator

2
2. SYSTEM ARCHITECTURE DESIGN

2.1 ARCHITECTURES USED

2.1.1 RESNET 50

ResNet is for Residual Networks. It is classic form of neural network which is widely used for the backbone
in many tasks that are associated with computer visiontasks.

It is a residual neural network having 50 layers. RESNET model won the ImageNet (ILSVRC) competition
which was held in 2015 and it is a very popular model which is used in classification of images. Res Net was
the first one to introduce the conceptof skip connection.

Architecture Diagram

Fig 3.1.1

2.1.2 INCEPTION V3

It is a model which is widely for image recognition. Inception V3 is pre-trained CNN model which has 48
layers.

InceptionV3 is the model by Google which is 3rd edition of the Inception CNN and the training of
Inception V3 was done on ImageNet.

3
Architecture Diagram

Fig 3.1.2

2.1.3 VGG 16

It is a CNN model and have 16 layers . One of the most important point of this algorithm is that it have 3*3
filter convolution layer

VGG 16 is considered to be one of the best vision model till date.


It follows the arrangement of the convolution along with the max pool layers consistently along the process
throughout whole architecture. In the end it has2 FC(fully connected layers) followed by a soft max for
output.
Architecture Diagram

Fig 3.1.3

4
3. METHODOLOGY

3.1 PROCEDURE OF DISEASE PREDICTION

We studied many literature reviews out of which we figured out CNN, ANN, SVM, KNN and naive image
recognition algorithms are used.
So for our model creation we proceeded with using Transfer learning concepts and making use of keras.
Importing: Keras, tensorflow.

Image Acquisition

Acquiring the image from the dataset and from the internet browsing through various websites of different
topologies.

This is the basic step in any of the computer vision projects. We have taken the images from the online
internet sources along with including new ones from and including some of the images that are really hard
to differentiate.

Digitizing images

Digitized form of the images are generated which can further be used in processing and analysis part .

Image rescaling

Rescaling an image according to the size and flattening it and basically manipulatingthe physical attributes of
the image so that it becomes easy for the computer to process further with the images.

Image pre processing

It expands the size of the training dataset to create more datasets for the training process of the model,
indefinitely generate batches of augmented image data. Image assignment to different classes.
Also along with it vectorization of the images and also resizing of the dataset according to the desired needs
.Image augmentation is one of the crucial steps to which deals with further generating more datasets from the
available ones, which helps in training of our model.

Feature extraction

Extracting features from the images that can be used by the model for prediction .This is a crucial step as
in the processes that we have done before all of the processes aims to provide proper and the adequate amount
of input data all of the processes aredone so that when we come to the feature extraction part it can be done
with best accuracy. Layers of the model are used to learn important information that is available in the data
set. Making use of filtering mechanism the datasets are used tomatch and extract the values.

5
Model creation

Before model creation we need to classify our dataset .After this we have to select an accurate algorithm
for our model creation since we were using the concept of transfer learning we had gone forward with
creating an accurate model using Kerasand compiling it.

Model training

Model Training is one of the most important and time taking process out of the whole processes. Since our
project deals with computer vision along with machine learning implementations we now need to put our
GPU into use and train our model by utilizing the training data. The model training is based on the number
of the epochs that we run our model for. Mainly we decide the number of batches for the initialization of
epochs. Saving trained model to file so that we can further, use it in our web app (whiledeploying web
app through ‘Spyder’).

Graph Plotting

Plotting the graph to show accuracy and loss in the model. Some important point:

1. Fetching images from various sources.


2. Going forward with image augmentation for manipulating the images.
3. Using self-trained layers (top and bottom layers) for desired output.

Going forward in details we are using transfer learning Machine learningalgorithms.


The following are the machine learning algorithms we are using in our project:
1. Resnet 50
2. Inception V3
3. VGG 16
Designing the frontend with the help of HTML, C S S to make a user friendly interface

6
3.2 DESIGN OF MODULE (flow diagram)

Fig 4.2

7
4. RESULTS AND DISCUSSION

4.1 RESULTS ACHIEVED

We developed a ML model, in which we used the transfer learning concept using which we were able to
achieve good accuracy in disease prediction.
In our project we used three main algorithms: Resnet 50Inception V3VGG 16
We decided to go forward with the VGG 16 algorithms as it was using less parameter for training and was
able to achieve more accuracy arrangement of the convolution along with the max pool layers.
To back up our research we have compared the existing and the proposed algorithms that can be brought into
use Comparison between proposed and existing models.

Fig 6.1

8
Fig 6.2
The above output is for DISEASED COTTON PLANT

Fig 6.3
This output is for HEALTHY COTTON PLANT but our model is also telling that the leaves of the cotton
plant have some chemical fallen on it.

9
Fig 6.4

This output is for HEALTHY COTTON PLANT

4.2 DISCUSSION

Getting to know whether the plant is diseased or not and its diagnosis is an important process. Many factors
such as yield capacity, the richness of the grains, nutrition retention are evaluated and are fully based on
crop health. This paper aims to make a comprehensive study on various computational methods and models
embedded in the plant disease identification and classification system. There were many algorithms played
a vital role in achieving the desired task. The outcome of this study reveals the significance of technology
nowadays that makes it possible for us to develop and deploy automated tools for assisting the end-users to
find the plant disease without human intervention , hence we studied all the pre-existing model and came
up with the new and improved way of helping the users (cultivators).We came to the conclusion that till
now still some algorithms that are being used can be changed and should be changed in order to get results
more accurately helping the end-users with more accurate results and hence we brought out machine
learning model.

10
5. CONCLUSION AND FUTURE ENHANCEMENT

5.1 CONCLUSION

Recognition of disease is the main purpose of the proposed approach which helps to recognize the leaf
diseases with little computational effort. The plant disease not only affects the production of cotton, it also
reduces the quality of the produced crop. The model proposed for the detection of disease in cotton plants
also serves farmers to increase the productivity of crops by classifying the disease type. It is created to
detect if the plant is diseased, healthy or its leaf is burned due to excessive use of chemicals. The proposed
system also recommends medicine with the dosage whichcan be used to cure the plants.

5.2 FUTURE ENHANCEMENT

At present time the model can only identify limited number of diseases for the crop which further can be
enhanced.
Going forward and developing android application for the ML model can be of great use.
If seeing and looking in a broader perspective fully deploying the model in cloud environment can increase
the utilization of the proposed ML model.
We can also go forward in increasing the number of images hence enabling our model to predict results
with much more efficiency and making are model morecompatible to noise.
Also we can implement a feedback system for our ML model ex: the working will be if anyone gives the
feedback that the result was not correct we can follow up with the user and try to fed more plant images to
our model so that our model becomes capable to find the disease.

11

You might also like