Crop-Disease-Detection

A Deep Learning Project

Introduction

• Agricultural production rate plays a vital role in the economic development of a country.
• The identification of plant diseases at an early stage is crucial for global health and wellbeing. So, controlling on the diseased leaves during the growing stages of crops is a crucial step.
• Moreover, increasing crop production is essential to areas where food is scarce.
• Loss of crops from plant diseases would result in reduction of income for crop producers, higher prices for consumers and significant economic impact.
• The access to disease-control methods is limited which results in annual losses of 30 to 50 percent for major crops in various countries.
• Hence, detection of crop diseases is very crucial for economic development.

Problem Motivation

• Agriculture is a very important sector of the Indian economy.
• The share of agriculture in GDP increased to 20.2 per cent in 2020-21 from 18.4 per cent in 2019-20.
• So, for the identification and detection of plant diseases, human raters are employed.
• This process is very time consuming and expensive and sometimes may lead to poorly identify the crop disease.
• Thus, continuous monitoring must be done which is a repetitive process which involves large group of experts costing very high when dealing with large farms.
• Therefore, this motivated us to automate the process and perform evaluation metrics, based on a deep learning classifier.

Problem Statement

• Detect the crop diseases.
• Differentiate between various crop diseases for a particular plant and then for many various plants.
• Using object detection algorithms to find the diseased area in the crop on the basis of the features such as color, wilt, leaf spots, unusual size of the leaf.
• Choosing appropriate neural network for classification.

Objective

1. Our Main Objective is that given the images of a specific crop, it should classify it as a healthy crop or the disease it may be infected with.
2. We intend to use deep convolutional neural networks (D-CNN) and some concepts of image processing to reach our goal of identifying the disease of a plant using color, leaf spots, etc.
3. We developed a web based application so that it can be used widely by large number of people to determine crop diseases.

For more details about the project, please refer to "Crop Disease Detection PPT.pdf".

Dataset Description

Old Dataset

The dataset has different types of diseases for tomato leaves. There are 10 classes present which is shown below.

• Here goes the list:

  • Tomatomosaicvirus
  • Target_Spot
  • Bacterial_spot
  • TomatoYellowLeafCurlVirus
  • Late_blight
  • Leaf_Mold
  • Early_blight
  • Spidermites Two-spottedspider_mite
  • Tomato___healthy
  • Septorialeafspot

• The total number of images present are :

  • 10000 images for training data
  • 1000 images for validation data

• The data set contains the top 10 classes of diseases which are highly occured on tomato plant.

• The images had taken with different angles, with different backgrounds, and in different lighting conditions with an image size is 255 X 255 pixels.

For more details about the dataset, please refer the "Old Dataset".

New Augmented Dataset

• Now, the datatset is augmented and the total images for training data are 18,345 images and for testing data are 4,585 images.

• Image Size 255 X 255

• A combination of picture flipping, rotation, blur, relighting, and random cropping, image augmentation artificially builds training images. In this, we scale, shear, zoom, and horizontally flip the photos as part of the image augmentation process.

• The images were stored in the RGB image by

For more details about the dataset, please refer the "New Augmented Dataset".

Models Implemented

We have implemented 5 models successfully on tomato plant.

1. CNN (with 3 different variants)
2. INCEPTION-v3
3. VGG-16
4. VGG-16-Fine-Tuning
5. VGG-19

For more details, about the models and their code, please refer to "Codes".

Moreover, to get an overview of the entire project, please go through the "Crop Disease Detection Report.pdf".

Group Members

Anirudh Jakhotia  - S20190010007
Khushi Pathak     - S20190010091
V.Naveen Kumar    - S20190010192