Plant Disease Detection Using Deep Learning

Abstract

Crop diseases are one of the major issues, but their identification is difficult due to the lack of required infrastructure. Plant diseases affect farmers whose livelihood depends on the crops and also it increases the vulnerability of food security at the large scale. Plant disease identification is very important because it affects the growth of the plant species. Usage of pesticides reduces the ability to fight back. The plant diseases are detected using Deep Convolutional Neural Network trained and added into the database of leaves from different plants. The dataset consists of 38 disease classes and one background class stanford open dataset [1]. The proposed model predicts from the image of a leaf if it is diseased or not and also gives the name of the disease predicted. In Deep learning, CNN is an algorithm that takes an image as an input and it assigns importance to the objects in the image and we can distinguish one image from the other. Architecture used is resnet architecture. We have many types of resnet architecture i.e. concept is same but with different number of layers, for example we have ResNet-34, ResNet-50, ResNet-18, ResNet-101 etc. Here, In this paper we are using ResNet-50 architecture, since it is a variant of Resnet model that has 48 convolution layers along with one average pool and one max pool layer. The results are deployed into the cloud (AWS) where the data can be fetched whenever required. Accuracy of the system is around 97–98%.