Convolution Neural Network Based Classification of Plant Leaf Disease Images

Abstract

In a country like India, agriculture has been the mainstay. In recent years, there has been a fall in the agricultural produce, mainly due to the lack of immunity in crops. The plant leaf diseases and destructive insects are major challenges to the economic growth of the agricultural sector. Early disease detection helps in enabling the control of plant diseases and improve the harvest rate tremendously. The existing methods are time-consuming and not efficient. In an effort to solve this issue, we develop an ideal system, in which plant leaf diseases are identified using the pre-trained VGG16 Convolution Neural Network model in deep learning. Our proposed work is also compared with detection and identification using Alexnet and resnet50. The network model is retrained by transfer learning approach in our work. A total of 1750 images, under seven different classes, is used for training and testing the system. These seven classes include five disease classes and two non-disease classes. Our system can effectively identify four diseases of the tomato plant and one disease of pepper bell plant. The five different diseases classified are pepper bell bacterial spot, mosaic virus, target spot, early blight, and tomato bacterial spot. Our proposed system achieves an accuracy of 97% on an average for all the seven classes compared to the other two Alexnet and resnet50.