Rice plant diseases detection using convolutional neural networks

Abstract

Rice is one of the main crops grown in India and it is complicated for farmers to accurately classify rice diseases manually with their imperfect information. Thus, the automatic recognition of rice plant diseases is highly desired. Many methods are available and have been proposed for the rice plant diseases detection. The latest advances indicate that the use of CNN models can be very beneficial in such troubles. In this paper we have explored and trained various CNN models with the unique combinations of training and learning methods to enhance the accuracy. The most advanced large-scale architecture, such as VGG19, XceptionNet, ResNet50, DenseNet, SqueezeNet, and CNN are implemented with the baseline and transfer learning methods. These models are trained and tested on datasets collected from various sources. Experimental results show that the ResNet50 architecture achieved the highest accuracy of 97.5% as compared to other CNN architectures and existing literature.