Research on Integrated Algorithm Based on Convolutional Neural Network for Rice Disease Identification

Abstract

Rice is the staple food in many Asian countries. The effective intelligent identification of rice diseases is helpful to improve the industrialized management of rice, reduce the input of manpower and improve the efficiency. In 2018, China’s first intelligent identification and service system for rice pests and diseases was released, but its average identification rate is only 80%. This paper takes 4 diseases of rice as the research object, adopts supervised learning, uses convolutional neural network to extract the characteristics of each disease, and integrates the extracted characteristics with AdaBoost algorithm to improve the accuracy. According to the research quantity and the characteristics of the research object, a 10-layer convolutional neural network and three convolution kernels are designed to miss the important information of the image as much as possible. According to the method in this paper, the correct recognition rate of the four diseases of Bacterial Blight, Flax Spot, Sheath Blight and Leaf Sheath Spot was 95.64% on average, which is higher than 5.56% and 14.69% respectively when the convolution neural network and AdaBoost algorithm are stable. Therefore, compared with the classical convolutional neural network and AdaBoost, the integrated algorithm of convolutional neural network proposed in this paper has a higher accuracy in rice disease research.