Rice disease identification and classification based on matlab and android platform

Abstract

To achieve rapid and portable automatic identification and classification of rice diseases, this paper proposes texture feature value comparison, the multiple linear regression equation, the BP (back propagation) neural network three-step recognition method using the log-sigmoid function, and the convolutional neural network (CNN) comprehensive recognition method. First, the background, the leaves, and the lesion areas are thresholded according to the color histogram. After morphological processing, the texture, morphological, and color features are extracted and used as the input parameters of the three-step recognition model. The third step constructs a 3 × 6 × 2 three-layer BP neural network prediction model with the tan-sigmoid function as the hidden layer neuron activation function to identify the remaining similar diseases and finally transplants the algorithm to the mobile phone Android platform. The comparative analysis of the visual experiments reveals that the proposed method has an average recognition accuracy of 95.5% for the five types of diseases, an average recognition accuracy increase of 3.1%, and an average classification accuracy of 88.9% compared with direct classification using only CNNs. This research on rice disease identification can provide reference on multiple types of disease identification and disease degree classification.