A novel capsule neural network for identification of rice leaf disease

Abstract

A novel network model (RESCapsNet) for identification of rice leaf diseases was proposed by combining the traditional convolutional neural network and capsule networks. The results show the identification accuracy and F1_score of the proposed RESCapsNet is 99.63% and 99.6% on the public dataset from Kaggle, respectively. The accuracy of RESCapsNet is improved by 12.54% compared to the capsule network (CapsNet), although CapsNet only required 110 epochs to reach its optimal state, RESCapsNet required training 140 epochs to reach its optimal state. This method is effective in identifying and classifying rice leaf diseases, as shown by the experimental results, and can achieve early defense against rice leaf diseases.