关键词: 家蚕, 病害识别, 卷积神经网络, 深度学习

Abstract:  Diseases are one of the main threats facing the healthy development of the sericulture industry in China. In order to study methods of prevention and control of silkworm disease under mechanized sericulture mode, a convolutional neural network is used to carry out recognition of silkworm disease images in this paper. First of all, by breading and feeding pathogens, healthy samples and samples of silkworm species Fang·Xiu·Bai·Chun with nuclear polyhedrosis disease, nosema bombycis, beauveria bassiana, bacterial disease, and pesticide poisoning in the stage of larva were obtained. Images collection was conducted to construct a silkworm disease image dataset. Secondly, the method of feature fusion and structure reduction was adopted to modify the residual neural network to avoid unnecessary calculation loss caused by the direct use of the algorithm. Finally, a silkworm disease recognition test was performed. The results show that the convolutional neural network can efficiently and accurately recognize silkworm disease images. The accuracy rate of the modified algorithm on the test set reaches 94.31%, which is equivalent to the standard residual neural network accuracy rate. However, the amount of training parameters is only 1/3 of the original network, which is more conducive to training and deployment of the network, which has a wide range of application prospects.

Key words: silkworm, diseases control;Convolution Neural Network, deep learning