Improved model for identifying rice panicle disease based on MobileNetV2

Rice plays a crucial role in agriculture, but a major issue is that various disasters and diseases of rice will greatly reduce rice production, especially affecting rice required for human consumption, and the rice seeds sown in the next year will also encounter problems. The learning nature of convolutional neural networks is used to identify rice ear diseases, and rice ear disease recognition rate is improved by modifying the network structure and integrating other network structures that can enhance deep learning for picture recognition. In this study, MobileNetV2 is used as the main network and trained on ImageNet using migration learning. The underlying convolutional layer weights are frozen to conserve resources. Then, the pre-trained MobileNetV2 network is fused with BAM blocks to develop a new network. Experiments show that the efficiency and recognition rate of this method are improved, with an average recognition rate of 98.18%. The generalisation ability of the model is then tested on the PlantVillage data set, with an average recognition rate of 98.7%. The results of the experiments show that the model can effectively improve image recognition, and the generalisability of the model is also guaranteed.