Deep residual neural networks with feature recalibration for crop image disease recognition

Nowadays, half of the global population depends on crop as their staple food, and crop yield is related to the food security of all mankind. crop pests and diseases are important factors affecting crop yield, and it is important to effectively prevent and efficiently detect crop pests and diseases. crop pests and diseases are diverse, previously mainly relying on manual experience to distinguish pests and disease types. With the extremely rapid development of deep learning technology, it is possible to recognise crop pests and diseases by technical methods. The diversity of pests and diseases makes higher requirements on the generalisation ability of the recognition model. In this paper, FRseNet is constructed based on ResNet-50 by introducing the SENet concept, and the experiments on the self-constructed crop pest and disease dataset show that it is capable of the task of recognising 15 kinds of diseases and 21 medium insect pests, and the performance is competitive.