Multi-scale Parallel Fusion Convolution Network for Crop Disease Identification

In order to identify crop disease accurately and efficiently, we propose a multi-scale parallel fusion convolution network called MSPFNet for the automatic crop disease identification. It combines multi-scale convolution and attention mechanism. By mixing convolution kernels of different sizes in the same convolution unit, multi-scale convolution can identify crop disease features at different scales better. In this paper, Convolutional Block Attention Module is introduced to improve the representation ability of the model. By analyzing the information of spatial attention and channel attention, model can pay more attention to important features. In addition, we design a feature fusion module to realize the fusion of features at different depths. And feature fusion module can solve the problem of the disappearance of disease features in the process of image feature extraction. We validate the proposed model on the publicly available apple leaf pest data set. Experimental results show that the proposed MSPFNet can achieve higher recognition accuracy.