An attention-based lightweight residual network for plant disease recognition

Aiming at the problems of large computational resource and low precision in most Convolutional Neural Network models, an attention-based lightweight residual network is proposed for plant disease recognition in this paper. In order to reduce model parameters without degrading network performance, the proposed network model uses depthwise separable convolution to replace the conventional convolution on the basis of traditional residual neural network. At the same time, the attention module is introduced, which can effectively prevent the overfitting problem of the network and enrich local feature learning. To verify the effectiveness of the proposed model for plant disease recognition, the proposed model is compared with several commonly deep neural network models. The results show that the proposed model can identify plant diseases well, which achieves a mean accuracy of 98.89% on the public datasets, and has the size of only 18.9MB. So the proposed model will help to achieve real-time diagnosis of plant diseases on low-performance terminals in the future.