Plant pest and Disease Identification Based on Multiscale Dual Attention Network
Plant pest and disease problems are a major challenge in agriculture,and accurate identification of plant pests and diseases is a key step in the prevention and management of agricultural pests and diseases.Experi-enced plant pathologists make diagnoses by observing leaf states,which is not only time-consuming and labor-inten-sive,but also requires a significant cost for farmers to contact experts.Therefore,this paper designs an efficient Multi-scale Dual Attention Network(MDANet)plant pest identification method based on the ResNet model.First,sub-fea-ture maps at different scales were obtained by multiscale convolution,and then,the important features of input leaves were weighted using spatial attention and channel attention.The important global features and local features in the leaf images are extracted in depth for fast and accurate plant disease recognition.The experimental results show that MDA-Net obtained 90.2%accuracy in the plant disease dataset of AI Challenge 2018,which has a significant advantage o-ver other convolutional neural network models.
NETL
NSTL
万方数据
