Due to low accuracy of current crop drought detection methods based on remote sensing images,this paper proposed an image detection method based on the coding-decoding neural network.In this meth-od,the deep residual neural network was used as the main network for feature extraction,and multi-scale attention pooling and multi-scale atrous convolution techniques were combined to reduce information loss and enhance effects of feature extraction.By effectively fusing high-level and low-level feature informa-tion,the recognition effect of crop drought boundary improved.The experimental results showed that this method achieved pixel accuracy of 91.05％and average pixel accuracy of 76.19％in drought detection based on remote sensing images,which was obviously superior to other existing models.
维普
万方数据