基于自适应边缘引导网络的遥感影像道路分割OA

Aiming at the problems of spatial feature and edge detail information loss,weak anti-interference ability in road segmentation algorithms in remote sensing images,a remote sensing image road semantic segmentation method based on adaptive edge guidance network is proposed.This method is based on an improved U-net network framework,combined with edge detection attention mechanism,target boundaries are used as assistance to guide the network model to automatically adjust learning parame-ters and adaptively capture effective feature information of interest.At the same time,multi-scale feature fusion technology is adopt-ed to aggregate information from multiple kernels to achieve adaptive receptive field size for neurons,and to adaptively screen useful feature information under different receptive fields from both channel and spatial dimensions,highlighting road edge features and improving the accuracy of road classification.This method is tested on the publicly available dataset of Deep Globe Road Dataset,and the Dice segmentation coefficient is 0.867,Jaccard segmentation coefficient is 0.765,accuracy is 97.26%,precision is 86.73%,recall is 79.23%,and average IoU is 78.15%.The experimental results show that this method can effectively improve the accuracy of road segmentation in remote sensing images.