Road extraction from remote sensing images based on residual dense blocks and mixed attention mechanism

Abstract

In order to solve the problem of road edge blur in remote sensing image road extraction, a road edge feature information enhancement algorithm based on residual dense blocks and mixed attention mechanism is designed in this paper. At present, semantic segmentation algorithm based on deep learning has been able to extract relatively complete and accurate road network, but the segmentation of road edge is not accurate enough, and the extraction of road edge details is often incomplete. To this end, the ADRU-net road extraction network is designed in this paper, and experimental verification is conducted on data set Zimbabwe_Data_Roads.