Research on road extraction network of high-resolution remote sensing images improved with attention mechanism

On the basis of the existing road extraction methods and the characteristics of high-resolution remote sensing images, this paper proposes a multispectral-guided hybrid attention mechanism road extraction model (MS-DANET), which introduces a parallel hybrid attention mechanism to process the dual-branch image features between the encoder and decoder structure. It is shown from experimental results that MS-DANET outperforms current road extraction networks and exhibits excellent results with higher road extraction accuracy and better integrity.