A Road Extraction Method Based on Improved Residual Network

In this paper, we propose a method of remote sensing image road extraction based on improved residual network. We first introduce the road extraction network structure of improved residual network, deeply study and improve the residual-based network model, analyze the atrous spatial pyramid pooling to increase the sensitivity field of convolution, obtain multi-scale feature information fusion, and recover the original size of the image based on the upsampling method of sub-pixel convolution. Then, the method of enhancing high-resolution remote sensing image data based on random image clipping is studied to extend the training set and prevent overfitting. Then, a post-processing technology is introduced, which can extract the road edge automatically by extracting the road skeleton and restoring the road edge on the basis of road pixel segmentation. Finally, the feasibility of the proposed method is verified, and the experimental comparison and example analysis are carried out with many methods.