Road network information extraction from high-resolution remote sensing images based on improved U-Net

In order to solve the problem of low accuracy of road network information extraction in high-resolution remote sensing images due to complex ground object environment, this paper proposes an improved deep learning semantic segmentation model CP-Unet. In this model, the CBAM full-connection layer module is used to enhance the feature fusion of the model. At the same time, the subpixel convolution up sampling module is introduced to reduce the loss of definition caused by the amplification of the dimension of the feature map in the up sampled convolution. Finally, the model is more suitable for road network extraction in high-resolution remote sensing images. In order to verify the reliability of CP-Unet model, an area of Xinjiang Road network was taken as the object of the experiment. The overall extraction accuracy index IoU score of the model in this paper is 81.73%, which is 6.66% higher than that of U-Net. It can better overcome the complex environmental interference and extract the road network in a more complete way. It provides method reference for road network information checking and updating.