An Approach for Extracting Road Network from Remote Sensing Images

In recent years, the neural network architecture has developed rapidly and has been widely used in the semantic segmentation of remote sensing images. In this paper, we apply the neural network to the road network extraction of high-resolution remote sensing images. Subsequently, a series of single-pixel coordinate points are obtained by using the refinement method. Given the different lengths of road sections, our algorithm uses a double-loop mechanism to perform multi-scale fitting of them, which improves the rough results of the neural network and enhances the accuracy of road network extraction. For multiple line segments that may be on the same road, we also propose appropriate rules to classify and merge them. Our experimental results show that compared with other methods of road network extraction, our approach can obtain better results.