Abstract: Due to the complexity of road structure in remote sensing satellite images, it is difficult to extract small road information. To solve this problem, a residual segmentation network based on multi-scale feature extraction is proposed. The method used Resnet34 as the encoder of the network to ensure the depth of the network and the robustness of the neural network. The multi-scale feature extraction structure of ASPP was used to further extract semantic features, which improved the ability of the network to capture small targets. The decoder structure of Unet was used to ensure the integrity of the semantic segmentation task on the input and output scales. The method was verified on the dataset of CVPR DeepGlobe 2018 Road Extraction Challenge, and the three indexes of mIoU, dice similarity coefficient and recall rate reached 69.76%, 81.60% and 80.25%, respectively, which all exceeded DLinknet34, the champion of the challenge, and improved the effect of road extraction.