Road Extraction by Deep Residual U-Net

Road extraction from aerial images has been a hot research topic in the field remote sensing image analysis. In this letter, a semantic segmentation network which combines the strengths of residual learning and U-Net is for road area extraction. The network is built with residual units and similar architecture to that of U-Net. The benefits of this model is-fold: first, residual units ease training of deep networks. Second, the skip connections within the network could facilitate information, allowing us to design networks with fewer parameters however performance. We test our network on a public road dataset and compare it U-Net and other two state of the art deep learning based road extraction. The proposed approach outperforms all the comparing methods, which its superiority over recently developed state of the arts.