This paper proposes a deep residual U-Net (ResUnet) for road extraction from high-resolution remote sensing images. The proposed network combines the strengths of residual learning and U-Net architecture. The network uses residual units to ease training and facilitate information propagation through skip connections, allowing for better performance with fewer parameters. The ResUnet is tested on the Massachusetts roads dataset and outperforms existing methods such as U-Net, Mnih's method, and Saito's method in terms of relaxed precision and recall. The network achieves better results in road segmentation, especially in complex scenarios like multi-lane roads and intersections. It also demonstrates robustness to occlusions by considering contextual information, which helps distinguish roads from similar objects like building roofs and airfield runways. The proposed network has a smaller parameter count (7.8M) compared to U-Net (30.6M) and achieves superior performance. The results show that the ResUnet is effective in road extraction, with clean and accurate segmentation, and is capable of handling challenging scenarios in remote sensing image analysis.This paper proposes a deep residual U-Net (ResUnet) for road extraction from high-resolution remote sensing images. The proposed network combines the strengths of residual learning and U-Net architecture. The network uses residual units to ease training and facilitate information propagation through skip connections, allowing for better performance with fewer parameters. The ResUnet is tested on the Massachusetts roads dataset and outperforms existing methods such as U-Net, Mnih's method, and Saito's method in terms of relaxed precision and recall. The network achieves better results in road segmentation, especially in complex scenarios like multi-lane roads and intersections. It also demonstrates robustness to occlusions by considering contextual information, which helps distinguish roads from similar objects like building roofs and airfield runways. The proposed network has a smaller parameter count (7.8M) compared to U-Net (30.6M) and achieves superior performance. The results show that the ResUnet is effective in road extraction, with clean and accurate segmentation, and is capable of handling challenging scenarios in remote sensing image analysis.