Semantic–Spatial Feature Refinement Network for Road Extraction From Remote Sensing Images

Extracting precise road information from remote sensing images remains challenging due to the interference from similar objects and occlusion from surroundings. To alleviate these issues, we propose a novel road extraction network to enhance both the precision and topological connectivity of extracted road networks, dubbed as CRNet. Specifically, a Global-Local Context Decoupling Module (GLCDM) is introduced to explicitly model long-range contextual dependencies while preserving fine-grained local road features, thereby improving the model’s inference capability in occluded regions. Furthermore, a Semantic-Spatial Feature Refinement Module (SSFRM) is integrated into the skip connections, which leverages deep semantic features to guide the suppression of background noise in shallow feature maps across both channel and spatial dimensions, ensuring the decoder receives structurally accurate road representations. Experimental results on the remote sensing road datasets demonstrate that CRNet achieves state-of-the-art performance in terms of both segmentation accuracy and road connectivity. The source code is publicly available at https://github.com/CVer- Yang/CRNet.