Cross-Scale Coupled Semantic Segmentation Network for Crack Detection

Crack segmentation plays a vital role in guiding the maintenance of concrete damage. While existing semantic segmentation algorithms perform well on natural image datasets, they often lack edge sensitivity in the fine-grained representation of cracks. To address this limitation, this paper proposes a Cross-Scale Coupled Semantic Segmentation Network for crack detection. The network captures boundary information across multiple scales through a multi-path architecture and employs low-rank coupling at low resolutions to facilitate cross-scale semantic interaction and integration. Experimental results on the CFD dataset demonstrate the effectiveness and robustness of the proposed method.