Crack segmentation network via difference convolution-based encoder and hybrid CNN-Mamba multi-scale attention

Jianming Zhang, Shigen Zhang, Dianwen Li, Jianxin Wang, Jin Wang

Abstract: Highlights•Adopt or design five difference convolution operators from the perspectives of Cartesian and polar coordinate system.•Design an enhanced convolution module based on the five difference convolution operators to extract crack edges.•Design a mixed convolution and Mamba attention module to enhance spatial and global feature extraction.•Design a feature fusion module to integrate and restore crack details layer by layer.