A Polarimetric Multi-Modal Sensing Approach for Crack Detection in Underwater Infrastructure

Crack detection in underwater infrastructure serves as a crucial technology for safeguarding hydropower and natural gas systems, yet its reliability is severely compromised by challenging underwater optical conditions. This study tackles the image degradation issues caused by light attenuation, scattering effects, and suspended particles through a novel crack detection framework based on polarimetric multimodal sensing. First, we developed a polarization-sensitive array sensor capable of precise spatial and polarization information acquisition, supported by a multi-layered media imaging model. Subsequently, we constructed a specialized database containing diverse water turbidity levels and crack morphological patterns specific to underwater infrastructure. Leveraging this database, we designed a unsupervised multi-feature fusion crack detection network that integrates intensity, depth, and polarization characteristics, achieving robust performance in complex optical degradation scenarios. Experimental validation on the custom dataset demonstrated the method's superior detection accuracy, highlighting its potential for real-world applications in infrastructure maintenance.