Cable Tunnel Waterlogging Detection for Low-Light and Interference Based on Three-Stage Recognition

Ensuring the stability of cable tunnels is crucial for power safety and urban reliability in light of the increasing demand for urban electricity. However, frequent waterlogging incidents within tunnels pose a significant threat to city safety by disrupting the power supply. Current methods have proven insufficient in timely and effective water detection within cable tunnels. While visual methods show promise, the complexity of cable tunnel environments degrades image quality under low-light conditions, intensifying the challenge of detection. To address these issues, this study proposes a novel hierarchical approach for waterlogging detection. The approach decomposes the problem into three subproblems: low-light image enhancement, image segmentation, and detection. Firstly, low-light image enhancement techniques are employed to improve image quality and enrich details for subsequent analysis. Next, image segmentation accurately delineates waterlogged road areas while mitigating interference from complex backgrounds. Finally, Faster R-CNN with ResNet as its backbone and integrated attention mechanisms enhances the model’s capacity to identify essential features amidst complex backgrounds, significantly improving accuracy in detecting waterlogged areas. Experimental results demonstrate our method’s superiority in accuracy and robustness for tunnel waterlogging detection tasks compared to traditional approaches. This proposed approach provides effective technical support for cable tunnel safety monitoring, contributing to ensuring urban power safety and reliability in the face of waterlogging challenges.