Research on Identifying Cracks in Urban Sewers Based on GCANet-YOLOv11

Aiming at the problem of accuracy degradation due to fog interference and low-contrast features in urban underground pipeline crack detection, this study proposes the GCANet-YOLOv11 fusion model. In the image preprocessing stage, real-time defogging of the video stream is achieved by adding temporal convolution and inter-frame attention mechanism to the GCANet network; in the detection stage, deformable large kernel attention (DLKA), wavelet convolution, and spatial pooling modules are innovatively integrated into the YOLOv11 detector head to strengthen the ability of detecting small targets with irregular shapes. Experiments show that the model achieves 80.6% F1-Score and 89.7% [email protected] on the crack detection dataset, a 2.3<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">%</sup> improvement over the baseline model. The program combines both detection efficiency and engineering practicality, providing reliable technical support for intelligent inspection of urban underground facilities.