RDLK-YOLO enhanced pipeline defect detection in uneven illumination

Effective detection and timely treatment of drainage pipeline defects are crucial for maintaining underground pipeline systems and urban environments. Traditional image processing methods often struggle under complex lighting conditions. This study proposes a lighting compensation and feature enhancement algorithm to address feature degradation in uneven illumination, offering high-quality preprocessing for defect detection. We further develop an efficient object detection network, Rectangular Deformable Large Kernel-YOLO (RDLK-YOLO), based on YOLOv11. Key innovations include a Deformable Multi-scale Axial Calibration Module (DMACM), a Dynamic Large-Kernel Attention (DLKA) mechanism, Wavelet Transform Convolution (WT-Conv), and a hybrid loss function. Experimental results demonstrate that RDLK-YOLO achieves 94.7% precision, 88% recall, and 92.5% mAP@50, outperforming existing methods in accuracy and robustness under uneven illumination. This study highlights the practical utility of RDLK-YOLO in real-world pipeline defect detection scenarios. The code and dataset required to reproduce these findings are available in the following GitHub repository: https://github.com/aussup/RDLK-YOLO.git.