Construction defect detection of intermediate joints of urban underground cables based on improved YOLO11

Aiming at the problems of insufficient feature extraction of small targets in the construction defect detection of urban underground cable intermediate joints, a detection algorithm based on improved YOLO11 is proposed. A panoramic imaging framework is implemented through multi-camera collaborative acquisition and SIFT-based image stitching, effectively resolving the issue of defect omission in single-view imaging. Key enhancements to YOLO11 include the integration of deformable convolution (DCNv2) to improve geometric adaptability for modeling misaligned semiconductive layer stripping defects, the incorporation of large kernel attention (LSKA) to strengthen global contextual awareness of construction anomalies, and the addition of a P2 small-target detection layer to refine localization accuracy for main insulation contamination, scratches, and burrs on compression sleeves. Experimental results demonstrate that the proposed algorithm achieves a detection accuracy of 80.3% and an mAP@0.5 of 70.3% for four typical defect categories, representing improvements of 4.1% and 16.6%, respectively, over the baseline YOLO11s. The algorithm outperforms mainstream lightweight models such as YOLOv8s and YOLOv10n, providing a high-precision automated solution for quality inspection of urban underground cable joints.