Abstract
Pavement defects pose serious threats to traffic safety, pavement durability, and operational efficiency. To achieve accurate and real-time identification of pavement defects, this study proposes an enhanced lightweight model, YOLO11-WLBS, which integrates four improved modules-Wavelet Transform Convolution, Lightweight Adaptive Extraction, Bidirectional Feature Pyramid Network, and Simple Attention-into the YOLO11 framework. Each module's contribution is verified through ablation experiments. The proposed model achieves a precision of 0.947, recall of 0.895, F1-score of 0.895, mAP@0.5 of 0.944, and mAP@0.5-0.95 of 0.703, demonstrating high accuracy and efficiency. Compared with the baseline YOLO11, YOLO11-WLBS improves precision by 6.4%, recall by 15.8%, and mAP@0.5 by 12.2%, while reducing parameters by 25.5%. The model maintains excellent detection performance under extreme lighting and blurring conditions and exhibits strong generalization in cross-dataset applications. These results indicate that YOLO11-WLBS provides an efficient and robust solution for intelligent pavement defect detection and offers practical potential for real-time deployment on edge devices in pavement maintenance and infrastructure monitoring systems.