Lightweight YOLO-RDD pavement distress detection algorithm

In response to the cumulative increase in pavement distresses induced by prolonged road service life, as well as the prevalent issues of missed detections and false positives concerning subtle defects (e.g., fine cracks) amid complex pavement backgrounds, this paper proposes an improved lightweight object detection algorithm, YOLO-RDD, based on YOLOv11n.Within the backbone network of YOLOv11n, the RepGELAN module—incorporating structural reparameterization—replaces the conventional C3k2 module. This modification enhances feature extraction capability, strengthens the representation of low-contrast distresses such as fine cracks, and suppresses interference from irrelevant background information.Furthermore, an optimized DySlim-Neck lightweight neck structure is designed by integrating the DSU dynamic alignment module and depthwise separable convolution. This design not only reduces the computational overhead of feature fusion but also reinforces the propagation of small-target information.The adoption of the DynamicHead dynamic detection head mitigates missed detections caused by insufficient coupling between classification and regression tasks for subtle distress targets. The improved YOLOv11n algorithm achieves significant enhancements in detecting pavement cracks and potholes. Quantitative evaluations using metrics including F1-score, mean average precision (mAP), and model parameters demonstrate that: compared to the baseline, the proposed algorithm reduces model parameters by 19.7% while increasing the F1-score and mAP by 1.0 and 2.3 percentage points, respectively.When compared with the benchmark YOLOv11n and other state-of-the-art object detection models, the proposed YOLO-RDD algorithm exhibits superior comprehensive performance, offering an efficient solution for pavement distress detection.