YOLO-RDD: An Efficient Lightweight Model for Road Defect Detection

Detecting road defects is crucial for traffic safety and infrastructure maintenance, yet existing methods are often constrained by poor generalization and high computational costs, which hinder practical deployment. To address these challenges, this study proposes YOLO-RDD, a lightweight detector derived from YOLOv11. The model incorporates a WaveletPool module for high-frequency edge preservation, a C3k2-CF-CGLU module for global-local feature interaction, a context-aware anchor attention pyramid for precise multi-scale fusion, and a lightweight decoupled detection head for reduced computational redundancy. Experiments on a dual-view dataset demonstrate that YOLO-RDD achieves 83.0% mAP@0.5, improving by 2.0 percentage points over the baseline, while reducing parameters by 48.4%, FLOPs by 14.2%, and model size by 39.6%. The study verifies the model’s ability to sustain high detection precision while reducing computational cost, and further confirms that YOLO-RDD maintains this balance consistently. Its compact and efficient design ensures strong adaptability, making it suitable for both real-time applications and edge-driven road maintenance tasks.