Abstract
Road damage detection is crucial for ensuring road safety and minimizing maintenance costs. However, detecting small damage, managing complex backgrounds, and identifying irregular damage shapes remain significant challenges. To address these issues, we propose YOLO-RD, an advanced detection framework that integrates innovative modules for feature enhancement, multi-scale robustness, and detail preservation. Specifically, the Star Operation Module (SOM) improves sensitivity to small-scale damage, the Multi-dimensional Auxiliary Fusion (MAF) module strengthens robustness in complex environments, and the Wavelet Transform Convolution (WTC) enables adaptive focus on irregular shapes. On the Japanese road dataset in RDD2022, YOLO-RD achieves a detection accuracy of 25.75%, with a notable 4.93% improvement in small object detection over the baseline YOLOv8. These results demonstrate the effectiveness and practicality of YOLO-RD in addressing diverse and challenging real-world scenarios, establishing it as a robust solution for automated road condition monitoring.