A road damage detection model based on improved YOLOv11s

Abstract Rapid and accurate road damage detection is a critical task for road maintenance departments. Existing detection methods primarily adopt semi-automated approaches that rely on human intervention, leaving room for improvement in both efficiency and accuracy. This study utilizes vehicle-mounted cameras for pavement data collection and achieves automated detection of road damage through an improved YOLOv11s model. Key model improvements include: Firstly, replacing the SPPF module with SPPF_LSKA to enhance recognition capability in complex scenarios; Secondly, designing the M-Head-T4 structure to significantly boost detection accuracy; Thirdly, adopting the Focal-DIoU loss function to optimize the training process; Finally, conducting comparative experiments across multiple public road damage datasets. Experimental results demonstrate that the improved model achieves a 5.2% improvement in [email protected] and a 4.7% increase in F1-score compared to the original YOLOv11s baseline on the SVRDD dataset.