HPSM‐YOLO: A Multi‐Scale Feature Aggregation Network for Pavement Distress Detection

In order to overcome the difficulties of multi‐scale target recognition, background complexity, and varying environmental conditions in pavement damage identification using street‐view imagery, we introduce HPSM‐YOLO, a multi‐scale enhanced detection model built upon YOLO11. First, we introduce the HFA module and optimize the C3k2 module to facilitate the detection performance and improve the accuracy of localizing small‐scale pavement damage. Second, the detection structure is augmented by incorporating the P6 layer, which improves the model's ability to identify large‐scale damage. In addition, we propose the SDAF module to achieve efficient cross‐scale feature fusion. As a complementary strategy, the MPDIoU loss function is employed to constrain and optimize bounding box regression accuracy. The experiments conducted on the SVRDD dataset indicate that HPSM‐YOLO attains mAP50 and mAP50:95 values of 0.785 and 0.511, which correspond to performance gains of 3.7% and 3.8% compared with the baseline YOLO11s model. The proposed a practical accuracy‐optimized solution for intelligent pavement maintenance and holds significant practical value.