Improved YOLOv12s-based small defect detection algorithm for pavement in UAV aerial images

Abstract

In UAV aerial photography-based pavement defect detection, several challenges remain, including large variations in defect scales, complex background interference, difficult extraction of effective features, and the inability of existing methods to balance detection accuracy and real-time performance. To address these issues, this paper proposes an improved lightweight detection algorithm based on YOLOv12s.A hierarchical extended path aggregation network (HEPAN) is designed to enhance cross-scale feature fusion. A lightweight module named C2fDCB, which integrates depthwise separable convolution and reparameterization strategies, is developed to reduce model parameters while maintaining feature extraction capability. Frequency-Domain focused Downsampling Module (FD) is introduced to avoid the loss of detailed information. Furthermore, the MPDIoU loss function is adopted to optimize bounding box regression. Experimental results demonstrate that the proposed model achieves an mAP@0.5 of 47.8% on the VisDrone2019 and UAV-PDD2023 datasets, which is 6.3 percentage points higher than that of the baseline model. The number of parameters is reduced by 18.5%, and the detection speed reaches 128 FPS. The improved method effectively balances detection accuracy and inference speed, indicating that the proposed algorithm is suitable for pavement defect detection in UAV road inspection tasks.