Abstract:With the continuous increase in the number of motor vehicles, road loads are intensified, resulting in increasingly severe road damage issues. Road damage detection is crucial for ensuring road safety and sustainability. However, UAV aerial imagery-based road damage detection still faces challenges such as difficulty in detecting small targets, complex background interference, and diverse damage morphologies, which significantly limit detection performance. To address these issues, this study proposes an improved UAV aerial road damage detection algorithm based on YOLO11n, termed YOLO11-DPSS. A dynamic snake convolution (DSConv) module is introduced to accurately capture slender and curved topological features of road damage, such as cracks and potholes, by exploiting adaptive deformation capability. A high-resolution P2 tiny object detection layer is constructed to alleviate feature loss of extremely small damage under high-altitude aerial views. In addition, the Slim-neck architecture is adopted to enhance feature fusion efficiency, preserving edge details while reducing computational complexity. Furthermore, the Shape-IoU loss function is employed to incorporate shape and scale factors into bounding box regression, thereby improving geometric perception and localization accuracy for road damage with extreme aspect ratios. The experimental results show that YOLO11-DPSS achieves a 4.4% improvement in mAP@0.5 and a 2.5% improvement in mAP@0.5:0.95 on the UAV-PDD2023 dataset, along with 3% higher precision and 7.8% higher recall, compared with YOLO11n. Meanwhile, the parameter count (Params) and computational complexity (GFLOPS) remain unchanged, indicating that improved detection accuracy is achieved while maintaining low computational overhead, which demonstrates strong overall detection performance.