Complex pavement distress detection using YOLOv11-EfficientRepBiPAN with cross-level structural feature fusion

Abstract

Pavement surface distress detection is an important task in road maintenance and intelligent infrastructure inspection. In practical vehicle-mounted inspection images, cracks and other distress targets often present weak edges, irregular shapes, large scale variations, and strong background interference, which makes stable recognition difficult. To address this problem, this paper proposes YOLOv11-EfficientRepBiPAN, an optimized object detection model enhanced by crosslevel structural feature fusion. The method takes YOLOv11 as the baseline detector and introduces EfficientRepBiPAN into the feature fusion stage. Shallow detail features, middle structural features, and deep semantic features are aligned and fused through a bidirectional progressive aggregation mechanism, so that crack edges, weak textures, boundary morphology, and semantic information can be jointly represented. Experimental results show that the proposed YOLOv11-EfficientRepBiPAN improves mAP@0.5, mAP@0.5:0.95, and F1-score by 6.22 percentage points, 4.34 percentage points, and 0.0520, respectively, compared with the baseline model, while maintaining a real-time inference speed of 202.5 FPS. The proposed method provides a feasible solution for vehicle-mounted pavement inspection and automatic distress recognition.