YOLOv11n-CDL: accurate and lightweight pavement defect detection via enhanced multi-scale attention and feature fusion

Pavement defect detection requires both high accuracy and real-time performance in complex road environments, yet existing lightweight models often struggle with blurred textures, background interference, and small cracks. To address these limitations, this study proposes YOLOv11n-CDL, an enhanced lightweight detector integrating three targeted improvements. First, the ConvSmart module expands the receptive field and strengthens multi-scale feature extraction, improving the representation of defects of varying sizes. Second, a Double-Stage Attention (DSA) mechanism, embedded at the deepest backbone stage, iteratively highlights discriminative crack patterns while suppressing shadows, markings, and texture noise. Third, a P2-level small-object detection path provides high-resolution features that significantly improve sensitivity to fine cracks and micro-potholes. Experiments on IRRDD show that YOLOv11n-CDL achieves 75.3% mAP@0.5 and 44.6% mAP@0.5:0.95, outperforming the baseline by 3.0 and 1.1 percentage points, and exceeding YOLOv8n and YOLOv7-tiny in both precision and recall. Additional results on RDD2022 and low-power devices confirm strong generalization and real-time deployability. These improvements demonstrate that YOLOv11n-CDL offers an effective balance between accuracy, robustness, and efficiency for practical pavement inspection applications.