Research on Road Defect Detection Technology Based on Improved YOLOv11n

To address the poor performance of current detection models in complex road defect detection tasks, particularly the significant shortcomings in small-object detection and the frequent occurrence of missed and false detections, this paper proposes an improved YOLOv11n model for road defect detection. First, the C3k2 module in the backbone is enhanced using the Frequency-Aware Cascade Attention mechanism, resulting in the design of the C3k2-FCA-CTA module, which improves the model's feature extraction ability for small targets. Second, the model's multi-scale feature fusion capability is improved by substituting the MAFPN for the original network in the neck. Finally, the inner-MPDIoU loss function is adopted to optimize the algorithm, thereby improving the model's generalization performance. The experimental findings indicate that the enhanced version of YOLOv11n exhibits notable improvements in detection accuracy, with precision, recall, and mAP50 increasing by 1.2%, 2.4%, and 1.2% respectively when compared to the original model. These findings confirm the success of the proposed modifications.