Research on Road Defect Detection Based on Improved YOLO11n

Accurate identification of road defects is of great significance for maintaining the safety of the transportation system, extending the lifespan of roads, and reducing costs. To improve the accuracy of road defect detection, reduce false positives and false negatives, and lower the computational complexity of algorithms, an algorithm based on the optimized and improved YOLOv11n architecture is proposed. The algorithm introduces deep convolutional techniques to restructure convolutional operations, reducing the number of model parameters, lowering costs and hardware burdens, enhancing generalization and robustness, and accelerating inference. A C3K2-DCNV2-Dynamic module is designed to integrate deformable and dynamic convolutions, enabling adaptive focus on key areas and improving the accuracy of feature extraction. The neck network architecture is restructured with the EMBSFPN, incorporating channel shuffling and multi-scale feature fusion strategies to optimize computational efficiency and enhance the ability to capture multi-scale defect features. Experiments show that the improved model's precision, recall, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\text{mAP} {@} 0.5$</tex> increased by <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$6.2 \%, 1.7 \%$</tex>, and 5.9 %, respectively. The number of parameters is reduced, and detection accuracy is significantly improved. The model has important application value and further meets the high-precision recognition requirements of road defect detection tasks.