Small Object Detection Algorithm for Road Images Based on Improved YOLOv11n

Accurate detection of small targets is crucial for improving traffic efficiency in road safety scenes. To address the issues of high missed detection rates and insufficient bounding box regression accuracy in existing detection models for small targets in road scenes, this paper proposes a small target detection algorithm based on improved YOLOv11n. A lightweight MLCA module is designed in the backbone network, combining local and global features to optimize small target feature extraction capabilities. The Wise-IoUv3 loss function with dynamic gradient allocation is adopted to improve the robustness of the bounding box regression. Experimental results on the KITTI autonomous driving dataset show that the improved model achieves an improvement of 3.3% in [email protected], with a minimal parameter increment of 10 and a computational overhead increase of merely 0.1 GFLOPs, significantly outperforming existing baseline models in detection accuracy. The proposed method demonstrates superior performance in small object detection under complex road scenarios while maintaining comparable parameter counts and computational costs.