WHF‐YOLOv11: Traffic Cone Detection Algorithm in Complex Scenes Based on Wavelet Convolution and Hierarchical Feature Attention
Ming Zhao, Yange Chen, Baohua Guo, Xiaoyu Zhang, Weifan Gu · Journal of Advanced Transportation · 2025
Aiming to address the issues of missed detection and low accuracy in detecting small, dense traffic cones in complex traffic scenarios, this study proposes an improved traffic small target detection method based on YOLOv11, referred to as WHF‐YOLOv11. In terms of network structure, large receptive field wavelet convolution is introduced into the YOLOv11 network (WTConv). By decomposing and reconstructing the image at different scales, the model can capture the local and global features of the image more accurately, so as to effectively extract the key details such as texture and edge. At the level of feature map extraction, the hierarchical multiscale feature fusion network (HiFuse) was introduced into the neck network. By utilizing a three‐branch HiFuse, the local perception capabilities of CNNs and the global modeling strengths of transformers were combined to enhance the image classification accuracy. In terms of loss function, Focaler‐IoU is used as the bounding box loss function to improve the detection ability of small target difficult cases. The experimental results show that for the traffic cone dataset obtained on the Roboflow platform, compared to the benchmark model YOLOv11 n, the improved model improves the accuracy rate P and recall rate R by 2.8% and 1.7%, respectively, and mAP50 and mAP50∼95 by 1.6% and 3.3%, which verifies the effectiveness of the model and provides technical support for the intelligent detection of small targets in traffic scenes.