多尺度融合的轻量化交通标志检测算法OA

To address the challenges in traffic sign detection for autonomous driving,such as small target sizes,large scale variations,low detection accuracy,excessive parameters,and limited deployability,a lightweight traffic sign detection algorithm with multi-scale feature fusion is proposed based on YOLOv8n.The algorithm constructs a lightweight feature extraction module using partial convolution and gated linear units to enable dynamic feature selection.A multi-scale feature fusion module is designed by incorporating dynamic sampling and a channel attention mechanism to effectively fuse features at different levels.In addition,a lightweight downsampling module is built using group convolution and local attention,achieving a balance between computational efficiency and detection accuracy.Experiments on the TT100K traffic sign detection dataset show that,compared with the baseline algorithm,the proposed method improves precision and mean aver-age precision by 2.1%and 3%,respectively,while reducing the number of model parameters by 50%.On the CCTS-DB2021 traffic sign detection dataset,the proposed algorithm achieves improvements of 2.1%in precision and 1.3%in mean average precision over the baseline,demonstrating the effectiveness of the proposed approach.