河北省科学院学报

针对复杂环境下交通目标漏检率高的问题，提出一种融合注意力机制的实时目标检测方法。首先，构建以YOLOv11架构为核心的基础检测网络；其次，将SimAM注意力模块集成于YOLOv11网络，建立跨层注意力融合机制，提取关键特征信息，增强模型对多尺度目标的表征能力。实验结果表明，融合注意力机制的YOLOv11模型在自建交通数据集上mAP指标达到91.38%，较基准模型提高了2.8%，在计算效率与检测精度之间表现出优异的平衡性；而且在公开数据集上验证了模型的泛化能力。本研究构建的智能检测框架可为城市交通管理系统提供具有工程实用价值的解决方案。

To address the challenge of high miss rates for traffic targets in complex environments,this article proposes a real-time target detection approach incorporating an attention mechanism.First,a baseline detection network is constructed using the YOLOv11 model.Second,the SimAM attention module is integrated into the YOLOv11 network to establish a cross-layer attention fusion mechanism,which extracts critical feature information and enhances the model′s representation capability for multiscale targets.Experimental results show that the YOLOv11 model with integrated attention mechanisms achieves a mean Average Precision(mAP)of 91.38% on a self-built traffic dataset,a 2.8%improvement over the baseline,demonstrating an excellent balance between computational efficiency and detection accuracy.Additionally,the model′s generalization capability is validated on public datasets.The intelligent detection framework developed in this study provides a practically valuable solution for urban traffic management systems.