为解决雾天道路目标特征模糊、提取困难等问题，提出了一种基于双通道注意力机制的目标检测算法GMA-YOLO。通过特征融合的方式直接检测，其设计思路：首先，将局部混合特征通道注意力模块（global & local channel attention, GLCA）改进主干和颈部特征融合的C2f模块，提升模糊目标的局部特征提取能力；然后，设计双路径结构的混合池化卷积（mixed pooling convolution, MPConv）改进下采样模块，提升特征保留的多样性与处理复杂场景的鲁棒性；接着，引入辅助检测头（auxiliary head, AuxHead），提升远景小目标的检测效果；最后，引入Focaler-CIoU优化损失函数，降低易分类样本的权重，强调受阻严重的样本。实验表明，在RTTS数据集上相较于YOLOv8s基线模型，GMA-YOLO模型在参数量减少的情况下，mAP@0.5提升了3.2%，达到78.6%，召回率提升了3.8%，达到71.7%，该算法有效提高雾天场景的目标检测精度。

In order to address the issues of blurred features and difficult extraction of road objects in foggy weather, a target detection algorithm based on a dual-channel attention mechanism was proposed. The algorithm detects through feature fusion, and its design approach is as follows. Firstly, GLCA(global & local channel attention) was used to improve the C2f module for backbone and neck feature fusion, enhancing the ability to extract local features of blurred targets. Then, a dual-path structured MPConv (mixed pooling convolution) was designed to improve the downsampling module, increasing the diversity of feature retention and robustness in handling complex scenes. Subsequently, AuxHead(auxiliary detection head) was introduced to enhance the detection effect of distant small targets. Finally, Focaler-CIoU was introduced to optimize the loss function, reducing the weight of easily classified samples and emphasizing severely obstructed samples. Experiments show that on the RTTS dataset, compared to the YOLOv8s baseline model, the GMA-YOLO model achieved a 3.2% increase in mAP@0.5, reaching 78.6%, and a 3.8% increase in recall rate, reaching 71.7%, while reducing the number of parameters. It is concluded that this algorithm effectively improves the accuracy of target detection in foggy scenes.