针对高速公路施工场景中背景干扰复杂、车辆种类多样以及目标特征稀缺等问题，提出了一种改进的YOLOv5高速公路施工车辆检测算法。首先，引入空间感知通道注意力机制（spatial-aware channel attention, SCA），通过动态调整通道权重，增强与施工车辆相关的特征表达能力，同时抑制背景干扰；其次，采用改进的加权跨层特征融合网络结构（weighted cross-layer feature fusion network, WCL-FFN）替代原有的路径聚合网络（path aggregation network, PAN），优化了特征融合过程，减少了计算冗余，并显著提高了多尺度目标检测性能；此外，结合幻影操作降低网络计算量与参数量，提高检测效率；最后，使用高效交并比损失函数（efficient intersecion over union, EIoU）替代传统交并比损失函数（intersecion over union, IoU），改善了边界框回归精度，进一步提升了检测精度和网络收敛速度。为验证算法的有效性，在施工车辆数据集上进行了实验，结果表明：本文提出的改进YOLOv5模型的均值平均精度（mean average precision, mAP）达到90.5%，与YOLOv5s相比，mAP提高了8.9%；与YOLOx相比，提高了6.2%；与YOLOv8相比，提高了3.5%。该方法在复杂场景中表现出更强的鲁棒性，提供了高速公路施工车辆检测的高效解决方案。

In response to the challenges posed by complex background interference, diverse vehicle types, and scarce target features in highway construction scenarios, an improved YOLOv5-based highway construction vehicle detection algorithm was proposed. First, a spatial-aware channel attention (SCA) mechanism was designed to dynamically adjust channel weights, thereby enhancing feature representation related to construction vehicles and effectively suppressing background interference. Secondly, an improved weighted cross layer feature fusion network (WCL-FFN) was introduced to replace the original path aggregation network (PAN), optimizing feature fusion, reducing computational redundancy, and improving multi-scale object detection performance. Furthermore, phantom operations are incorporated to reduce network computation and parameter size, thereby increasing detection efficiency. Lastly, an efficient intersection over union (EIoU) loss function was used to replace the traditional IoU loss function, improving bounding box regression accuracy, detection performance, and network convergence speed. To verify the effectiveness of the algorithm, experiments were conducted on a highway construction vehicle dataset. The results show that the average detection accuracy (mAP) of the improved YOLOv5 model reaches 90.5%. Compared to YOLOv5s, mAP is improved by 8.9%; compared to YOLOx, it is improved by 6.2%; and compared to YOLOv8, it is improved by 3.5%. The method demonstrates greater robustness in complex scenarios and provides an efficient solution for highway construction vehicle detection.