无线电工程

针对高速公路交通违法行为识别中存在的目标尺度小、动态特征敏感度不足及多目标空间关系建模薄弱等问题，提出一种基于改进YOLOv8的无人机智能识别算法——GDsec＿YOLOv8。通过引入卷积块注意力模块(Convolutional Block Attention Module, CBAM)双重注意力机制增强模型对小目标(分心驾驶、车辆违规变道等)的特征提取能力，结合Wise-IoU损失函数优化边界框回归过程，有效提升了检测精度与收敛效率。实验表明，改进后的GDsec＿YOLOv8模型在参数量为28.9 MB时，mAP@0.50和mAP@0.50:0.95分别达到99.73%和93.45%,较主流算法性能提升显著。算法以无人机动态巡检为数据源，对所涉及的目标违法行为进行识别，实现地空联动执法，为低空经济与交通治理协同发展提供了技术支撑。

To solve the problems of small target scale, insufficient dynamic feature sensitivity and weak modeling of multi-target spatial relationship in the identification of highway traffic violations, a UAV intelligent identification algorithm, GDsec＿YOLOv8, is proposed based on the improvement of YOLOv8. The feature extraction capability of the model for small targets(distracted driving, vehicle lane changing violations, etc.) is enhanced by introducing the Convolutional Block Attention Module(CBAM) dual attention mechanism, and the bounding box regression process is optimized by combining the Wise-IoU loss function, which effectively improves detection accuracy and convergence efficiency. Experimental results demonstrate that the improved GDsec＿YOLOv8 model reaches 99.73% and 93.45% respectively for mAP@0.50 and mAP@0.50:0.95 when the number of parameters is 28.9 MB, significantly outperforming mainstream algorithms in performance. The algorithm utilizes drone patrols as its data source to identify target traffic violations, enabling ground-air coordinated law enforcement and providing technical support for the synergistic development of the low-altitude economy and traffic governance.