基于改进YOLOv5的小目标交通标志检测算法

Small Target Traffic Sign Detection Algorithm Based on Improved YOLOv5

针对小目标交通标志检测精度低的问题,提出基于改进YOLOv5的小目标交通标志检测算法。优化YOLOv5的主干网络结构,采用微小目标检测层替换原始大目标检测层,调整下采样倍数,将MobileViT Block融入颈部网络,采用离线增强与在线增强的方法对数据集进行处理。实验结果表明:与其他目标检测算法相比,改进算法的小目标交通标志检测均值平均精度为85.3%,参数量降低了39%,FPS为70,满足实时检测的要求。

In order to improve the low detection accuracy of small target traffic signs,a small target traf⁃fic sign detection algorithm based on improved YOLOv5 was proposed.The backbone network structure of YOLOv5 was optimized,and the original large target detection layer was replaced with a small target detection layer.In addition,the down-sampling factor was adjusted,and the MobileViT Block module was integrated into the neck network.Then,the dataset was processed using offline enhancement and online enhancement methods.The experimental results show that compared with other target detection algorithms,the average accuracy of the improved algorithm for small target traffic sign detection is 85.3%;the parameter quantity is reduced by 39%,and the FPS is 70.Therefore,it meets the require⁃ments of real-time detection.