一种改进多尺度特征融合的交通标志识别算法  

An Improved Traffic Sign Recognition Algorithm Based on Multi-scale Feature Fusion

作　　者：余翔[1] 靳闪闪 杨路[1] YU Xiang;JIN Shanshan;YANG Lu(School of Communications and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《电讯技术》2024年第12期1955-1962,共8页Telecommunication Engineering

基　　金：国家自然科学基金资助项目(62176035)。

摘　　要：为了进一步提高在背景复杂且目标距离远的情况下交通标志识别算法的识别准确率,在YOLOv5s算法的基础上提出了一种改进的交通标志识别算法MAFM-YOLO。首先,在颈部网络设计了基于空洞混合注意力机制的多尺度注意力特征融合模块,使网络在特征融合阶段能够高效保留图像中的细节信息,对小目标更加的敏感。其次,在回归阶段采用归一化Wasserstein距离改进原有的损失函数,提高了边界框的回归性能,从而进一步提高网络的识别性能。在TT100K数据集上的实验结果表明,MAFM-YOLO较基准模型在精确率、召回率和平均精度均值上分别实现了9.4%、3.3%、6.3%的提升。To improve the recognition accuracy of traffic sign recognition algorithm in the case of complex backgrounds and long distances,an improved traffic sign recognition algorithm MAFM-YOLO is proposed based on YOLOv5s algorithm.Firstly,a multi-scale attentional feature fusion module based on the atrous blend attention mechanism is designed in the neck network,so that the network can efficiently retain the detail information in the image in the feature fusion stage,and is more sensitive to the small targets.Secondly,the normalized Wasserstein distance is used to improve the original loss function in the regression stage,which improves the regression performance of the bounding box,thus further improving the recognition performance of the network.Experimental results on the TT100K dataset show that MAFMYOLO achieves 9.4%,3.3%,and 6.3%improvement in precision,recall,and mean average precision,respectively,over the baseline model.

关 键 词：交通标志识别 YOLOv5s 多尺度特征融合 混合注意力机制 归一化Wasserstein距离 

分 类 号：TN911.73[电子电信—通信与信息系统] TP391.4[电子电信—信息与通信工程]