YOLOv5s模型的改进及其在交通标志检测上的应用  

Improvement of YOLOv5s Model and Its Application in Traffic Sign Detection

作　　者：傅融 逯洋[1] 彭淼 FU Rong;LU Yang;PENG Miao(College of Mathematics and Computer Science,Jilin Normal University,Siping,Jilin 136000,China)

机构地区：[1]吉林师范大学数学与计算机学院,吉林四平136000

出　　处：《遥感信息》2024年第6期87-93,共7页Remote Sensing Information

基　　金：吉林省创新创业人才基金(2023QN31);吉林省自然科学基金(YDZJ202301ZYTS157);吉林省发展和改革委员会创新项目(2021C038-7)。

摘　　要：正确检测交通标志是智能驾驶和无人驾驶的关键性技术。针对交通标志目标小且精度低的问题,提出了一种基于改进YOLOv5s模型的算法。首先,采用加权双向特征金字塔结构进行跨尺度特征融合,很好地整合了语义信息与定位信息,提高了检测精度;其次,采用EIoU损失函数代替YOLOv5s的原始损失函数,加快收敛速度;最后,在BiFPN的基础上增加检测层,更好地帮助模型预测准确结果。改进后的YOLOv5s模型与基础模型在TT100K数据集上进行对比实验,识别的平均精度均值由73.6%提高到80.8%,在CCTSDB数据集上的识别平均精度均值由97.4%提高到97.8%,具有更好的识别性能。Correct detection and recognition of traffic signs is the key technology of intelligent driving and driverless driving.An algorithm based on improved YOLOv5s model is proposed to solve the problem of small target and precision of traffic signs.Firstly,the weighted bidirectional feature pyramid structure is used to carry out the cross-scale feature fusion,which integrates the semantic information and the location information well and improves the detection accuracy,the EIoU loss function is used to replace the original loss function of YOLOv5s to accelerate the convergence rate,and the detection layer is added to BiFPN to help the model predict the accurate results.Compared with the basic model on the TT100K data set,the average precision of recognition is improved from 73.6%to 80.8%,and the average precision of recognition on CCTSDB data set is improved from 97.4%to 97.8%.

关 键 词：YOLOv5 交通标志识别 BiFPN 损失函数 检测层 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]