基于YOLOv4改进的交通标志检测算法  

AN IMPROVED TRAFFIC SIGN DETECTION ALGORITHM BASED ON YOLOV4

作　　者：李丹阳 刘卫光[1] 强赞霞[1] 肖顺亮 Li Danyang;Liu Weiguang;Qiang Zanxia;Xiao Shunliang(School of Computer Science,Zhengzhou University of Technology,Zhengzhou 451191,Henan,China)

机构地区：[1]中原工学院计算机学院,河南郑州451191

出　　处：《计算机应用与软件》2024年第11期327-334,共8页Computer Applications and Software

基　　金：河南省科技攻关项目(182102210126)。

摘　　要：自动驾驶中要尽早检测到交通标志及时作出行车决策,其中的交通标志属于小目标。针对当前小目标检测精度低的情况,提出基于YOLOv4改进的交通标志检测算法。主要有以下几部分:改进骨干网络,嵌入融合注意力机制模块,加强对通道和空间信息的关注;将交叉熵损失函数改为Focal Loss损失,解决样本分布不平衡的问题;利用图片的多尺度信息进行特征提取,空洞卷积增加感受野。在TT00k数据集上进行实验,结果表明改进之后的网络模型总mAP提升14.16%,其整体性能胜过其他检测方法。It is necessary to detect traffic signs as early as possible and make driving decisions in a timely manner in the automatic driving scene,traffic signs at this point are small targets.The detection accuracy of small traffic signs is low.In order to solve this problem,an improved traffic sign detection algorithm based on YOLOv4 is proposed.The improvement of the algorithm mainly included the following parts:the integrated attention module was embedded into the backbone network to strengthen the attention of channel and spatial information;the binary cross entropy loss function was changed to focal loss to solve the imbalance problem of positive and negative samples;multi-scale information of picture was used for feature extraction and dilated convolution increase receptive field.The proposed methods were trained and tested on TT00K dataset respectively.The experimental result shows that the total mAP of the improved network model is improved by 14.16%compared with the original YOLOV4,and its overall performance outperforms other detection methods.

关 键 词：交通标志检测 注意力模块 损失函数 多尺度特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]