摘 要：现实中交通标志的检测和识别具有环境多变的特点，交通标志长时间暴露在外经常会出现损坏情况，对检测的精 度和速度产生较大影响。利用最新的 YOLO 系列算法——YOLOX，对网络结构的加强特征提取层进行改进，引入 OPA-FPN网络，相较于原来的 PANet 网络，后者精度提升 2.2%。在交通标志识别过程，对经典的卷积神经网络模型 LeNet-5 进行改进，在数据集 TT100K 中进行实验，相较于其他交通标志识别模型，使用改进的模型可以使识别正确率提升 2.31%，识别时间减少了 13.02 ms。
关键词：单步路径聚合网络；YOLO；卷积神经网络；FPN；LeNet-5
DOI:10.19850/j.cnki.2096-4706.2022.02.025
基金项目： 安徽省自然科学基金项目（1908085MF189）
中图分类号：TP273+.4 文献标识码：A 文章编号：2096-4706（2022）02-0101-04
Traffic Sign Detection and Recognition Based on Improved YOLOX
CHEN Min, WU Guanmao
(School of Computer Science and Engineering, Anhui University of Technology, Huainan 232001, China)
Abstract: In reality, the detection and recognition of traffic signs have the characteristics of changeable environment. Traffic signs are often damaged after being exposed for a long time, which has a great impact on the accuracy and speed of detection. Using the latest YOLO series algorithm—YOLOX, the enhanced feature extraction layer of the network structure is improved, and the OPA-FPN network is introduced. Compared with the original PANet network, the accuracy of the latter is improved by 2.2%. In the process of traffic sign recognition, the classical convolutional neural network model LeNet-5 is improved, experiments are carried out in the data set TT100K. Compared with other traffic sign recognition models, using the improved model can improve the recognition accuracy by 2.31% and reduce the recognition time by 13.02 ms.
Keywords: single-step path aggregation network; YOLO; convolutional neural network; FPN; LeNet-5
参考文献：
[1] 陈维馨，李翠华，汪哲慎 . 基于颜色和形状的道路交通标志检测 [J]. 厦门大学学报（自然科学版），2007（5）：635-640.
[2] REN S，HE K，GIRSHICK R，et al. Faster R-CNN： Towards Real-Time Object Detection with Region Proposal Networks [J].IEEE Transactions on Pattern Analysis & Machine Intelligence，2017，39（6）：1137-1149.
[3] REDMON J，FARHADI A. YOLOv3：An Incremental Improvement [J/OL].arXiv：1804.02767 [cs.CV].（2018-04-08）.https://arxiv.org/abs/1804.02767v1.
[4] BOCHKOVSKIY A，WANG C Y，Liao H Y M. YOLOv4：Optimal Speed and Accuracy of Object Detection [J/OL].arXiv：2004.10934 [cs.CV].（2020-04-23）.https://arxiv.org/abs/2004.10934.[5] YI Z T，WU G，PAN X L，et al. Research on Helmet
Wearing Detection in Multiple Scenarios Based on YOLOv5 [EB/OL].[2021-11-16].https://mall.cnki.net/magazine/Article/IPFD/KZJC202105007096.htm.
[6] 赵志宏，杨绍普，马增强 . 基于卷积神经网络 LeNet-5 的车牌字符识别研究 [J]. 系统仿真学报，2010（3）：638-641.
作者简介：陈民（1997—），女，汉族，安徽濉溪人，硕士在读，研究方向：目标检测；吴观茂（1965—），男，汉族，安徽歙县人，副教授，博士，研究方向：深度学习。