摘 要：文章提出一种基于 YOLOv5 改进的雨天环境交通标志识别检测方法。首先采用渐进递归网络（PRN）对摄像头采集到的画面进行去雨处理；其次通过加深网络深度，提取更深层次的小目标特征；然后在减少残差网络深度以减少计算量的基础上，加快模型检测的速度；最后以控制下采样倍数的方式解决小型目标难以识别的问题，并且引入 K-means++ 先验框到模型。实验结果表明，YOLOv5 改进模型的 F1-score 为 0.923，AP@0.5 为 0.96，mAP@0.5:0.95 位为 0.759，且 FPS 高达 71，能够很好地满足实时检测的需求。
关键词：交通标志识别；YOLOv5；渐进递归网络；雨天
DOI:10.19850/j.cnki.2096-4706.2022.20.018
中图分类号：TP391.4 文献标识码：A 文章编号：2096-4706（2022）20-0071-06
An Improved Traffic Sign Recognition and Detection on Rainy Environment Based on YOLOv5
WANG Xu
(College of Computer Science and Technology, Taiyuan Normal University, Jinzhong 030619, China)
Abstract: This paper presents an improved traffic sign recognition and detection method on rainy environment based on YOLOv5. Firstly, Progressive Recursive Network (PRN) is used to remove rain from the images collected by the camera. Secondly, through deepening the depth of the network, the deeper small target features are extracted. Then on the basis of reducing the residual network depth to reduce the amount of computation, the speed of model detection is accelerated. Finally, the problem that small targets are difficult to recognize is solved by the method of controlling the lower sampling multiple, and the K-means++ prior box is introduced into the model. The experimental results show that the F1-score of YOLOv5 improved model is 0.923, AP@0.5 is 0.96, mAP@0.5:0.95 bit is 0.759, and FPS is up to 71, which can well meet the requirements of real-time detection.
Keywords: traffic sign recognition; YOLOv5; Progressive Recursive Network; a rainy day
参考文献：
[1] REDMON J，DIVVALA S，GIRSHICK R，et al. You Only Look Once: Unified，Real-Time Object Detection [C]//2016 IEEE Conference on Computer Vision and Pattern Recognition （CVPR）. IEEE，2016：379-387.
[2] GIRSHICK R，DONAHUE J，DARRELL T，et al. Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation [C]//IEEE Conference on Computer Vision & Pattern Recognition. IEEE Computer Society，2014：580-587.
[3] 李志刚，张娜.一种轻量型YOLOv5交通标志识别方法 [J].电讯技术，2022，62（9）：1201-1206.
[4] 尹靖涵，瞿绍军，姚泽楷，等 . 基于 YOLOv5 的雾霾天气下交通标志识别模型 [J]. 计算机应用，2022，42（9）：2876-2884.
[5] 柴国强，王大为，芦宾，李竹 . 基于注意机制的轻量化稠密网络单幅图像去雨 [J/OL]. 北京航空航天大学学报：1-8[2022-02-28].DOI:10.13700/j.bh.1001-5965.2021.0294.
[6] 江智呈，李志伟，陈晨，等 . 基于多尺度前馈结构的单幅图像去雨算法 [J]. 激光与光电子学进展，2022，59（12）：244-255.
[7] 韩冉，曾广淼，王荣杰 . 基于残差块网络的图像去雨算法[J]. 仪器仪表学报，2021，41（8）：175-182.
作者简介：王旭（1997—），男，汉族，安徽合肥人，硕士在读，研究方向：智能数据研究与应用。