基于改进的YOLOv5s的夜间行人目标识别算法研究

Research on night pedestrian target recognition algorithm based on improved YOLOv5s

刘文骄 ¹廖义奎 ¹梅欢子 ¹胡昌瑞 ¹徐钲槟¹

扫码查看

作者信息

1. 广西民族大学电子信息学院,南宁 530006
折叠

摘要

针对传统夜间行人识别过程中存在的方法识别速度慢、精度低、识别效果差的问题,提出了一种改进的YOLOv5s夜间行人识别算法.首先采用C3CSGC模块替换YOLOv5s原网络模型中的C3模块.其次,将YOLOv5s的损失函数CIoU换为EIoU.最后,将YOLOv5s模型的特征金字塔换成加权双向特征金字塔BiFPN.实验结果表明,对于夜间行人识别算法的改进,针对原始的YOLOv5s模型准确率,召回率分别提升了4.1%和5.9%,mAP_0.5值提升了7.2%,参数量由7012825变为3604758,模型大小由14.4 M变为7.5 M,说明了改进算法对夜间行人识别的有效性.

Abstract

Aiming at the problems of slow speed,low precision and poor recognition effect in the process of traditional night-time pedestrian recognition,an improved YOLOv5s nighttime pedestrian recognition algorithm was proposed.Firstly,C3CSGC mod-ule was used to replace C3 module in the original YOLOv5s network model.Secondly,the loss function CIoU of YOLOv5s is re-placed by EIoU.Finally,the feature pyramid of YOLOv5s model is replaced by weighted bidirectional feature pyramid BiFPN.Ex-perimental results show that for the improved nighttime pedestrian recognition algorithm,the Precision(P)and Recall(R)of the original YOLOv5s model are increased by 4.1%and 5.9%,and the values of mAP_0.5 is increased by 7.2%,respectively.The num-ber of parameters changed from 7012825 to 3604758,and the model size changed from 14.4 M to 7.5M,indicating the effectiveness of the improved algorithm for nighttime pedestrian recognition.

关键词

深度学习/夜间行人识别/YOLOv5s/C3CSGC/BiFPN/损失函数

Key words

deep learning/pedestrian identification at night/YOLOv5s/C3CSGC/BiFPN/loss function

引用本文复制引用

出版年

2024

现代计算机

中大控股

现代计算机

影响因子：0.292

ISSN：1007-1423

段落导航