基于轻量化的指针仪表检测算法研究

Research on pointer gauge detection algorithm based on lightweighting

骆东松 ¹张杰锋 ¹魏羲民¹

扫码查看

作者信息

1. 兰州理工大学电气工程与信息工程学院,甘肃兰州 730050
折叠

摘要

通过轻量化的神经网络算法部署在边缘计算设备是解决老式仪表读数问题的主流方向.该文提出了 YOLOv5s-Pointer轻量化的指针仪表检测网络模型,以YOLOv5s网络模型作为基础,通过引入Mixup数据增强技术,使用MobileNetV3网络替换CSPDarkNet53特征提取网络,采用SLoU Loss定位损失函数,并引入动态样本权重思想,对网络进行改进.实验结果表明,相较于YOLOv5s网络在验证集上的参数量和计算量分别减少了 78％和57％,精确率提升了1.3％.

Abstract

Deployment of neural network algorithms via lightweight neural network algorithms to edge computing devices is the mainstream direction to solve the old meter reading problem.In this pa-per,we propose the YOLOv5s-Pointer lightweight network model for pointer meter detection,using the YOLOv5s network model as the basis,replacing the CSPDarkNet53 feature extraction network with the MobileNetV3 network by introducing the Mixup data enhancement technique,using the SLoU Loss to localize the loss function,and introducing the dynamic sample weighting idea to improve the net-work.The experimental results show that compared to YOLOv5s network the number of parameters and computation on the validation set are reduced by 78％and 57％,respectively,and the accuracy rate is improved by 1.3％.

关键词

指针仪表检测/YOLOv5/s/数据增强/MobileNetV3/轻量化

Key words

pointer meter inspection/YOLOv5s/data enhancement/MobileNetV3/lightweighting

引用本文复制引用

出版年

2024

工业仪表与自动化装置

陕西鼓风机（集团）有限公司

工业仪表与自动化装置

CSTPCD

影响因子：0.393

ISSN：1000-0682

参考文献量6

段落导航