摘要
在汽车液晶仪表的自动化测试领域,机器视觉学习取代人工检测液晶仪表已成为趋势。针对现有方法存在检测环境苛刻、准确度低以及通用性差等问题,提出了一种基于YOLOv5的汽车液晶仪表图标符号检测方法,实现对仪表常用指示及报警图标符号的自动化检测工作。该方法通过在YOLOv5网络模型基础上添加ECA注意力机制以减少图标符号周围冗余信息的干扰、采用轻量级通用上采样算子CARAFE扩大特征图的感受野、采用多尺度检测结构增加小图标符号的检测层、引入Dynamic Head模块提高对图标符号的感知能力,从而提高网络模型的检测准确度;同时使用SIoU Loss作为回归定位损失函数,以加快网络模型的收敛速度。与原始YOLOv5模型进行对比实验,结果显示改进模型在多种类的汽车液晶仪表图标符号检测上表现更好,mAP0.5和mAP0.5:0.95分别提高了3.1%和3.6%,达到了95.9%和76.8%,这为汽车液晶仪表图标符号的自动化检测工作提供了方法上的参考。
In the field of automated testing of automotive LCD instruments,machine vision learning has become a trend to replace manu-al detection of LCD instrument.In view of the problems of harsh detection environment,low accuracy and poor versatility of existing methods,a vehicle LCD instrument icon symbol detection method based on YOLOv5 is proposed to realize the common use of indicator and alarm icon symbols for instruments automated inspection work.This method is improved on the basis of the original YOLOv5 net-work model.An ECA attention mechanism is added to reduce the interference of redundant information around icon symbols,lightweight general-purpose up sampling operator of CARAFE is used to expand the sensing field of feature map,the multi-scale detection structure is used to improve the detection ability of small icon symbols,the dynamic head module is introduced to improve the perception of icon symbols.SIoU Loss is used as the loss function to accelerate the convergence of network models.The proposed model is compared with the original YOLOv5 model,the results show that the improved model performs better in the detection of various types of automotive LCD instrument icon symbols,and the mAP0.5 and mAP0.5:0.95 are improved by 3.1%and 3.6%respectively,reaching 95.9%and 76.8%,which provides a methodological reference for the automatic detection of automotive LCD instrument icon symbols.
作者
郭健忠
肖庆
谢斌
闵锐
丁宁
GUO Jianzhong;XIAO Qing;XIE Bin;MIN Rui;DING Ning(School of Automotive and Transportation Engineering,Wuhan University of Science and Technology,Wuhan Hubei 430065,China;Wuhan Baohua Display Technology Co.,Ltd.,Wuhan Hubei 430082,China)
出处
《电子器件》
2025年第4期853-862,共10页
Chinese Journal of Electron Devices
