基于改进YOLOv4的柱形锂电池缺陷检测研究.pdfVIP

机械工程师

MECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEERMECHANICALENGINEER

基于改进YOLOv4的柱形锂电池缺陷检测研究

田杰，胡昊，周华健，邹润

（合肥工业大学机械工程学院，合肥230009）

摘要：

为了实现柱形锂电池缺陷检测的实时性与高精度，提出一种基于改进YOLOv4的柱形锂电池表面缺陷检测算法。

将主干网络由CSPDarkNet53替换为轻量化网络MobileNetv1，使用K-means++算法对锂电池缺陷数据集先验框进行重

新聚类，同时构建新的注意力机制ECSA模块关注重要信息。改进后的模型检测精度与检测速度均得到提升。

关键词：

柱形锂电池；缺陷检测；深度学习；YOLOv4

中图分类号:TM912文献标志码:粤文章编号：员园园圆原圆猿猿猿（圆园23）03原园016原园3

SurfaceDefectDetectionofCylindricalLithiumBatteryBasedonImprovedYOLOv4

TIANJieHUHaoZHOUHuajianZOURun

(SchoolofMechanicalEngineering,HefeiUniversityofTechnology,Hefei230009,China)

Abstract:Inordertoachievereal-timeandhigh-precisiondefectdetectionofcylindricallithiumbattery,thispaper

proposesasurfacedefectdetectionalgorithmofcylindricallithiumbatterybasedonimprovedYOLOv4.Thebackbone

networkCSPDarkNet53isreplacedbyMobileNetv1,whichisalightweightnetwork.K-means++algorithmisusedto

reclusterthepriorframesoflithiumbatterydefectdataset.Atthesametime,anewattentionmechanismECSAmoduleis

constructedtopayattentiontoimportantinformation.Thedete