改进YOLOv8n的织物疵点检测算法

打开文本图片集

关键词：织物疵点；目标检测；YOLOv8n算法；GSConv；三重注意力机制DOI：10.15938/j. jhust.2025.06.006中图分类号：TP391 文献标志码：A 文章编号：1007-2683（2025）06-0051-10

Abstract：Toaddressthechallngesofdiversedefectshapesandisuessuchasmissedandfalsedetectionsinfabricdefect detection，weproposeanimprovedfabricdefectdetectionmethodbasedonYOLOvn.Firstly，aslimneck moduleisintroducedinthe featurefusionpart，eplacingtheoriginalconvolutionkernelswithGSConv，alowingachfeaturelayertosimultaneouslyconsiderboth deepandshalowfeatureinformation，therebyenhancingthedetectioncapabilityforsmalltargetdefectswhilereducingmodel complexity.SecondlyaTripletAtentionmechanismisembeddedinthebackbonenetwork，whichcanasigndiferent weightstoach featurechannelinordertoimprovethedetectionacuracyofmodelforfabricdefects.Finaly，theShape-IoUfunctionisusedto replacethecross-entropylossfunction，focusingontheshapeandscaleoftheboundingboxes tocalculateloss，whichenhancesthe model'sconvergence spedandacuracyWhendetectingeighttypesoffabricdefects，theimprovedmodelachievesaprecisionof （20号 87.1% ，a recall of 89.2% ，an mAP @ 0.5 of 92.2% ，and an of 50.3% .Experimental results indicate that the improved model better meets the precision requirements for fabric defect detection.

Keywords：fabric defects；object detection；YOLOv8n algorithm；GSConv；triplet attention mechanism

0 引言

储等环节的影响，纺织品表面可能会出现形状各异的疵点（如破洞、污渍、粗纬等）1]。（剩余11905字）