混合先验分析的零件表面缺陷检测方法研究

打开文本图片集

中图分类号：TP391 文献标识码：A文章编号：1006-8228（2026）01-74-07

Abstract：Toadressissesarisingfrominacuratedatasetconstructionprolongedmodeltrainingtimesndlowfeatureetraction precisionduetothediversetypes，varyingscales，andinconsistentshapesofsurfacedefectsoncomponents，thispaperproposesa hybridprioranalysismethodforsurfacedefectdetection.Firstly，byrevealingthepriordistibutioncharacteristicsofomponent surfacedefects，themethodemploysarotmeansquarefilterenhancementmodeltosimultaneouslyremovenoiseandenhancelowcontrastfeaturesfrominputdefectimagedata.Thisenhancesdatasetacuracywhilereducingmodelcomputationaltime.Secondly byintegratingthedep-learningYOLOv5modeltoperformhybridprioranalysisondefectfeaturesofdiferenttypes，scales，and shapes，theprecsionfdefectfeaturextractionisimproederebyhancingthaccuracyofefectdtectionFinallsiulation experimentsoncomponentsurfacedefectdetectiondemonstratethattheproposedmethodcanidentfylocalizeandclasifysurface defects incorporating diverse prior information，achieving an average detection confidence exceeding 79% .Thisvalidates the feasibility of the proposed method in terms of surface defect detection performance.

Keywords：Surface Defect Detection；Hybrid Prior Analysis;；Image Processing;Deep Learning;YOLOv5

0引言

零件表面品质对其机械性能、装配精度和工作稳定性至关重要。（剩余9972字）