資源描述:
《[IJCNN 2012] Steel Defect Classification with Max-Pooling Convolutional Neural Networks》由會員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、SteelDefectClassi?cationwithMax-PoolingConvolutionalNeuralNetworksJonathanMasci,UeliMeier,DanCiresan,GabrielFricoutJurgenSchmidhuber¨ArcelorMittalIDSIA,USIandSUPSIMaizieresResearchSA,`Galleria2,6928Manno-Lugano,FranceSwitzerlandfgabriel.fricout@arcelormittal.comgfjona
2、than,ueli,dan,juergeng@idsia.chAbstract—WepresentaMax-PoolingConvolutionalNeuralsi?cation.Thesystemisusuallybasedonasetofhand-Networkapproachforsupervisedsteeldefectclassi?cation.Onawiredpipelineswithpartialornoself-adjustableparametersclassi?cationtaskwith7defects,co
3、llectedfromarealproductionwhichmakesthe?ne-tuningprocessofthisindustrialsystemsline,anerrorrateof7%isobtained.ComparedtoSVMcumbersome,requiringmuchmorehumaninterventionthanclassi?erstrainedoncommonlyusedfeaturedescriptorsourbestnetperformsatleasttwotimesbetter.Notonly
4、wedoobtaindesired.Inthisworkwefocusonthetwolastpipelinestagesmuchbetterresults,buttheproposedmethodalsoworksdirectlyandproposeanapproachbasedonMax-PoolingConvolutionalonrawpixelintensitiesofdetectedandsegmentedsteeldefects,NeuralNetworks(MPCNN)[1],[2],[3],[4],[5],that
5、learnavoidingfurthertimeconsumingandhardtooptimizead-hocthefeaturesdirectlyfromlabeledimagesusingsupervisedpreprocessing.learning.Weshowthattheproposedmethodachievesstate-of-the-artresultsonrealworlddataandcompareourapproachI.INTRODUCTIONtoclassi?erstrainedonclassicfe
6、aturedescriptors.MachinevisionbasedsurfaceinspectiontechnologieshaveThereisnotmuchliteratureaboutsteeldefectdetectiongainedalotofinterestfromvariousindustriestoautomatein-[6].However,inabroadercontexttheproblemcanbespectionsystems,andtosigni?cantlyimproveoverallproduc
7、tviewedasdefectdetectionintexturedmaterialwhichhasquality.Atypicalindustryadoptingthesere?nedinspectionreceivedconsiderableattentionincomputervision[7],[8],toolsistherolledsteelstripmarket.Real-timevisualinspec-[9].Inclassicalapproaches,featureextractionisperformedtio
8、nofproductionlinesiscrucialtoprovideaproductwithusingthe?lter-bankparadigm,resultinginanarchitectureeverfewersurfacedefects.