资源描述：
《Feature extraction and selection for defect class...》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。
1、ARTICLEINPRESSNDT&EInternational41(2008)467–476ContentslistsavailableatScienceDirectNDT&EInternationaljournalhomepage:www.elsevier.com/locate/ndteintFeatureextractionandselectionfordefectclassificationofpulsededdycurrentNDTa,ba,abTianluChen,GuiYunTian,Al
2、iSophian,PeiWenQueaSchoolofComputingandEngineering,UniversityofHuddersfield,Queensgate,HuddersfieldHD13DH,UKbDepartmentofInstrumentScienceandEngineering,ShanghaiJiaotongUniversity,Shanghai200240,ChinaarticleinfoabstractArticlehistory:Pulsededdycurrent(PEC)
3、isanewemergingnondestructivetesting(NDT)techniqueusingabroadbandReceived12July2007pulseexcitationwithrichfrequencyinformationandhaswideapplicationpotentials.ThistechniqueReceivedinrevisedformmainlyusesfeaturepointsandresponsesignalshapesfordefectdetectio
4、nandcharacterization,12November2007includingpeakpoint,frequencyanalysis,andstatisticalmethodssuchasprincipalcomponentanalysisAccepted18February2008(PCA).ThispaperintroducestheapplicationofHilberttransformtoextractanewdescendingfeatureAvailableonline4Marc
5、h2008pointandusethepointasacutoffpointofsamplingdatafordetectionandfeatureestimation.TheKeywords:responsesignalisthendividedbytheconventionalrising,peak,andthenewdescendingpoints.SomePulsededdycurrentshapefeaturesoftherisingpartanddescendingpartareextrac
6、ted.ThecharactersofshapefeaturesareHilberttransformalsodiscussedandcompared.VariousfeatureselectionandintegrationsareproposedfordefectFeatureextractionclassification.Experimentalstudies,includingblindtests,showthevalidationofthenewfeaturesandFeatureselect
7、ionDefectclassificationcombinationofselectedfeaturesindefectclassification.Therobustnessofthefeaturesandfurtherworkarealsodiscussed.&2008PublishedbyElsevierLtd.1.Introductiondevelopmentofeddycurrentinnondestructivetesting(NDT)frommultilayerstructurestoon-l
8、inemonitoring[2,3].ThesePulsededdycurrent(PEC)sensingisanewandemergingfeaturesarefarfromenoughforfastandaccuratedefecttechnique[1,2]thathasbeenparticularlydevelopedanddevisedclassificationandquantificationinrealtime[11].fors
