资源描述:
《基于显著性特征的钢坯表面缺陷检测技术》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、ᜧắᦻ᤺⌕ᨬLhe℉ឋᱯ©îᵨüý⊤☢,be℉ឋᱯ©Ḅüý⊤☢◍g4,Ïl,Éüý⊤☢Ḅᐰ¢º¢»e℉ឋᱯ©,;Gaborÿᱯ©@Aᔠ,Ç]âª◍IJ,ᨵᑭEF¿。ÏlqAÔ⊤,e℉ឋᱯ©Ḅüý⊤☢◍g4᳛,ឋ:,ᓣe,ᐹᨵãḄq▭îᵨst。ᐵ:℉ឋᱯ!,#$⊤☢'◍)*,+,-.,ᱯ!/ᔠ,℉ឋᱯ!12úIV⚓万方数据ᜧắᦻABSTRACTTheBasedonImprovedImageSaliencyCharacter
2、isticDetectionBilletSurfaceDefectDetectingTechnologyABSTRACTWiththecontinuousdevelopmentofcomputervisiontechnology,industry-orientedvisualdefectdetectiontechnologyhasattractedmoreandmoreextensiveattentionincreasingly.However,astheproductiontechnology’srapid
3、ascension,manytraditionaltechnologieshavebeenunabletomeettherequirementsofindustrialproduction.Ifthedefectdetectiontechnologycansimulatethewayofvisionsystemworking,whichprioritizedlocatingsuspecteddefectareasandthencentralizedprocessingwhichcanavoidcomputin
4、gresourceswasteandimprovedetectionaccuracy.Therefore,howtodesignthevisualdefectdetectionprocessthatconformstothesaliencyfeaturesofvisualattentionhasanimportantpracticalvalue.Firstofall,thisthesiselaboratesthecomposition,attentionmechanismandformingprocessof
5、thehumanvisualsystem.Thenthethesisintroducestheexistingclassicalsaliencymodelindetail,includingthebasedonbiologicalvisionIttimodel,thebasedonfrequencydomainfastprocessingHoumodel,thebasedoncontext-awareGofermanmodelandthebasedonimagesegmentationandareacontr
6、astChengmodel.Thesemodels,however,alsohavelimitations:lowefficiency,featureinformationsimplificationandpoorrobustnesstocomplexbackgroundandsoon.Synthesizedtherequirementsofvisualdefectdetection,thisthesispresentsanovelsaliencyalgorithmbasedonthefusionofglob
7、alchromaticdifferenceandlocallow-levelfeature.Simulatedthevisualsystemmechanism,themethodcalculatesthecolorsaliencyinnaturesceneimagetogettheglobalsaliencyfeaturemapatfirst,whichcombinesthehistogramquantizationandsaliencysmoothing;thenextractavarietyoflow-l
8、evelfeatures,makethemfusiononmulti-scalefordifferentfeaturesandgetfinallocalsaliencymapinlinearweightedsynthesisway;Finallythroughthefusionofglobalandlocalfeaturesandúv⚓万方数据ᜧắᦻABSTRACTthestrateg