机器视觉英文资料

IdentificationofcrackfeaturesinfreshjujubeusingVis/NIRhyperspectralimagingcombinedwithimageprocessingKeqiangYua,YanruZhaoa,XiaoliLia,b,YongniShaoa,b,FengleZhua,YongHea,b,⇑aCollegeofBiosystemsEngineeringandFoodScience,ZhejiangUniversity,866YuhangtangRoad,Hangzhou310058,PRChinabKeyLaboratoryofEquipmentandInformatizationinEnvironmentControlledAgriculture,MinistryofAgriculture,PRChinaarticleinfoArticlehistory:Received31July2013Receivedinrevisedform23December2013Accepted24January2014Keywords:HyperspectralimagingCrackfeatureIdentificationChemometricsFreshjujubeabstractDetectionofcrackdefectinfreshjujubeisacriticalprocesstoguaranteejujubequalityandmeetprocessingdemandsoffreshjujubefruit.Thisstudypresentedanovelmethodforidentificationoffreshjujubecrackfeatureusinghyperspectralimaginginvisibleandnearinfrared(Vis/NIR)region(380–1030nm)combinedwithimageprocessing.Hyperspectralimagedataofsampleswereusedtoextractthecharacteristicwavebandsbychemometrics,whichintegratedthemethodofpartialleastsquaresregression(PLSR),principalcomponentanalysis(PCA)ofspatialhyperspectralimage(SPCA)andindependentcomponentanalysis(ICA)ofspatialhyperspectralimage(SICA).Onthebasisoftheselectedwavebands,least-squaressupportvectormachine(LS-SVM)discriminationmodelswereestablishedtocorrectlydistinguishbetweencrackedandsoundfreshjujube.Theperformanceofdiscrim-inatingmodelwasevaluatedusingreceiveroperatingcharacteristics(ROC)curveanalysis.TheresultsdemonstratedthatPLSR–LS-SVMdiscriminationmodelwithahighaccuracyof100%hadtheoptimalperformanceof‘‘area’’=1and‘‘std’’=0.Foracquiringrichcrackfeatureinformation,SPCAwasalsocar-riedonimagesatthefivecharacteristicwavebands(467,544,639,673and682nm)selectedbyPLSR.Finally,theSPC-4imagewasexploredtoidentifythelocationandareaofcrackfeaturethroughadevelopedimageprocessingalgorithm.Theresultsrevealedthathyperspectralimagingcombinedwithimageprocessingtechniquecouldachievetherapididentificationofcrackfeaturesinfreshjujube.2014ElsevierB.V.Allrightsreserved.1.IntroductionJujube(ZizyphusjujubeMiller)ismainlydistributedinthesub-tropicalandtropicalregionsofAsia,andithasbeencultivatedforoverthreethousandyears.Jujubeisconsideredtobefruit,foodadditivesandpharmaceuticalsforalongtime(Lietal.,2011b).FreshjujubeisnotonlyjuicyanddeliciousbutalsorichinvitaminCamongallfruits.However,freshjujube’smaturationstageisusu-allyinSeptemberorOctober,inwhichitoftenrainsinnorthofChina.Oncejujubefruitiscaughtinrainfall,itspeel’sgrowthstresswouldreducewiththeincreaseofwater-solublepectin.Alongwiththerapidexpansionofpulp,freshjujubepeelwouldcrackwithdifferentsizeandshape.Crackscouldexposetheinternalfleshofthejujubesandleadtofreshjujuberotinseverecases,whichwoulddowngradingtofruitquality.Inotherwords,crackcoulddeterminethequalityoffruitandreducetheeconomicvalueoffreshjujube.Rapidandaccurateidentificationoffreshjujubecrackfeaturehassignificantimplicationsforscreeningdefectivefruits,rankingfruitsindifferentgradesandprocessingjujubeproducts.Qualitativeandquantitativeanalysisofjujubecrackscouldsatisfydifferentneedsofjujubeprocessing.Atpresent,freshjujubegradingmostlyreliesonmanualoperationofexperiencedworkers,whichistime-consumingandlaborious.Inrecentyears,severalresearchresultswerereportedtodetectjujubeusingspectroscopictechnology.Wangetal.(2011a)com-pletedrapiddetectionofinternalinsectinfestationwithdifferentdamagedlevelsinjujubesusingthreespectroscopicmeasurements(interactance,reflectanceandtransmittance)intheVis/NIRrange.Wangetal.(2011b)comparedpredictionabilityofinteractance,transmissionmeasurementsofvisibleandnear-infrared(Vis/NIR)spectroscopyindetectingthesolublesolidscontent(SSC)ofju-jubes.Althoughafewfindingshavebeenprovedondetectionofjujubeusingspectroscopictechnology,nostudyhasspecificallymadequalitativeandquantitativeanalysisofjujubecrackusinghyperspectralimagingtechnique.Hyperspectralimagingtechniqueisanemergingtechnologythatintegratesbothdigitalimageandspectroscopytechnologyin2014ElsevierB.V.Allrightsreserved.⇑Correspondingauthorat:CollegeofBiosystemsEngineeringandFoodScience,ZhejiangUniversity,866YuhangtangRoad,Hangzhou310058,PRChina.Tel./fax:[phone].E-mailaddress:[e-mail](Y.He).ComputersandElectronicsinAgriculture103(2014)1–10ContentslistsavailableatScienceDirectComputersandElectronicsinAgriculturejournalhomepage:(ZhangandHe,2013;WuandSun,2013a,b,c,d).Ithasattractedtheinterestofresearchersasapowerfultoolfordetectingavarietyofagriculturalandsidelineproducts(FernándezPiernaetal.,2012;WuandSun,2013a,b,c,d).Inrecentyears,researchersconductedalargenumberofstudiesusinghyperspectralimagingtechnology.Forexample,hyperspectralimagingtechnologywasemployedtodetectexter-nalinsectdamageinjujubefruitsandestablishedanimagingalgo-rithmtorecognizetheinsectdamage(Wangetal.,2011c).Baranowskietal.(2012)appliedahyperspectralandthermalimagingsystemtodetectearlybruiseinappleandobtainedthepositionanddepthofdamagedregion.Hypersp