搜索
GrapematurityestimationbasedonseedimagesandneuralnetworksAlexZuñigac,n,MarcoMoraa,c,nn,MiguelOyarcec,ClaudioFredesb,caDepartmentofComputerScience,UniversidadCatólicadelMaule,ChilebDepartmentofAgriculturalScience,UniversidadCatólicadelMaule,ChilecLaboratoryofTechnologicalResearchonPatterRecognition,1UniversidadCatólicadelMaule,ChilearticleinfoArticlehistory:Received5December2013Receivedinrevisedform12April2014Accepted10June2014Availableonline9July2014Keywords:GrapematurityestimationNeuralnetworksAppearancedescriptorsSeedimagesabstractThegrapephenolicmaturityisoneofthemostimportantparameterstodeterminetheoptimaltimeforharvest.Inthispaperweproposeaninnovativemethodologyfortheproblemofhowthistaskisperformedtoday.Inparticular,themethodconsistsinanalyzingseedimagesusingpatternrecognitionmethodology,andclassifyingtheminimmature,matureandovermaturestatesthroughasupervisedlearningneuralnetwork.Themethodologypresentedgivesobjectiveinformationaboutmaturity,whichisusefulfordecidingthemomentwhentheharvestshouldbeperformed.&2014ElsevierLtd.Allrightsreserved.1.IntroductionThewinemarketisverydynamic;therefore,itisrelevanttodevelopappliedresearchinthisfieldinordertoimprovethelevelofcompetitivenessinademandingmarket.Toobtainagoodqualitywine,itisnecessarytoharvestthegrapesatanoptimallevelofmaturity.Traditionally,thematurityestimationisper-formedthroughchemicalanalysisinalaboratory,whichentailsahighcost.Moreover,itispossibletocountonthejudgmentofanexpert(oenologist),butthisalternativehasahighsubjectivitydegreeandlowrepresentativity.Theresultingwinequalityisrelatedtothetimingofthegrapeharvest.Inparticular,itistheOenologistwhomustdetermine,byaquantitativeorqualitativemethod,ifgrapesfromavineyardareatitsharvestpoint.FromatechnicalpointofviewthetimeofharvestisestimatedbystudyingthedegreeofPhenolicMaturityofthegrapes.TheoptimalPhenolicMaturityfortheharvestiswhentheseedtanninsareataminimumandtheskinphenolsatamaximum,resultinginminimalbitterness.TodeterminethePhenolicMaturitydegreeobjectivemethodsexist,correspondingtochemicalanalysis,thatrelatetothemeasurementofindicatorssuchaspH,acidity,solublesolids,amongothers(GilandPszczólkowski,2007).Thesemethodsaregenerallyveryaccurate,butinvolvelaboratoryanalysis.Therearealsosubjectivemethods,whicharebasedontheexperienceoftheOenologist,andPhenolicMaturityisestimatedbyinspectingthearoma,flavorandappearanceofbothfruitandwine.ThesetypesofanalysesarecalledOrganolepticanalysis.Traditionally,organolepticanalysesareperformedonthefruitofthegrape,orthewine.Unlikepreviouswork,averyrecentlineofresearchexplorestheseedofthefruittodeterminethePhenolicMaturity.Suchindicatorsascolor,shapeandtextureoftheseedswerestudied,findingahighcorrelationbetweenappearanceindicatorsandthetasteofwines(RisticandIland,2005;Fredesetal.,2010).ThesubjectivemethodsfordetermininggrapePhenolicMaturitydependontheexperienceofahumanexpert,andtheyareknowntobesubjecttoerrorsofconsistencyintheevaluation(GawelandGodden,2008).Whentheseedappearancemethodisused,theexpertvisuallycomparestheseedsurfacecolorwithacolorchart.Thedisadvantagesofanalyzingseedsthroughitsappearanceandorganolepticmethodsgenerallyincludethefol-lowing:first,thevisualdeterminationofcolorisverysubjective,andcaneasilyincurintomisjudgments.Second,itisnotpossibletoinferinformationregardingthetotalityofthesitewithasmallgroupofseeds,whichforcestheexperttoanalyzealargevolumeofseedstoobtainrepresentativeinformationoftheentireprop-erty.Finally,asubjectiveanalysisrequiresahighlytrainedexpertawareofrecentprogressinthisfield,anexpertwhoisnotalwaysavailable.Asmentionedabove,thesubjectivemethodshavecertaindisadvantages,butduetotheirlowcost,theyarepossibletoimplementinsmallvineyards.ContentslistsavailableatScienceDirectjournalhomepage:ficialIntelligence:LaboratoryofTechnologicalResearchonPatternRecognition,UniversidadCatólicadelMaule,Chile.E-mailaddresses:azuniga@litrp.cl(A.Zuñiga),marcomoracofre@gmail.com,mora@spock.ucm.cl(M.Mora),moyarce@litrp.cl(M.Oyarce),cfredes@ucm.cl(C.Fredes).URL:(M.Mora).1Website:(2014)95–104Arecentlineofresearchexaminestheseedofthefruittoestimateitslevelofmaturity.InRisticandIland(2005)astudyoftherelationshipbetweenthedevelopmentoftheseedandberryofVitisviniferaL.cvShirazispresented.Seedwereremovedtoanalyzetheirphysicalpropertiessuchasweight,colorandmoisture,inadditiontoitsphenoliccomponents.Thestudyidentifiedthreephasesofgrowthanddevelopmentoftheseed,whichdeterminedspecificstagesinthedevelopmentoftheberry.Also,changesinthephenoliccompositionoftheseedsinrelationtothedifferentstagesofdevelopmentandmaturityoftheseedsandberrieswerestudied.Itprovidedevidencethatchangesinthephenoliccompositionoftheseedscanbereflectedbychangesincolor,andthatthiscanbeusedasanindicatorofmaturity.InFredesetal.