聚類與曲線進化方法及在農(nóng)產(chǎn)品圖像分割中的應(yīng)用研究.pdf

1、江南大學(xué)博士學(xué)位論文聚類與曲線進化方法及在農(nóng)產(chǎn)品圖像分割中的應(yīng)用研究姓名：謝振平申請學(xué)位級別：博士專業(yè)：輕工信息技術(shù)與工程指導(dǎo)教師：王士同20080301AbsIractAbstractNondesn’uctiVedetectionofagriculturalproductisabasictechniqueforquali夠assurance，soitisanimportantcomponentofagriculturalmodemiz

2、atjonAIongwiththedeVelopmentofcomputerVisiontechniques，nowmanycorrespondingnondestructiVedetectionmethodsareintroducedandsomegoode何’ectareobtained1nthesemethods，aprimalytaskistogaintheexactobjectimages，ie，imagesegmentati

3、on，withwhichthequantitatiVeanalysesonproductquali妙mightbedoneHoweVer，imagesegmentationisoneofthedi仟icultproblemsincomputerVision；toimproVethepowerofcorrespondingnondeStluctivedetection，theimagesegmentationproblemsonagric

4、ultumlproductsdeserVetomnherexploration1nthispaper，twotypesofagriculturalproductsaremainlyexaminedincludingpeachshadeimagesandnatural仇litimagesClusteringandcuⅣeeV01ution(1eVelset)techniquesaret、voprominentimagesegmentati

5、ontools，whicharealsousedtoimagesegmentationsofagricultLnalproductsinthisp印erInch印ter2andchapter3，wefocusoncJusteringbasedimagesegmentationmethods，wheret、^，Onewclusteringalgorithmsarepresentedincluding：anewdiVisiVehierarc

6、hicalclusteringalgorithmSHPDHCandanenhancedpossbjlisticclusteringalgorjthm(EPCM)丁hesetwonewimagesegmentationmethodsareallbasedonthesoRhypersphericpanitionstrategiesproposedinthech印ter2ofthispaperComparedwithseVemlpopular

7、clusteringalgorithms，SHPDHCandEPCMhaVethebetter(orequaJ)clusteringrobustnessandtheabili夠tolabeltheoutliersofthedatasetThepeachshadeimagesaremainlyexaminedinthesetwochapters，andtheexperimentalresultsdemons協(xié)牡ethatthebetter

8、segmentationresuItscanbeobtainedbyaboVeMonewalgorithmsthanseVeralclassicalcIusteringbaSedimagesegmenta_tionmethodsforthemostofsuchtypeofimagesHoweVeLforthecompIexnaturalfrujtimages，thesatisaedresultscannotbeobtainedbyabo

9、Ve鉚oalgorithmsThesecasesare如rdherresearchedinthenextt、voch印ters1nchapter4，anewimagesegmentationmodelintegratingthemzzycmeansclusteringintoMumford—Shahmodel(FCMMS)ispresentedNext，acoupledleVelsetmethodintegratingwithGauss

10、ianmixturemodel(GMMLS)forimagesegmentationisproposedinch印ter5AboVe鉚onewmodelscanbeVjewedasacombinationofclusteringandleVelsetimagesegmentationmethodsfromtheViewpointofmodelobjectfunctionalTheycalle膿ctiVeIyinheritthemerit

11、soft、Vooriginalmethods，whiIetherespectiVeshortcomingsareweakenedThe印plicationperfomlanceofthesetwonewmethodsonpeachshadeimagesandnatural行uitimagesareVe哆satisfiedwithourexpectationsFurthermore，inmanycaSes，GMMLScangetthebe

12、ttersegmentationperfornlancethanFCMMSduetoitstheoreticaImamri哆NeVenheless，accordingtothesamefhmeworkofFCMMS，manyextendedVersionsintegratingwithcurrentfhzzycJusteringalgorithmsforimagesegmentationmightbeputfonVard，whichw川

13、jmproVeitspracticalapplicationworthinessInchapter6，someadVancedtechniquesoftwonewimagesegmentationmethodsarediscussedInmanyc硒es，theseadVancedtechniquesmightpaythecruciaIrolesandimproVethesegmentationperfbnllanceingreatpa