基于混合像元模型的高光譜數(shù)據(jù)分類.pdf

1、西北工業(yè)大學(xué)碩士學(xué)位論文基于混合像元模型的高光譜數(shù)據(jù)分類姓名：宋江紅申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：信號(hào)與信息處理指導(dǎo)教師：馮燕20060301西北一L：業(yè)人學(xué)碩十學(xué)位論文AbstractABSTRACTHyperspectraldataclassificationisoneofthemostimportantproblemsinthefieldofremotesensingtechnologyThebasicpremiseofhyperspe

2、ctraldataclassificationisthatthespectralfeaturesofvariousmaterialsaredifferentfromeachotherWimthedevelopmentofsensortechnology。hyperspectralsensorcancollectinasmanyasseveralhundredsspectralbandsatonceBecauseofthelowspaci

3、alresolution，therearemanymixedpixelsinthehyperspectralimageThespectraofthemixedpixelsandindividualpurespectraaremeasurablydifferentIgnoringthemixedpixels，traditionalalgorithmsinhyperspectraldataclassificationaredeteriora

4、tedTosolvethemixedpixelproblem，thispaperputsemphasesonbothfeatureabstractionandelassifierdesignFirst，theindependentcomponentanalysisOCA)algorithmisusedforfeatureabstractionAccordingtothelinearmixingmodel，thespectraofmixe

5、dpixelsareconsideredalinearcombinationofvariousindividualpurespectraIntllispaperdifferentmaterialtypesarealsoconsideredtobestatisticalindependentComparedwithprincipalcomponentanalysis(PCA)algorithm，thenumberofindependent

6、componentsismuchsmaller，meanwhiletheindependentcomponentsareinformativeforclassificationThentheneuralnetworkisusedforclassificationBoththeback—propagationalgorithmbasedneuralnetworkandRBFneuralnetworkareappliedtotheclass

7、ificationTheclassificationprecisionof220bandhyperspectraldataisover82％，whichisremarkablysuperiortothatoftheconventionalbayesclasssifierIntheend，fourdifferentwaysareresearchedforneuralnetworkclassifierdesignTheBPneuralnet

8、workclassifiersbasedonone—versusrestalgorithmandmultilayeralgorithmworkwellAccordingtotheexperimentsuponthe220一bandhyperspectraldatatheclassificationprecisionisapproachingto90％Keywords：hyperspectral，classification，mixed—