SVM多類分類及其在遙感圖像中的應(yīng)用.pdf

1、分類號(hào)UDC密級(jí)F“786781學(xué)校代碼：10561學(xué)號(hào)：20024000020103i華南理工大學(xué)學(xué)位論文SVM多類分類及其在遙感圖像中的應(yīng)用王曉峰指導(dǎo)教師：堂圄基熬援堡直堡王丕堂麴堂型堂堂暄申請(qǐng)學(xué)位級(jí)別：亟專業(yè)名稱：廑廈數(shù)堂論文提交日期：2QQ：論文答辯日期：2QQ：學(xué)位授予單位和日期：坐直堡王態(tài)堂答辯委員會(huì)主席：叢望應(yīng)教攔論文評(píng)閱人：熬空虛熬撞宣厶盤(pán)蜃_數(shù)絲AbstractSupportVectorMachine(SVM)isan

2、ewmethodforpatternrecognitionbasedonthestatisticallearningtheoryIttakesstructureriskminimization(SRM)asprincipleBymappinginputdataintoahighdimensionalcharacteristicspaceinwhichanoptimalseparatinghyperplaneisbuiltSVMprese

3、ntsalotofadvantagesforresolvingthesmalisamples，nonlinearandhighdimensionalpatternrecognition，aswellasothermachine一1earningproblemssuchasfunctionfittingAtthepresenttimeapplYingSVMtopatternrecognitionareaissetasafocusinthe

4、studyofSVMInthebeginningSVMisusedtosolvethetwo—categoryclassificationbutthemulti—categoryclassificationHowtoextendittomulti—categoryclassificationeffectivelYisstillbeingstudiedAtpresentmanyalgorithmsthatextendSVMtosolvem

5、ulti—categoryclassificationaredevelopedwhichcalledMulti—categorySupportVectorMachimes(M—SVMs)InthispaperaM—SVMsnamedCB—SVMsispresentbasedonbintree—codingThisalgorithmcanrejecttheredundanceinformationbetweensubclassifiers

6、andmakeuseoftheclassificationresultfromothersub—classifiersAsaresult，Itcandeeplyreducetheamountofsub—classifiersComplexityanalysisandtheresultrunningonstandardtestingdatashowthatcomparetootherM—SVMsCB—SVMscanmarkedlYenha

7、ncethetrainingandtestingspeedAtthesametimeithavetheadvantagesofsimplestructureandgoodstabilityAt1astthehasictheoryandalgorithmaboutremotesensingimagerecognitionareintroducedApplyingCB—SVMstoremotesensingimagerecognitionc

8、anindicatethealgorithmhasahighprecisionofclassificationandagoodgeneralizationabilityA11theseshowthatSVMhaveagoodfutureinremotesensingimagerecognitionKeywords：patternrecognition：SVM：Multi—categoryclassificatiOnremotesensi