BP神經網絡與支持向量機二級識別手寫體數字.pdf

1、北京郵電大學碩士學位論文BP神經網絡與支持向量機二級識別手寫體數字姓名：王樹海申請學位級別：碩士專業(yè)：信號與信息處理指導教師：盛立東20070308HandwrittenDigitRecognitionBasedonBPNeuralNetworkandSVMABSTRACTH柚曲rincnDj百tRec0印itionisam血脅chofopticalChmcterRccognjti加1tisoneofthemostd勰sicsubjcc

2、tsinpattemre∞gtIitionre∞afch，卸d徹eofthemost蛐cccssfIIl印pljcatio璐nh笛been印pliedinmany6eldsDigitReoo罟面ti衄h勰fewpattems，wJlichhelpsto蛆alyzc卸dverifysome眥wtheod髂觚dmethodsFunllcrmorc，themethodsinh柚dwrittend蟛t蚴gnitionameasytogcnera

3、ljzein∞merclativcpmblems’Ihe佗fo陀，itissi印ifi伽ttomal【edc印rcsearchcsoniththjspapcrah柚dwrinendjgn耵ecognitionsystembascd帆BPncuralⅡet’Ⅳork柚dSupportVccIorMachmc(sⅥ田培bumup，whichismade叩of伽oparts，includinglcaming舡Idre∞gnitionDurin

4、gthe刪izingofthesystem，thefonowingw勰d咖eintllispapcL1Make∞meimp0vements仰BPtrainingaIgoritllmt0quick鋤thenet、)lrorkco璐婦gcncyspecdandavoid伍kc刪[IImtionph如omen叩2Probabilitys詛tisti岱ofmi虻l罄si6cati∞pane儺is百伽in恤leamingph蛾ofBPNcunlN

5、ct、】lrorksmsystcmwiⅡsct∞mec0：Ilfidenccp踟ct哪accord噸tothisProbabilitystatisti岱hre∞鰣ti∞ph撇，mcsystemwilldccidcifmc伽tputofBPneuralne啊orksneedtop獬theSVMaccordiⅡgtotllis∞娟den∞pa衄etcL3ApplySM0algorithmandimpfoVcDir鰍SVMalgomminth

6、cle帥ingph螄ofSvlⅥF∞theSVMspcnds10ngtim囂in仃ainillgthel雄e蛐plcsct’weapplicdDircdSVMalgorithm柚dilnprovcditAndwegct柚appmximatesupportVecIor∞tinmuchshortcrtimcThenwccan璐etllisvector辯ttotraintheSVMbySM0a190ritllIn，avoidingdcprc輜

7、ofthewholepcrfb肋柚cc0fthesystcmhthjswaywegctaf笛terSvMda輻i丘eL4Ah粕dwrittcndjgjtrc∞印iti仰systcmoft帥lcvelsisbuiltupiIIthispapcL1nheBPNeuralNet、ⅣOrl【sserves勰thefirstlcvelcla跖jfieLandtheSVMserv鼯嬲Ihesecondlevelcl笛s塌eLⅡthere∞gniti