模式识别中的支持向量机方法82007.pdfVIP

37 5 浙　江　大　学　学　报( ) Vol 37 No 5 2003 9 J o ur nal o f Zhejiang U niv ersity( Eng ineer ing Science) Sep 200 3 杜树新, 吴铁军 ( , , , 3 10027) : , , - SV M SV M SVM SV M , , , M -a ry , : ; ; : T P 181: A : 1008-973X( 200 3) 05 -052 1-07 Support vector machines for pattern recognition DU Shu -x in , WU T ie-jun ( N ational K ey Laboratory of I ndustrial Control Technology , I nstitute of I ntelligent Sy stems and D ecis ion M ak ing , Zhej iang Uni ersity , H angz hou 310027 , China) Abstract: A n ov erv iew o f t he basic ideas under lying Support V ect or M achine ( SV M ) f or pat t ern recog ni- tio n w as given N ew met hods su ch as -SV M , L S-SV M , w eight ed SV M and direct SV M , t raining alg o- rithm s incl uding chunking m et hod and decom posit io n met hod f or t he sake of f ast comput at ional sp eed and ease of im plement at ion , model selection approaches minim izing t he g eneralization error, and mult iclas s clas sif icat io n met ho ds such as o ne-again st -t he-rest met ho d, one-ag ain st-o ne method , M -ar y clas sification w er e concentr at ed Finally, an ex ampl e of mult iclass classif icat ion fo r mo nit oring operat ion st at us of w ast ew at er treat ment pr ocesses w as given A s a direct im plementat ion of t he str ucture risk minim iz ation ( SRM ) induct iv e principle, SV M pr ovides go od perf ormances su ch as g lob al opt im ization and goo d gener- aliz ation abilit y : ; ; Key words suppor t vect or mac