一种基于旋转最小-最大超盒的聚类算法

200355:100026788(2003)052011320421,2,3(1.,410073;2.,430010;3.784,610051):,2L,2,,L,,,FCAL:2(RBCA);;:TP315:AANovelClusteringAlgorithmBased2onRotatedSuper2boxHUANGXiaobin1,MAXiao2yan2,PENGZhi2gang3(1.NationalUniversityofDefenseTechnology,Changsha410073,Chian;2.DapartmentofInformationEngineering,AirforceAcademy,Wuhan430010,China;3.MilitaryRepresentativeOfficeof784Factory,Chengdu610051,China)Abstract:Toresolvetheproblemthatthetraditionalclusteringalgorithmscannotdealwiththecomplicateddataset,anovelclusteringalgorithmbased2onrotatedsuper2boxispresentedinthispaper.Thisalgorithmfirstuniformsthedataset,thenpartitionsthesamplespacewithrotatedsuper2boxtogetaseriesofpreliminaryclasses,finallyfusesthepreliminaryclasseswiththefusiblefactorstoformtheclusteringresult.ThesimulationresultsshowthatthisalgorithmhasbetterclusteringperformancethanthetraditionalfuzzyC2meanalgorithmindealingwiththedatasetwhichhasunknownpatternnumberandcomplicateddistributionshapes.Keywords:clusteringalgorithmbased2onrotatedMin2Maxsuper2box(RBCA);super2box;fusiblefactor:2002203211:(1978-),,,,;(1962-),,1P.K.Simpson[1]-,(),,L[1],,,,L,[2],2,,,-L,,,,,,,,,©1995-2005TsinghuaTongfangOpticalDiscCo.,Ltd.Allrightsreserved.,L,,FCAL2RBCARBCA1L1RBCA2.1S={s1,s2,,sh,,sN}NM,shh(1FhFN);shihi(1FiFM)Z(1)S,SZshi=shimax1FhFN(shi)-min1FhFN(shi),1iM(1)shiShiZ2.2,Z,ZBj={7j(sûsS),Vj,Wj,Qj,bj(sh,Vj,Wj,Qj)}(2)Bjj;7j(sûsS),7j;Vj=(vj1,vj2,,vjM)(,);Wj=(wj1,wj2,,wjM)(,);Qj;bj(sh,Vj,Wj,Qj)sh(,),,bj(sh),:bj(sh)=mini=1,,M(min([1-f(shi-wji,Ci)],[1-f(vji-shi,Ci)]))(3),f(A,B)=1AB1AB0FABF10AB1;shish=Qjsh;Ci4ZMatlab:(,,);L=0;B=B0ööL,B,B0(0,1)fori=1:Nööi,Nifi=171={s1};V1=W1=(s11,s12);Qi=I;B1={71,V1,W1,Q1,bi(sh)};öö;L=L+1;öö1elesbT(si)=max1FjFLbj(sj;ööi,TendifbT(si)EB7T=7T{si};VT=(min(vT1,si1),min(vT2,si2));WT=(max(wT1,si1),max(wT2,si2));öö,C=cov(7T);ööMatlabcov7[QT,D]=eig(C);ööMatlabeigTQTBT={7T,VT,WT,QT,bT(sh)};ööT41120035©1995-2005TsinghuaTongfangOpticalDiscCo.,Ltd.Allrightsreserved.elseL=L+1;7L={si};VL=(si1,si2);QL=I;BL={7L,VL,WL,QL,bL(sh)};ööendend,B,,(B),Z,B,,,,Z,,,,B;,B,,Z,,,0.4-0.8Z2()2-,Z2.3,J,JZ:1(4)JD={dpdû1Fp,qFJ}Zdpd=dqp=max(max1FhFPbp(sh),max1FhFQbq(sh))(4)Pp,QqZ,0FdpqF1Z2JJ,rjZG=(R,D),R={r1,r2,,rJ)J,D1,dpqrprqZ(5),A={apqû1Fp,qFJ},apq=maxLAF(min(S))(5)Frprq,SFZ3apqp,qZKZ33.1,[3],[4]L3FCA(,34),,FCAL4RBCA,(),1LRBCA,,,2L3.23FCA,,FCA,,FCAL4RBCA,,,L1,51152©1995-2005TsinghuaTongfangOpticalDiscCo.,Ltd.Allrightsreserved.3FCA4RBFA,L3,,RBCAFCAL1RBCA121,234191018192324282RBCA12BK0.50.40.750.73FCARBCA12FCA33.3%8%RBCA004RBCA,(),,L:[1]PatrickKSimpson.Fuzzymin2maxneuralnetworks2part2:clustering[J].IEEETransactionsonFuzzySystems,1993,1(1):32-45.[2]FMFrattaleMasciilietal.Scale2basedapproachtohierarchicalfuzzyclustering[J].SignalProcessing,2000,80:1001-1006.[3]LiuYuTsengetal.AGeneticAlgorithmforDatawithNon-spherical-shapeclusters[J].PatternRecognition,2000,33:1251-1259.[4]KazuoHattori,MasahitoTakahashi.Aneweditedk2nearestneighborruleinthepatternclassificationproblem[J].PatternRecognition,2000,33:521-528.61120035©1995-2005TsinghuaTongfangOpticalDiscCo.,Ltd.Allrightsreserved.