基于脉冲耦合神经网络的空中扩展目标检测

1912Vol.19,No.12200712HIGHPOWERLASERANDPARTICLEBEAMSDec.,2007:100124322(2007)12220112063,,(,610054):,,,,,,,,:;;;;:TP391:A(PCNN),PCNN,[1],PCNN[2]PCNN,[328][9][10][11212],PCNN,PCNN[4,13],,,,[14],[15]PCNN,PCNN:;(T);()(N);PCNN,,,,PCNN11.1PCNNFig.1SinglenervecellmodelcomposingPCNN1PCNNPCNN,Eckhorn[1]PCNN1Fij[n]i,jn,Iij,Lij[n],,,Tij[n],Yij[n]PCNNPCNNIijLij[n];Uij[n];Uij[n]Tij[n],PCNNYij[n]1,PCNNPCNNFij[n]=exp(-F)Fij[n-1]+F6MijklYkl(n-1)+Iij(1)3:2007204223;:2007211213:(20060112116);:(1966),,,,,;[e-mail][n]=exp(-L)Lij[n-1]+L6WijklYkl(n-1)(2)Uij[n]=Fij[n](1+Lij[n])(3)Yij[n]=1,Uij[n]Tij[n]0,otherwise(4)Tij[n]=exp(-T)Tij[n-1]+T6Ykl(n-1)(5):Mijkl,Wijkl;;PCNN,,Fij[n]=Iij(6)Lij[n]=L6WijklYkl(n-1)(7),Iij,0,,UijIij,Tij,Uij,(),,Yij=1,,Tij,,Tij,TijUij,,,(Lij)(),,,,1.2PCNNPCNN,PC2NN,PCNN,,Lij[n]=L6WijklYkl(n-1),PCNN:,L=1,,L=033,K,Y,K=111111111,L=(YáK)(8):á;Mijkl,Wijkl,F,L2PCNN2.1PCNN,;,[425],[7],,,(),(,),,;(),,,PCNN,,,,,nmax,nmax,nmax,,,nmax,,,,2.222.3PCNN2.3.1210219Fig.2Flowchartoftheproposedalgorithmfordetectingtargets2PCNNUij[n]=Fij[n](1+Lij[n])=Fij[n]+Fij[n]Lij[n],L,,(F),(Fij[n]Lij[n]),,,,,Uij[n]=Fij[n]+Lij[n](9)2.3.2PCNNTij((5)),,,,,,,,,Tij[n]=Tij[n-1]-+T6Ykl(n-1)(10)2.3.3,[425],0.05:(1),;(2),,,,=0.3,,,:1/,,,,:=1,1=0.7,10.7;2,0.70.4;3,0.40.1;4,0.10;5,0.80.5;6,0.50.2;7,0.20,1/,=0.2,0.70.5,;,01,,,,2.3.4PCNN,,:,(,),,,,,:Yij[n]=255-Yij[n],,,:(),,,2.3.5=0.3,10,,,10,,,102.4PCNN,PCNN,,,,PCNN,,,310212:2.4.1PCNN,,,,2.4.2PCNN,,,,,,,,:(1),,(),;(2)(),,2.4.3[16],,():,,,,,2.4.4Hough,:(1),;(2);(3);(4),,[17]S=Ps-PeLen(Ps-Pe)(11):Ps-PePsPe;Len(Ps-Pe)PsPe;S,S,SSth,,,Lth3,VisualC++PCNN:=0.2;=0.3;N=10;Lth=15;Sth=0.913,4(),5PCNN,6,,,PCNN,,,,Fig.3Initialimagescontainvariousaerialextendedtargetsundercomplexbackground3410219Fig.4IterativeresultsusingPCNNwithcross2entropyandmaximumedgenumbers4PCNNFig.5Optimaliterativeresultsusingunit2linkingPCNN5PCNNFig.6Finalresultsfortargetsdetectionusingtheproposedmethods6,,,,4PCNN,,,,,,,,PCNN:[1]EckhornR,ReitboeckHJ,ArndtM,etal.Featurelinkingviasynchronizationamongdistributedassemblies:simulationofresultsfromcatcortex[J].NeuralComput,1990,2(3):2932307.510212:[2]JcaufieldH,KinserJM.FindingshortestpathintheshortesttimeusingPCNNs[J].IEEETransNeuralNetworks,1999,10(3):6042606.[3]LiuX,WangDL.Rangeimagesegmentationusingarelaxationoscillatornetworks[J].IEEETransNeuralNetworks,1999,10(3):5642573.[4]GuXD,GuoSD,YuDH.Anewapproachforautomatedimagesegmentationbasedonunit2linkingPCNN[C]//ProceedingofIEEEtheFirstInternationalConferenceonMachineLearningCybernetics.2002:1752178.[5]MaYD,LiuQ,QingZB.AutomatedimagesegmentationusingimprovedPCNNmodelbasedoncross2entropy[C]//ProceedingofIEEE2004InternationalSymposiumonIntelligentMutimedia,VideoandSpeechProcessing.2004:7432746.[6]SkourikhineAN,PrasadL,SchleiBR.Neuralnetworkforimagesegmentation[C]//ProceedingsofSPIE.2000,4120:28235.[7],,.PCNN[J],2005,33(7):134221344.(ZhaoSJ,ZhangTW,ZhangZH.AstudyofanewimagesegmentationalgorithmbasedonPCNN.AataElectronicaSinica,2005,33(7):134221344)[8],,.PCNN[J].,2005,32(5):93296.(WangHM,ZhangK,LiYJ.Imagesegmen2tationmethodbasedonPCNN.Opto2ElectronicEngineering,2005,32(5):93296)[9],,.PCNN[J].,2005,24(4):2912295.(FangY,QiFH,PeiBZ.PCNNimplementationandapplicationsinimageprocessing.JInfraredandMilli2meterWaves,2005,24(4):2912295)[10]XuBH,ChenZ.AmultisensorimagefusionalgorithmbasedonPCNN[C]//ProceedingsofIEEEthe5WorldCongressonIntelligentControlandAutomation.2004:367923682.[11]RanganathHS,KuntimadG.Objectdetectionusingpulsecoupledneuralnetworks[J].IEEETransNeuralNetworks,1999,10(3):6152620.[12]ChenJ,KoseiI,WadaM.Movingobjectextractionusingmulti2tieredpulse2coupledneuralnetwork[C]//ProceedingofIEEESICEAnnualConference.2004:284322848.[13]GuXD,ZhangLM,YiDH.Generaldesignapproachtounit2linkingPCNNforimageprocessing[C]//ProceedingsofIEEEInternationalJointConferenceonNeuralNetworks.2005:183621841.[14],,,.[J].,2004,16(3):2812285.(PengZM,ZhangQH,WeiYX,etal.Imagematchingbasedonmulti2featuresfusion.HighPowerLaserandParticleBeams,2004,16(3):2812285)[15],,,.Mumford2Shah[J].,2007,19(4):5662570.(FangL,LuJJ,YeYT,etal.NewmethodforedgedetectionofinfraredimagesbasedonMumford2Shahmodel.HighPowerLaserandParticleBeams,2007,19(4):5662570)[16].[J].,1994,(10):49252.(LuZQ.Contourtrackinginimageprocessinganditsap2plication.ChinaComputerUsers,1994,(10):49252)[17],,.[J].,2006,(2):2562258.(SunH,RenMW,YangJY.Afastandpracticalalgorithmforlinedetection.ComputerApplication,2006,(2):2562258)Aerialextendedtargetdetectionbasedonunit2linkingpulsecoupledneuralnetworksPENGZhen2ming,JIANGBiao,XIAOJun(SchoolofOpto2ElectronicInformation,UniversityofElectronicScienceandTechnologyofChina,Chengdu610054,China)Abstract:Anovelmethodofaerialextendedtargetdetectionbasedonpulsecoupledneuralnetwork(PCNN)ispresented.Somekeytechniquesareimprovedandnetworkmodelissimplified,whicharemoresuitableforimageprocessing,suchaslinearmodulation,linearattenuationofdynamicthresholdandstepselection,anditerationtimecontrol.Meanwhile,maximumlinecon2tourpointnumberandcolornegativeareadoptedtodeterminetheoptimal