无源声探测目标跟踪与分类递推方法研究

杭州电子科技大学硕士学位论文无源声探测目标跟踪与分类递推方法研究姓名：申屠晗申请学位级别：硕士专业：控制理论与控制工程指导教师：薛安克;郭云飞20091101I(1)(2)(3)(4)IIABSTRACTPassiveacousticdetectionsystem(PADS)utilizesacousticsensorstoreceivebearing-onlydatafortargettrackingandclassification.Comparedwithactiveradarsystem,ithastheadvantageofgooddetectingability,immunetoelectromagneticinterference,highbattlefieldsurvivability,lowcostandallweatheroperations.Therefore,thePADSisregardedasausefulsupplementfortheactiveradarsystem.However,challengeremainsandwearelookingforwardtotacklewiththenuisancesofnonlinearity,incompleteobservationandsignaltimedelaywhichmakethetrackingandclassificationproblemsmorecomplicated.Inthispaper,westudiedtheproblemsoftargetstrackingandclassificationforthePADS.Themainresultsandachievementsaregivenasfollows:(1)Unifiedmathematicalmodelsoftargetstateandclassattributesaswellastimedelaymeasurementswerebuiltforsingleandmultipletargetstrackingandclassificationproblem.(2)Bearing-onlyacousticdatacannotbeuseddirectlyfortheeffectsofsignaltimedelayinthePADS.Inordertomakethedelayedmeasurementsavailablefortracking,synthesizedmeasurementscouldbegetthroughposteriorlyoptimizingthejointmeasurementswhichderivedfromeverysensorsindifferentscansusingthemethodsofprobabilisticmodeling.Then,aTD-PMHTalgorithmcanbeconstructedbyembeddingaBayesiansmootherwithinExpectationandMaximization(EM)frameworkforsingletargettrackingproblem.(3)Anapplicationofthejointtrackingandclassificationalgorithmbasedonmixtureparticlefiter(MPF-JTC)inthePADSwasrealized.Furthermore,anewmixtureunscentedparticlejointtrackingandclassificationalgorithm(MUPF-JTC)wasproposedbyadoptingthemethodsofunscentedtransform(UT).SincebetterparticleproposaldistributioncanbecomputedbyembeddingunscentedKalmanfilters(UKFs),theMUPF-JTCcanachievebettertrackingandclassificationresultsthanacommonMPF-JTCalgorithm.(4)Thedataassociationthatwasbroughtbymulti-targetfurthercomplicatesthetrakcingandclassificationproblemsinthePADS.Tosloveit,amulti-targettimedelayprobabilisticmultiplehypothesistrackingalgorithm(MTD-PMHT)andanothermulti-targettimedelayprobabilisticmultiplehypothesisjointtrackingandclassificationalgorithm(MTD-PMHT-JTC)wereproposedbasedontheTD-PMHTandMUPF-JTCalgorithmsaswellasthePMHTtechnology.TheMTD-PMHTcantackletheproblemofdataassociation.TheMTD-PMHT-JTCcanaccomplishtheclassificationworkofmulti-target.Simulationresultsconfirmtheefficiencyoftheproposedalgorithms.Keywords:passiveacousticsensornetworks,timedelaytracking,multipleprobabilistichypothesis,jointtrackingandclassification511.1[1,2]1234PALSGR-8SAR-2xxxxxxx1.261.2.1[3,4][8,9][5][6,7](weightedleastsquares,WLS)[11](PLE)[12](maximumlikelihoodestimate,MLE)[13,14](modifiedinstrumentalvariable,MIV)[15,16]WLSPLEMLE60[17,18]1969Murphy(extendedKalmanfilter,EKF)[18]1983Aidala(modifiedpolarextendedKalmanfilter,MPEKF)[19]1995N.Peach7(rangeparametrisedextendedKalmanfilter,RPEKF)[43]EKFEKFRPEKFEKF1)2)1995Julier(unscentedtransform,UT)(unscentedKalmanfilter,UKF)[20-22]EKFUKF,2003B.RisticUKF[23]EKFUKF[24](PF)[25]2050(MonteCarlo)[26-28][34-37]1993Gordon(resampleing)[29][30-33]2000ArulampalamRisticPFRPEKFMPEKF[38]2003PF[39]2005KarlssonPFPF(marginalizedparticlefilter)RPEKF[40]2006Djuric(generalizedparticlefilter,GPF)PF[41]1.1[42-45]1.2[46-48]8α1.1αβ121.2340m/s1.39[49-51]α1.3[52-54][55-57]1kHz[48-60]()[61-62][63]1.2.2[103][104][[105][106-107]10[108][103]Cune[109][110-111][94,111]1.2.3[64]1995Miller(jointtrackingandclassification,JTC)1.4JTC[65]JTC[65-73,113]K-1K-1K-1KK-1KK1.4JTC11JTC[66,71-72]JTC90[66]1999LantermanJTC[67]A.FarinaJTC[68]JTCJTC[69,70]S.Maskell[70]2004B.RisticJTC[71](EKF,UKF,PF)2006D.AngelovaL.Mihaylova(mixtureparticlefilterjointtrackingandclassification,MPF-JTC)[72]JTC1.3123121322.1(MMSE)[101](EKF)[18-19](UKF)[20-22](PF)[24-28,30-40]2.2)())(()1(,1kvkxfkxkk+=++(2.1))(kxkkkf,1+)(kx)(kv)(kQ)(])()([kQkvkvET=⋅(2.2))())(()(kwkxhkzk+=(2.3))(kzkkh)(kx)(kw)(kR)}(,),2(),1({)(kzzzkZ⋅⋅⋅=k)(])()([kRkwkwET=⋅(2.4))(kR)0(∧x)0(ˆP)0(P)(kx∧)](|)([)(kZkxEkx≈∧(2.5)142.3(extendedKalmanfilter,EKF)(2.1)(2.2)EKFk)(kx∧)(kP1+k)1(+∧kx)(kx∧)k(v)]()([)())(())(()1(,1,1+−∂+≈+∧∧∧+∧+∧kxkxkxkxfkxfkxkkkk(2.6))(/))((,1kxkxfkk∧∧+∂))((,1kxfkk∧+))(()1(~,1kxfkxkk∧+=+(2.7)TkkkkkxkxfkPkxkxfkP))(/))((()(ˆ))(/))((()1(~,1,1∧∧+∧∧+∂⋅⋅∂=+(2.8)k(2.3))(kx∧)k(w))(~)(())(ˆ/))(ˆ(())1(~()1(+−⋅∂++≈+kxkxkxkxhkxhkzkk(2.9))(ˆ/))(ˆ(kxkxhk∂))(ˆ(kxhk))1(~()1(~+=+kxhkzk(2.10)1+k))1(~)1(()1(~)1(ˆ+−+++=+kzkzKkxkx(2.11)1+k)1(~)(ˆ/))(ˆ()1(~)1(ˆ+⋅∂⋅−+=+kPkxkxhKkPkPk(2.12)K1))1())(ˆ/))(ˆ((()1(~−++∂⋅+=kRkxkxhkPKTk(2.13)2.4EKF[81]Julier15(unscentedKalmanfilter,UKF)δnxxxPmx)(xfz=(2.14)xx12+nδfγzzPδ++=⋅+−==⋅++==nnniPnxniPnxxixiixi2,...,2,1,))((,...,2,1,))((0λδλδδ(2.15)n⋅=2αλα=+=+=ninni2,...,2,1,)/(5.0)/(0λωλλω(2.16)γnifii2,...,1,0,)(==δγ(2.17)∑=⋅=niiiz20γω(2.18)TiniiizzzP)()(20−⋅−⋅=∑=γγω(2.19)TiniiixzzxP)()(20−⋅−⋅=∑=γδω(2.20)UKF)(ˆkx)(ˆkP)1(ˆ+kx)1(ˆ+kPkδ()()0ˆ()ˆˆ()()(),1,2,...,ˆˆ()()(),1,2,...,2kikiikixkxknPkinxknPkinnnδδλδλ==++⋅==−+⋅=++(2.21)16δ++−⋅+−⋅=+⋅=+==∑∑=++=+++niikkikkiniikkiikkkikkkQkxkxkPkxnif20|1|120|1,1|1)())1(~()