简介:BasedonthesinglesensorKalmanfilteringequations,thispaperpresentstwo-levelandthree-leveloptimalcentralizedanddistributedestimationalgorithmsforhierarchicalmultisensorsystems.Thesolutionshowsthatwhenthecorrelatedmatrix,themeanofnoise,thecontrolinput,andthemeasurementerrorareallzero,theresultinthispapertumsouttobethestandardalgorithmdiscussed.Simulationshowsthatthemeanofnoise,thecontrolinput,andthemeasurementerrorwillnotchangetheestimationcovarianceandtheestimationcovariancefluctuatesgreatlywhenthecross-correlatedmatrixissimilartothecovarianceofprocessnoise.
简介:信息熔化是在多模式的生物测定的系统的关键步。特征级的熔化是比由于原来的特征集合关于生物测定的数据包含更富有的信息的事实的分数级、决定级的方法更有效的。在这份报纸,我们在场当时,为特征级的多模式的生物测定的信息熔化的概括正规判别式设计(MGCDP)方法,它最大化intra班的关联展示的multiset最小化在班之间的关联。另外,连续MGCDP(S-MGCDP)和策略也是的平行MGCDP(P-MGCDP)求婚了,它能多于二种生物测定的信息熔化,以便完成更好的鉴定效果。实验在MGCDP方法超过的各种各样的生物测定的数据库表演表现了另外的最先进的特征级的信息熔化途径。
简介:Informationfusioninbiometricsystems,eithermultimodalorintramodalfusion,usuallyprovidesanimprovementinrecognitionperformance.Thispaperpresentsanimprovedscore-levelfusionschemecalledboostedscorefusion.Theproposedframeworkisatwo-stagedesignwhereanexistingfusionalgorithmisadoptedatthefirststage.Atthesecondstage,theweightsobtainedbytheAdaBoostalgorithmareutilizedtoboosttheperformanceofthepreviouslyfusedresults.Theexperimentalresultsdemonstratethattheperformanceofseveralscore-levelfusionmethodscanbeimprovedbyusingthepresentedmethod.
简介:Gabor特征被显示了为手掌静脉识别有效。这份报纸论述一个新奇特征表示方法,实现本地Gabor直方图(FLGH)的熔化,以便改进手掌静脉识别系统的精确性。称为本地Gabor主要差别模式(LGPDP)的一个新本地描述符用本地最大的差别(LMD)编码Gabor大小操作员。相应Gabor阶段模式被独占的本地Gabor编码或(XOR)模式(LGXP)。Fishers线性判别式(FLD)方法然后被实现减少特征表示的维数。低维的Gabor大小和阶段特征向量最后被熔化提高精确性。从自动化的研究所的试验性的结果,建议FLGH方法更好完成的中国科学院(CASIA)数据库表演由利用分数级的熔化的性能。相等的错误率(无论何时)是0.08%,它超过另外的常规手掌静脉识别方法(从2.87%~0.16%的无论何时范围),例如,拉普拉斯算符手掌,琐事特征,麻袋布阶段,Eigenvein,本地不变的特征,相互的前景本地人二进制代码模式(LBP),并且多采样特征熔化方法。
简介:DecisionfusionrulesforWirelessSensorNetworks(WSNs)underNakagamifadingchannelsareinvestigatedinthispaper.ConsideringtheapplicationlimitationofLikelihoodRatioTestfusionrulebasedoninformationofChannelStatisticsusingSeriesexpansion(LRT-CSS),andthedetectionperformancelimitationoftheCensoringbasedMixedFusionrule(CMF),anewLRTfusionrulebasedoninformationofchannelstatisticshasbeenpresentedusingLaplaceapproximation(LRT-CSL).TheoreticalanalysisandsimulationsshowthattheproposedfusionruleprovidesbetterdetectionperformancethantheCensoringbasedMixedFusion(CMF)andLRT-CSSfusionrules.Furthermore,comparedwithLRT-CSSfusionrule,theproposedfusionruleexpandstheapplicationrangeoflikelihoodratiotestfusionrule.
简介:Fiber-opticFabry-Perot(F-P)sensorsarewidelyinvestigatedbecausetheyhaveseveraladvantagesoverconventionalsensors,suchasimmunitytoelectromagneticinterference,abilitytooperateunderbadenvironments,highsensitivityandthepotentialformultiplexing.Anewmethodtofabricatemicro-cavityFabry-Perotinterferometerisintroduced,whichisfusionsplicingasectionofconventionalsingle-modefiber(SMF)andasectionofhollowcoreorsolidcorephotoniccrystalfiber(PCF)togethertoformamicro-cavityatthesplicejoint.Thetechnologyoffusionsplicingisdiscussed,andtwominiatureopticalfibersensorsbasedonFabry-Perotinterferenceusingfusionsplicingarepresented.Thetwosensorsarecompletelymadeoffusedsilica,andhavegoodhigh-temperaturecapability.
简介:Spectrumsensingisoneofthecoretechnologiesforcognitiveradios(CR),wherereliabledetectionofthesignalsofprimaryusers(PUs)ispreconditionforimplementingtheCRsystems.AcooperativespectrumsensingschemebasedonanadaptivedecisionfusionalgorithmforspectrumsensinginCRisproposedinthispaper.ThisschemecanestimatethePUpriorprobabilityandthemissdetectionandfalsealarmprobabilitiesofvarioussecondaryusers(SU),andmakethelocaldecisionwiththeChair-Varshneyrulesothatthedecisionsfusioncanbedonefortheglobaldecision.SimulationresultsshowthatthefalsealarmandmissdetectionprobabilitiesresultedfromtheproposedalgorithmaresignificantlylowerthanthoseofthesingleSU,andtheperformanceoftheschemeoutperformsthatofthecooperativedetectionbyusingtheconventionaldecisionfusionalgorithms.
简介:Basedonthetheoryoffuzzylogic,themethodofobfuscatingcoefficientandreliabilitytofusetheinformationofhandgeometryandpalmprintsforidentitydiscriminationisproposed.Theexperimentprovesthatthemethodisusefulandeffective.Itsidentificationrateisupto90%,whichis20%-30%higherthanthatofusinghandgeometryorpalmprintssingly,thusitcanbewidelyusedinhighlydemandedsecurityfield,suchasfinance,entranceguard,etc.
简介:AccordingtotheprincipleofpolarizationimagingandtherelationbetweenStokesparametersandthedegreeoflinearpolarization,therearemuchredundantandcomplementaryinformationinpolarizedimages.Sinceman-madeobjectsandnaturalobjectscanbeeasilydistinguishedinimagesofdegreeoflinearpolarizationandimagesofStokesparameterscontainrichdetailedinformationofthescene,thecluttersintheimagescanberemovedefficientlywhilethedetailedinformationcanbemaintainedbycombiningtheseimages.Analgorithmofadaptivepolarizationimagefusionbasedonregionalenergydynamicweightedaverageisproposedinthispapertocombinetheseimages.Throughanexperimentandsimulations,mostcluttersareremovedbythisalgorithm.Thefusionmethodisusedfordifferentlightconditionsinsimulation,andtheinfluenceoflightingconditionsonthefusionresultsisanalyzed.
简介:Ahierarchicalparticlefilter(HPF)frameworkbasedonmulti-featurefusionisproposed.TheproposedHPFeffectivelyusesdifferentfeatureinformationtoavoidthetrackingfailurebasedonthesinglefeatureinacomplicatedenvironment.Inthisapproach,theHarrisalgorithmisintroducedtodetectthecornerpointsoftheobject,andthecornermatchingalgorithmbasedonsingularvaluedecompositionisusedtocomputethefirstorderweightsandmakeparticlescentralizeinthehighlikelihoodarea.Thenthelocalbinarypattern(LBP)operatorisusedtobuildtheobservationmodelofthetargetbasedonthecolorandtexturefeatures,bywhichthesecond-orderweightsofparticlesandtheaccuratelocationofthetargetcanbeobtained.Moreover,abacksteppingcontrollerisproposedtocompletethewholetrackingsystem.Simulationsandexperimentsarecarriedout,andtheresultsshowthattheHPFalgorithmwiththebacksteppingcontrollerachievesstableandaccuratetrackingwithgoodrobustnessincomplexenvironments.
简介:Analgorithmofhighlymaneuveringtargettrackingisproposedtosolvetheproblemoflargetrackingerrorcausedbystrongmaneuver.Inthisalgorithm,anewestimator,namedasmulti-parameterfusionSinger(MF-Singer)modelisderivedbasedontheSingermodelandthefuzzyreasoningmethodbyusingradialaccelerationandvelocityofthetarget,andappliedtotheproblemofmaneuveringtargettrackinginstrongmaneuveringenvironmentandoperatingenvironment.ThetrackingperformanceoftheMF-Singermodelisevaluatedandcomparedwithothermanueveringtrackingmodels.ItisshownthattheMF-Singermodeloutperformsthesealgorithmsinseveralexamples.
简介:Inthispaper,weproposeadynamicmulti-descriptorfusion(DMDF)approachtoimprovingtheretrievalaccuracyof3-dimensional(3D)modelretrievalsystems.First,anindependentretrievallistisgeneratedbyusingeachindividualdescriptor.Second,weproposeanautomaticrelevant/irrelevantmodelsselection(ARMS)approachtoselectingtherelevantandirrelevant3Dmodelsautomaticallywithoutanyuserinteraction.Aweighteddistance,inwhichtheweightassociatedwitheachindividualdescriptorislearntbyusingtheselectedrelevantandirrelevantmodels,isusedtomeasurethesimilaritybetweentwo3Dmodels.Furthermore,adescriptor-dependentadaptivequerypointmovement(AQPM)approachisemployedtoupdateeveryfeaturevector.Thissetofnewfeaturevectorsisusedtoindex3Dmodelsinthenextsearchprocess.Four3DmodeldatabasesareusedtocomparetheretrievalaccuracyofourproposedDMDFapproachwithseveraldescriptorsaswellassomewell-knowninformationfusionmethods.ExperimentalresultshaveshownthatourproposedDMDFapproachprovidesapromisingretrievalresultandalwaysyieldsthebestretrievalaccuracy.
简介:Anup-taperedMach-Zehnderinterferometer(MZI)isproposedanddemonstrated.Byapplyingamodificationsettingofanexistingprogramofthefusionsplicer,twoup-tapersarefabricatedbypushingbothsidesofthefibertothemiddle.Itisfoundthatthemaximumextinctionratioreachesupto20dB.Withthefiberlengthof6.3-49cmbetweenthetwoup-tapers,thefreespectralrange(FSR)changesfrom10.7nmto1.3nm.Besides,thewavelengthofmaximumextinctionratioshiftstotheshorterwavelengthinthescopeoftensofnanometers,whiletheelongationchangesfrom0%to0.23%.
简介:Theethanolcontentinethanol-gasolineisrespectivelydetectedbythefirst-orderderivativeUV/visabsorptionspectrum,thefirst-orderderivativenearinfrared(NIR)absorptionspectrumandtheinformationfusionmethod.Thebackwardintervalpartialleastsquares(BiPLS)algorithmisusedasthefeatureextractionmethod,whichisestablishedbythepartialleastsquares(PLS)regressionmodel.Basedontheinformationfusiontheory,thelowleveldatafusion(LLDF)andmid-leveldatafusion(MLDF)modelsareestablishedbythefirst-orderderivativeUV/visandNIRspectra.Theanalytical-resultsarecomparedwiththerelatedtextualreferences.Thereby,thesingle-spectralmodelbasedonthefirst-orderderivativeNIRabsorptionspectrumhastheoptimalresults,whereR^2p=0.9991andRMSEP=-0.3245,whiletheLLDFaftervectornormalization(LLDF-VN2)istheoptimalmulti-spectralfusionmodel,whereR^2p=0.9983andRMSEP=-0.4982.Theproposedmethodcanbeusedtodetecttheethanolcontentinethanol-gasolinerapidlyandprovidesabetterchoiceforthecomponentdetectioninmixedoils.
简介:ThispaperproposesaPCAandKPCAself-fusionbasedMSTARSARautomatictargetrecognitionalgorithm.Thisalgorithmcombinesthelinearfeatureextractedfromprincipalcomponentanalysis(PCA)andnonlinearfeatureextractedfromkernelprincipalcomponentanalysis(KPCA)respectively,andthenutilizestheadaptivefeaturefusionalgorithmwhichisbasedontheweightedmaximummargincriterion(WMMC)tofusethefeaturesinordertoachievebetterperformance.Thelinearregressionclassifierisusedintheexperiments.Theexperimentalresultsindicatethattheproposedself-fusionalgorithmachieveshigherrecognitionratecomparedwiththetraditionalPCAandKPCAfeaturefusionalgorithms.
简介:Whitenoisedeconvolutionorinputwhitenoiseestimationproblemhasimportantappli-cationbackgroundsinoilseismicexploration,communicationandsignalprocessing.Bythemoderntimeseriesanalysismethod,basedontheAuto-RegressiveMovingAverage(ARMA)innovationmodel,underthelinearminimumvarianceoptimalfusionrules,threeoptimalweightedfusionwhitenoisedeconvolutionestimatorsarepresentedforthemultisensorsystemswithtime-delayedmeasurementsandcoloredmeasurementnoises.Theycanhandletheinputwhitenoisefusedfiltering,predictionandsmoothingproblems.Theaccuracyofthefusersishigherthanthatofeachlocalwhitenoiseestimator.Inordertocomputetheoptimalweights,theformulaofcomputingthelocalestimationerrorcross-covariancesisgiven.AMonteCarlosimulationexampleforthesystemwith3sensorsandtheBernoulli-Gaussianinputwhitenoiseshowstheireffectivenessandperformances.