人工神经网络杂波抑制应用研究(IJCNIS-V4-N10-6)

I.J.ComputerNetworkandInformationSecurity,2012,10,55-62PublishedOnlineSeptember2012inMECS()DOI:10.5815/ijcnis.2012.10.06Copyright©2012MECSI.J.ComputerNetworkandInformationSecurity,2012,10,55-62ApplicationofArtificialNeuralNetworkforClutterRejectionPriyabrataKarmakar1,SouravDhar2,MithunChakraborty3,TirthankarPaul41,3Electronics&CommunicationEngineeringDepartment,SurendraInstituteofEngineering&Management,Siliguri,India2Electronics&CommunicationEngineeringDepartment,SikkimManipalInstituteofTechnology,Sikkim,India4Electronics&CommunicationEngineeringDepartment,SwamiVivekanadaInstituteofScience&Technology,[e-mail]—ThispaperdealswithapplicationofArtificialNeuralNetwork(ANN)forRadarClutterrejection,functionapproximationmethodofsupervisedANNisappliedhereusingbackpropagationalgorithm.ThedatabaseusedfortrainingandtestingtheANNhasbeencollectedfromsimulatingamovingvehicleinMATALAB(version7.9)toobtaintheRCSvaluesatrangeandcross-rangeprofiles.Thisworkisvalidatedbycomparingreceivedsignalafterclutterrejectionwiththereceivedsignalinnocluttercondition.IndexTerms—ArtificialNeuralNetworks,Radar,ClutterI.INTRODUCTIONRadarreturnsareproducedfromnearlyallsurfaceswhenilluminatedbyradar.Therefore,incompetitionwiththereturnfromatarget,therearemanysourcesofunwantedsignals.Unwantedsignalsinsearchradararegenerallydescribedasnoiseandclutter.Clutteristhetermusedandincludesgroundreturns,seareturns,weather,buildings,birdsandinsects.WeiYouetal[4]proposedafastclutterrejectionmethodforultrasoundcolorflowimagingbasedonthefirst-orderperturbationasanefficientimplementationofeigen-decompositionandthemethodwasverifiedbysimulateddata.R.Vicen-Buenoetal[5]proposedanestimationoftheshipsizeusinganANN-basedclutterreductionsystemfollowedbyafixedthreshold,highclutterreductionrateswereachievedusing1-dimensional(horizontalorvertical)integrationmodes,althoughinaccurateshipwidthestimationswereachieved.Theseestimationswereimprovedusinga2-dimensional(rhombus)integrationmode.ProposedsystemwascomparedwithaCA-CFARsystem,denotingagreatperformanceimprovementandagreatrobustnessagainstchangesinseaclutterconditionsandshipparameters,independentlyofthedirectionofmovementoftheoceanwavesandships.P.KVermaetal[6]appliedthroughwallimaging(TWI)techniqueforclutterrejection,firstlyobserveddataarepreprocessedforimagingandthendifferenttypesofclutterreductiontechniqueslikePrincipalComponentAnalysis(PCA),IndependentComponentAnalysis(ICA),FactorAnalysis(FA)andSingularValueDecomposition(SVD)havebeenapplied,andresultswereanalyzed.Signaltonoiseratio(SNR)ofthefinalimageshadbeencomputedtocomparetheresultsandknowtheeffectivenessofindividualclutterremovaltechniques.S.LDurdenetal[7]foundthatDopplerfilteringcansignificantlyreducethesurfacereturn,bringingsurfacecluttertoacceptablelevels.KeerthiRametal[8]presentedanewapproachforautomaticmicroaneurysmdetectionfromdigitalcolourfundusimages,theyformulateMAdetectionasaproblemoftargetdetectionfromclutter,wheretheprobabilityofoccurrenceoftargetisconsiderablysmallercomparedtotheclutter.WeiYouetal[9]proposedaclutterrejectionmethodbasedontherecursiveeigen-decompositionalgorithm.Inthismethod,thecurrenteigenvectormatrixoftheultrasoundechocorrelationmatrix,whichwillbeusedtoconstructthecluttersubspace,isdeterminedbypreviouseigenvectormatricesandthecurrentinput.JinYonggaoetal[10]proposednewclutterrejectionmethodnamed“nonlinearprojection”toimprovetheSNRofthetargetasseacluttermaskssomeweaktargetsignals,cluttermodelingisdoneasanonlineardeterministicdynamicalsystem.Afterapproximatingthemultidimensionalreconstructionoftheclutterbyalow-dimensionalattractor,projectionsontothisattractorcanseparatetheclutterfromothercomponents.Realseaclutter,simulatedtargetdataandrealtargetdataareusedtoshowthatanonlinearclutterrejectionmethodisapromisingtechniquetosuppressseaclutterandenhancestargetdetection.Thedefinitionofclutterdependsonthefunctionoftheradar.Sincevehiclesusuallymovemuchfasterthanweatherorsurfaceslowmovingtargets,velocity-sensitiveradarcaneliminateunwantedclutterfromtheradarindicator.InRadarsystemsArtificialNeuralNetworkcanbeappliedtorejectunwantedClutterbyapplyingtheconceptsofSupervisedartificialNeuralnetworkusingBack-Propagationasthelearningalgorithm.RestofthepaperisorganizedassectionIIdescribeshowdatabaseismadefortheexperiment,insectionIII,IVandVdetailsofArtificialNeural56ApplicationofArtificialNeuralNetworkforClutterRejectionCopyright©2012MECSI.J.ComputerNetworkandInformationSecurity,2012,10,55-62Network,ANNarchitectureandpreandpostprocessingofdataaregivenrespectively,CluttermodelandClutterrejectiondescribedinsectionVIandsectionVIIrespectively,insectionVIIIresultsaregivenandfinallysectionIXconcludesthepaper.II.DATABASETocollectdatafortrainingthenetworkwehavesimulatedalightvehicleusingtheirstructureinMATLABtogettheRCSvaluesusingthebelowfigure,Fig1.LightVehicleAbovefigureisthesideviewofthecorrespondingvehicle.wecanconsiderthissideviewimageofthetargetasthecombinationofseveralrectangular,triangularandcircularflatpl