Featurereductionisakeyprocessinpatternrecognition.Thispaperdealswiththefeaturereductionmethodsforatime-shiftinvariantfeature,powerspectrum,inRadarAutomaticTargetRecognition(RATR)usingHigh-ResolutionRangeProfiles(HRRPs).Severalexistingfeaturereductionmethodsinpatternrecognitionareanalyzed,andaweightedfeaturereductionmethodbasedonFisher'sDiscriminantRatio(FDR)isproposedinthispaper.AccordingtothecharacteristicsofradarHRRPtargetrecognition,thisproposedmethodsearchestheoptimalweightvectorforpowerspectraofHRRPsbymeansofaniterativealgorithm,andthusreducesfeaturedimensionality.Comparedwiththemethodofusingrawpowerspectraandsomeexistingfeaturereductionmethods,theweightedfeaturereductionmethodcannotonlyreducefeaturedimensionality,butalsoimproverecognitionperformancewithlowcomputationcomplexity.Intherecognitionexperimentsbasedonmeasureddata,theproposedmethodisrobusttodifferenttestdataandachievesgoodrecognitionresults.
