AnewfilteringmethodforSARdatade-noisingusingwaveletsupportvectorregression(WSVR)isdeveloped.OnthebasisofthegreyscaledistributioncharacterofSARimagery,thelogarithmicSARimageasanoisepollutedsignalistakenandthenoisemodelassumptioninlogarithmicdomainwithGaussiannoiseandimpactnoiseisproposed.Basedonthebetterperformanceofsupportvectorregression(SVR)forcomplexsignalapproximationandthewaveletforsignaldetailexpression,thewaveletkernelfunctionischosenassupportvectorkernelfunction.ThenthelogarithmicSARimageisregressedwithWSVR.Furthermoretheregressiondistanceisusedasajudgmentindexofthenoisetype.Accordingtothejudgmentofnoisetypeeverypixelcanbeadaptivelyde-noisedwithdifferentfilters.Throughanapproximationexperimentforaone-dimensionalcomplexsignal,thefeasibilityofSARdataregressionbasedonWSVRisconfirmed.AfterwardtheSARimageistreatedasatwo-dimensionalcontinuoussignalandfilteredbyanSVRwithwaveletkernelfunction.Theresultsshowthatthemethodproposedherereducestheradarspecklenoiseeffectivelywhilemaintainingedgefeaturesanddetailswell.
