FastICA Algorithm plus GAs
Hualin Liu · 2006
componentanalysis (ICA)is a statistical methoddeveloped fromtheseparation ofblind signal, andnowithasbeensuccessfully usedinmanyfields. In this paper, wepresent aneffective technique combined witha modifiedFastICA(M-FastICA) algorithm plusgenetic algorithms (GAs)forradarhigh-resolution rangeprofiles (HRRPs)feature extraction. Asweallknowthatthemosttimeconsuming courseinFastICAistocomputetheJacobian matrix. So inthismodified version, several iterations of FastICAaremergedintooneiteration butonlyneedsto computetheJacobianmatrixoncetime.Therebythe convergence velocity ofFastICAisaccelerated whilethe performance isnotdegraded. To demonstrate theabove feature extraction algorithm, theclassification experiment on three typesofradartargets areevaluated. First M-FastICA is applied toextract theindependent components fromthe HRRPs.ThenGAsisusedtoselect theoptimal basis vectors andthusbuild afeature subspace. Theresults showthatthe introduced method can achievebetterclassification performance thanbothPCAandICA.