The Application Of Support Vector Machine In Electromagnetic Inverse Scattering Of The Rough Surface With/Without Target

The application of Support Vector Machine (SVM) in electromagnetic (EM) inversescattering from the one-dimensional (1-D) rough (sea) surface with or without target isstudied in this paper. SVM is applied to inversion of the radius and dielectric constant of2-D cylinder, windspeed above1-D sea surface, the dielectric constant, the root meansquare (RMS) height and correlation length of the rough (sea) surface, as well as theheight and radius of2-D cylinder above1-D rough (sea) surface. The main works are asfollows:The principle of SVM method and the algorithm in the classification problems andregression problems are given. The principle of Method of Moments (MoM) isintroduced, and the applications of MoM in the EM scattering from1-D conducting ordielectric rough (sea) surface with or without2-D target are described, which includeestablishment of integral equations, the realization of incident tapered wave and thecalculation of the bistatic scattering coefficient.The EM inverse scattering problems of2-D cylinder and1-D rough (sea) surface areinvestigated based on SVM method, the inversion results for different polarization arepresented. With increase of nonlinearity for the curve of scattering coefficient in theprocess of retrieving the parameters, it is difficult to achieve good inversion results ifthe scattering coefficient is calculated with a single incident frequency. Thus, the vectorthat consists of the scattering coefficients produced in multi-frequency are considered tobe feature value, which is proved to be effective.The problem of EM inverse scattering is examined, which mainly covers the retrievalof height and radius of the2-D cylinder above1-D rough (sea) surface. SVM is appliedto inversion of the height and radius of2-D cylinder in different frequencies, differentincident angles and different polarization modes. Comparing the different inversionresults under different conditions, the influences of incident frequency, the incidentangle, as well as the polarization mode on the inversion results are analyzed.