Three-dimensional Parametric Scattering Character Modeling And Feature Extraction Of High-resolution SAR Targets

With the rapid development of the technology of Synthetic Aperture Radar(SAR)imaging and the increasing need of automatic SAR image interpretation,so the research on effective feature extraction method,towards SAR image automatic interpretation,is very important.The scattering center feature becomes the inherent character of high-resolution SAR target for the microwave illuminating imaging mode.Given that the mainstream of scattering feature extraction is based on parametric scattering center model at present,this paper studies the feature extraction method based on the three-dimensional parametric scattering center model of high-resolution SAR target.Modeling scattering character is fundamental the feature extraction of high-resolution SAR target based on the three-dimensional parametric scattering center model.In terms of how to model the target's scattering character in bistatic and three dimensions,this paper derives and modifies the 3-D attributed scattering center model of bistatic SAR target.The category scattering response,polarization scattering response and differential range in the general expression of 3-D attributed scattering center model of bistatic SAR target are derived carefully here.In terms of the category scattering response in the model,based on the analysis of planar scattering mechanisms from flat plate,right angle,circular and relationship of angles in SAR imaging,the category scattering response is divided into azimuthal and elevational planar scattering approximately.Simultaneously,its scattering amplitude is modified according to the practical peak Radar Cross Section(RCS)of the scattering centers.The experimental results verify that the model can fit scattering response of target well.The parameter estimation is the key in the feature extraction of high-resolution SAR target based on 3-D parametric scattering center model.In terms of the condition that the objective cost function has a lot of local optimum,this paper proposes a parametric estimation method combining parameter initialization and optimization.The scattering response of single scattering center in image domain and frequency domain is utilized to achieve the parameter initial value according to initial formula of scattering center location,size,orientation and amplitude during initialization.The Sequential Quadratic Programming(SQP),under the scattering response of single scattering center in frequency domain,is applied to implement parameter iterative optimization of scattering center according to non-coherent minimum cost objective function during optimization.The experimental results verify that the estimated value can approach the truth-value well.The structure selection is the important step in the feature extraction of high-resolution SAR target based on 3-D parametric scattering center model.In terms of the condition that structure selection can mix different scattering centers,the paper proposes a method that combines multiple classifiers to accomplish the structure selection of scattering center.Under the scattering response of single unknown scattering center and the estimated value of candidate scattering center,the category of scattering center is settled by the classification strategy which combines multiple classifiers.Further more,on the foundation of the proposed parameter estimation method and structure selection method,this paper accomplishes the feature extraction based on 3-D parametric scattering center model of high-resolution SAR target according to the RD-AML-CLEAN.Based on the feature extraction method,a software that extracts 3-D attributive scattering center features of high-resolution SAR target is developed here.The experimental results verify that the structure selection method can classify the scattering center effectively and the feature extraction method can extract scattering center 3-D feature well respectively.It also verify that the effectiveness of the parameter estimation and modeling method.