High Resolution Isar Imaging Based On Bayesian Nonparametrics

At present,the traditional high-resolution ISAR imaging methods based on fourier analysis have been relatively mature,which could obtain well-focused imaging results of stationary targets under the circumstance of high signal-to-noise ratio.However,the target and observation environment faced by ISAR are becoming increasingly complex.When the target is non-stationary,and the signal-to-noise ratio of the echo is low or there are losses of data in the echo,the performance of existing imaging methods will degrade rapidly or even completely fail,resulting in great difficulties in feature extraction and recognition based on ISAR images.Because the parameter space of bayesian nonparametric method can adapt to the change of observation data and has strong flexibility,a new idea is provided for solving the problem of high-resolution ISAR imaging under complex environment.This thesis is aiming at the problem of high-resolution ISAR imaging of space target under complex environment.Firstly,the ISAR observation model and the bayesian probabilistic graphical model are established,and the mechanisms of high-resolution ISAR imaging of space target based on bayesian nonparametric methods such as Beta process linear regression and Gaussian process regression are explored.Secondly,high-resolution ISAR imaging method of space target under strong noise environment based on the advanced bayesian learning method is studied.Finally,by building a graphical user interface(GUI),high-resolution ISAR imaging of typical targets under complex environment is achieved.The related research work of this thesis will provide theoretical and technical support for improving the ability of space target detection of imaging radars under complex conditions.The main work of this thesis is summarized as follows:In the first part,aiming at complex observation conditions such as low signal-to-noise ratio and data losses,a high-resolution ISAR imaging method based on Beta process linear regression is proposed.First,the sparse signal observation model of ISAR imaging is established.Then,the imaging model based on unidimensional Beta process linear regression is constructed,and the effective prior parameter selection method and model parameter solving method based on Gibbs sampling are proposed.On this basis,the unidimensional Beta process linear regression(UBPLR)model is improved,and the ISAR imaging model based on multidimensional Beta process linear regression(MBPLR)is proposed,and the optimal parameters of the model are determined.Finally,the effectiveness of the algorithm is verified by simulation and measured data.In the second part,aiming at the failure of traditional ISAR image denoising methods based on data domain and image domain under the condition of low signal-to-noise ratio,a RID image sequence denoising method of space target based on Gaussian process regression is proposed.Firstly,the Gaussian process regression model of ISAR image sequence is established.Secondly,the model parameters and kernel function parameters solving methods based on maximum likelihood estimation and gradient descent method are proposed.Finally,the effectiveness of the proposed method is verified by point simulation and electromagnetic simulation data,and the robustness of the method is verified by Monte Carlo experiments.In the third part,the user interface based on bayesian nonparametric high-resolution ISAR imaging is programmed.At first,the basic concepts,basic structure and design principles of the MATLAB GUI are introduced.Then,GUI is programmed by GUIDE,and instances of GUI for high-resolution ISAR imaging based on Beta process linear regression and RID image sequence denoising based on Gaussian process regression are given respectively.At last,high-resolution ISAR imaging and noise reduction processing of typical space target under complex environment are realized.