Investigation on SAR ground moving target imaging under sparse Bayesian learning framework

Abstract

In this paper, we investigate ground moving target imaging (GMTIm) by synthetic aperture radar (SAR) under sparse Bayesian learning (SBL) framework. To automatically determine the parametric dictionary used in the framework, an novel time-frequency representation method, known as Lv's distribution (LVD), is adopted, which is superior to represent multiple moving targets on the Doppler centroid frequency and chirp rate (CFCR) domain. A remarkable advantage of the SBL formulation is that a full posterior distribution can be provided for the SAR moving target image, instead of a simple point estimate as in the reported conventional methods. High order statistical information can be therefore exploited, and the imaging performance in terms of accuracy can be accordingly enhanced. To achieve an efficient Bayesian inference for the SBL implementation, an emerging technique, variational Bayesian expectation maximization (VB-EM), is employed. Both additive and multiplicative perturbations are considered in the SBL formulation, which improves the applicability of the proposed algorithm in practice. Simulation with isotropic point targets is presented to validate the effectiveness and superiority of the proposed algorithm.