Three-dimensional Imaging Method of Target Based on Time-domain Sparse Representation of Multi-view SAR Data

Abstract

Three-dimension (3-D) images provide additional information of targets for automatic target recognition (ATR) and 3D scattering model generation. Methods based on sparse representations can reconstruct extreme resolution 3D images from sparse measurements, but suffer from the huge dimension of separable dictionaries. This paper presents a time-domain sparse representation method for 3-D target imaging from multi-view synthetic aperture radar (SAR) data, including a basic method and two improved ones. The time-domain framework uses time-domain responses to build a separable dictionary and a sparse representation model. In the time-domain framework, the basic approach is to transform the dictionary into a rather sparse matrix via a low-energy threshold that shrinks the spatial region of the 3D imaging based on multi-aspect 2D images. By exploiting the properties of multi-aspect SAR data in the time domain, one modification makes the sparse representation model more compact, leading to a reduction in dimensionality, and another additional modification splits a high-dimensional large-scale model into a set of very low-dimensional small-scale models. They overcome the curse of dimensionality and improve the efficiency of sparse representation-based 3D imaging to varying degrees. Experimental results show the effectiveness and great efficiency of the proposed method.