Radar Image Reconstruction Based on Compressive Sensing

Abstract

This dissertation discusses a study on radar image reconstruction using sparse recovery based on compressive sensing (CS) theory. The organization of this study is as follows. First, we compare the performance of sparse recovery algorithms (SRAs) for the reconstruction of a two-dimensional (2D) inverse synthetic aperture radar (ISAR) image from incomplete radar-cross-section (RCS) data. The three methods considered for the SRA include the basis pursuit (BP), the basis pursuit denoising (BPDN), and the orthogonal matching pursuit (OMP) methods. The performance of the methods in terms of the reconstruction accuracy of the ISAR image is compared using the incomplete RCS data. In addition, traditional interpolation methods such as nearest-neighbor interpolation (NIP), linear interpolation (LIP), and spline interpolation (SIP) are applied to the incomplete RCS data to reconstruct ISAR images, and their performance is compared to that of the SRAs. Consequently, the relaxed constraint rather than the strictly equality constraint for the SRA is more adequate for SRA applied to radar imaging. Next, we proposes a one-dimensional (1D) scattering center extraction (SCE) method that includes mainly two steps: coarse estimation of scattering centers using the iteratively reweighted least square (IRLS) coupled with a peak-finding algorithm, and fine estimation of scattering centers using the OMP procedure from the adaptively sampled Fourier dictionary. Consequently, the proposed method can achieve high SCE accuracy regardless of whether data are missing in the collected RCS dataset or not. Next, we also propose a 2D SCE method using sparse recovery based on CS theory regardless of whether data are missing in the received RCS data or not. The proposed method first generates a 2D grid with adaptive discretization that has a much smaller size than the fully sampled fine grid. As a result, this adaptive grid generation enables us to significantly reduce the computational complexity of the proposed method. Then, coarse estimation of 2D scattering centers is implemented using the IRLS coupled with a general peak-finding algorithm. Finally, fine estimation of 2D scattering centers is performed using the OMP procedure from an adaptively sampled Fourier dictionary. Measured RCS data as well as simulation data using the point scatterer model are used to evaluate the 2D SCE accuracy of the proposed method. The results show that the proposed method can achieve high SCE accuracy for both the complete RCS dataset in a noisy environment and the incomplete RCS dataset with missing data compared with the conventional OMP and existing Fourier-transform (FT)-based discrete spectral estimation (DSE) techniques such as CLEAN and RELAX. Finally, we develop a new method based on an OMP-type group-searching scheme for bistatic ISAR (Bi-ISAR) imaging using incomplete bistatic RCS (Bi-RCS) datasets with missing data. If the conventional FT-based algorithm is applied to the incomplete dataset, the resultant Bi-ISAR image is usually corrupted, leading to severe deterioration in image quality. To overcome this problem, the parameters that are related to the bistatic angle in the Bi-ISAR signal model are estimated suboptimally by using a combination of the OMP-type searching scheme and rank-based group selection. Next, a clear Bi-ISAR image, without any corruption caused by the missing data, can be obtained from the reconstructed Bi-ISAR signal using the estimated parameters. To validate the reconstruction capability of the proposed method, bistatic scattered field data using the physical optics (PO) technique as well as the point-scatterer model are used for Bi-ISAR image reconstruction. The results show that the proposed method can give high reconstruction accuracy for the incomplete Bi-RCS dataset compared to conventional reconstruction methods.