An Improved Structured-Dictionary-Learning-Based Sparse SAR Imaging Algorithm

Abstract

Existing structured-dictionary-learning (SDL)-based sparse synthetic aperture radar (SAR) imaging methods suffer from insufficient artifact suppression, limited parameter adaptivity, and inadequate structure preservation during refined imaging under low-sampling-rate conditions. To address these limitations, an improved SDL-based sparse SAR imaging method is proposed, which integrates an iteratively reweighted ℓ1-norm (IRL1) constraint with a total-variation (TV) regularization safety mechanism. Within the framework of structured dictionary learning, sub-dictionary discrimination, and refined imaging, the proposed method focuses on improving the refined-imaging procedure. Specifically, an IRL1 sparsity constraint is introduced to enhance the representation of dominant scattering components; a regularization-parameter continuation strategy is adopted to improve early-stage optimization stability and late-stage detail recovery; an adaptive parameter-updating mechanism driven by the ratio between the measurement residual and the dictionary residual is designed to improve reconstruction adaptivity; and a TV-regularization safety mechanism with objective-function backtracking protection is incorporated to suppress background artifacts and local noise while preserving target structural edges. Experiments on structured simulation scenes and the real Moving and Stationary Target Acquisition and Recognition (MSTAR) SAR target dataset show that, compared with the K-singular value decomposition (K-SVD) method and the original SDL method, the proposed method achieves better imaging reconstruction performance at different sampling rates and provides stronger structure-preserving and artifact-suppression capabilities. These results demonstrate the effectiveness of the proposed method for low-sampling-rate SAR target imaging.