Preserving 2-D Structural Features in ISAR Images via Arbitrary Pattern-Coupled Sparse Bayesian Learning

For inverse synthetic aperture radar (ISAR) imaging under incomplete data conditions including sparse aperture (SA) and limited bandwidth (LB), the high-quality focused ISAR images cannot be obtained by traditional methods. Although various sparse signal processing methods have been proposed to improve ISAR imaging quality, they often neglect the intrinsic structural features of these images. Conventional pattern-coupled (PC) sparse Bayesian learning algorithms attempt to capture these structures by constructing a PC prior. However, these priors are often sub-optimal and computationally expensive. This study proposes a two-dimensional (2D) arbitrary PC prior that captures the 2D structural features to obtain high-resolution ISAR images within a Bayesian framework. Model parameters are iteratively estimated using variational Bayesian inference. Computational efficiency is improved by employing relaxed evidence lower bound for inversion-free Bayesian inference in complex number field and exploiting the Kronecker structure of the observation matrix. The noise distribution is extended to gaussian mixture model to improve the robustness in real-world with complex noises. Experiments with both simulated and measured data demonstrate that the proposed algorithm better retains the structural features of ISAR images under incomplete data, low signal-to-noise ratio and complex noises distribution conditions with high computational efficiency.