Auto-focused Sparse Bayesian Learning for ISAR Imagery Based on Spike-and-Slab Prior Via Variational Approximation

Weitian Sun, Lei Yang, Yuchen Dou, Xuan Li, Cheng Fang

Abstract: For inverse synthetic aperture radar (ISAR) imagery, the scatterers of target of interest are inherently sparse in spatial domain. In order to encode the sparsity, the sparse Bayesain model is a popular choice as a statistical manner, where Laplace distribution is imposed on the signal. Meanwhile, the sparse Bayesian model provides a confidential knowledge in the result. Due to the inaccurate compensation of the translational motion of interested target, the phase error correction is incorporated in the sparse Bayesian model to achieve high-resolution for ISAR imaging. In this papar, a novel antofocus sparse Bayesian Learning based on the spike-and-slab prior (SSP-AFSBL) is proposed, which can induce and enhance the sparse feature more flexibly. Due to the nonconjugacy of the likelihood and prior, a hierarchical Bayesian model is performed to provide the full Bayseian inference for the sparse coefficients. To avoid the calculation of a multidimensional integral, the variational Bayesian (VB) approximation is adopted to derive the analytical solution to hyperparameters. Finally, qualitative and quantitative analysis based on simulated and raw datasets have demonstrated the effectiveness of the proposed algorithm.