A variational Bayesian approximation approach via a sparsity enforcing prior in acoustic imaging

Acoustic imaging is an advanced technique for acoustic source localization and power reconstruction from limited noisy measurements at microphone sensors. To solve this ill-posed inverse problem, the Bayesian inference methods using proper prior knowledge have been widely investigated. In this paper, we propose to use a hierarchical Variational Bayesian Approximation for the robust acoustic imaging. And we explore the Student's-t priors with heavy tails to enforce source sparsity and non-Gaussian noises, so that we can achieve the super spatial resolution and wide dynamic range of source powers. In addition, proposed approach is validated by simulations and real data from wind tunnel in automobile industry.