Sparse Bayesian Learning Approach for Wide-band Acoustic Imaging

Acoustic imaging with a microphone array is a crucial technology for diagnosing abnormal sounds in mechanical device, which typically exhibit wide-band characteristics. Despite this, the full potential of these wide-band features has not been fully exploited. Hence, this paper introduces a sparse Bayesian learning approach for wide-band acoustic imaging. An over-complete observation model is built and sparse priors are independently assigned to signals to encourage sparse solutions, resulting in high resolution performance. By sharing hyper-parameters across different frequency sub-bands, wide-band characteristics are utilized for acoustic imaging. Due to the auto-regressive nature and uncertainty estimation character, the proposed method achieves better performance than state-of-the-art methods. Numerical simulations exhibit that the proposed method can be utilized for improving the accuracy performance of wide-band acoustic imaging.