N. Chu (L2S): Bayesian approaches via sparsity enforcing priors for acoustic source imaging with robustness\, super resolution and wide dynamic range




Date(s) : 21/03/2013   iCal
14 h 00 min - 15 h 00 min

Bayesian approaches via sparsity enforcing priors for acoustic source imaging with robustness\, super resolution and wide dynamic range\n\nBy Ning Chu\, L2S.\n\n Acoustic imaging is a standard technique for mapping the positions and powers of acoustic sources with microphone arrays\, which provides the insight into the mechanisms and properties of acoustic sources\, especially for acoustic comfort in automobile and aircraft industries. However\, acoustic imaging often involves the ill-posed inversions and becomes too time-consuming to obtain high spatial resolutions and wide dynamic range.\n In this talk\, we first improve the forward model by considering background noises\, as well as forward model uncertainty caused by multipath propagation. Then we propose several robust super-resolution approaches for acoustic imaging via the following techniques:\n 1 Sparsity constraint\,\n 2 Sparsity regularization\,\n 3 Bayesian inference via Double Exponential prior under joint Maximum A Posteriori (MAP) optimization \,\n 4 Variational Bayesian Approximation (VBA) via Student’s-t priors for both source powers and colored noises.\n Finally\, for efficient acoustic imaging\, we propose a shift-invariant 2D convolution model for acoustic power propagation.\n Our approaches are compared with some of the state-of-the-art methods on simulations\, real data from wind tunnel experiments of Renault S2A\, as well as hybrid data. The main advantages of proposed approaches are robust to background (colored) noise\, wide dynamic range of acoustic powers\, super spatial resolutions\, efficient to use\, and not need for the prior knowledge of the source number or SNR. It is feasible to apply in near-field wideband monopole and extended source imaging based on the 2D non-uniform microphone array for wind tunnel tests.\n\n Download slides

Catégories



Retour en haut 

Secured By miniOrange