We develop an efficient Bayesian algorithm for solving the inverse problem of classifying and locating certain two dimensional objects using noisy far field data obtained by illuminating them with a radiating wave. While application of Bayesian algorithms for wave-propagation inverse problems is itself innovative, the principal novelty in this work is in using (i) a surrogate Bayesian posterior distribution computed using a generalised polynomial chaos approximation; and (ii) an efficient wave-propagation-specific reduced order model in place of the full multiple scattering forward model. We demonstrate the capability of this approach with simulations in which we accurately detect two dimensional objects, with shapes motivated by safety and security applications.
|Name||MATRIX Book Series|
|Conference||2018 Matrix Annals|
|Period||8/01/18 → 21/12/18|