Signature Prediction and Uncertainty Analysis for Recognition Applications

Development of target recognition algorithms require 1) predicting accurate physics-based signatures, and 2) characterizing signature sensitivity to various field parameters in order to identify robust and stable signature features. HyPerComp Inc. in collaboration with SAIC-DEMACO proposes to build a computational platform using its highly accurate Maxwell’s equations solver TEMPUS (Time-Domain EM Parallel Unstructured Simulator) a procedure to model uncertainty that enables quantifying the sensitivity of solutions and derived quantities (RCS, SAR, etc.) to key parameters affecting signatures observed in the field. The uncertainty model is based on the recent work of Prof. Hesthaven of Brown University and his colleagues that uses a so-called chaos expansions to represent the functional dependence of the solution on parameters that can only be described in a statistical sense. This approach is far more efficient than the traditional method of Monte Carlo sampling and it naturally provides a probability distribution for the solution space.