Improved Information Extraction Using Prior Knowledge to Satisfy Multiple Objectives

"Existing signal processing algorithms for remote sensing were typically derived, first, by formulating the problem in the contextof one of the following: inverse theory (e.g. integral equations, integral transforms, deconvolution), singular valued decomposition,filter theory, estimation theory, information theory, detection theory, or beamforming; and then, second, by deriving a solution thatmeets a single optimization criterion. For example, based on filter theory, the generalized sidelobe canceller selects the unbiased filterhaving minimum output power consistent with constraints. We discovered a new problem formulation that provides an optimum solution with additional degrees of freedom. For certain,specific choices of these additional degrees of freedom, the new solution reduces to known algorithms. For other choices, our approachprovides new algorithms with added capability that allows multiple processing objectives to be met optimally and simultaneously. We propose to demonstrate the feasibility of this new approach in an application of programmatic interest to DARPA,in particular, improved detection and classification in FTIR spectroscopy. Our approach simultaneously provides: adaptivity; incorporation of prior information; sensor fusion; physics-based processing;high resolution; imaging; improved classification; multipath propagation; monostatic and multistatic operation; mitigation ofnon-Gaussian, non-stati