Adaptive Multi-Modality Inverse Scattering for Targets Embedded in General Stochastic Environments

Abstract : The program addressed development of adaptive detection and classification algorithms for multi-modal inverse problems. The research focused on the general problem of detection and classification of targets surrounded by dielectric layers and stochastically distributed scattering centers. Within the context of this overarching theme, the program addressed the particular applications of detecting and classifying obscured ground targets, landmines, and subsurface structures. In the research "inverse scattering" was defined broadly to represent an algorithm that infers the target and environmental characteristics using data from a multiplicity of active and passive sensors. The program considered two classes of inverse-scattering algorithms. One class was based on a direct use of the associated underlying wave equations, often employing a forward solver as an integral component of the inversion process itself. Methods such as reverse-time migration fall under this class of approaches. The second class of inversion schemes used the forward algorithm and available measured data for "training" a statistical model, and during the subsequent inversion the trained algorithm no longer need employ a forward solver. Bayesian and mutual-information-based algorithms fall under this latter class.