Statistical and Adaptive Signal Processing for UXO Discrimination for Next-Generation Sensor Data

摘要

To address the limitations associated with current sensors, SERDP and ESTCP have been supporting efforts to develop a new generation of UXO sensors that produce data streams of multi-axis vector or gradiometric measurements, for which optimal processing has not yet been carefully considered or developed. Here, we outline a research program where our goal was to address this processing gap, employing a synergistic use of advanced phenomenological-modeling and signal-processing algorithms. The focus of the research was: (i) exploitation and refinement of Duke's existing phenomenological models to accurately predict the underlying target signatures for the new sensor modalities, with the goal of pinpointing physical parameters that can be utilized within the signal processing architecture; (ii) development of physics-based statistical signal processing approaches applicable to the problem in which vector data is available from such sensors; (iii) development of the theory of optimal experiments to guide the optimal deployment of sensor modalities; and (iv) development of active learning and kernel-based algorithms that yield target classification based upon all available data (not based on individual feature vectors from isolated targets), as well as allowing data collected across multiple sites to be integrated within the classifier.