Data and evidence fusion from multiple independent Decision Theoretic sources: Hybrid Decision Networks

We will develop an innovative approach to the detection of subpixel missile threats in hyperspectral images. The new approach will use a generalized likelihood ratio statistic to define decision surfaces in terms of spectral/spatial subspace models for missile targets and backgrounds. Missile targets will be represented by invariant spectral subspaces that characterize target properties over a range of conditions. Background clutter will be characterized using adaptive spectral/spatial subspaces. The resulting detection algorithm will have the ability to detect subpixel threats in cluttered environments under unknown atmospheric conditions. The spectral subspace representation will allow the use of high-speed techniques for the computation of the likelihood ratio decision statistic. We will demonstrate the feasibility of the new approach using experiments with missile spectral signatures measured using a chromotomographic sensor embedded in cluttered hyperspectral images. An important goal is to determine the fundamental bounds on hyperspectral missile detection performance. We will also assess algorithm performance and computational requirements as a function of the amount of spectral and spatial information that is provided to the algorithm. The general approach will be applicable to any spectral region, but algorithm assessment during Phase I will focus on the mid-wave infrared.