Advanced Algorithms for Exploitation of Space-Based Imagery

Physical Sciences Inc. proposes to develop statistically-based algorithms for detection of spectrally-structured targets via unsupervised analysis of hyperspectral imagery and to develop a software tool for image analysts based on those algorithms. The proposed analysis algorithms are suitable for incorporation into an Automatic Target Recognition (ATR) formalism and will also be suitable for detection and tracking of subpixel targets from satellite imagery. The software tool will be developed as an add-on to the widely-used ENVI software package and will enable hyperspectral datacube analysis on a Pentium 4-level desktop computer on the timescale of seconds to minutes (datacubes ~256 x 256 pixels x 200 spectral bands x 14 bits/pixel for a total of some 170 Mbits/datacube). It will enable image analysts to generate high-level ‘target detection maps’ using a variety of detection algorithms as well as enable low-level, quantitative comparison of the results of different approaches using Receiver Operating Characteristic curves. The near-term application for the proposed algorithms is detection and tracking of militarily significant targets using visible/near-IR/SWIR hyperspectral imagery from the COMPASS sensor. Potential future applications include analysis of data from the ARTEMIS sensor.