Advanced Signature-Matched Hyperspectral Change Detection

The proposed program will investigate the physical basis for pixel-level hyperspectral change detection (HSI-CD), develop enhanced HSI-CD algorithms, and demonstrate proof-of-concept using real HSI data. This work will build on Space Computer Corporation’s existing real-time HSI-CD system by making novel improvements to the following key elements of the processing chain: (1) precision registration, (2) spectral normalization, and (3) target signature matching and detection filtering. The registration effort focuses on improving the accuracy of multi-look image alignment in complex backgrounds using fine-scale 3-D LIDAR elevation data and suppression of variable illumination effects that cause errors in scene-based registration measurements. Physics-based spectral normalization procedures that enable pixel-level comparisons between different HSI passes over the same area will be developed to augment or replace currently-used empirical transformations, providing more effective compensation of localized image variations introduced by the environment rather than by material changes. Similar methods will be used to obtain precise matching of material signatures to in-scene spectral observations, enabling the use of highly selective nonlinear spectral filters for improved discrimination of specific changes. A primary goal of the Phase I effort is to demonstrate tangible improvements in HSI change detection target sensitivity and false alarm rejection performance.