INTRINSIC FEATURES FOR AUTOMATED TARGET MODELING AND TRACKING USING HYPERSPECTRAL IMAGERY

In this project, we will develop and demonstrate new feature-based algorithms for target modeling and tracing across hyperspectral imagery acquired at different times under unknown condition. The problems addressed by this project are more general than traditional ATR problems in the sense that target properties are initially assumed to be unknown. The algorithms are derived from recently developed physics-based invariants of hyperspectral data that do not depend on the illumination environment or atmospheric conditions. These invariants reduce the high dimensionality associated with a hyperspectral data that do not depend on the illumination environment or atmospheric conditions. These invariants reduce the high dimensionality associated with a hyperspectral target signature to a low dimensionality intrinsic representation. The information contained in hyperspectral data enables the algorithms to discriminate similar materials as well as camouflaged objects from background. Hyperspectral data also enables local apporaches to be used for ATR making the algorithms relatively insensitive to partial obscuration. Since hyperspectral data is complementary to several other sensing modalities, we will investigate the fusion of information obtained from the hyperspectral algorithms with information obtained from other sensors. The performance of the new algorithms will be quantified relative to traditional approaches using imagery containing military vehicles and other objects under different illumination and atmospheric conditions with varying degrees of bacdground clutter and object obscuration.