Comparison of hyperspectral sub-pixel detection with and without a priori knowledge of target features

Hyperspectral sensors collect imagery simultaneously in hundreds of contiguous bands. They provide a unique mixture of spectral and spatial information to detect and identify materials and targets, even targets that are much smaller than a pixel. The ability to perform sub-pixel detection allows hyperspectral systems to perform wide-area search to direct other assets to areas of interest. For example HSI sensors can be used to cue narrow-field, high-spatial resolution E-O cameras or focus the attention of image analysts on target regions. Several algorithms have been developed at TRW to perform target detection based on the spectral information in a scene, each assuming a different level of a priori knowledge about materials present in the scene. This paper describes results obtained applying these algorithms to an HSI data set collected with the HYDICE sensor.