Target Identification and Detection Using LWIR Hyperpectral Signature Transformation of Multiple Missions without Registration

摘要

Changes in atmospheric conditions and sensor response for successive imaging sessions have limited the use of fixed target hyperspectral libraries, especially for multiple mission studies, to help identify and discriminate targets from cluttered backgrounds. The hyperspectral target signature instability has resulted in a dependence on anomaly detection algorithms in real time surveillance applications. These algorithms fail to meet some critical military requirements. This study examines a variety of mathematical transforms of the spectral signatures derived from missions flown on different days with starkly different weather conditions. The transforms use overlapping regions in the two data sets but avoid registering the image cubes. Some of the transforms use statistical features such as auto covariance matrices, means, and/or standard deviations of the image cubes. Other algorithms use spectral means taken from common features in the image cubes such as trees, roads, or blackbodies in both image cubes. Our study examines target spectra transformations in the long-wave infrared spectra of man-made targets and natural backgrounds obtained with the SEBASS (8-12 microns) imager as part of the Dark HORSE 2 exercise during the HYDRA data collection in November, 1998. This study computes the signal to clutter ratio (SCR) for transforms that required high accuracy registration, various spectral signature transformations that do not need any registration, and those transforms that used random, varying number of pixels in the overlap area. The transformed signatures were subsequently used in matched filter searches to successfully find targets with low false alarm rates (< 1 FA/Km2).