Statistics for the Relative Detectability of Chemicals in Weak Gaseous Plumes in LWIR Hyperspectral Imagery

Abstract

Executive SummaryThe detection and identification of weak gaseous plumes using thermal imaging data is complicated by many factors. These include variability due to atmosphere, ground and plume temperature, and background clutter. This report presents an analysis of one formulation of the physics-based model that describes the at-sensor observed radiance. The motivating question for the analyses performed in this report is as follows. Given a set of backgrounds, is there a way to predict the background over which the probability of detecting a given chemical will be the highest? Two statistics were developed to address this question. These statistics incorporate data from the long-wave infrared band to predict the background over which chemical detectability will be the highest. These statistics can be computed prior to data collection. As a preliminary exploration into the predictive ability of these statistics, analyses were performed on synthetic hyperspectral images. Each image contained one chemical (either carbon tetrachloride or ammonia) spread across six distinct background types. The statistics were used to generate predictions for the background ranks. Then, the predicted ranks were compared to the empirical ranks obtained from the analyses of the synthetic images. For the simplified images under consideration, the predicted and empirical ranks showed a promising amount of agreement. One statistic accurately predicted the best and worst background for detection in all of the images. Future work may include explorations of more complicated plume ingredients, background types, and noise structures.iii