Remote vapor detection and classification using hyperspectral images

Adaptive Infrared Imaging Spectroradiometer (AIRIS) is a longwave infrared (LWIR) sensor for remote detection of chemical agents such as nerve gas. AIRIS can be considered as a hyperspectral imager with 20 bands. In this paper, we present a systematic and practical approach to detecting and classifying chemical vapor from a distance. Our approach involves the construction of a spectral signature library of different vapors, certain practical preprocessing procedures, and the use of effective detection and classification algorithms. In particular, our preprocessing involves effective vapor signature extraction with adaptive background subtraction and normalization, and vapor detection and classification using Spectral Angle Mapper (SAM) technique, which is a signature-based target detection method for vapor detection. We have conducted extensive vapor detection analyses on AIRIS data that include TEP and DMMP vapors with different concentrations collected at different distances and times of the day. We have observed promising detection results both in low and high-concentrated vapor releases.