Hyperspectral imagers take pictures not in the traditional three colors of human vision, but with hundreds of spectral channels, sometimes extending far into the thermal infrared. Hyperspectral imagery is particularly valuable for detecting and characterizing weak gaseous plumes. These plumes have distinctive chemical signatures, and the large number of spectral bands permits the detection of even trace amounts of gas.
But plumes differ from other kinds of targets in remote sensing; you do not so much look *at* plumes as *through* them. The plume is a weak perturbation that is superimposed on the signal that comes from the underlying scene. An important consequence is that this underlying scene must be well characterized in order for a detection to be made.
Traditionally, this background has been modelled as a single multivariate gaussian distribution, and in that case the optimal detector is a linear matched filter. This talk will discuss the nature of hyperspectral clutter, and the implications of nongaussian clutter on the weak signal detection problem.
Presentation (PDF File)
Video of Talk (RealPlayer File)
Copyright. All Rights Reserved.