| Remote sensing of chemical vapor plumes is a difficult but important task with many military and civilian applications. Hyperspectral sensors operating in the long wave infrared (LWIR) regime have well demonstrated detection capabilities. However, the identification of a plume's chemical constituents, based on a chemical library, is a multiple hypothesis-testing problem that standard detection metrics do not fully describe. Our approach partitions and weights a confusion matrix to develop both the standard detection metrics and an identification metric based on the Dice index. Using the developed metrics, we demonstrate that using a detector bank followed by an identifier can achieve superior performance relative to either algorithm individually.;Performance of the cascaded system relies on the first pass reliably detecting the plume. However, detection performance is severely hampered by the inclusion of plume pixels in estimates of background quantities. We demonstrate that this problem, known as contamination, can be mitigated by iteratively applying a spatial filter to the detected pixels. Multiple detection and filtering passes can remove nearly all contamination from the background estimates, a vast improvement over single-pass techniques. |
