Band selection for gas detection in hyperspectral images

Abstract

The objective of this work is to find the minimal number of spectral bands needed to detect a specific gas in a hyperspectral cube, in order to develop a hyperspectral gas sensor in the future. In the framework of the research, a band selection method is developed in order to correctly detect a specific gas, using few of the hundreds of spectral bands available in the data. It shows that for a synthetic cube, using a correlation matrix built only from non-gas pixels, the detection improves as more spectral bands are in use, but with relatively few bands we still get good detection.