Enhanced classification performance from multiband infrared imagery

Abstract

Target classification using infrared data can be enhanced by using multiple spectral bands rather than a single band. Previously, algorithms have been developed and shown to provide detection enhancement with multiple bands. However, not all the bands produced by a hyperspectral infrared sensor are useful. This paper presents measures of performance that are useful for determining which bands to use and how many bands may be required to achieve reliable classification. These performance measures are applied to data collected by the spatially modulated inverse Fourier transform spectrometer (SMIFTS) hyperspectral infrared sensor to illustrate the advantages of increasing the number of spectral bands.