Automatic spectral classification of imaging spectrometer data

Multi-spectral imagery plays a significant role in Earth resource survey and evaluation and has been an essential part of terrestrial and planetary exploration. Multi-spectral imaging sensors-such as the Airborne Visible/Infrared Imaging spectrometer (AVIRIS), which images in 224 wavebands in the range 0.4-2.45 /spl mu/m giving 224 images, each of 614/spl times/512 pixels, and GER which images in 63 wavebands-are now routinely in use. With this increasing utilization of imaging spectrometer data and the constant improvement in instrument resolution and spectral discrimination, automatic identification of spectral signatures emanating from this imagery would be an invaluable facility as a precursor to classifying each pixel. Existing methods for identifying constituent spectra typically rely on spectra that are selected either manually or involve manual intervention. The aim of the work described in this paper is to devise techniques suitable for fully-automatic analysis. Two techniques are described. The first is artificial neural networks and the second is singular value decomposition (SVD). The ability of these methods to distinguish between up to 160 different spectra is assessed, as is their stability in the presence of noise and their capacity to identify correctly combinations of spectra, i.e., where a test spectrum is made up of more than one spectrum from a reference database.