Fast searching of spectral library database using variable interval spectral average method

Abstract

With ever-increasing number of spectral channels from space-borne hyperspectral instruments, demand on approaches for fast search schemes for matching hyperspectral pixel vector with standard spectral library database has increased proportionately. The present-day methods are tedious and time consuming to meet the above task. We propose a fast matching scheme based on bivariate short-interval local variance that can be used to capture the essence of reference materials in the spectral library. The variance of each selected window is computed across the spectral curve data and the peak variance above a threshold is taken as a spike. The position and linewidth of the spikes are shown to carry unique signatures of the given material spectral data, which can be stored and used as matching criteria. The choice of appropriate threshold is important; it has been found that the mean value of background variance signal could be used as the threshold value. The proposed method was successfully applied to identify some samples of the AVIRIS hyperspectral imagery to the standard JPL spectral library database.