Unmixing of measured spectra and reconstructed spectra from colored image of objects of known proportions

Abstract

Spectral mapping of benthic ecosystems using hyperspectral and colored images taken by remote sensors is hoped to replace the tedious and expensive method of mapping by field survey. These images, however, have pixel scales greater than the sizes of patches of any one benthic cover type. Here we employ the method of linear unmixing to recover a priori-known proportions of end member spectra from spectra measured using a spectroradiometer and spectra reconstructed from colored image. Obtained proportions agree with expected values despite inaccuracies in the reconstructed spectra. Moreover, it is found that as the number of end members is increased, the accuracy of the relative proportions of the end members is probable to decrease.