NONLINEAR SPECTRAL UNMIXING USING MANIFOLD LEARNING

Spectral mixtures of hyperspetral data often display nonlinear mixing effects. This paper develops locally linear weighted estimation (LLWE) based on two of the best known algorithms of manifold learning, Isomap and LLE. Studying on in situ spectral reflectance data, Spectral reflectance of four kinds of the mixed land-cover types in different percentages was measured and preliminarily analyzed. The model LLWE was verified by predicting the abundacne of main land-cover types. Compared with principal component regression (PCR) and partial least squares regression (PLSR), the results of the standard error of prediction show that the LLWE has better predictability. It's recommended that the proposed LLWE has the potential for the information extraction of mixed land cover types in hyperspectral remote sensing imagery.