High-Efficiency Hyperspectral Unmixing Based on Band Selection

Hyper spectral unmixing (HU) is important for ground objects identification. Due to the mass data hyper spectral sensors bring, band selection plays an important role in boosting efficiency of HU. This paper proposes a high-efficiency approach of HU that carries out two modified algorithms of band selection followed by nonnegative matrix factorization (NMF), which are linear prediction (LP) combined with K-L divergence and mutual information (MI). Experiment results based on simulated data and real hyper spectral imagery demonstrate that the proposed scheme is more efficient than initial NMF in HU.