A split bregman method for linear spectral unmixing

Jianjun Liu, Zebin Wu, Zhihui Wei, Le Sun

Abstract: Linear spectral unmixing is a popular tool to describe the remote sensing hyperspectral data. However, due to the huge size of hyperspectral data and the real-time processing, it needs a faster and more accurate algorithm. In this paper, we present a novel algorithm for linear spectral unmixing based on nonnegative matrix factorization (NMF), referred to as the split bregman method for NMF (SBNMF). The proposed algorithm takes advantage of the fast convergence of split bregman method, and at the same time optimizes the alternating update method, which help it get accurate results faster. The experimental results based on both synthetic mixtures and a real image scene demonstrate the superiority of our proposed algorithm.