Minimum Spectral Correlation Constraint Algorithm Based on Non-negativeMatrix Factorization for Hyperspectral Unmixing

This paper proposes minimum spectral correlation constraint algorithm based on non-nega-tive matrix factorization(MSCCNMF).According to the properties of non-correlation among the endmembers in a hyperspectral image,a spectral correlation function is proposed to measure the degree ofcorrelation between spectral signatures.The smaller the function value,the smaller the correlationbetween the spectra.By minimizing the spectral correlation function with the error function of non-negative matrix factorization,a set of spectra with smallest correlation can be obtained,considered asendmembers,and the corresponding abundances can be obtained simultaneously.Synthetic and realexperiments show the effectiveness of the proposed algorithm.