A novel initialization method for nonnegative matrix factorization and its application in component recognition with three-dimensional fluorescence spectra

Abstract

Nonnegative matrix factorization has been widely used in many areas and has been applied for component recognition with three dimensional fluorescence spectra recently. However, nonnegative matrix factorization is a nonconvex programming in the iteration process, thus the solution is dependent on the initial values and consequently not unique. Up to now, an effective global convergent algorithm is still absent. In this work, we propose an initialization scheme based on independent component analysis. Compared with other initialization schemes, the optimal solution of nonnegative matrix factorization based on independent component analysis is much better and it is demonstrated by typical experiments of component recognition with three-dimensional fluorescence spectra.