Constraint Non-Negative Matrix Factorization With Sparseness and Piece wise Smoothness for Hyperspectral Unmixing

摘要

The technique of Constrained Non-negative Matrix Factorization is widely used in hyperspectral image unmixing. Multiple constraints NMFs which add constraints to both end members and abundances have been proposed for spectral unmixing in recent years. This paper proposed a novel algorithm named Sparseness and Piecewise Smoothness constraint Non-negative Matrix Factorization (SPSNMF), in which both piecewise smoothness of end members and sparseness of abundance are added to NMF cost function simultaneously. The unmixing results of this method can satisfy the three facts: the non-negativity of both end members and abundances, the smoothness of end members and the sparsity of abundances. Aiming to minimize the improved cost function, the updating rules of end members and abundances are given, and the whole work flow of SPSNMF is presented. Two real data experiments proved that SPSNMF outperforms traditional algorithm and typical unmixing algorithms base on Non-negative Matrix Factorization for its lower reconstruction errors and better balance between end member purity and abundance sparsity.