Improvement Of NMF Algorithm Based On Endmember And Abundance Attributes

The non-negative matrix factorization(NMF)algorithm is an important branch of the blind linear spectrum solution.But,applying the original NMF algorithm directly to decomposition of mixed pixels will result in local minimum and slow convergence.In this paper,the attributes of the endmember and abundance are considered,and two non-negative matrix factorization algorithms are proposed based on the MDC-NMF.On the one hand,to preserve the local invariance of the hyperspectral remote sensing image,spectral information divergence(SID)is joined to the manifold regularization,which is to measure the similarity between the pixels.At the same time,the endmember distance is added to the algorithm.It can constrain the simplex volume to be small,and make the algorithm converge to the simplex made up by real endmembers.A new constrained objective function for non-negative matrix factorization algorithm named MMDC-NMF is proposed based on the endmember distance and manifold regularizationOn the other hand,the characteristics of pixel structure in the image are considered.The sparse constraint is joined into the algorithm,which may make the abundance matrix sparser following the structure of the image.The non-negative matrix factorization combining sparseness constraint and endmember distance named SMDC-NMF is proposed.As expected,the experimental results show the effectiveness of the proposed approach.The experimental results of simulated data and actual data show that the two methods are better than the results of the MDC-NMF algorithm,and MMDC-NMF is more suitable for the solution of hyperspectral image blending with low sparsity,and SMDC-NMF has obvious effect on the image of high sparsity.