Unsupervised Linear Spectral Unmixing of Multispectral Images using the NMF and Modified-Multilayer NMF Algorithms

Spectral Unmixing of satellite images plays a major role in preparation of accurate maps. Here, the pixels are considered as Linearly mixed. The endmembers which contributes the unmixing accuracy is extracted using the Principal Component Analysis(PCA) algorithm. Non-negative Matrix Factorization(NMF) had been introduced in the unmixing process. Endmember dissimilarity Constraint is imposed on NMF. Multilayer Non-negative matrix factorization(MLNMF) uses the Vertex component Analysis for finding the initial Spectral matrix. In this paper, the Modified ML-NMF algorithm is used for finding the fraction images. Results obtained from both NMF and Modified MLNMF are compared and it shows that the proposed Modified ML-NMF algorithm unmix effectively which reduces the Root Mean Square Error by 15% and Reconstruction Error by 20%.