Using non-negative matrix factorization with projected gradient for hyperspectral images feature extraction

Aiming at hyperspectral remote sensing images containing huge amounts of data, removing redundant information and reducing processing dimensions are the premise and foundation for hyperspectral remote sensing processing and applications. In this paper, a new feature extraction algorithm based on non-negative matrix factorization with projected gradient, PGNMF for hyperspectral remote sensing images is proposed. Experimental results on AVIRIS 220 bands data set of a mixed agriculture or forestry landscape in the Indian pine test site show that the proposed method achieved lower time complexity and more strong analysis capability than comparative algorithms. Compared with the PCA and ICA method, classification accuracy can be improved. The proposed hyperspectral feature extraction based on PGNMF balance algorithm efficiency and performance very well.