Investigation on constrained matrix factorization for hyperspectral image analysis

摘要: Abstract — Matrix factorization is applied to unsupervised linear unmixing for hyperspectral imagery. The algorithm, called non-negative matrix factorization, used. It imposes a constraint on the non-negativity of amplitudes recovered endmember spectral signatures as well their fractional abundances. This ensures recovery physically meaningful endmembers and algorithm further modified include sum-to-one such that sum abundances each pixel unity. Several practical implementation issues in image unmixng are discussed. Some preliminary results from AVIRIS experiments presented. We also discuss advantages possible limitations this method analysis. Keywords: factorization; nonnegative mixture model; unmixing; I. I NTRODUCTION Linear analysis well-known technique remote sensing based fact rough spatial resolution permits different materials present area covered by single pixel. model says reflectance visible-near infrared multispectral or all independent pure (endmembers) an scene [1]. Let