Low Resolutions Linear Unmixing-Nmf-Based Change Detection Approaches for Multiresolution Remote Sensing Data

In this work, new low resolutions algorithms to detect changes in multiresolution remote sensing data are proposed. Using the linear spectral unmixing concept, by means of nonnegative matrix factorization, these algorithms allow providing 2D variables with the same low spatial and spectral resolutions in order to detect changes in the considered data. The first algorithm is a gradient-based one, whereas the second algorithm uses multiplicative update rules. Obtained simulation results, using synthetic data, show the high ability of the designed algorithms to reveal changes occurred between multiresolution optical data.