A new multiplicative nonnegative matrix factorization method for unmixing hyperspectral images combined with multispectral data

Abstract

In these investigations, a novel algorithm is proposed for linearly unmixing hyperspectral images combined with multispectral data. This algorithm, which is used to unmix the considered hyperspectral image, is founded on nonnegative matrix factorization. It minimizes, with new multiplicative update rules, a novel cost function, which includes multispectral data and a spectral degradation model between these data and hyperspectral ones. The considered multispectral variables are also used to initialize the proposed algorithm. Tests, using synthetic data, are carried out to assess the performance of our algorithm and its robustness to spectral variability between the processed data. The obtained results are compared to those of state of the art methods. These tests prove that the proposed algorithm outperforms all other used approaches.Index Terms-Hyper/multispectral imaging, linear unmixing, multiplicative update rule, nonnegative matrix factorization, spectral degradation model

Using a Panchromatic Image to Improve Hyperspectral Unmixing

“…Since all the previously described methods show limitations due to the assumptions made and the data features, new hyperspectral unmixing strategies combining synchronized and distinct acquisitions with different characteristics may be relevant. For example, Uezato et al [48] used LIght Detection And Ranging (LIDAR) data to unmix hyperspectral images, and Karoui [33], Benkouider et al [49] developed two algorithms taking in account multispectral data with a smaller GSD. Multispectral/hyperspectral and panchromatic/multispectral data fusion may also use hyperspectral unmixing techniques in their process [33,[50][51][52][53][54].…”