Robust Coupled Non-Negative Matrix Factorization for Hyperspectral and Multispectral Data Fusion

Abstract

In recent time, Hyperspectral(HS) and multispectral(MS) data fusion based on spectral unmixing methods has become an active area of research. Coupled non-negative matrix factorization (CNMF), one among many unmixing-based data fusion approaches, performs alternating unmixing of the HS and MS data while connecting the results by point spread function and spectral response function of the sensors. However, CNMF operates exclusively on the spectral information of the HS and MS data and also disregards the spatial distribution of the data. In this paper, we propose an extended linear mixing model approach to enhance the spatial resolution of HS data. The proposed method extends the commonly used linear mixing model for data fusion by introducing an additional term that accounts for the non-linearity effects as well. The results of the simulation obtained from the analysis on various synthesized datasets suggest that the proposed method can significantly improve the Peak signal-to-noise ratio (PSNR), minimize the relative dimensionless global error (ERGAS) of fusion, and also competes with state-of-the-art approaches.