Journal of the Faculty of Engineering and Architecture of Gazi University, vol.38, no.4, pp.2385-2396, 2023 (SCI-Expanded)
Hyperspectral (HS) images have high spectral resolution, but their spatial resolution is low due to technological constraints. It is beneficial to have HS images that have high spatial resolution as well as high spectral resolution to increase classification accuracy or obtain more detailed content in this kind of image. Thus, HS and multispectral (MS) image fusion have become a very popular topic in recent years. In this study, spectral decomposition and neighborhood pixel relation-based hyperspectral and multispectral image fusion is proposed. Firstly, spectral decomposition is used to get endmembers and their abundance maps from the hyperspectral image. Hyperspectral endmembers are degraded to multispectral spectral resolution according to the spectral response model of the multispectral sensor. Then, abundance fractions of the endmembers and neighborhood pixels are estimated for each multispectral pixel. Finally, the abundance map of multispectral image and endmembers, and neighborhood pixels of the hyperspectral image are used to obtain an image with high spatial and spectral resolution. The proposed method is tested on real hyperspectral images and experimental results show that it gives the highest accuracy compared to studies in the literature.