Spectral Enhancement of Multispectral Imagery Using Partially Overlapped Hyperspectral Data and Sparse Signal Representation

Abstract

Spaceborne imaging spectroscopy enables the identification and quantification of different ground-surface materials with rich spectral information, but its swath width is much narrower than those of optical broadband satellites. We present a new approach that enhances spectral resolution of multispectral imagery by using partially overlapped hyperspectral data. The proposed technique assumes that all materials in the target scene exist in the partial area observed by the hyperspectral sensor and that any pixel spectral signature can be approximated by a sparse linear combination of observed spectral signatures. We evaluate our method using simulated EnMAP and Sentinel-2 images generated from HyMap data acquired over Costa Rica in 2005. Experimental results show that the proposed method is capable of reconstructing hyperspectral data for the coverage of Sentinel-2 when the assumptions are satisfied. Furthermore, we demonstrate the effectiveness of the proposed spectral enhancement technique for subsequent data analysis by presenting spectral unmixing for discriminating bare soil and nonphotosynthetic-active vegetation.