Use of Superresolution Enhanced Chris/proba Images for Land- Cover Classification and Spectral Unmixing

CHRIS/Proba represents a new generation of hyperspectral-oriented data that provides different viewing angles of the same scene. This multi-angle acquisition, with 18 bands between 0.4-1 μm at five different angles, can substantially improve characterization of various land-cover types. However, with around 18m pixel size, the spatial resolution is still too coarse for many applications. To improve the resolution of CHRIS data, we propose the use of a superresolution (SR) image reconstruction method utilizing the information present in the multi-angle imagery. CHRIS images were first pre-processed and atmospherically corrected. Then a selected superresolution method was applied to upscale the spatial resolution to around 10m. The SR enhanced images show significant improvements in terms of contrasts and detail. The added value of SR imagery for land-cover classification and spectral unmixing was evaluated. In terms of classification accuracy, the use of SR CHRIS images leads to marginal increase in classification accuracies. The large amount of additional classification detail requires extended ground truth for validation. Spectral unmixing applied on SR CHRIS shows that fraction images have more detail and are more accurate than the fraction images derived from the original CHRIS images. * Corresponding author.