SINGLE-FRAME SUPER-RESOLUTION OF REAL-WORLD SPACEBORNE HYPERSPECTRAL DATA

Single-frame super-resolution (SFSR) is a well-researched issue in inverse imaging with the latest developments in the field of hyperspectral remote sensing. Paucity of open-source very highresolution benchmark datasets restricts these algorithms from being applied to real-world scenes captured by hyperspectral sensors and the users have to contain with the outputs of the simulated trials. We attempt to narrow down this void by generating 15 m spatial resolution datasets from the 30 m spatial resolution Hyperion dataset of Ahmedabad, India through two classic SFSR algorithms: iterative back projection (IBP) and sparse representation (VSR). Visually inspecting land cover features and their spectral profiles in the superresolved results against open-source panchromatic data shows the successful preservation of the spectral and spatial content of the original Hyperion data by VSR. No-reference image quality metrics confirm this finding although the processing time remains quite high compared to IBP.

Automated summary

Single-frame super-resolution (SFSR) is an important research topic in inverse imaging, especially in the field of hyperspectral remote sensing. The lack of high-resolution benchmark datasets limits the application of these algorithms to real-world scenes captured by hyperspectral sensors. This study successfully generated 15m spatial resolution datasets from a 30m spatial resolution Hyperion dataset using two classic SFSR algorithms: iterative back projection (IBP) and sparse representation (VSR). Visual inspection and image quality metrics confirm the successful preservation of spectral and spatial content by VSR, although it has a longer processing time compared to IBP.