Accuracy of super-resolution for hyperspectral ocean observations
Joseph Landon Garrett, Dennis D. Langer, Karine Avagian, Annette Stahl · OCEANS 2019 - Marseille · 2019
Super-resolution, a class of techniques used to reconstruct a high-resolution image from one or more low-resolution observations, is a possible route to utilize the full remote imaging capabilities of small satellites and unmanned aerial vehicles. Here, we test two frequently-used variants, Robust Super-Resolution (RSR) and Projection onto Convex Sets (POCS), to see how accurately each technique reconstructs images from a small satellite. The two techniques are chosen because each utilizes a different kind of prior knowledge. RSR utilizes knowledge about the scene, while POCS utilizes knowledge about the imaging process. The algorithms are run on three bands of two hyperspectral images: one lab-acquired image of a wooden block and one simulated image of a remote sensing ocean scene. The superresolution reconstructions of the simulated image are evaluated by calculating the brightness error and spectral angle with respect to the original scene. Both super-resolution algorithms improve both metrics relative to the raw, registered data. RSR achieves more improvement overall, but POCS operates faster.