De-Striping and De-Smiling of Aerial Hyperspectral Image for Water Color Analysis

Abstract

Hyperspectral imagery is typically acquired in push-broom mechanism, which is prone to image artifacts such as striping and smile for the sensor array whose sensor elements are not perfectly calibrated. The best practice would be to calibrate the sensor elements before the flight, but post-correction is required when images are already acquired without calibration. While there are some studies that addressed those striping and smile effects for hyperspectral images acquired from satellite or aircraft platforms, few studies were done for hyperspectral images that are focusing water color analysis, where the radiance level is approximately only a tenth of terrestrial scenes. This study proposes a correction method specialized for water scenes that also may contain terrestrial objects together, and analyzes the results using real drone-borne hyperspectral imagery taken for an island area in Korea. The result revealed that the variation in columnar mean before the de-striping, which ranges 5-15%, reduced to under 2% after the correction, also exhibiting successful removal of striping in visual inspection. The smile effect that ranges approximately 1-2 mW/m2/nm/sr, which accounts for 30% of radiance from water area, also reduced to under 0.1 mW/m2/nm/sr after the smile correction.