Atmospheric correction of airborne infrared Hyperspectral images using neural networks

摘要

In this paper we present EARTH (surface Emissivity, temperature and Atmosphere Retrievals from Thermal infrared Hyperspectral image), a new method to extract spectral emissivity and surface temperature from infrared radiances measured by an airborne hyperspectral sensor. The method solves two underlying problems : atmospheric compensation and Temperature/Emissivity Separation (TES). The atmospheric correction scheme is based on sounding techniques and neural networks to extract principal components of temperature profile and water vapor content. A spectral Smoothness (SpSm) method is then used for TES and to improve water vapor estimation. Accuracy of atmospheric and surface retrievals has been evaluated on synthetic data. Finally the method has been applied to the S-HIS spectro-radiometer measurements of the EAQUATE campaign.