AVIRIS-NG-like smart virtual remote sensing via spectra-aware physics informed GANs

This paper aims to create a physics informed virtual replica of the
hyperspectral image captured by NASA’s Airborne Visible InfraRed Imaging
Spectrometer - Next Generation (AVIRIS-NG) sensor equipped on manned
aircraft. Few image samples are selected from study site around New
Mexico, USA from flight mission ran in 2019. Out of 425 bands, 8 bands are
utilized. For each band, reflectance spectra are chosen from United States
Geological Survey (USGS) based on site specific geographical features.
These spectras are infused during the image generation process with the
correction check using matched filters. Moreover, we include the methane
plumes along with other closely related hydrocarbons during the image
generation process. Generative Adversarial Networks (GANs) architecture is
employed with physics informed loss function for generating realistic and
physically plausible images. Additionally, we also propose a new light
weight dataset for creating the virtual replica of the AVIRIS-NG sensor on
selected 8 bands in the visible light spectrum and the short-wave infrared
region.