Gregory Badura, Georgia Tech Research Institute; Elena Plis, Georgia Tech Research Institute; Christopher Valenta, Georgia Tech Research Institute
Keywords: Spectral Bidirectional Reflectance Distribution Function (sBRDF), Spectroscopy, Modeling & Simulation, Machine Learning
Abstract:
The identification of unresolved resident space object (RSO) material composition via imaging spectrometry techniques requires knowledge of the constituent materials directional reflectance characteristics and knowledge of how these directional signatures will mix within the chosen sensor bandpass. In addition, knowledge is required regarding how the directional reflectance changes due to in-orbit aging in order to make use of these datasets in empirically collected measurement campaigns. In this study, we present the design of a principal plane goniometer that is responsive to spectral radiance between 400 and 2400 nm and a method for acquiring spectral bidirectional reflectance distribution function (sBRDF) measurements using the system. Initial sBRDF results for multiple spacecraft materials of interest are presented (including several samples recently returned from outside the International Space Station) for a variety of illumination and observational geometries as a first step towards building a robust database of sBRDF measurements as a function of in-orbit aging time. A two-step BRDF model inversion of the Theoretical Ward sBRDF model was performed using the sBRDF data-sets in order to derive model parameters that can be ingested by radiometric modeling tools such as the Digital Imaging and Remote Sensing Image Generation (DIRSIG) model. These radiometric modeling tools were then utilized to model the at-aperture spectral radiance when imaging unresolved mixtures of the materials as a function of varying observer and solar illumination angles for sensor bandpasses of interest to the Space Domain Awareness (SDA) community. Other applications of the methods developed towards efforts such as training machine learning frameworks on radiometrically-accurate resolved imagery and using unsupervised machine learning techniques to perform spectral un-mixing are discussed and presented.
Date of Conference: September 14-17, 2021
Track: Non-Resolved Object Characterization