Matthew Phelps, Space Systems Command, United States Space Force; Thomas Swindle, Space Systems Command, United States Space Force; J. Zachary Gazak, Space Systems Command, United States Space Force; Justin Fletcher, Space Systems Command, United States Space Force; Andrew Vandenberg, Air Force Research Laboratory
Keywords: spectroscopy, spectra, GEO, non-resolved, neural networks, pointing angle, pose, attitude, rotation, orientation
Abstract:
Accurately measuring the orientation of a given satellite can provide vital information in identifying the spacecrafts operational status and predicting its propagation into a future state. Measuring such a quantity using ground-based sensors has, however, proven to be an immense challenge for the majority of space objects of interest, especially those in geosynchronous orbit (GEO) with large orbital radii. In the GEO regime, conventional resolved imaging techniques are ill-equipped to resolve the spatial distribution of material about a satellite due to limited spatial resolution. In this work, we continue exploring the application of ground-based spectrometry to estimate the rotational state of a space object. Building on previous works of spectroscopy applied to tasks like positive-id and pose estimation, we train deep convolution-based neural networks on simulated spectroscopic renders to estimate the rotational state. By increasing the spectral range and extending the capacity of the pose-prediction head, we obtain an appreciable reduction in rotational error. We present current leading performance results on spectroscopic pose estimation and discuss inherent limitations.
Date of Conference: September 27-20, 2022
Track: Non-Resolved Object Characterization