Trevor Putman, Johns Hopkins Applied Physics Lab; Major Caleb McComas, Space Security Defense Program; Lt. Col.Julia Faustman, Space Security Defense Program
Keywords: fusion, Bayesian, classifier, characterization, identification
Abstract:
The space domain is becoming increasingly crowded with proliferated constellations and advanced, highly maneuverable satellites. New mission sets, such as on-orbit servicing, introduce space scenarios that stress our current Space Domain Awareness (SDA) approach. These complex space scenarios include large maneuvers, rendezvous proximity operations (RPO), and deployment events. Maintaining accurate identification of satellites is a critical component of the SDA mission. Traditionally, space operators identify satellites by associating their kinematic information with a known catalog. The trend of increasing complexity in the space domain suggests a kinematic approach may prove insufficient to maintain identification of space objects in the near future. There is a compelling need to provide space operators with the ability to leverage information beyond traditional kinematics when determining the identity of orbital objects.
As the complexity of space operations has increased, so has the quantity and availability of SDA data. Commercial SDA systems have enabled lower-latency, high-frequency collection of kinematics and unresolved signature information on satellites in all orbital regimes. Unresolved signature information refers to sensor measured information such as radar cross section, visual magnitude, and passive radiofrequency collections. While individual sensors may use non-resolved signature data to distinguish objects within their field of view, complex space scenarios indicate a need to incorporate multiple phenomenologies to achieve an accurate characterization.
This study explores the use of unresolved signature information across multiple phenomenologies to identify satellites with a calculated confidence. The result of this study will be a proposed framework and classification algorithm for using multiple phenomenologies to classify a satellite. Additionally, the study highlights future force design and data exposure recommendations to improve SDA systems’ capabilities for classifying satellites. The study defines an algorithmic and architectural approach to providing a classification capability as a decision aid to SDA Command and Control (C2) system operators.
This study explores multiple variations of Naïve Bayesian classifiers for use in satellite identification. Bayesian classifiers are particularly well suited to this problem set due to their ease of explanation, low computational complexity, and statistics-based confidence values. They are also robust to incomplete datasets which is a critical need in the data sparse space domain environment. We will train and evaluate the classification algorithms against real-world scenarios using historical datasets pulled from the Unified Data Library (UDL). Our algorithm’s objective is to label the orbital object with the correct satellite catalog number based on a comparison of current observed signature information with historical-based signature profiles.
Our results indicate significant utility in using unresolved signature information to identify satellites. Bayesian classifiers prove to be a simple yet effective method to leverage this information, providing a 95% accuracy across historical scenarios. Furthermore, analysis of the reported confidence identifies clear thresholds for using the output of the algorithm as a decision aid to space operators. A sensitivity analysis of contributing phenomenologies highlights the utility of each phenomenology for providing a classification call. Furthermore, the sensitivity analysis highlights key areas for investment to improve satellite identification capabilities.
In this study, we established the usefulness of unresolved signature information, but space operators need to be able to convert these signatures into actionable decision aids. Our study defines a clear path forward to provide operators with defendable identification calls in complex scenarios.
Date of Conference: September 17-20, 2024
Track: Satellite Characterization