Timothy Olson, Slingshot Aerospace; Clarice Reid, Slingshot Aerospace; Cam Key, Slingshot Aerospace; Brian Williams, Slingshot Aerospace; David Witman, Slingshot Aerospace; Jeff Shaddix, Slingshot Aerospace; Dylan Kesler, Slingshot Aerospace; Belinda Marchand, Slingshot Aerospace
Keywords: Space Situational Awareness, Space Domain Awareness, Machine Learning, Predictive Analytics, Photometric Change Detection, Satellite Characterization, Low Earth Orbit, Simulation, Photometric Fingerprinting
Abstract:
Space situational awareness (SSA) is becoming increasingly important as the number of objects in low Earth orbit (LEO) grows exponentially. Due to the congested and dynamic background of the LEO population, objects can be challenging to differentiate, and potential hazards can go unnoticed as satellites undertake activities that are not discernable from astrometric observations alone. These challenges can be addressed using photometric (brightness) information from optical measurements to enhance orbital intelligence by providing complementary signals about individual resident space object (RSO) characteristics. This work demonstrates a two-step machine learning (ML)-based approach for characterizing satellites in LEO via optical sensors. First, we present a novel photometric fingerprinting algorithm that enables object characterization, identification, and classification. Second, we examine methods for embedding and comparing fingerprints with potential applications to attitude change detection using photometry. We demonstrate performance using a wealth of real LEO observations. While versions of these ideas have proven effective for assessing RSOs in other orbital regimes, the use of photometric fingerprinting for object characterization and change detection has not been previously demonstrated at scale in LEO.
To generate photometric fingerprints for RSOs in LEO, we collect a diverse set of optical measurements for each object at a variety of viewing geometries and solar illumination angles. Effectively, each measurement represents a different projected view of the satellite body under specific illumination conditions. We combine these measurements into a high-dimensional photometric fingerprint for the object by applying a multidimensional embedding function to observed brightness measurements, thereby establishing a unique visual signature for each RSO. Our fingerprinting techniques differ from other “all-sky characterization” and “effective albedo” methods, though the approach bears some key similarities via the embedding of observations into a hyperspace defined by the viewing and illumination angles. However, we generate the embeddings by distinct methods and apply them to real LEO observations.
These fingerprints, when combined with existing astrometric data, strengthen the ability to differentiate RSOs, which helps retain custody of objects on orbit, an important component of SSA. Photometric characteristics are often underutilized in uncorrelated track processing; by comparing the measured light curve data to our fingerprint database, candidate associations are provided to improve the identification process. Likewise, when new observations of a candidate object are collected, the obtained measurements can be compared to the existing fingerprint for that object to validate that the observations match the expected signature at those viewing geometries. If they match, these observations are added to the fingerprint to improve the confidence of the fingerprint, whereas a mismatch could indicate a cross-tag with another object which can be resolved by downstream processes, thus reducing the likelihood of catalog missassociation.
However, a photometric mismatch could instead indicate that the spacecraft has physically reoriented. Attitude changes may occur as a routine mission component or as a precursor to significant events such as maneuvers or orbital changes. Possible extensions of the model to provide early detection of attitude changes via fingerprint shifts are discussed. An event-detection algorithm using these signals could generate important alerts for tracking and monitoring systems to prepare for significant events that may affect planning and object association algorithms. Related approaches have been studied extensively for GEO applications, but this is a new development for the LEO regime.
The ML models presented in this paper require large and diverse data sets of photometric measurements. To facilitate this project, we used real data collected by Slingshot’s Horus optical fence system, which provides global staring coverage for LEO objects, sampling photometric observations of LEO from a diverse set of sites. The efficacy of the new fingerprinting approach for distinguishing RSOs was then tested by evaluating its performance at separating objects from three closely related families of satellites.
Accurate detection, tracking, identification, and characterization of resident space objects (RSOs) is vital to reducing on-orbit hazards. The ML-based algorithms in this paper directly address some of the most salient challenges in this domain.
Date of Conference: September 17-20, 2024
Track: Satellite Characterization