Clarice Reid, Slingshot Aerospace; David Witman, Slingshot Aerospace; Max Geissbuhler, Slingshot Aerospace; Jeffrey Shaddix, Slingshot Aerospace; Belinda Marchand, Slingshot Aerospace; Adrian Thompson, Slingshot Aerospace
https://doi.org/10.64861/XDRJ3507
Keywords: RPO, GEO, ML, Correlation, Characterization
Abstract:
Rendezvous and proximity operations (RPO) events have become more commonplace in geosynchronous orbits (GEO) due to the rise in mission extension on-orbit servicing. Refueling, repairing, and/or repositioning GEO spacecraft may provide a better cost benefit than designing, constructing, and launching new satellites. The volume of such missions is expected to grow as more spacecraft are deployed to GEO, legacy satellites exhaust their fuel reserves, and operators look to extend their vehicles’ lifespans. As these RPO events unfold, it is essential for operators and situational awareness systems to maintain accurate state information between the multiple resident space objects (RSOs) involved. This can prove challenging for existing ground-based electro-optical space surveillance networks. Discriminating closely spaced objects, while improved by certain optical configurations and algorithms, will ultimately collapse if docking is achieved. Thus, maintaining proper correlation (or identification) of the RSOs as they enter and exit RPO is vital. In this paper, the authors propose a novel method to achieve consistent correlation through use of photometric fingerprinting.
From the perspective of a small aperture ground-based electro optical observer, objects detected in RPO conditions are typically unresolved, leaving only astrometric (“where” the objects are detected) and photometric (“how bright” the detected objects are) information to use for RSO identification. As the RSOs approach and separate during RPO events, the close relative separation and frequency of small maneuvers prohibit astrometric data alone to be sufficient for accurate RSO correlation. This challenge is exemplified by the recent on-orbit activities of SJ-21 and SJ-25. In July 2025, SJ-21 and SJ-25 docked while on-orbit in GEO and remained docked until January 15, 2026 (per public reporting sources). Initially when the objects separated, space-track.org cross-tagged the two objects, (i.e. correlating SJ-21 as SJ-25, and vice-versa), before updating its correlations on January 30, 2026.
This paper will discuss a method that uses a machine learning (ML) algorithm to classify and differentiate objects based on their photometric signatures in GEO orbits, extending on prior work. The algorithm uses historical photometric data and measurement geometries to build a unique representation (or fingerprint) of each object. Future tracks collected on RSOs can then be compared to these fingerprints for correlation. The features can also be compared across other objects to determine similar satellite bus geometries and materials. Patterns of life can be baselined to enable detection of changes in shape and attitude. We have previously demonstrated our model’s utility in accurately correlating previously uncorrelated tracks (UCTs) generated during RPO events when astrometric information alone was insufficient. We likewise presented on our model’s scalable application to identifying the satellite bus classes of LEO assets. By extending this work to GEO objects, ML-based photometric fingerprinting can provide operators deeper real-time scalable insights into ongoing and future operations. This paper will first provide a recap of the underlying models being used and associated feature processing required, with notation on the differences between this GEO approach and previous LEO work. Then, we will present performance metrics used for verification and validation of our approach on a representative set of data. Finally, we will investigate specific uses of the GEO ML fingerprinting capability with a case study focus on the SJ-21 and SJ-25 RPO event. We will consider the unique insights that such an algorithm can provide to improve correlation for GEO RSOs during RPO conditions and discuss the implications of each insight in real-world process flows.
Date of Conference: September 15-18, 2026
Track: Satellite Characterization