Kai Garcia, Cloudstone Innovations; Nathan Toner, Cloudstone Innovations
https://doi.org/10.64861/FEQX3431
Keywords: Photometry, characterization, wide field of view, machine learning
Abstract:
The growing population of resident space objects (RSOs) is increasing the demand for scalable, automated object characterization within space domain awareness (SDA). Although wide-field optical sensors routinely provide observations for object detection and astrometric tracking, the temporal photometric information contained within long-exposure satellite streaks remains largely underutilized. This work presents a framework for recovering calibrated high-rate photometric signatures from long-exposure wide field-of-view (WFOV) observations and demonstrates their application to automated resident space object type classification.
Our approach recovers calibrated high-rate visual magnitude time series from satellite streaks in WFOV imagery by estimating photometric measurements over temporal sub-segments along each streak. This allows for reconstruction of intra-exposure brightness variability for multiple objects simultaneously without requiring prior knowledge of their trajectories or dedicated tracking.
The recovered photometric signatures are used to train a supervised deep learning classifier using object-type labels derived from publicly available satellite catalogs. Evaluation using object-level grouped partitions, in which no catalog object appears in more than one split, achieved 62.4% classification accuracy, a macro F1 score of 54.1%, and a balanced accuracy of 57.2%. By enabling photometric time-series analysis from single-exposure WFOV imagery, this framework extends survey-mode optical sensing beyond detection and astrometry toward automated resident space object type classification within SDA systems.
Date of Conference: September 15-18, 2026
Track: Satellite Characterization