| |
 |
UNIVERSITY OF BUCHAREST
FACULTY OF PHYSICS Guest
2026-06-11 23:58 |
 |
|
|
|
Conference: Bucharest University Faculty of Physics 2026 Meeting
Section: Atmosphere and Earth Science; Environment Protection
Title:
All-Sky Camera Systems and Deep Learning for Space Object Detection: Implications for Ground-Based Astronomy
Authors:
Cristian OMAT (1, 2), Mirel BIRLAN (1, 2, 3)
*
Affiliation:
1) University of Bucharest, Faculty of Physics, Doctoral School of Physics, 405 Atomistilor street, Magurele, 077125, Ilfov, Romania
2) Astronomical Institute of the Romanian Academy, 5 Cutitul de Argint, Bucharest, 040557, Romania
3) LTE, Observatoire de Paris, Université PSL, Sorbonne Université, Université de Lille, LNE, CNRS, 61 Avenue de l’Observatoire, Paris, 75014, France
E-mail
cristian.omat@astro.ro
Keywords:
LEO, satellite, space debris, YOLO, deep learning
Abstract:
The exponential deployment of satellites in low Earth orbit (LEO) has created urgent challenges for ground-based astronomy and space situational awareness. This work, part of my PhD thesis, develops a YOLOv12-based deep learning approach to automatically detect satellites and debris trails in all-sky images, enabling existing all-sky camera networks to serve as cost-effective, passive SST contributors without hardware modifications.
We acquired and annotated 8,596 all-sky images from Berthelot Observatory (January–March 2025), covering 358 transits of active satellites and space debris. After augmentation, the dataset comprised 13,162 images split into training (82%), validation (11%), and test (7%) subsets.
We integrated the trained model into an operational SST workflow where incoming all-sky frames are automatically classified as "contaminated" (containing artificial-object trails) or "clean" (science-ready), simultaneously protecting scientific data quality and creating a timestamped archive of satellite and debris transits. This approach requires no new hardware, only existing all-sky cameras and modest cloud-based computational access.
The results demonstrate that all-sky camera networks, originally deployed for meteor detection and sky monitoring, can be repurposed as distributed, low-cost assets for Space Surveillance and Tracking. By automating frame triage and generating time-correlated overflight records, these systems enable small observatories and amateur stations to contribute meaningfully to space situational awareness at negligible operational cost. This represents a practical, scalable pathway for large participation in SST activities across the existing global network of all-sky cameras.
Acknowledgement:
This work is supported by the project ”Romanian Hub for Artificial Intelligence-HRIA”, Smart Growth, Digitization and Financial Instruments Program, MySMIS no. 351416
|
|
|
|