Aidan Lorenz, ARKA; Shawn Abernethy, ARKA; Mike Fischer, ARKA; Jacob Griesbach, ARKA; W. Jody Mandeville, InTrack Radar Technologies; Sid Arora, InTrack Radar Technologies; Tim McLaughlin, InTrack Radar Technologies; Harshitha Challa, West Virginia University; Piyush Mehta, West Virginia University
Keywords: machine learning, object detection, convolutional neural networks, CNNs, LNT, debris detection
Abstract:
As the space domain awareness community increasingly prioritizes the detection of dim, uncataloged resident space objects (RSOs), traditional techniques such as rate tracking and velocity-matched filtering face practical limitations, especially when object motion is unknown. In this paper, we present potential avenues to address these limitations using deep learning-based approaches for object detection near the noise floor in ground-based electro-optical (EO) imagery. Leveraging simulated datasets generated with SatSim– containing objects of known visual magnitude and precise ground-truth– we investigate two key objectives: (1) assessing the impact of incorporating temporal context through multi-frame input data, and (2) evaluating the generalization of models trained exclusively on synthetic data to real EO imagery without fine-tuning. Our results show that the models employed effectively leverage temporal frame stacking to significantly improve detection performance for low-SNR objects and that the simulated-data-trained models demonstrate strong recall on real data, in addition to successful detections of RSOs missed by traditional algorithms. These findings support the integration of deep learning into operational EO detection pipelines to enhance sensitivity and scalability.
Date of Conference: September 16-19, 2025
Track: Machine Learning for SDA Applications