Space Domain Awareness of Debris Evaluated Through Synthetic Data

Despite high expectations for artificial intelligence (AI) to improve on-orbit space domain awareness, the number of satellites equipped with AI has been limited. Existing projects demonstrated the possibility of AI deployment while facing constrained computational resources. One of the first uses of AI consisted of image quality analysis while on orbit, but other applications should be explored preceding novel technologies that will allow for future on orbit AI. This project aims to use AI to detect space debris that can pose the risk of a conjunction event. Data for this project is difficult to find since conjunction events are expected to be avoided in real satellites. Simulation software is employed to address the lack of available data. The simulation generates images of space debris detected using the You-Only-Look-Once object detection architecture. Object detection shows that space debris can be identified with more than 90% mean average precision even when using the simplest AI model, which occupies a memory size compliant with previous on orbit AI deployments such as the OPS-SAT. We conclude that AI integrated into space debris detection subsystems can be effective as a key component to mitigate and provide attribution for conjunction risks in future space missions.