学术文献库

An important task within the broader goal of Space Situational Awareness (SSA) is to observe changes in the orbits of satellites, where the data spans thousands of objects over long time scales (decades). The Two-Line Element (TLE) data provided by the North American Aerospace Defense Command is the most comprehensive and widely-available dataset cataloguing the orbits of satellites. This makes it a highly-attractive data source on which to perform this observation. However, when attempting to infer changes in satellite behaviour from TLE data, there are a number of potential pitfalls. These mostly relate to specific features of the TLE data which are not always clearly documented in the data sources or popular software packages for manipulating them. These quirks produce a particularly hazardous data type for researchers from adjacent disciplines (such as anomaly detection or machine learning). We highlight these features of TLE data and the resulting pitfalls in order to save future researchers from being trapped. A seperate, significant, issue is that existing contributions to manoeuvre detection from TLE data evaluate their algorithms on different satellites, making comparison between these methods difficult. Moreover, the ground-truth in these datasets is often poor quality, sometimes being based on subjective human assessment. We therefore release and describe in-depth an open, curated, benchmark dataset containing TLE data for 15 satellites alongside high-quality ground-truth manoeuvre timestamps.