(Cambridge University Press on behalf of International Astronomical Union, 2026-01-19) Drago González, Àlex; Núñez, Jorge
Optical observation is used to track objects in orbit, but it’s challenging for objects in Low-Earth Orbit (LEOs) due to high-speed satellites leaving long trails. We present a study that uses a multilayer neural network software called Source-Extractor (SExtractor) to accurately determine the position of stars and orbital objects. By modifying input parameters, we can draw ellipses around star trails, including very long ones. Using images from the TFRM telescope and Fabra Observatory, we accurately determine the position of objects in all types of orbits, including of 6000 km-high 1U CubeSats and up to magnitude 18.0 on 10-second exposure images following the object showing star trails over 400 pixels long. This technique can recognise all types of objects, including other satellites or debris, if bright enough. This technique can help study the negative impact on astronomical observations caused by megaconstellations of satellites like Starlink or OneWeb, for professional and amateur observatories.