Decentralized Global Satellite Tracking Network: Leveraging Crowd-sourced Astronomical Observations and Machine Learning for Cost Effective Orbital Monitoring

Abstract

With 50,000 current actively tracked objects in orbit and a clear upwards trajectory, especially as launch vehicle technologies improve, orbital density and the amount of assets in space is increasing, making space debris a more and more pressing issue. As active debris removal is difficult to make sustainable, active prevention through precise satellite tracking becomes necessary. While current ground based solutions are limited in update rate and are incredibly expensive making them hard to scale, this paper investigates the feasibility of a decentralized alternative where data gathered from a global network of amateur astronomers, classified using a machine learning model that automatically detects satellite trails in telescope imagery, providing position updates at higher frequency and lower costs than conventional systems. A ResNet-50 convolutional neural network was used for trail detection, trained on more than 11,000 images and achieving an overall accuracy of 98% and F1 score of 0.95 on a validation set demonstrating feasibility of use with a full network. A separate global network simulation using TLE orbital data modelling variations of a global network over a 48 hour period found that performance scaled almost linearly, with the best configuration of 50 stations globally and 5 observers at each station achieving an update rate more than 11x higher than the U.S. Space Force’s Space Surveillance Network baseline at a fraction of the cost. The results demonstrate both technical and economic viability of the decentralized network, with the potential to play a very meaningful role in preserving sustainable access to Earth orbit.