A Wavelet–Random Forest Framework for Automated Detection of Satellite Trails

Artificial satellite trails are becoming an increasingly important challenge for ground-based telescopes, as they affect the quality of the observations needed for scientific analysis. In this work, we present a machine learning approach based on a wavelet-transform encoder whose features are used as input to a Random Forest binary classifier that detects the presence of trails in astronomical images. Using both real and simulated data, the method achieves an accuracy of 0.9721 with a low false-positive rate (0.0070), which is crucial for subsequent operational analysis and for addressing space debris in Earth’s orbit. This approach not only enables the automation of the analysis, considering the large datasets involved, but also provides a robust technique that leverages the benefits of the 2-level wavelet decomposition of the input images.