In the current context of the rapid growth in satellite launches and orbital objects, it is crucial to precisely understand the space object population and its evolution over time. Satellites and space debris represent a growing concern for astronomers and an increasing risk for space operators.

This work presents a pipeline developed to detect space objects in archival data from large field-of-view telescopes, focusing specifically on the VST/OmegaCAM dataset. Our aim is to reliably identify objects serendipitously captured by the OmegaCAM imager, assess their impact on VST’s scientific operations, and prepare detections for photometric studies to characterize the objects.

We describe the adapted deep learning architecture used for streak detection in astronomical images, along with the datasets employed for training, validation, and testing. Additionally, we introduce a convolutional neural network classifier trained to distinguish true detections from false positives. Using this pipeline, we analyze five years of OmegaCAM data.

The detection algorithm achieves an F1-score exceeding 95%, and when combined with the classifier to discard false positives, the overall F1-score improves beyond 97%, demonstrating impressive reliability and detection capability. Over the five-year period, our study reveals a clear trend of increasing impact of space object streaks on scientific images, especially following the deployment of mega-constellations.

Understanding and characterizing the space object population is now more important than ever. Our pipeline serves as a valuable tool for assessing the impact of streaks on astronomical images and for detailed studies of the detected objects. It is easily adaptable to other telescopes, and in support of open science, we are making our detection results freely accessible through a web interface.