Elisabeth Rachith, Laboratory of Astrophysics EPFL LASTRO, Observatoire de Sauverny; Belén Yu Irureta-Goyena, Laboratory of Astrophysics EPFL LASTRO, Observatoire de Sauverny; Jean-Paul Kneib, Laboratory of Astrophysics EPFL LASTRO, Observatoire de Sauverny
Keywords: satellites, space debris, Hough transform, machine learning, streak detection
Abstract:
A growing number of satellites and space debris orbit the Earth. This increasing population represents both a danger to space operations and a disruption for astronomical observations by adding unwanted signal on top of the astrophysical data. It is therefore critical to monitor and quantify the population of satellites and space debris to understand and prevent the consequences of their increasing number.
A way to identify satellites and space debris in the visible band is to look at the characteristic streaks of light they leave in astronomical images taken from ground-based telescopes and compare them to existing satellite catalogs. This allows for the assembly of a comprehensive overview of the global satellite and space debris population, as well as for the identification of unlisted resident space objects. Orbital data, as well as satellite and space debris ephemerides, can also be corrected and their accuracy improved when compared with the actual measurements collected. The first step in this direction is to detect the traces left by satellites and space debris on astronomical images and the work presented here evaluates two detection methods.
The first method uses in particular the Hough transform algorithm for the detection of streaks in raw astrophysical images. The input images raw unstacked astronomical images are first cleaned of elements that could hinder the detection of satellites and space debris using Gabor filters, a top-hat transform and various thresholding carefully optimized to prevent unwanted detection with the Hough transform while ensuring a minimum of information loss. The elements that need to be removed from the images are bad columns, persistence effects and saturated bands among others, as well as very bright stars that can interfere with the detections. A Canny filtering algorithm is then applied on the images to detect the edges of the streaks left by the satellites and space debris crossing the field of view and a Hough transform algorithm finally differentiates the traces of the other elements of the image.
The second detection method evaluated is based on the adaptation of a machine learning algorithm initially proposed by Lin et al. for line detection in an urban environment. It is composed of a neural network of the LCNN type (Lookup based Convolutional Neural Network) enhanced with a trainable Hough transform prior block. The ability of the algorithm to correctly detect satellite tracks and space debris in astronomical images is evaluated after different training phases and the algorithm is finally applied, as for the detection method using the Hough transform, on a subset of images obtained with the 2.5m VLT Survey Telescope (VST).
The benefits and drawbacks of both methods are presented and discussed and ultimately, using the most accurate and complete method for streak detection and thanks to the sensitivity of VST and its comprehensive archive, we expect to detect more satellites and space debris than predicted by existing catalogs. Our goal is to give a quantitative overview of the impact of light streaks of satellites and space debris on astronomical images as a function of time, thus providing statistical monitoring of the global satellites and space debris situation.
Date of Conference: September 27-20, 2022
Track: Space Debris