Refined STACK-CNN for Meteor and Space Debris Detection in Highly Variable Backgrounds
Leonardo Olivi,Antonio Montanaro,Mario Edoardo Bertaina,Antonio Giulio Coretti,Dario Barghini,Matteo Battisti,Alexander Belov,Marta Bianciotto,Francesca Bisconti,Carl Blaksley,Sylvie Blin,Karl Bolmgren,Giorgio Cambiè,Francesca Capel,Marco Casolino,Igor Churilo,Marino Crisconio,Christophe De La Taille,Toshikazu Ebisuzaki,Johannes Eser,Francesco Fenu,George Filippatos,Massimo Alberto Franceschi,Christer Fuglesang,Alessio Golzio,Philippe Gorodetzky,Fumiyoshi Kajino,Hiroshi Kasuga,Pavel Klimov,Viktoria Kungel,Vladimir Kuznetsov,Massimiliano Manfrin,Laura Marcelli,Gabriele Mascetti,Włodzimierz Marszał,Marco Mignone,Hiroko Miyamoto,Alexey Murashov,Tommaso Napolitano,Hitoshi Ohmori,Angela Olinto,Étienne Parizot,Piergiorgio Picozza,Lech Wiktor Piotrowski,Zbigniew Plebaniak,Guillaume Prévôt,Enzo Reali,Marco Ricci,Giulia Romoli,Naoto Sakaki,Sergei Sharakin,Kenji Shinozaki,Jacek Szabelski,Yoshiyuki Takizawa,Valerio Vagelli,Giovanni Valentini,Michal Vrabel,Lawrence Wiencke,Mikhail Zotov
DOI: https://doi.org/10.1109/jstars.2024.3397734
IF: 4.715
2024-06-04
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Abstract:In this article, we present cutting-edge machine learning-based techniques for the detection and reconstruction of meteors and space debris in the Mini-EUSO experiment, a detector installed on board of the International Space Station, and pointing toward the Earth. We base our approach on a recent technique, the STACKing method plus Convolutional Neural Network (STACK-CNN), originally developed as an online trigger in an orbiting remediation system to detect space debris. Our proposed method, the refined-STACKing method plus convolutional neural network (R-Stack-CNN), makes the STACKing method plus convolutional neural network (STACK-CNN) more robust, thanks to a random forest that learns the temporal development of these events in the camera. We prove the flexibility of our method by showing that it is sensitive to any space object that moves linearly in the field of view. First, we search small space debris, never observed by Mini-EUSO. Due to the limiting statistics, also in this case, no debris were found. However, since meteors produce signals similar to space debris but they are much more frequent, the R-Stack-CNN is adapted to identify such events while avoiding the numerous false positives of the Stack-CNN. Results from real data show that the R-Stack-CNN is able to find more meteors than a classical thresholding method and a new method of two neural networks. We also show that the method is also able to accurately reconstruct speed and direction of meteors with simulated data.
imaging science & photographic technology,remote sensing,engineering, electrical & electronic,geography, physical