Ensemble Methods for Automated Asteroid Detection

Raul Urechiatu

Abstract

Near-Earth Objects (NEOs) and especially Near-Earth Asteroids (NEAs) are potential threats due to their proximity to Earth. For this reason, we need to constantly survey the nearby space and continuously monitor NEOs and NEAs to prevent future impacts with our planet. While some automated methods exist they focus on image processing and complement the work of human image reducers. Their main task is to identify asteroids from stars and consists of visually analyzing hundreds if not thousands of images taken during sky survey. Since these asteroids need to be reported quickly to take credit for them the analysis usually happens in parallel with the image capturing over the course of several nights. In this work we present a machine learning algorithm combining three neural networks designed to identify these objects and to reduce the effort of human image reducers. The algorithm will be integrated into an existing software called NEARBY developed by a consortium made of UTCN, UCV, and UVT. For validation, we rely on several thousands of images taken by the La Palma astronomical observatory.

This work is in collaboration with Calin Dumitrescu and dr. Marc Frincu and was conducted in the framework of the CERES project (Classification software module of asteroids in satellite imagery using machine learning - PNIII-PED, 2020-2022).