Speaker
Description
Konkoly Observatory is conducting the most successful NEO survey project in Europe with a total number of NEOs found in the past four years in excess of 250, with three imminent impactors discovered between 2022 and 2024. Recently, supported by the European Space Agency, we started the implementation of a new search technique that is using machine learning algorithms to accelerate real-time image analysis with the scope of finding extreme trailed images of the smallest and nearest NEOs passing by. We have created a custom deep-learning model that was trained on a large dataset of astronomical images and their associated annotations. In addition to the real observations from the Piszkesteto Mountain Station of the Konkoly Observatory, we have also created a huge synthetic photorealistic training dataset to improve the precision and accuracy of the neural network. As a result, the model successfully learnt to recognise patterns and features in the images that are indicative of NEOs and space debris. The main goal was to have an optimized deep learning model to perform this analysis in real-time, providing quick and reliable detection that is made possible by the AI-based robust image-artifact decomposition for false positive suppression. The outcome of this project is a service that can quickly and accurately detect NEOs and space debris on astronomical images, potentially increasing the number of discoveries and improving the speed and reliability of the discovery process. The system has been evaluated using a set of rigorous tests and is benchmarked against existing methods. We provide valuable insights into the feasibility of using deep learning techniques for this type of image analysis problem and will lay the groundwork for future work in this field.