Enhancing Meteor Detection with Machine Learning
Abstract
Introduction: The use of artificial intelligence (AI) is growing across various fields, including meteor science. This study focuses on applying AI via several tools to improve the detection and identification of meteors from single-station observations, which often struggle with high rates of false positives. Methods: We used data from the Meteorites Orbits Reconstruction by Optical Imaging Network (MOROI) in Romania, covering events from 2017 to 2020. Fifteen ML models were trained on features derived from the movement of meteors across the camera's CCD. These models were selected for their effectiveness in tabular data classification and feature independence, to enhance scalability. Results: Out of the 24 features designed from the detection data, 7 significantly improved the classification accuracy. The best models achieved an accuracy of 98.2% and a recall of 96%, which increased to 99.92% when combining detections. These methods and features can be applied to other optical detectors, being compatible with a large diversity of cameras. Conclusions: ML models significantly improve the identification of real meteors in single-station observations. Performance varied among models due to differences in sensitivity to outliers and noise. Ensemble models like Gradient Boost and Random Forest showed better performance in dealing with imbalances and noisy data. This study highlights the potential of ML in refining meteor detection data, especially from single stations. The approach is scalable and applicable to other networks, improving meteor analysis as a whole.
- Publication:
-
EAS2024
- Pub Date:
- July 2024
- Bibcode:
- 2024eas..conf..891A