Machine learning methods applied to filter meteor detections.

The rise in data collection and storage capabilities introduced a challenge for both finding connections between data sets and extracting the valuable information. This phenomenon is also found in meteor science, when false meteor detections make their way in trajectory computations, and will lead to erroneous orbits. During this talk, a set of machine learning (ML) methods will be presented, along with their capability to classify the meteors detected in Meteorites Orbits Reconstruction by Optical Imaging network (MOROI) between 2017-2020. First, a set of regression based features were extracted from the meteor centroid in each frame, next the ML models were tested and tuned to obtain the highest score. The Neural networks method was found to best filter out the false meteors, with a recall score of 96%, followed by 95% for Gradient Boost and Random Forest algorithms. When combining this with the spatio-temporal data from other stations, the recall increases to 99.92%. The results entail follow-up studies on the larger FRIPON network, which is already collecting data in realtime.