Classification and Characterization of "Ugly" Specular Radar Meteors using Machine Learning

Specular meteors, characterized by their reflective properties, offer valuable insights into atmospheric composition, dynamics, and meteor properties. Traditional methods of classifying meteors rely on manual analysis or mathematical signal processing techniques. We have identified two problems with this approach. Firstly, these methods exclude all except a very small selection of “perfect” underdense and high-SNR meteor trails. These “perfect” meteor trails are useful for determining atmospheric winds, meteor radiant distributions, velocities, and more, but there may be additional valuable insights about the atmosphere that are overlooked by rejecting many of the observed meteors. Secondly, the signal processing techniques must be designed for specific radar systems operating with their own settings, such as power, frequency, PRF, and more. Thus, the data processing techniques must be modified to analyze meteor detections from different systems, costing time and energy. By leveraging advancements in machine learning, this study aims to develop an automated classification framework for specular meteors detected by all-sky radars, with the goal of extracting new information about the atmosphere and meteor properties from these detections, as well as offering a fast and effective method of signal processing that is applicable to specular meteor radar data regardless of the system used to make the observations. This tool will also be made available for community use as an open-source program.