A self-organizing map-based approach to automatic meteor detection in radio spectrograms

The automatic meteor detection solution presented in this paper uses a self-organizing map to analyze radio spectrogram data and detect the meteor samples found within. This artificial neural network is trained using data samples extracted from spectrograms of radio recordings using a rectangular sliding window. Several tests were run to find the optimal neural network topology and duration of training. The trained network is then analyzed using a new set of data and its performance is manually validated. Testing has shown that the proposed self-organizing map solution produces significant meteor detection results.