Automatic detection of meteors in spectrograms using artificial neural networks

Abstract

Artificial neural networks are widely used in classification problems due to their adaptability. In this paper, an original multi-layer perceptron is used to automatically detect meteors within radio recordings. The approach presented can be divided into two stages: data preparation and meteor detection. In the data preparation stage, samples are built from a number of the spectrogram's vertical lines. During the meteor detection stage, neural networks are trained using the inputs previously extracted, and their meteor detection capabilities are tested. The rate of correctly detected meteor samples by the neural networks trained in this paper is over 80%.