Solar Radio Spectrum Classification Based on ConvLSTM

Abstract

AbstractSolar radio bursts are a kind of instantaneously enhanced radio waves from the Sun. The high-intensity signals can seriously interfere with radio communication and navigation systems. At present, the traditional method is to manually analyze the radio frequency spectrum to find out the solar radio bursts, which has low efficiency and time lag. If the solar radio spectrum can be automatically classified in real time, it will have positive significance for space weather early warning and solar physics research. An automatic classification method of solar radio spectrum using deep learning technology is proposed in this paper. First of all, due to sample imbalance and small number of samples, the data set is expanded according to the characteristics of solar radio spectrogram. Then, spectrum normalization and down-sampling are used to preprocess the solar radio spectrum. Finally, considering the radio spectrum has both the spatial characteristics of the image and the time–frequency characteristics of the sequence signal, convolutional long short-term memory network (ConvLSTM) is used to classify the solar radio spectrum in the paper. ConvLSTM also takes into account the spatial and temporal characteristics of the solar radio spectrogram, which is useful for classifying radio spectrograms that have both of these features. Compared to the method that only considers one of the characteristics, the experimental results show that the method achieves better results than the method using only a single feature.KeywordsSolar radio burstConvLSTMDeep learningImage classification