Deep Active Learning–Based Classification of Solar Radio Spectrogram Data

Abstract The study of solar burst activity can provide early warnings for the environmental protection of the solar–terrestrial space environment. With the improvement of solar radio observation techniques, observation devices have generated enormous amounts of observation data. To solve the shortcomings of time-consuming and error-prone manual recognition, researchers have begun to use deep learning to recognize and automatically classify solar radio outbursts. Deep learning will depend on a large number of labeled samples; however, the labeling of samples requires a lot of time and manual labor. This leads to low efficiency. In addition, the labeled samples are not all valuable samples, so it is necessary to improve the effectiveness of the labeled samples and select the high-value samples. The occurrence of active-learning techniques provides an opportunity to solve this problem. In this study, we developed a progressive deep convolutional generative adversarial network model. Then, we combined it with deep active learning to complete the automatic classification of solar radio spectrum data. We used solar radio spectrum data from the Chashan Observatory (CSO) of Shandong University and Learmonth Observatory in Australia. The results show that the method proposed in this paper can achieve high accuracy in the automatic recognition of solar radio spectrum data and solve the time-consuming problem of labeling a huge number of data samples. Finally, we applied the results to the CSO and realized the automatic recognition of solar radio spectral data.