Imaging and representation learning of solar radio spectrums for classification

In this paper, the authors make the first attempt to employ the deep learning method for the representation learning of the solar radio spectrums. The original solar radio spectrums are pre-processed, including normalization, enhancement and etc., to generate new images for the next processing. With the expertise of solar radio astronomy for identifying solar radio activity, we build a solar radio activity database, which contains solar radio spectrums as well as their labels indicating the types of solar radio bursts. The employed deep learning network is firstly pre-trained based on the available massive of unlabeled radio solar images. Afterwards, the weights of the network are further fined-tuned based on the labeled data. Experimental results have demonstrated that the employed network can effectively classify the solar radio image into the labeled categories. Moreover, the pre-training process can help improve the classification accuracy.