Convolutional neural network for classification of solar radio spectrum

Abstract

This paper makes the first attempt to utilize convolutional neural network (CNN) for classification of solar radio spectrums. The solar radio spectrum is a two-dimensional gray-scale image with one dimension of frequency and the other of time. Taking the advantages of CNN, we can efficiently learn the distinct characteristic of different types of spectrum, and further classify them even more accurate. The proposed CNN-based network consists of four convolution layers, four pooling layers and one fully connected layer. Its input is spectrums of the size 120×120. The output gives the type of each spectrum among “burst”, “non-burst” and “calibration”. Experimental results demonstrate that the proposed CNN can achieve more accuracy of classification of solar radio spectrum beyond our previous efforts by employing deep belief network (DBN) and autoencoder (AE).