Identification and Extraction of Solar Radio Spikes Based on Deep Learning

Abstract

Solar radio spikes are short-duration, narrow-band burst signals, which are a fine structure of solar radio bursts. The processing and analysis of their observed data are of great significance in the study of electron acceleration in the process of solar flares and electron acceleration during the explosion and diagnosis of corona parameters. Deep learning interprets data by mimicking the mechanism of the human brain. Faster Region-based Convolutional Neural Network (Faster R-CNN) is a branch of deep learning based on region nomination, and its classification results have considerable advantages in accuracy. In this paper, Faster R-CNN will be used to identify and extract solar radio spikes. In order to improve the detection ability of small events, a multi-scale detection frame and a multi-layer feature fusion training method are used. The analysis results show that the Average Precision (AP) value of the improved network is close to 91%, which is nearly 10% higher than the original network. So the improved Faster R-CNN method can also be used for the identification and extraction of small-scale fine structures in other fields.