Research On The Method For Identification And Parameter Extraction Of Solar Radio Spikes

Solar radio spikes have the characteristics of short lifetime and narrow bandwidth in the solar radio signal,which is the most special radio fine structure that can be observed so far.The observation data of solar radio spikes can reflect the process of tiny energy release in the solar corona and small-scale changes in the magnetic field during the solar radio burst,and its average frequency bandwidth can also indirectly reflect the spatial scale of the burst source region.At present,most of the extraction methods for solar radio spikes are based on data methods,which not only require a lot of manpower and time to process,but there may also be problems such as missed inspections.This thesis proposes two automatic extraction methods of data processing and deep learning to detect solar radio spikes and extract their parameters.According to the criteria for determining solar radio spikes,this thesis uses the data extraction method such as outlier analysis method,the Mean shift algorithm and the Gaussian fitting algorithm in turn to extract the bounding box of solar radio spikes.The experimental results show that the data extraction method has high positioning accuracy and can basically meet the detection requirements.However,the data extraction method has problems such as the inability to detect the frequency drift rate of spikes and the time-consuming detection.In this thesis,the improved YOLOv5s lightweight network model is used to locate,identify and analyze the characteristics of spikes.According to the characteristics of small target,irregular shape and frequency drift rate of solar radio spikes,multi-scale fusion and attention mechanism module are integrated into the network,and the angle information of the bounding box is added to improve the original structure of the network.The experimental results show that the AP value detected by the improved YOLOv5s network proposed in this thesis is 74.033%,which is nearly 14%higher than that of the YOLOv5s network that only adds angle information,so it can better detect and mine solar radio spikes and their characteristics.Finally,this thesis summarizes the extraction process of solar radio spikes.The events observed in the 1.1-1.34 GHz and 150-500 MHz frequency bands are used as datasets to perform statistics and analysis on the detection results of the solar radio spikes.The comparison of the statistical results of other researchers proves the effectiveness of the method used in this thesis.