Magnetic Type Classification of Sunspot Groups Based on Deep Learning

Solar activity, as a significant manifestation of energy release and material movement in the solar atmosphere, is the main disturbance source of space weather. The violent solar activity represented by sunspots may lead to drastic changes in the near-earth space environment, and then have a profound impact on human production and life. Accurate and efficient prediction of space weather is helpful to reduce its impact on human production. In this paper, a magnetic type classification model of sunspot Mount Wilson based on squeeze-and-excitation module and deep residual network is established by using the continuum map and magnetogram map data observed by the HMI instrument on the Solar Dynamics Observatory (SDO) from 2010 to 2017. In order to effectively avoid the problem of model overfitting caused by the continuity of time series, this paper uses the time series segmentation method to divide the data set, and applies the data augmentation strategy combined with the characteristics of sunspot images to improve the generalization ability of the model. The experimental results show that the model proposed in this study can perform the task of sunspot classification accurately, especially in the recognition of complex sunspots, and its recognition ability has been significantly improved compared with traditional methods. In addition, this paper uses the class activation mapping method to visualize the test set samples, analyzes the feature images extracted from the model and the classification basis, so as to improve the interpretability of the model.