Multiple CNN Variants and Ensemble Learning for Sunspot Group Classification by Magnetic Type

Abstract A solar active region is a source of disturbance for the Sun–terrestrial space environment and usually causes extreme weather, such as geomagnetic storms. The main indicator an sunspots. Certain types sunspots are related to weather caused by eruptive events coronal mass ejections or flares. Thus, automatic classification sunspot groups helpful predict activity quickly accurately. This paper completed group data set based on Mount Wilson scheme, which contains continuum magnetogram images provided Solar Dynamics Observatory’s Helioseismic Magnetic Imager SHARP from 2010 May 1 2017 December 12. After applying some preprocessing steps image cropping standardization, features magnetic type in more obvious, amount increased. processed spliced into two frames single-channel neural network perform 3D convolution operations. constructs variety convolutional networks with different structures numbers layers, selects 10 models representatives, chooses XGBoost, commonly used ensemble-learning algorithms, fuse results independent models. We found that XGBoost effective way models, proved relatively balanced high scores three types. accuracy ensemble model above 92%. F1 Alpha, Beta, Beta-x reached 0.95, 0.91, 0.82 respectively.