A Magnetic Classification Method of Sunspot Group Based on Dual-stream CNN

Most solar flares occur over sunspot groups with complex magnetic polarities, so the magnetic type of sunspot groups can be used as an important basis for predicting solar flares. A sunspot group magnetic classification method, dual-stream CNN (Convolutional Neural Network) is proposed for sunspot classification with both continuum and magnetogram. The method simultaneously extracts the continuum and the magnetogram features through two network branches, then fuses the two types of features in the fully connected layer, and finally uses them for training and classification. Experimental results show that the method can avoid the information loss of single-channel network model and the interaction of continuum and magnetogram in the dual-channel image, and the weighted F₁ score is 84.87%, and the average accuracy is 84.93%.