基于CornerNet-Saccade的太阳黑子群磁分类研究

太阳黑子与太阳活动联系紧密,是太阳表面强磁场的典型表现,如大部分太阳耀斑爆发在复杂的太阳黑子群上方。研究太阳黑子群磁分类对太阳耀斑的预测具有重要意义,因此针对黑子群分类问题制作了太阳黑子群磁分类数据集,基于深度学习目标检测算法CornerNet-Saccade设计了CNSF(CornerNet-Saccade-Fast)算法,改进了CornerNet-Saccade的边界框回归(Bounding Box Regression)和边界框融合阶段,提高了在太阳黑子群磁分类数据集上的检测精度。此外,通过精简骨干网络提高了检测效率。实验结果显示,CNSF在太阳黑子群磁分类上具有良好的性能,其Precision和Recall均达到了0.93,mAP达到了0.91。相较于CornerNet-Saccade,CNSF在检测精度和效率上分别提升了约4%和13%。

Research on the Magnetic Classification of Sunspot Groups Based on CornerNet-Saccade

Sunspots are closely related to solar activities.They are typical manifestation of the strong magnetic fields on the solar surface.Such as most solar flares erupt above the complex sunspot groups.Research on the magnetic classification of sunspot groups is of great significance to the prediction of solar flares.Aiming at the problem of sunspot groups classification,a magnetic classification data set of sunspot groups is produced,and CNSF(CornerNet-Saccade-Fast)algorithm is designed based on the deep learning object detection algorithm of CornerNet-Saccade,which improves the stages of Bounding Box Regression and bounding box fusion of CornerNet-Saccade,the detection accuracy on the sunspot groups magnetic classification data set is improved.Not only that,by streamlining the backbone network,but also improving the detection efficiency.The experimental results show that this method has a good performance in the magnetic classification of sunspot groups,through which the Precision,Recall and mAP reach up to 0.93,0.93 and 0.91,respectively.Compared with the CornerNet-Saccade algorithm,CNSF has improved detection accuracy and detection efficiency by approximately 4%and 13%,respectively.