RUPDA: An Unsupervised Method for Magnetopause Tangent Direction Detection With Multiviewpoints in Low SNR

Initiated by the Chinese Academy of Sciences and the European Space Agency, the Solar Wind Magnetosphere Ionosphere Link Explorer (SMILE) mission aims to provide soft X-ray imager (SXI) images of the large-scale magnetopause near the subsolar region, offering valuable insights into understanding the space weather phenomena. Nonetheless, the SXI images acquired by the SMILE satellite have a low signal-to-noise ratio and multiple viewpoints, so how to quickly and accurately detect the magnetopause tangent directions from these images is the focus of the current study. In this article, we introduce a novel method for high-accuracy magnetopause tangent direction detection from low signal-to-noise ratios and multiviewpoints SXI images. A residual attention U-Net architecture first is proposed to overcome the detection difficulties due to SXI image characteristics and improve the detection accuracy. In addition, the proposed two-stage training strategy based on unsupervised domain adaptation (UDA) using auxiliary classifiers, pseudo-label learning, and other techniques makes the model transfer from theoretical Jorgensen and Sun model data to magnetohydrodynamic (MHD) model data, thus enhancing the practical applicability of the model in the SMILE mission. Finally, through a series of comparison and ablation experiments coupled with tangent fitting approach (TFA) experiments, we validate that our proposed network and strategy effectively detect the tangent direction of the magnetopause in SXI images.