Finite-Angle 2D Reconstruction Of Magnetosphere Based On Neural Network

There are high-valence ions in the solar wind.These high-valence ions can directly enter the polar cusp from the magnetic sheath region and collide with the neutral atmosphere that escapes from the earth's atmosphere.These radiant energy is superimposed on the radiation energy of the solar,and has a great influence on the Earth's atmosphere.They also carry information about the large-scale structure of the Earth's magnetosphere,and they play an important role in the study of magnetosphere physics.In recent years,soft X-ray imaging detection of the Earth's magnetosphere is the frontier direction of the magnetosphere research.Research on the method of reconstructing the boundary of the magnetosphere from two-dimensional X-ray imaging is an important research topic related to imaging detection.Therefore,the Chinese Academy of Sciences and the European Space Agency jointly proposed the China-Europe Joint Space Science Satellite Project: Solar wind Magnetosphere Ionosphere Link Explorer(SMILE).The X-ray imager on the satellite will image the Earth's magnetosphere.The method of graphics can be used to reconstruct opaque objects,but the magnetosphere is a transparent object,the visual reconstruction theory is not applicable here,so it is necessary to use Computer Tomography(CT)to reconstruct the magnetosphere structure.For the orbit design of the current SMILE satellite mission is difficult to meet the full coverage of the scanning angle,the magnetosphere can only be observed from a finite angle,which is a type of CT reconstruction problem with a finite angle.The traditional CT reconstruction method cannot obtain good reconstruction results when there are a few projection data,so it is not applicable here.Since the current deep learning methods have significant effects in many aspects such as feature extraction,depth estimation,and image completion,as a basis for threedimensional reconstruction research,this paper proposes a simplified two-dimensional magnetosphere structure reconstruction method based on the Generative Adversarial Network(GAN).First,an improved generative adversarial network is proposed to perform image completion on the magnetosphere energy integral image with a finite angle.In the experiment,we define a context loss and a priori loss to constrain the completed image.Considering that the pixels around the missing position of the image have the greatest correlation with the pixels of the missing part,a weight term is set in the context loss to distinguish the importance of each pixel.This can ensure that the completed image is closer to the real situation.Finally,the Algebraic Reconstruction Technique(ART)is used to reconstruct the boundary of the two-dimensional magnetosphere with the integral image,and the Peak Signal-to-Noise Ratio(PSNR)is used to measure the reconstruction effect of the integral image after completion.In the experiment,we choose the magnetosphere MHD model and another simplified version named Jorgensen radiation model to generate the dataset.The experiment shows that when the scanning angle is greater than 50%,the image completion network can effectively and accurately complete the missing image,and the reconstruction effect is better.When the scanning angle is less than 50%,the general characteristics of the magnetosphere can still be learned.Compared with the traditional ART algorithm,it also has a great improvement.This method also has a certain ability to learn from simplified data to complex data.After testing on a more complex MHD model,it is shown that there is still a very significant reconstruction effect when the coverage angle is greater than 50% without training the MHD model.It is confirmed from the side that learning the feature data distribution of the simplified magnetosphere model has a great effect on the reconstruction of the real magnetosphere.It is of great significance for the magnetosphere reconstruction of the true energy integral image with finite angle collected by the SMILE satellite in the future.