基于对抗神经网络有限角度的磁层边界CT重构技术

对地球磁层进行软X射线成像探测是近年来磁层研究的前沿方向.由二维X射线成像图重构三维磁层边界的方法研究是与成像探测相关的重要研究课题.传统的计算机断层成像技术（CT）重构方法在图像数据很少时无法得到较好的重构结果，甚至无法重构三维结构.考虑到地球磁层空间分布范围方面的约束，当前卫星任务的轨道设计很难满足扫描角度的全方位覆盖，只能从有限角度观测磁层，这给磁层的CT重构带来困难.作为三维重构研究的基础，本文考查简化的二维磁层结构重构方法，采用基于对抗神经网络的CT技术对简化的地球磁层边界结构进行重构.首先使用改进的生成式对抗网络（GAN）对有限角度的卫星扫描图像进行图像补全，进而使用代数迭代重建方法（ART）重构磁层.实验表明，当扫描角度大于90°时，生成式对抗网络能有效准确地补全缺失图像，重构效果较好. 

Finite-angle Magnetosphere Boundary CT Reconstruction Technique Based on Generative Adversarial Networks

Soft X-ray imaging detection of the Earth's magnetosphere is the frontier direction of recent magnetosphere research. Research on the method of reconstructing 3D magnetic layer boundary from 2D X-ray image is an important research topic related to imaging detection. Traditional Computer Tomography (CT) reconstruction methods cannot obtain good reconstruction results when image data is small, or even fail to reconstruct 3D structures. Considering the constraints on the spatial distribution of the Earth's magnetosphere, the orbital design of current satellite missions is difficult to meet the full angle coverage of the scanning angle, and the magnetosphere can only be observed from a finite angle, which brings problem to the CT reconstruction of the magnetosphere. As the basis of the 3D reconstruction research, this paper examines a simplified 2D magnetospheric reconstruction method, and uses CT technology based on adversarial neural network to reconstruct the simplified magnetosphere layer boundary structure. First, we use an improved Generative Adversarial Networks (GAN) to complete the finite-angle satellite scanned image, and then the magnetosphere layer is reconstructed using an Algebraic Reconstruction Technique (ART) reconstruction method. Experiments show that when the scanning angle is greater than 90°, the Generative Adversarial Networks can effectively and accurately complete the missing image, and the reconstruction effect is better.