跨尺度极光事件演化过程分析与建模

Aurora is a most common and important natural phenomenon on the polar region. The research on aurora can help people to understand the influence way and degree of solar activity on the earth. At this time, it has important significance in guiding to space physics and polar expedition. There are many difficulties in perceiving auroras, such as complex morphology, dynamic characteristics, no obvious shape, no color features and quick dynamic changes, and so on. To this end, the project intends to study visual cognitive characteristics of aurora. The project will combine the knowledge on ionosphere during aurora evolution progress with novel methods of visual cognition computing and machine learning, to develop: 1) Synergy perception and coupling cerrelation on the cross-scale auora data obtained from space-based,ground-based,RADAR and geomagnetometer sensor network 2) the aurora events are labled by Eyemovement of experts from Eyelink and effective visual fearture are extracted by deep learning method to help other researcher to label unlabed aurora data; 3) extract the effective feautures of aurora events and establish the model of aurora evolution pragress; 4) study the evolution law of aurora morphology combining the characteristics of polar region environment, forecast the occurrence of auroras. The project provides the new research ideas for aurora phenomenon occurance and forecast, which also provide a solid theoretical foundation and technical support to the polar scientific research tourism development.

极光是极地上空最常见也是最重要的一种自然现象，对它的研究可以帮助人们深入了解太阳活动对地球的影响方式和程度，同时在空间物理、极地科考研究等方面也有着重要的指导意义。由于极光形态和动态极为复杂，极光图像没有明显的形状、颜色等特征且动态变化快速等多方面原因，对极光事件演化过程的分析、建模以及预测也变得相当的困难。本项目根据极光事件演化过程的特点，结合电离层相关知识，以视觉认知计算、机器学习等领域的新方法为基础：1）基于天基-地基以及雷达、地磁仪等传感器网络实现跨尺度极光数据耦合关联； 2）对极光事件形态进行基于眼动仪辅助的视觉语义信息标定学习；3）极光事件特征提取与跨尺度极光事件聚合分析；4）研究极光形态演化规律并结合极区环境特点，对不同形态的极光发生进行预测。该项目研究成果为极光现象发生与预测提供线索，同时为发展旅游事业及极地科考研究等提供研究思路和技术保障。

太阳活动与日地空间环境的变化息息相关，太阳风能够直接引起日地空间中的电磁扰动，进而可能引发空间天气灾害，影响人类的日常活动。极光作为太阳活动对地球磁层扰动的直观表现形式，是国内外日地空间研究的重要方向。本项目利用认知心理学、计算机视觉和深度学习等领域知识完成了对极光事件的发生、演化过程的研究。具体来说，本项目首先构建跨尺度多传感器耦合的极光事件数据库，并利用眼动仪辅助极光事件的语义信息标定。提出了多种基于深度卷积网络、生成判别网络和主成分分析网络等深度学习框架的极光影像特征提取、极光卵分割、极光亮度建模和极光卵边界建模等算法，并利用形态学描述、动态纹理模型和粘滞流体运动模型对极光的运动过程进行建模。从极光事件检测、分析、建模和预测等四个方面对极光事件分析研究。本项目的成果将为后期日地空间环境监测和极地科学考察研究等提供研究思路和技术保障。