基于离群检测的日冕高温区分量提取与分析

YNAO OpenIR > 天文技术实验室

	基于离群检测的日冕高温区分量提取与分析
其他题名	Extraction and analysis of coronal high-temperature components based on outlier detection
	孙莉焰
学位类型	硕士
导师	刘辉
	2023-07-01
学位授予单位	中国科学院大学
学位授予地点	北京
学位专业	天文技术与方法
关键词	日冕活动区日冕高温区域离群检测残差网络微分发射度测量
摘要	日冕高温区域的定位对于日冕加热和其它太阳物理活动的研究具有重要的意义。在AIA 的窄带极紫外观测数据基础上，科学家们提出了采用AIA 171 Å，193 Å，211 Å 三个低温波段来对日冕高温区域进行线性诊断的方法，然而这些方法不能很好去除以Fe xiv 为主的低温发射线，得到的结果与最新的理论预测存在不同。为此，本文针对AIA 多通道复杂的非线性关系，利用深度神经网络建立AIA 94 Å 与其它多个波段观测之间的非线性回归模型，并通过对回归离群区域的提取和分析，获取日冕图像中真正代表高温辐射的区域。本文主要通过以下工作对该方法进行了相关研究：（1）对全文所采用的数据进行预处理并设计后续使用的数据集。在获取到的AIA 一级极紫外观测数据进行一系列预处理的基础上计算各波段的发射测量并重新反演生成AIA 数据，通过筛选出的日冕活动区信息对新生成的数据进一步处理为标准的数据格式用于后续神经网络的训练。（2）为离群检测搭建了日冕多波段的非线性低温回归网络模型。AIA 多通道间具有复杂的非线性关系，基于此设计了基于深度神经网络的非线性低温回归网络。以残差网络为基础，以171Å，193Å，211Å 为输入集，94Å 为标签，搭建了三对一的非线性低温回归网络。（3）设计了基于离群检测的日冕高温区域提取方法。高温区域相比整个相对稳定的日冕层可以视为一个异常对象也即离群点，将这个离群对象检测出来就是高温区域的离群检测。为了避免神经网络对离群数据的拟合，本文提出了平均绝对误差损失函数和基于离群阈值的采样筛选，最后通过基于距离的比较获取离群高温分量。（4）对非线性低温回归网络模型进行了实验与结果分析。采用测试集对非线性低温回归的结果进行检验，与原始94 Å 观测数据以及线性拟合方法对比发现非线性低温回归网络拟合得到的低温分量更多，强度更高。（5）对日冕高温区域离群检测的结果进行分析。离群分检测获取到的高温区域与DEM 反演生成的高温区域对比发现有较好的一致性，表明该方法能够实现日冕高温区域的检测，进一步的残差分析表明离群检测获取到的高温区域中包含有更少的Fe xiv 低温分量成分。因此，本论文得到的日冕高温区域检测结果具有较为明确的物理意义，该基于离群检测的日冕高温区分量提取与分析方法也为太阳观测数据处理提供了一种新的思路。
其他摘要	The localization of coronal high-temperature regions is important for the study of coronal heating and other solar physical activities. Based on the narrow-band extreme ultraviolet observations of the AIA, scientists have proposed the use of three lowtemperature bands of the AIA 171 Å, 193 Å, and 211 Å for linear diagnosis of the coronal high-temperature region, however, these methods cannot well remove the lowtemperature emission lines dominated by Fe xiv, and the obtained results differ fromthe latest theoretical predictions . For this reason, in this paper, we use deep neural networks to establish a nonlinear regression model between AIA 94 Å and other multi-band observations for the complex nonlinear relationship of AIA multi-channels, and obtain the regions in the coronal images that really represent the high-temperature emission by extracting and analyzing the regression outlier regions. In this paper, the method is related by the following works:(1) The data used in the full paper are pre-processed and the dataset for subsequent use is designed. Based on a series of preprocessing of the acquired AIA level 1 extreme ultraviolet observations, emission measurements in each band are calculated and re-inverted to generate the AIA data, and the newly generated data are further processed into a standard data format by filtering the coronal activity zone information for subsequent neural network training.(2) A nonlinear cryo-regression network model for coronal multi-band is built for outlier detection.The AIA multi-channel has complex nonlinear relationship among the channels, based on which a nonlinear cryo-regression network based on deep neural network is designed. Based on the residual network, a three-to-one nonlinear cryogenic regression network is built with 171Å, 193Å, 211Å as input sets and 94Å as labels.(3) An outlier detection-based method for coronal high-temperature region extraction is designed. The high-temperature region can be regarded as an anomalous object compared with the whole relatively stable coronal layer. The detection of this outlier object is the outlier detection of the high temperature region. In order to avoid the fitting of neural networks to outlier data, this paper proposes a mean absolute error loss function and sampling screening based on outlier thresholds, and finally obtains the outlier high temperature component by distance-based comparison.(4) Experiments and results were analyzed for nonlinear low-temperature regression network models. A test set was used to test the nonlinear The results of the nonlinear cryogenic regression were examined and compared with the original 94 Å observation data and the linear fitting method. The nonlinear cryogenic regression network fits more cryogenic components with higher intensity.(5) The results of coronal high-temperature region outlier detection are analyzed. It is found that the high-temperature region obtained by outlier detection is in good agreement with the high-temperature region generated by DEM inversion, indicating that the method can achieve the detection of the coronal high-temperature region, and further residual analysis shows that the high-temperature region obtained by outlier detection contains less Fe xiv low-temperature fraction components.Therefore, the coronal high-temperature region detection results obtained in this paper have a clear physical meaning, and the outlier-based coronal high-temperature region component extraction and analysis method also provides a new idea for solar observation data processing.
学科领域	天文学
学科门类	理学 ; 理学::天文学
页数	0
语种	中文
文献类型	学位论文
条目标识符	http://ir.ynao.ac.cn/handle/114a53/26388
专题	天文技术实验室
作者单位	中国科学院云南天文台
第一作者单位	中国科学院云南天文台
推荐引用方式 GB/T 7714	孙莉焰. 基于离群检测的日冕高温区分量提取与分析[D]. 北京. 中国科学院大学,2023.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
基于离群检测的日冕高温区分量提取与分析.（18985KB）	学位论文		开放获取	CC BY-NC-SA	浏览请求全文