基于全天空图像的日侧极光分类方法研究

项目名称： 基于全天空图像的日侧极光分类方法研究

项目编号： No.60872154

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 梁继民

作者单位： 西安电子科技大学

项目金额： 32万元

中文摘要： 主要针对中国北极黄河考察站的全天空光学成像系统获取的全天空极光观测数据，从极光能量分布、形态特征、动态特征等方面，研究全天空极光图像和图像序列的特征描述和自动分类方法，探索了利用图像处理和模式识别方法进行极光自动分类的可行性。首先利用极光纵观分布数据，结合物理机制将极光卵划分为四个区域，并对每个区域的极光形态特征进行了定性分析。在极光图像静态分类研究中，采用局部二元模式方法对极光形态特征进行描述和有监督分类，实验结果表明自动分类与人工标记结果基本吻合；利用该特征描述，进一步对极光图像进行了聚类分析，聚类分析结果很好地验证了我们对极光类型划分的有效性。针对极光的纹理特性，提出了两种纹理描述算法，并应用于极光图像的有监督分类。在极光图像序列动态分类研究中，首先采用流体光流场描述极光序列，进行有监督的分类和序列分割。其次，针对基于隐马尔可夫模型的图像序列描述方法进行了研究，并应用于极光图像序列建模，对极光极向运动事件进行自动检索和有监督的分类。发表国际刊物SCI论文7篇，国内刊物论文4篇，另有4篇国际刊物论文在审。研究成果为进一步研究极光发生机制及其与磁层边界层动力学过程的对应关系奠定基础。

中文关键词： 极光分类;局部二元模式;纹理描述;聚类分析;隐马尔可夫模型

英文摘要： This project investigated the morphology and dynamic process classification of the all sky imager (ASI) observations at the Yellow River Station of China by exploring the energy synoptic distributions, the morphology and dynamic characteristics of aurora, with the attempt of applying image processing and pattern recognition techniques to automatic auroral image analysis. We used the synoptic distribution of dayside aurora, combining with physical mechanism analysis, to partition the aurora oval into four active regions and summarized their corresponding morphology characteristics qualitatively. For static aurora image analysis, we first used the local binary patterns (LBP) for auroral image representation and supervised classification. The automatic classification results fit the manual labels well. Using the LBP based representation, we further analyzed the aurora morphology characteristics via unsupervised data clustering. The experimental results confirmed our aurora classification scheme. In consideration of the texture characteristic of auroral image, we proposed two texture descriptors for aurora feature extraction, and their performance were testified by supervised classification experiments. For dynamic auroral image sequence analysis, we first adopted the multi-scale fluid flow method to describe the auroral image sequence and applied it in supervised classification and sequence segmentation. Then, we studied the image sequence representation methods based on hidden Markov models (HMM) and applied them to polar moving auroral forms detection and supervised classification. In the project duration, we have published 7 international journal papers and 4 domestic journal papers. There are 4 international journal papers are under review with 3 of them had been revised. The achievements of this study established preliminary basis for the research of aurora occurrence mechanism and its relationship with the dynamic process of magnetosphere boundary regions.

英文关键词： Aurora classification; Local binary patterns; Texture description; Clustering analysis; Hidden Markov model