极地研究 ›› 2015, Vol. 27 ›› Issue (3): 255-263.DOI: 10.13679/j.jdyj.2015.3.255

• 研究论文 • 上一篇    下一篇

基于全天空极光图像方向能量表征方法的极光事件分类

张军1 胡泽骏2 王倩2,3 梁继民1   

  1. 1 西安电子科技大学,陕西 西安 710071;
    2 国家海洋局极地科学重点实验室,中国极地研究中心,上海 200136; 3 西安邮电大学,陕西 西安 710121
  • 收稿日期:2014-04-10 修回日期:2014-05-05 出版日期:2015-09-30 发布日期:2015-09-30
  • 通讯作者: 张军
  • 基金资助:

    国家自然科学基金;国家自然科学基金,南北极环境综合考察与评估专项资助项目;海洋公益性行业科研专项

AURORAL EVENT CLASSIFICATION USING ORIENTED ENERGY-BASED REPRESENTATION

Zhang Jun1, Hu Zejun2, Wang Qian2,3, Liang Jimin1   

  1. 1 Xidian University, Xi'an 710071, China;
    2. SOA Key Laboratory for Polar Science, Polar Research Institute of China, Shanghai 200136, China;
    3. Xi'an University of Posts & Telecommunications, Xi'an 710121, China
  • Received:2014-04-10 Revised:2014-05-05 Online:2015-09-30 Published:2015-09-30

摘要: 针对基于全天空极光图像的极光事件自动分类问题,提出一种基于方向能量二元编码重组表征的自动分类方法。首先,通过对多个方向上能量分解来描述极光事件中的局部纹理和各个方向上的运动信息,并且结合分块策略获得极光事件的全局形态信息;然后,借鉴一种二元编码重组的方式对多个方向能量进行融合,从而使得极光事件的表征具有同时表征局部纹理、全局形态和运动信息的能力。该表征方法完全不依赖于极光事件的长度,可用于表征不同持续时间的极光事件,并且不需要复杂的训练过程。利用最近邻和支撑向量机分类器分别对从中国北极黄河站拍摄到的极光图像中挑选的特定极光事件进行自动分类,结果表明,与其他两种典型的动态纹理描述方法相比,本文所提出的表征方法结合最近邻分类器得到了最好的分类效果,能有效用于极光事件的分析,为海量数据中的极光事件自动分类提供了一种新方法。

关键词: 极光事件, 有监督分类, 方向能量, 动态纹理, 二元编码

Abstract: The auroral event is a physical phenomenon with rich information of texture, morphology and motion. Therefore, there is an urgent need to have a representation which captures these information simultaneously. Addressing to this problem, a three dimensional dynamic texture representation method based on oriented energy with binary coding is proposed for auroral event representation and automatic classification. At first, the local texture and oriented motion are described by the decomposition of energy into different orientations. Secondly, combining with the block partition strategy, the global morphology information is obtained as well. In order to obtain the statistical histogram, the technique of binary coding is applied for the fusion of energies with different orientations. Finally, the classifiers of nearest neighbor and support vector machines are used for classifying the auroral events from Chinese Arctic Yellow River Station. The classification results demonstrate that the proposed method achieves superior classification performance using nearest neighbor classifier compared with other two representative dynamic texture representation methods. The proposed method is specifically designed for auroral event representation, which is independent to the duration and captures the local texture, global morphology and motion simultaneously. It provides a feasible method for automatic classification of massive auroral events.

Key words: auroral event, supervised classification, oriented energies, dynamic texture, binary coding