周朋辉
陕西师范大学物理学与信息技术学院, 西安 710019
基金项目: 国家自然科学基金项目（41504122）、中央高校基本科研业务费专项基金项目（GK201903021）和陕西省高校科协青年人才托举计划项目（20160211）共同资助
详细信息
作者简介: 杨秋菊, 女, 1986年生, 副教授, 主要从事极光图像自动分析研究工作.E-mail:yangqiuju@snnu.edu.cn
中图分类号: P352 收稿日期:2018-07-27
修回日期:2019-03-15
上线日期:2020-01-05
Automatic segmentation of auroral images using machine learning techniques
YANG QiuJu,ZHOU PengHui
School of Physics and Information Technology, Shaanxi Normal University, Xi'an 710019, China
MSC: P352
--> Received Date: 27 July 2018
Revised Date: 15 March 2019
Available Online: 05 January 2020
摘要
摘要:极光是太阳风能量注入到极区的指示器，从观测视野中准确分割出极光区域对研究极光演变如亚暴过程有非常重要的意义.本文基于全卷积神经网络提出了一种弱监督极光图像自动分割策略，数据标记时仅需指定极光区域的一个像素点即可，极大解决了机器学习人工标注数据的压力.首先利用简单单弧状极光图像训练一个初始分割模型Model 1，然后基于该模型，结合热点状和复杂多弧状极光图像获得一个增强的分割模型Model 2，最后对分割结果做进一步优化.本文对2003-2007年北极黄河站越冬观测的2715幅极光图像进行了分割，并和最新论文结果及人工标签进行了定量和定性比较，其中分割结果与人工标签的"交并比"高达60%，证明了本文方法的有效性.
关键词: 全天空极光图像/
分割/
全卷积神经网络
Abstract:Aurora is an indicator of the injection of solar wind energy into the Polar regions. It is of great importance to accurately segment the aurora from the field-of-view for aurora study, such as when studying the evolution of substorms. In this paper, based on deep Fully Convolutional Networks, a weakly supervised auroral image segmentation strategy is proposed. In the process of data labelling, only one pixel in the region of auroras needs to be specified, which greatly solves the pressure of manual labelling for segmentation models. Specifically, we first train an initial segmentation model, called Model 1, using the simple single-arc auroral images (only one distinct auroral arc) and their segmentation labels obtained by the seeded region growing method. Then, an enhanced segmentation model, called Model 2, is learned with (1) the hot-spot auroral images and their segmentation labels based on the Model 1, and (2)as well as the complicated multi-arc images (two or more arcs coexist)and their segmentation labels obtained by the seeded region growing method. Finally, the segmentation results are further optimized by the fully connected conditional random field model. Altogether 2715 auroral images captured at the Arctic Yellow River Station during years 2003-2007 are segmented in this paper. The experimental results are quantitatively and qualitatively compared with the state-of-the-art results and manual labels, of which the intersection-over-union value between the ground truth and the segmented results is as much as 60%, which proves the effectiveness of the proposed method.
Key words:All-sky auroral images/
Segmentation/
Fully convolutional network
PDF全文下载地址:
http://www.geophy.cn/data/article/export-pdf?id=dqwlxb_15288
