数据与计算发展前沿 ›› 2024, Vol. 6 ›› Issue (4): 128-138.

CSTR: 32002.14.jfdc.CN10-1649/TP.2024.04.011

doi: 10.11871/jfdc.issn.2096-742X.2024.04.011

• 专刊：面向国家科学数据中心的基础软件栈及系统 • 上一篇    下一篇

基于深度学习的低纬F区电离层场向不规则体事件识别

方少峰1,2(),邹自明1,2,*(),胡晓彦1,2   

  1. 1.中国科学院国家空间科学中心，北京 100190
    2.国家空间科学数据中心，北京 100190
  • 收稿日期:2024-02-04 出版日期:2024-08-20 发布日期:2024-08-20
  • 通讯作者: *邹自明（E-mail: mzou@nssc.ac.cn）
  • 作者简介:方少峰，博士，助理研究员，中国科学院国家空间科学中心，中国科学院青促会成员，主要研究方向为数据挖掘与空间天气学的交叉融合研究。
    本文中负责论文撰写与实验论证。
    FANG Shaofeng. Ph.D., research assistant, is a member of the National Space Science Center CAS, and the Youth Innovation Promotion Association CAS. His main research interests include the intersection of data mining and space weather.
    In this paper, he is responsible for the writing of the paper and the experimental demonstration.
    E-mail: fangsf@nssc.ac.cn|邹自明，博士，研究员，博士生导师，中国科学院国家空间科学中心，国家空间科学数据中心主任。主要研究方向为空间科学与数据科学交叉领域研究。
    本文中负责论文架构设计，提出修改意见。
    ZOU Ziming, Ph.D., is a Professor and Ph.D. supervisor at the National Space Science Center, CAS, and director of National Space Science Data Center. His main research interests include the intersection of space science and data science.
    In this paper, he is responsible for designing the structure of the article and making suggestions for revision.
    E-mail: mzou@nssc.ac.cn
  • 基金资助:
    国家重点研发计划“空间科学大数据智能管理与分析挖掘关键技术及应用”(2022YFF0711400);中国科学院网信专项“空间天气典型事件知识挖掘与智能建模研究”项目(CAS-WX2022SF-0103)

Automatic Recognition of Low Latitude Ionospheric Irregularities In F Section Based on Deep Learning

FANG Shaofeng1,2(),ZOU Ziming1,2,*(),HU Xiaoyan1,2   

  1. 1. National Space Science Center, Chinese Academy of Sciences, Beijing 100190, China
    2. National Space Science Data Center, Beijing 100190, China
  • Received:2024-02-04 Online:2024-08-20 Published:2024-08-20

RichHTML

7

PDF

189

摘要：

【目的】 利用子午工程甚高频（very high frequency，VHF）相干散射雷达观测数据，基于深度学习技术对低纬F区电离层3m尺度不规则体进行识别与特征提取。【方法】 本文基于CSPDarknet神经网络技术构建了电离层不规则体事件智能识别模型，并基于预训练好的CSPDarknet为骨干网络以及Yolo目标检测算法，构建了电离层不规则体事件定位模型。【结果】 所构建不规则体识别模型能自动从整天甚高频相关散射雷达观测数据当中挑选出低纬电离层不规则体，并根据不规则体定位模型提取出不规则体的高度和持续时间。实验结果表明，不规则体识别模型的F1得分达到了85.89%，比EfficientNet模型的F1得分高5.68%；不规则体定位模型的平均精度指标mAP可以达到87.22%，比Yolov5s模型的mAP高4.32%。【局限】模型训练过程中主要利用了海南富克站单台站的观测数据，为提升模型的泛化性能需进一步引入更多台站观测数据。【结论】 本文基于深度学习技术首次提出了一套电离层不规则体事件的智能识别与定位方案，极大改善了传统基于阈值法识别不规则体效率低下且依赖专家的问题，提升了电离层不规则体的研究效率。

关键词: 电离层不规则体, VHF雷达, CSPDarknet, YOLO算法

Abstract:

[Objective] Based on the observation data of very high frequency (VHF) coherent scattering radar from the Chinese Meridian Project, the 3m scale ionospheric irregularities in the low-latitude F region are identified and extracted by applying deep learning techniques. [Methods] An intelligent recognition model of ionospheric irregularities is constructed based on the CSPDarknet neural network, and an ionospheric irregularities location model is constructed by using the pre-trained CSPDarknet as the backbone of the Yolo object detection algorithm. [Results] The constructed recognition model can automatically select the low-latitude ionospheric irregularities from the all-day VHF observation data, and extract the height and duration of the irregularities according to the event location model. According to the evaluation of the test set, the F1-score of the recognition model is 85.89%, which improves 5.68% compared to the classical EfficientNet model; the average precision index mAP of the location model can achieve 87.22%, which is 4.32% better than the Yolov5s model. [Limitations] In the process of training the model, we mainly use the observation data from the Hainan Fuke station. To improve the generalization performance of the model, observation data from other stations should be introduced as much as possible. [Conclusions] In this paper, we first propose an intelligent identification and location scheme for ionospheric irregularities by applying advanced deep learning techniques. Compared to traditional threshold methods, the proposed method greatly improves the recognition efficiency, mitigates the dependency on experts, and helps elevate the research efficiency of ionospheric irregularities.

Key words: ionospheric irregularities, VHF radar, CSPDarknet, Yolo algorithm