Scholarlink

Context. <p>Currently coronal mass ejections (CME) detection and measurements heavily depend on time-consuming human observations. Existing automated catalogs are based on manually crafted mathematical algorithms and struggle to provide the performance of a human observer (Lamy et&#xa0;al. <CitationRef CitationID="CR24">2019</CitationRef>). Recent advances in machine learning (ML) have opened up new avenues for the early detection and analysis of coronal mass ejections (CMEs), which pose significant risks to space weather and technological systems on Earth.</p> Aim. <p>This study presents a novel methodology utilizing deep learning algorithms to analyze images from SOHO/LASCO coronagraphs, with the objective of enhancing the identification of CMEs by classifying the coronagraph frames into two categories: CME and non-CME.</p> Method. <p>We created a dataset based on level 0.5 SOHO LASCO (Solar and Heliospheric Observatory Large Angle Spectrometer COronagraph) C2 observations correlated with CDAW (Coordinated Data Analysis Workshop) CME measurements. Then we developed two convolutional neural network (CNN) models for binary classification: the standard AlexNet and a modified AlexNet for action recognition (Spatiotemporal AlexNet). We tested various data augmentation techniques and evaluated various approaches to clip creation from static images.</p> Result. <p>Initial results are very promising and show improvements in detection accuracy with the spatiotemporal model and suggest areas for further research in CME detection using machine learning. This research provides methodological insights that could aid the development of more effective CME detection systems and, ultimately, improved space weather preparedness. Further work will focus on replacing the automatically generated CDAW-based dataset with a manually curated version, as improved labeling quality is likely to enhance detection accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Spatiotemporal Machine-Learning Detection of Coronal Mass Ejections in SOHO/LASCO C2 Images

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月22日

查看全文
  • Tomasz J. Łaguz,
  • Grzegorz Michałek,
  • Adam A. Zychowicz,
  • Seiji Yashiro

摘要

Context.

Currently coronal mass ejections (CME) detection and measurements heavily depend on time-consuming human observations. Existing automated catalogs are based on manually crafted mathematical algorithms and struggle to provide the performance of a human observer (Lamy et al. 2019). Recent advances in machine learning (ML) have opened up new avenues for the early detection and analysis of coronal mass ejections (CMEs), which pose significant risks to space weather and technological systems on Earth.

Aim.

This study presents a novel methodology utilizing deep learning algorithms to analyze images from SOHO/LASCO coronagraphs, with the objective of enhancing the identification of CMEs by classifying the coronagraph frames into two categories: CME and non-CME.

Method.

We created a dataset based on level 0.5 SOHO LASCO (Solar and Heliospheric Observatory Large Angle Spectrometer COronagraph) C2 observations correlated with CDAW (Coordinated Data Analysis Workshop) CME measurements. Then we developed two convolutional neural network (CNN) models for binary classification: the standard AlexNet and a modified AlexNet for action recognition (Spatiotemporal AlexNet). We tested various data augmentation techniques and evaluated various approaches to clip creation from static images.

Result.

Initial results are very promising and show improvements in detection accuracy with the spatiotemporal model and suggest areas for further research in CME detection using machine learning. This research provides methodological insights that could aid the development of more effective CME detection systems and, ultimately, improved space weather preparedness. Further work will focus on replacing the automatically generated CDAW-based dataset with a manually curated version, as improved labeling quality is likely to enhance detection accuracy.