Skip to main content
SearchLoginLogin or Signup

Development of deep learning models to estimate three-dimensional parameters of coronal mass ejections

Presentation #110.16 in the session Data Analysis Techniques Posters.

Published onSep 18, 2023
Development of deep learning models to estimate three-dimensional parameters of coronal mass ejections

In this study, we present a new method to determine 3-D parameters of coronal mass ejections (CMEs) using convolutional neural network (CNN) methods, a commonly used deep learning algorithm in image recognition. As the first step, we apply CNN models to synthetic CME images. A synthetic CME image is generated by a full ice-cream cone model for given three-dimensional parameters (radial height, angular width, latitude, longitude). The CNN models are developed based on VGG network and Residual network. We generate 391,000 synthetic CME images with different three-dimensional parameters: 312,800 for training, 39,100 for validation, and 39,100 for test. For the test dataset, we estimate root mean square errors between three-dimensional parameters of the test dataset and those from CNN models. The best results are 0.4Rs for radial height, 2.3° for angular width, 0.8° for latitude, and 1.5° for longitude. We apply these models to several halo CME events observed by SOHO/LASCO C3. And we compare these results with three-dimensional parameters obtained from a full ice-cream cone model.

Comments
0
comment
No comments here