Deep Learning for Active Region Classification: A Systematic Study from Convolutional Neural Networks to Vision Transformers

TL;DR

It is observed that combining magnetogram and continuum image types enhances model performance by leveraging complementary features from diverse inputs, and when considering only magnetograms, data-efficient image transformers achieve the best performance, suggesting that these models can better capture the spatial complexity of magnetograms.

Abstract

A solar active region can significantly disrupt the Sun Earth space environment, often leading to severe space weather events such as solar flares and coronal mass ejections. As a consequence, the automatic classification of active region groups is the crucial starting point for accurately and promptly predicting solar activity. This study presents our results concerned with the application of deep learning techniques to the classification of active region cutouts based on the Mount Wilson classification scheme. Specifically, we have explored the latest advancements in image classification architectures, from Convolutional Neural Networks to Vision Transformers, and reported on their performances for the active region classification task, showing that the crucial point for their effectiveness consists in a robust training process based on the latest advances in the field.

Figures (5)

• Figure 1: Histogram displaying the distribution of aspect ratios (width/height) of the images in the SOLAR-STORM1 dataset.
• Figure 2: Temporal distribution of Active Regions (ARs) and the K-Fold split visualization for the dataset. The upper panel shows the number of ARs per day present in the dataset, while the lower one illustrates how the data is divided into training (blue), validation (red), and test (yellow) sets across the five different folds.
• Figure 3: Effects of the normalization process on the images. The AR considered as an example is a Beta-X one at time 2014-03-17 16:00. The first row shows the original continuum and magnetogram images, while the second one shows, on the left, the normalized continuum image with Min-Max normalization (\ref{['eq: MinMax']}) and, on the right, the magnetogram image normalized using the HardTanh function (\ref{['eq: HardTanh']}) with constant $d=800$. The last line shows the cropping of the AR done to filter out noise and allow the model to concentrate on the significant features relevant to the classification task.
• Figure 4: Illustration of data augmentation techniques. The original image (top left) undergoes a series of transformations to enhance the model’s ability to generalize. The first row shows the effect of horizontal and vertical flips. The second row demonstrates perspective transformations, altering the viewing angle. The third row applies affine transformations, including scaling, rotation, and shear, to further diversify the dataset.
• Figure 5: A summary of the workflow followed in this study.