Automatic Classification of Magnetic Chirality of Solar Filaments from H-Alpha Observations

2025-09-21

AI Summary

This study addresses the automated classification of solar filament chirality (sinistral/dextral), a task previously hindered by small-scale, non-reproducible datasets. To overcome these limitations, we introduce MAGFiLO—the first publicly available, large-scale, and meticulously annotated Hα observational dataset for filament chirality analysis—and establish a reproducible benchmark framework. Methodologically, we conduct a systematic evaluation of state-of-the-art convolutional architectures—including ResNet, WideResNet, ResNeXt, and ConvNeXt—incorporating data augmentation and class-balanced loss weighting to mitigate label imbalance. Experimental results demonstrate that ConvNeXt-Base achieves 69% accuracy on sinistral and 73% on dextral filaments on MAGFiLO, substantially outperforming prior approaches. This work pioneers a large-scale, data-driven paradigm for solar filament chirality classification, advancing the field toward standardization, cross-study comparability, and improved generalizability.

Technology Category

• Machine Learning: Multi-class/Multi-label Learning & Extreme Classification
• Computer Vision: Segmentation
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

In this study, we classify the magnetic chirality of solar filaments from H-Alpha observations using state-of-the-art image classification models. We establish the first reproducible baseline for solar filament chirality classification on the MAGFiLO dataset. The MAGFiLO dataset contains over 10,000 manually-annotated filaments from GONG H-Alpha observations, making it the largest dataset for filament detection and classification to date. Prior studies relied on much smaller datasets, which limited their generalizability and comparability. We fine-tuned several pre-trained, image classification architectures, including ResNet, WideResNet, ResNeXt, and ConvNeXt, and also applied data augmentation and per-class loss weights to optimize the models. Our best model, ConvNeXtBase, achieves a per-class accuracy of 0.69 for left chirality filaments and $0.73$ for right chirality filaments.

Problem

• Classifying magnetic chirality of solar filaments from H-Alpha observations
• Establishing first reproducible baseline for filament chirality classification
• Addressing limited generalizability from small datasets in prior studies

Innovation

• Fine-tuned pre-trained image classification models
• Applied data augmentation and weighted loss
• Used ConvNeXtBase achieving high chirality accuracy