Speaker
Description
The McIntosh sunspot classification system categorises sunspots according to their morphology and size. Because this process is typically performed manually, it is susceptible to labeling errors. In contrast, deep learning methods learn features directly from the data and classify samples based on these learned representations. In this paper, an unsupervised approach to McIntosh sunspot classification is presented, and how the models group sunspot data into clusters is examined. These clusters are compared with reference ground-truth labels to determine how strongly they correlate with the ground truth. To assess performance, the supervised models are trained using both the ground-truth labels and the unsupervised labels, and their accuracies are compared. The results indicate that the unsupervised approach outperformed the conventional supervised approach trained on the ground-truth labels. Finally, Local Interpretable Model-agnostic Explanations (LIME) is used to interpret the models' decisions on the unsupervised dataset. The findings highlight the potential for developing an AI-based McIntosh classification framework.
| Apply for student award at which level: | MSc |
|---|---|
| Consent on use of personal information: Abstract Submission | Yes, I ACCEPT |