Neural Network-based Sunspot Detection with Dynamic Thresholding Method

Accurate identification of sunspots is essential for understanding solar activity and predicting space weather. Existing detection methods, ranging from traditional image processing to advanced deep learning models, face limitations particularly the inability of fixed confidence thresholds to reliably detect small or low-contrast sunspots, leading to higher false positive and false negative rates. To address this issue, this research develops an enhanced detection pipeline incorporating high-resolution image processing, gamma correction, disc segmentation, and Sauvola thresholding for dynamic thresholding. The optimized pipeline is used to train a YOLOv8n model, resulting in improved detection accuracy for challenging sunspot cases. The study also evaluates scalability, dataset variability, and performance metrics, while proposing future improvements through adaptive thresholding, hybrid deep learning strategies, and data augmentation. This approach offers a more robust framework for automated sunspot detection, paving the way for more precise and reliable solar monitoring systems.