Computer Vision Framework for Near Real-Time Solar Feature Detection in SDO/AIA Images

Abstract

This article presents an automated image-processing framework for near real-time detection of solar features from the Atmospheric Imaging Assembly (AIA) onboard the Solar Dynamics Observatory (SDO). It addresses the critical challenge of processing the massive data flow generated by SDO—approximately one high-resolution image every ten seconds—which overwhelms traditional manual analysis methods. The proposed framework implements a three-stage processing pipeline: preprocessing to standardize and prepare images, feature classification using both histogram-based multilevel thresholding and K-Nearest Neighbor approaches, and spatial validity assessment to refine segmentation results. This article effectively identifies three primary solar features—Active Regions, Coronal Holes, and Quiet Sun regions—which are essential for space weather forecasting. Validation against existing methods demonstrates high agreement rates with expert classifications while maintaining computational efficiency suitable for real-time operations. Time series analysis confirms that detected features exhibit expected correlations with solar cycle indicators, with Active Region areas showing strong positive correlation with sunspot numbers and Coronal Hole areas displaying moderate negative correlation. The framework's cross-mission compatibility enables creation of standardized feature catalogs spanning multiple solar observation platforms, providing a foundation for both immediate space weather applications and long-term solar physics research.