A deep learning-based solar extreme ultraviolet image quality repairing method and system

Abstract

This invention discloses a method and system for quality restoration of solar extreme ultraviolet (EUV) images based on deep learning. First, considering the basic periodic characteristics of solar motion, Earth motion, and instrument operation, EUV data from other sensors with similar center wavelengths and high quality are selected as reference images. Then, time matching, radiometric normalization, and geometric registration strategies are applied to the image to be restored and the reference image to eliminate systematic differences in imaging time, radiometric distribution, and field of view structure among different instruments, constructing a large-scale, long-term, spatiotemporally aligned paired dataset. Based on this dataset, a conditional generative adversarial network (GAN) is constructed and trained to learn complex nonlinear radiometric mapping relationships. By implementing spatiotemporal joint loss constraints across multiple dimensions such as pixels, time, and features, adaptive restoration of complex degradation and quality deterioration is achieved. This invention can significantly improve the quality of solar EUV images and has advantages such as high accuracy, fast computational efficiency, ease of implementation, and strong generalization.

Images