A group of Russian and international astronomers has developed an innovative machine learning system capable of automatically processing and integrating the results of long-term observations of the Sun. The new method will maximize the effective use of archival data, according to the Skoltech press service.
According to Tatyana Podladchikova, director of the Skoltech Center for System Design, the work of astronomers is not limited to improving old images. Scientists are creating a universal language for studying solar evolution.
High-performance computing systems have helped train AI models that identify hidden connections in data and detect patterns over several solar cycles.
Over the past decades, specialists around the world have accumulated a huge amount of information about solar activity. However, differences in resolution and observation methods make it difficult to combine this data. To solve this problem, scientists from Skoltech, together with colleagues from the University of Graz, Austria, and the High Altitude Observatory in the USA, have developed an algorithm based on GAN neural networks.
It allows one part of the neural network to degrade the quality of images, while another restores them to a reference level. This method makes it possible to combine images taken by different telescopes, even without reference images.
Scientists have already been able to combine data from 24 years of observations, increase the resolution of solar images, and estimate the magnetic fields on its far side. The new algorithm promises many discoveries about the properties and behavior of our luminaries.
Read more on the topic:
Tornadoes may occur at the poles of the Sun, scientists have found out
Three powerful flares in two hours: another burst occurred on the Sun
