INCREASING THE OBSERVATION CAPABILITIES OF SMALL SOLAR TELESCOPES USING NEURAL NETWORKS

Small solar telescopes play an important role in monitoring dynamic heliophysical phenomena on the Sun, such as solar flares, which affect satellites and ground-based electronic systems. Unfortunately, the quality of the data they obtain is significantly affected by atmospheric turbulence, so they require further processing. Currently, this problem is mainly solved by using a state-of-the-art deterministic algorithm, multi-frame blind deconvolution (MFBD). However, this approach is sensitive to manually adjusted system parameters and requires high computational power and long processing time. In this paper, we explore the possibility of using fully convolutional networks (FCNs) as an alternative to MFBD. In our research, we focus on the data processing time, the quality of the results obtained, as well as the hardware resources required for the practical application of the FCNs by small observatories. We show that the proposed networks can achieve results comparable to MFBD, but in almost real time and using more energy-efficient equipment.