Fully convolutional neural networks for processing observational data from small remote solar telescopes

Area of Science:

• Heliophysics
• Solar Physics
• Image Processing

Background:

• Solar phenomena impact satellites and electronics, necessitating high-resolution solar imaging.
• Small solar telescopes have resolution limitations due to aperture size and atmospheric turbulence.
• Current image processing methods like multi-frame blind deconvolution (MFBD) can be computationally intensive.

Purpose of the Study:

• To explore the use of fully convolutional networks (FCNs) for enhancing solar chromosphere images from small telescopes.
• To compare the performance of FCNs against MFBD in terms of image quality and processing time.
• To investigate the influence of data volume and FCN complexity on results.

Main Methods:

• Utilized chromosphere data from a 50mm Hα Telescope.
• Applied fully convolutional networks (FCNs) for image enhancement.
• Compared FCN results with multi-frame blind deconvolution (MFBD) processing.

Main Results:

• FCNs achieved comparable image quality to MFBD.
• FCNs demonstrated significantly faster processing times (orders of magnitude).
• FCNs proved to be more energy-efficient than MFBD.

Conclusions:

• Fully convolutional networks (FCNs) offer a highly efficient and effective method for improving solar image resolution from small telescopes.
• FCNs present a viable and attractive alternative to traditional image deconvolution techniques for heliophysics applications.
• The study highlights the potential of deep learning in advancing solar observation capabilities.