# Enhancing SDO/HMI images using deep learning

**Authors:** C. J. Diaz Baso, A. Asensio Ramos  
**arXiv:** [1706.02933](https://arxiv.org/abs/1706.02933)  
**Date:** 2017-06-09  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/1706.02933v2/

## Key points

- The Helioseismic and Magnetic Imager (HMI) has been observing the Sun for 7 years, capturing continuum images and magnetograms with high cadence.
- HMI is limited in analyzing small-scale solar events due to a trade-off between full disk observations and spatial resolution.
- A new method called Enhance uses deep fully convolutional neural networks to deconvolve and super-resolve HMI images and magnetograms.
- Enhance mimics observations from a diffraction-limited telescope twice the diameter of HMI by inputting patches of HMI data into the neural networks.
- Neural networks are trained on synthetic data from simulations of solar active regions emergence to improve image quality.
- Prior information from simulations is incorporated using a neural network framework to address the ill-defined nature of the problem with infinite solutions.
- Testing against degraded Hinode data shows promising consistency, demonstrating improved resolution and quality in solar imaging data.
- The code for Enhance is open source, promoting transparency and reproducibility in research efforts to enhance solar observations using deep learning techniques.

## Summary

The Helioseismic and Magnetic Imager (HMI) has been providing continuous observations of the Sun for the past 7 years. It captures continuum images and magnetograms with a cadence better than one per minute. However, HMI is limited in its ability to analyze the smallest-scale events in the solar atmosphere due to a trade-off between full disk observations and spatial resolution. To address this limitation, a new method called Enhance has been developed. This method utilizes deep fully convolutional neural networks to simultaneously deconvolve and super-resolve images and magnetograms obtained from HMI. By inputting patches of HMI observations into the neural networks, the resulting output mimics observations from a diffraction-limited telescope twice the diameter of HMI. The neural networks are trained on synthetic data generated from simulations of solar active regions emergence. Through this innovative approach, deconvolved and super-resolved HMI images have been successfully obtained. To tackle the inherently ill-defined nature of this problem with infinite solutions, prior information from simulations is incorporated using a neural network framework. Testing against Hinode data that has been degraded to simulate observations from a 28 cm diameter telescope shows promising consistency. The code for Enhance is open source, allowing for transparency and reproducibility in research efforts aimed at enhancing solar imaging data. This work by authors C.J. Diaz Baso and A. Asensio Ramos showcases how deep learning techniques can be leveraged to improve the quality and resolution of solar observations, opening up new possibilities for studying intricate details within the solar atmosphere.