Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Improving the Spatial Resolution of GONG Solar Images to GST Quality Using Deep Learning

Chenyang Li, Qin Li, Haimin Wang, Bo Shen

TL;DR

This work addresses the challenge of extracting high-spatial-resolution information from full-disk solar images by cross-instrument super-resolution. It employs Real-ESRGAN with RRDB blocks and a Relativistic Average GAN discriminator, trained on carefully aligned LR GONG Hα and HR GST observations, and guided by perceptual and L1 losses with network interpolation to balance perceptual quality and quantitative fidelity. The results demonstrate improved recovery of fine-scale chromospheric structures such as sunspots, filaments, and fibrils, achieving $\text{MSE}=467.15$, $\text{RMSE}=21.59$, and $\text{CC}=0.7794$, though residual misalignment limits accuracy. This approach enables GST-like insights from widely available GONG data, with potential for extended-range monitoring, while future work aims to improve alignment, expand the dataset, and incorporate physics-informed refinements.

Abstract

High-resolution (HR) solar imaging is crucial for capturing fine-scale dynamic features such as filaments and fibrils. However, the spatial resolution of the full-disk H$α$ images is limited and insufficient to resolve these small-scale structures. To address this, we propose a GAN-based superresolution approach to enhance low-resolution (LR) full-disk H$α$ images from the Global Oscillation Network Group (GONG) to a quality comparable with HR observations from the Big Bear Solar Observatory/Goode Solar Telescope (BBSO/GST). We employ Real-ESRGAN with Residual-in-Residual Dense Blocks and a relativistic discriminator. We carefully aligned GONG-GST pairs. The model effectively recovers fine details within sunspot penumbrae and resolves fine details in filaments and fibrils, achieving an average mean squared error (MSE) of 467.15, root mean squared error (RMSE) of 21.59, and cross-correlation (CC) of 0.7794. Slight misalignments between image pairs limit quantitative performance, which we plan to address in future work alongside dataset expansion to further improve reconstruction quality.

Improving the Spatial Resolution of GONG Solar Images to GST Quality Using Deep Learning

TL;DR

This work addresses the challenge of extracting high-spatial-resolution information from full-disk solar images by cross-instrument super-resolution. It employs Real-ESRGAN with RRDB blocks and a Relativistic Average GAN discriminator, trained on carefully aligned LR GONG Hα and HR GST observations, and guided by perceptual and L1 losses with network interpolation to balance perceptual quality and quantitative fidelity. The results demonstrate improved recovery of fine-scale chromospheric structures such as sunspots, filaments, and fibrils, achieving , , and , though residual misalignment limits accuracy. This approach enables GST-like insights from widely available GONG data, with potential for extended-range monitoring, while future work aims to improve alignment, expand the dataset, and incorporate physics-informed refinements.

Abstract

High-resolution (HR) solar imaging is crucial for capturing fine-scale dynamic features such as filaments and fibrils. However, the spatial resolution of the full-disk H images is limited and insufficient to resolve these small-scale structures. To address this, we propose a GAN-based superresolution approach to enhance low-resolution (LR) full-disk H images from the Global Oscillation Network Group (GONG) to a quality comparable with HR observations from the Big Bear Solar Observatory/Goode Solar Telescope (BBSO/GST). We employ Real-ESRGAN with Residual-in-Residual Dense Blocks and a relativistic discriminator. We carefully aligned GONG-GST pairs. The model effectively recovers fine details within sunspot penumbrae and resolves fine details in filaments and fibrils, achieving an average mean squared error (MSE) of 467.15, root mean squared error (RMSE) of 21.59, and cross-correlation (CC) of 0.7794. Slight misalignments between image pairs limit quantitative performance, which we plan to address in future work alongside dataset expansion to further improve reconstruction quality.

Paper Structure

This paper contains 18 sections, 4 equations, 7 figures.

Table of Contents

  1. Introduction
  2. Dataset Description
  3. LR Dataset (GONG H$\alpha$ Images)
  4. HR Dataset (BBSO GST Images)
  5. Dataset Splits and Usage
  6. Data Alignment
  7. Initial Cropping and Region Selection
  8. Temporal Alignment of GONG Image Sequence
  9. Geometric Co-Alignment with GST Observations
  10. Final Cropping and Pair Generation
  11. Methodology
  12. Network Architecture
  13. Relativistic Discriminator
  14. Perceptual Loss
  15. Network Interpolation
  16. ...and 3 more sections

Figures (7)

  • Figure 1: Original dataset examples: (a) Full-disk GONG image with GST coverage marked in red; (b) Cropped GONG region corresponding to GST; (c) Original GST HR observation.
  • Figure 2: Initially aligned data examples: (a) Cropped and aligned GONG LR image; (b) GST HR image after rotation adjustment and preprocessing.
  • Figure 3: Final aligned dataset pair example: (a) GONG LR image fully prepared; (b) Corresponding GST HR image fully aligned and cropped.
  • Figure 4: Architecture of Real-ESRGAN used in this study. It consists of an initial convolutional layer, multiple Residual-in-Residual Dense Blocks (RRDB), followed by upsampling and additional convolutional layers to reconstruct the HR solar image.
  • Figure 5: Example of the generated HR image from GONG LR image of 2023/08/31 21:35:02 (a) GONG LR testing image; (b) Generated HR image; (c) The ground truth of the GST HR image.
  • ...and 2 more figures