Segmentation of coronal holes in solar disk images with a convolutional neural network

TL;DR

This paper presents a neural network-based method for segmenting coronal holes in solar images, offering improved accuracy and stability over traditional thresholding techniques, with applications in analyzing solar cycle variations.

Contribution

The study introduces a neural network approach for coronal hole segmentation that outperforms traditional methods and simplifies the analysis process.

Findings

Neural network accurately isolates coronal holes without complex pre- and post-processing.

The method demonstrates stability across different datasets.

Coronal hole areas increase threefold from solar maximum to minimum.

Abstract

Current coronal holes segmentation methods typically rely on image thresholding and require non-trivial image pre- and post-processing. We have trained a neural network that accurately isolates CHs from SDO/AIA 193 Angstrom solar disk images without additional complicated steps. We compare results with publicly available catalogues of CHs and demonstrate stability of the neural network approach. In our opinion, this approach can outperform hand-engineered solar image analysis and will have a wide application to solar data. In particular, we investigate long-term variations of CH indices within the solar cycle 24 and observe increasing of CH areas in about three times from minimal values in the maximum of the solar cycle to maximal values during the declining phase of the solar cycle.