This repository contains the source code to reproduce the coronal holes segmentation model from the paper Segmentation of coronal holes in solar disk images with a convolutional neural network (published in MNRAS).
Note that the code was updated to be compatible with helio framework. Checkout to the branch mnras2018 for original version.
Try a demo running directly in the browser to see how the model will process solar disk images you will feed to it. Note that the model was optimized to SDO/AIA images obtained from SunInTime website.
Clone the repo
git clone --recursive https://github.com/observethesun/coronal_holes.git
The code is based on helio framework. See the API documentation to learn more about its features.
A dataset proposed for model training consists of SDO/AIA 193 Angstrom solar disk images in 1K resolution obtained from SunInTime website and a dataset of coronal holes regions provided by the Kislovodsk Mountain Astronomical Station.
The notebook Train_segmentation_model contains data preprocessing, neural network architecture and model training pipeline. The notebook Apply_segmentation_model demonstrates inference pipeline.
Illarionov E., Tlatov А., 2018, MNRAS, 481, 4.
