Synthetically generated clouds on ground-based solar observations
- 1. Université de Toulon, Aix Marseille Univ, CNRS, LIS, Marseille, France
- 2. Observatoire de Paris/PSL, Paris, France
Description
A dataset consisting of Ca II & H-alpha images taken at the Paris Meudon Observatory. Synthetically generated cloud coverage has been applied to clean images, thereby creating an (cloudy, clean) pair--facilitating the training of cloud-removal algorithms.
Data description
The Ca-II and H-α synthetic dataset comprise respectively 319 and 367 pairs of shadow/shadow-free images, split into 223/96 and 256/111 training/testing pairs.
Listed here are two zip archives:
- filament-bounding-boxes.zip -- bounding boxes of filaments that were used to compute the patched metrics.
- synthetic-clouds.zip -- the cloudy input/clean output images that are used to train machine learning algorithms.
A PyTorch dataset has been created that handles the download, importing, and usage of this dataset. You can find this code at the github repository: https://github.com/jaypmorgan/cloud-removal
Pre-processing routines
To generate this set of data, we have applied a series of pre-processing routines. These are:
- Correct determination of the solar limb (source code can be found at: https://gitlab.lis-lab.fr/presage/solar-limb-detection).
- Scaling the solar disk to 420 pixels, and centring it at 511.5 pixels in the x and y dimensions.
- Setting background values outside the solar disk to 0.
- Normalising the disk intensity values into the range of 0-1.