suvi-trainer: DEPRECATED
SolarAnnotator instead of this tool.
Please use theThe Solar Ultraviolet Imager (SUVI) aboard the National Oceanic and Atmospheric Administration's Geostationary Operational Environmental R-Series satellites is used in machine learning applications to create thematic maps, images showing where different features are on the Sun. This tool allows domain experts to load images, manipulate them, and create labeled maps, which are used in training solar image classifiers.
Getting Started
These instructions will get you a copy of the project up and running on your local machine
for development and testing purposes. At the moment, it is advised to run this code in a fresh virtual environment with
Python 3.5. Installation should only require running the setup.py
script.
Installing
pip3 install suvitrainer
You will need to download and edit the configuration file to include your name and the upload password. Please contact me for the password and further information.
Please note that the extra scripts may require packages that are not automatically installed. They are auxiliary and not fundamental for the annotation tool to run. Everything needed for run_suvitrainer.py and the main annotation tool should be automatically installed. If you want to use an auxiliary script download the repo.
Running
The run_suvitrainer.py script should provide all needed functionality for the average user.
It takes a couple optional arguments: verbosity and dates.
The verbosity, -v
or --verbose
argument will print helpful status information while running.
The date
option allows three methods of specifying which date to run on:
simply a date string (2018-08-05T17:52), a path to a local file that contains a list of date
strings where each one is on a different line, or a url to an online list of dates.
The default is to pull using the url for dates.txt stored in this repository. This is preferred to create
a large curated data-set with some repeats for validation.
The list of options for input images:
-
halpha
(from GONG) - any of the AIA EUV channels listed as
aia-[WAVELENGTH]
with the wavelength in angstroms, e.g.aia-131
- any of the SUVI EUV L1B channels listed as products, e.g.
suvi-l1b-fe131
(see the FTP site for a list) - any locally saved SUVI l2 composite images (only available internally to NOAA employees at this time), e.g.
suvi-l2-ci094
For more detailed usage, see the user guide.
Data
The output of training is saved as a FITS file which is later converted into a labeled png. Labeled data will be available shortly. A couple examples are available here. This labeled data is then used in machine learning classification of solar images.
Contributing
Please contact me (email or just through pull/issue requests) for updates. If you're interested in labeling images or would like access to the labeled database, please contact me through email.
Authors
- J. Marcus Hughes - Initial work
License
This project is licensed under the MIT License - see the LICENSE file for details
Acknowledgments
- Guidance from Dan Seaton, Jon Darnel, and Vicki Hsu