REIMEI, a Japanese satellite mission, collected a lot of data from the northern lights (or aurora) from 2005 to 2012. There are different types of auroral phenomena, some include Alfvenic, Diffuse, and Inverted-V aurora. I am aiming to use some of the data that has already been identified and a machine learning algorithm to categorize the data that hasn't been identified yet, so that science can be done with a greater sample size.
Most training and files are not included, as there are too many files and I am not sure I am allowed to share them all publicly. Models are not included because GitHub was giving me HTTP code 500s when I would try to push them.
At the moment, the models seem to have a tough time differentiating between Inverted V and Diffuse types, as shown with the sample Inverted_V.png file sorted to the ./Files/Guessed/Diffuse folder. This project will help categorize more auroral data correctly, thus allowing there to be more training data for a more sophisticated algorithm, such as a RESNET 50 algorithm.
crop_images.py: script to crop REIMEI EISA QuickLook plots in./Files/Uncroppedto just the plot regions. Outputs to./Files/Croppedtrain_model.py: script to train the models. Won't be accurate with the quantity of training data I provided in this repouse_model.py: script to use multiple models to automate classification of cropped plots in./Files/Unlabeled
So far, this has been developed in Summer 2024 @ NASA GSFC and October 2024 @ the Technica hackathon.
A mamba environment was used, with keras, matplotlib, numpy, pillow (PIL), tensorflow, and tqdm installed.
Mentors: Emma Mirizio (UMD, NASA GSFC code 673), Marilia Samara (NASA GSFC code 673)