Data and Codes for the article: "Developing an Automated Detection, Tracking and Analysis Method for Solar Filaments Observed by CHASE via Machine Learning"

Codes

data_preparation.ipynb

We use this to get our dataset. This includes the function `readchase` for reading CHASE Hα spectral files and the function `keams_process` for preprocessing the spectra. Within this code segment, we employ 'sklearn.cluster.KMeans' for unsupervised clustering of the spectral data. Subsequently, we apply morphological closing operation on the results of K-means to obtain our dataset.

You can get the file list of our dataset in this code, and then download the files from Solar Science Data Center of Nanjing University. The file is too large to be conveniently uploaded here.

train_unet.py

You can run `python train_unet.py` to train the unet model. `UNet` defines our U-Net model. The original code is from labml.ai, and we have made some modifications to their model. 

In addition, the code for model prediction and evaluation is also included in it.

track.ipynb

This is our code for automated filament tracking. You can find the code flow of our tracking algorithm with helpful comments for better understanding.

feature_extraction.py

You can run `python feature_extraction.py` to obtain the Hα line central imaging of the Chase Hα file, the detection result of filaments, the cloud model inversion results of filaments, and the results of straightening the filaments along the main axis. The function `inversion_cloud` is used for cloud model inversion and `straightening_img` is used for straightening the filaments along the main axis.

These codes are provided on GitHub and Zenodo under the MIT license.

Data

detection/data

This folder comprises the training set `train.zip`, validation set `valid.zip`, test set `test.zip`, and the state dictionary of our U-Net model `model.zip`.

tracking

This folder includes 10 groups of results for tracking filaments, along with our accuracy log `result.xlsx`.