FCD: Fourier Convolutional Decoder

FCD is designed to work with STIX data; it takes 48 Fourier components (24 real, 24 imaginary) and reconstructs the image corresponding to them.

Below is the comparison of the FCD with other popular STIX reconstruction alorithms.

Running the Project

pip install -r requirements.txt to install all the necessary packages.

Project Bits and Pieces

• nn includes training and testing of the FCD.
• metrics, data, and utils include relevant utility functions of the project.

Pretrained model of the FCD and a demo of the model are available on HuggingFace.

Data

• Refer to STIX data center to use observational STIX data.
• Refer to STIX data generator to generate your own simulated data.

Citation

If this repository proves useful for your research, please cite our work as follows.

@article{SelcukSimsek2025,
 author    = {Selcuk-Simsek, Merve and Massa, Paolo and Xiao, Hualin and Krucker, S{\"a}m and Csillaghy, Andr{\'e}},
 title     = {Fourier convolutional decoder: reconstructing solar flare images via deep learning},
 journal   = {Neural Computing and Applications},
 pages     = {1--32},
 year      = {2025},
 publisher = {Springer}
 doi       = {10.1007/s00521-025-11283-6},
 url       = {https://doi.org/10.1007/s00521-025-11283-6}
}