SecNet: A Second Order Neural Network for MI-EEG

This repository contains the code for the paper:

Liang, W., Allison, B. Z., Xu, R., He, X., Wang, X., Cichocki, A., & Jin, J. (2025).
SecNet: A second order neural network for MI-EEG.
Information Processing & Management, 62(3), 104012.

Overview

SecNet is a neural network model designed to enhance motor imagery (MI) decoding from EEG signals by leveraging second-order information.

Requirements

• Python 3.10
• PyTorch 2.2.2
• Numpy 1.26.3 (Numpy 2.x.x may cause errors in EEG processing)
• Other dependencies can be found in requirements.txt.

Installation

1. Clone the repository:

   git clone https://github.com/SecNet-mi/SecNet.git
   cd SecNet

2. Install the required packages:

   pip install -r requirements.txt

Usage

Download Dataset and Prepare Data

The data used in this study are already publicly available at the following: OpenBMI in http://dx.doi.org/10.5524/100542; BCI competition IV 2a in https://bbci.de/competition/iv/.

1. Prepare your EEG dataset in the required format.

RawData/
├── BCICIV_2a/A01E.gdf...
└── OpenBMI/
    ├── session1/sess01_subj01_EEG_MI.mat ...
    └── session2/sess02_subj01_EEG_MI.mat ...

After downloaded the dataset, you need to run:

python DataLoader\LoadData.py

To run the model:

2. Train the model using the provided script:

   python ho.py 

For more details on usage, check the documentation in the codebase.

Citation

If these codes help you, please cite:

@article{liang2025secnet,
  title={SecNet: A second order neural network for MI-EEG},
  author={Liang, Wei and Allison, Brendan Z and Xu, Ren and He, Xinjie and Wang, Xingyu and Cichocki, Andrzej and Jin, Jing},
  journal={Information Processing \& Management},
  volume={62},
  number={3},
  pages={104012},
  year={2025},
  publisher={Elsevier}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.