This repository contains a CNN-based neural network model intended to classify motor imagery (MI) EEG data. It is developed using the Physionet MI/ME dataset.
- Physionet data
- use the script
eegmmidb/download.sh(requires wget) OR - download from https://www.physionet.org/pn4/eegmmidb/
- adjust the file path in
util.pyto the location of the files in your system
- use the script
- Python 2 environment with:
- numpy, scipy, matplotlib
- tensorflow & keras
- pyedflib (for data input)
The main.ipynb is an iPython notebook representing the entry point. I recommend running it in an Anaconda evironment (which also includes numpy/scipy/matplotlib). Anaconda can be downloaded from: https://www.anaconda.com/download/
The code represents a starting point for data preparation and training the neural network models. It does not provide methods for plotting/evaluation/visualization.
Note: The code is no longer maintained and comes without warranty for correctness. It can be freely used, changed, and distributed. If it was helpful to your work, consider citing
Dose, H., Møller, J. S., Iversen, H. K., & Puthusserypady, S. (2018). An end-to-end deep learning approach to MI-EEG signal classification for BCIs. Expert Systems with Applications, 114, 532–542. https://doi.org/10.1016/j.eswa.2018.08.031