You will find here all the codes and instructions needed to reproduce the experiments performed in "A realistic MI-EEG data augmentation approach for improving deep learning in BCI decoding", by Catalina M. Galván, Rubén D. Spies, Diego H. Milone and Victoria Peterson.
Here, all the functions to generate user-specific MI-EEG data with different characteristics are provided in src.
In lib/FBCNet you will find all the codes for using the FBCNetToolbox models, which have been adapted from their original implementation to include the possibility to employ data augmentation strategies.
Notebooks with detailed examples are included in notebooks.
- Example_generate_data_for_augmentation: a notebook in which extraction of periodic and aperiodic parameters from real MI-EEG data is implemented and then data that can be used for data augmentation is generated using these user-specific parameters.
- Example_generate_data_fatigue: a notebook that shows the simulation of MI-EEG data with fatigue effects.
- Example_generate_data_different_user_capabilities: a notebook that illustrates how different user capabilities to control a MI-BCI can be simulated.
- Example_cross_session_data_augmentation: a notebook that shows how simulated MI-EEG data can be employed for data augmentation. FBCNet model is trained without and with data augmentation in a cross-session scenario.
- Example_cross_validation_simdataset: a notebook that implements a 10-fold cross-validation scenario with simulated MI-EEG data.