Motor Imagery Classification
Dec 10, 2024·
,,,·
1 min read
Venkatarami Reddy
Mukesh Mann
Alok Sharma
Rakesh P. Badoni
WGANAbstract
Motor imagery (MI) is a brain-computer interface (BCI) technique enabling robotic control and neurological restoration using brain signals, particularly EEG data. Despite advancements in deep neural networks (DNNs), collecting large-scale, high-quality MI-EEG data remains challenging. This study explored data augmentation techniques such as geometric transformations and noise addition, combined with generative adversarial networks (GANs), to enhance MI-EEG data. Time-frequency data from the “BCI Competition IV 2a” dataset were processed using continuous wavelet transforms to improve data representation for GANs. Artificially generated samples were combined with actual data, and a convolutional neural network (CNN) classified MI signals into four categories. The approach significantly improved classification accuracy.
Stage
Plan:
- Data Collection: Used MI-EEG signals from the “BCI Competition IV 2a” dataset.
- Preprocessing: Applied continuous wavelet transform to convert raw EEG signals into time-frequency representations.
- Data Augmentation: Used GANs to generate artificial EEG samples and combined them with real samples using techniques like geometric transformations and noise addition.
- Classification: A CNN model classified the augmented dataset into four MI categories.
- Results: Augmentation significantly enhanced the model’s classification accuracy compared to using raw or unaugmented data.
