Investigation of Convolutional Neural Network with Motor Imagery-based Brain-Computer Interface Decoding

In recent years, Brain-Computer Interfaces (BCI) have become increasingly popular for assisting individuals with disabilities in restoring communication and mobility. The area of research focuses on utilizing convolutional neural networks (CNNs), specifically deeper architectures with high dropout, to decode brain signals and enable users to interact with the world more naturally, particularly through motor imagination. Motor imagination involves analyzing brain signals to decipher a person's intention to perform a motor task. CNNs are well-suited for BCI decoding due to their ability to learn intricate features from raw EEG data. However, the challenge lies in preventing overfitting in deeper CNN architectures, which have a larger number of parameters. To overcome this, the proposed approach involves a high truncation CNN architecture, a regularization technique that randomly turns off neurons during training to mitigate overfitting