Hybrid Deep Learning-Based EEG BCI for Accurate Mental Intention Recognition

Brain-Computer Interface (BCI) systems are changing the manner in which severe motor or speech impaired persons communicate, so that thoughts may be directly converted to actions. This paper describes a cheap, real-time EEG-based BCI that are able to identify three major mental intentions that are yes, no, and help. We have an integrated EEG acquisition system using a mixed deep learning model—Convolutional Neural Networks (CNN) for spatial network features, Recurrent Neural Networks (RNN) for time patterns and a Support Vector Machine (SVM) for final classification. This combination captures both the where and when of brain signals, which deals with the drawbacks of less complex and single-model methods. Experiments demonstrate that our system attains more than 95% classification and is useful in daily life, as it can be run in real time. Combining low costs with high technology but simple to use machine learning approaches, this work shows that there is potential to move towards available and viable assistive communication tools.