Brain-Computer Interface of Motor Imagery and Emotion using Multiple Recurrent Neural Networks

Abstract

Brain-Computer Interface (BCI) can control external devices without directly moving from processed brain signals. The performance of the BCI was determined mainly by the device used, one of which is the Electroencephalogram (EEG). There are variables of EEG signals commonly used as BCI actions, such as emotion, motor imagery, and concentration. These variables can be used single or multiple. Multivariable BCI actions add features to drive external devices from the brain directly. Each variable EEG signal has its characteristics, including the frequency band. Therefore, processing each variable as a separate network is an appropriate choice. One method often used to identify data series such as EEG signals is Recurrent Neural Networks (RNN). This paper proposed multiple RNN in motor imagery and emotion of EEG signal to drive BCI. The EEG signal was filtered at frequencies 8 – 30 Hz for the motor imagery and emotion variables. Both use the Wavelet transform. The experiment results gave 91.59% accuracy when using Multiple RNNs compared to a single RNN, which obtained an accuracy of 76.18%. Moreover, the use of Wavelets in filtering EEG signals increased the accuracy by 21.84%.