Motor Imagery Recognition Using Filter Bank Common Spatial Pattern and Recurrent Neural Networks

A Brain-Computer Interface (BCI) technology allows a person to communicate with hardware without involving physical movement. The instrument that connects brain commands to device movements is called an intermediate, one of which is an Electroencephalogram (EEG). EEG captures brain signals by placing electrodes on the scalp and translating the recorded signals to external commands. In BCI, one of the variables that can be used is motor imagery (MI), where a person can control an external device by imagining its movements. EEG signal processing is important in BCI performance. There are feature extraction and classification. This paper proposed the Common Spatial Pattern (FBCSP) technique for feature extraction and Recurrent Neural Network (RNN) to recognize BCI competition data sets. FBCSP is effective in MI recognition because it extracts frequency and spatial features. RNN is used for the classification by considering the data sequence of each spatial feature of FBCSP extraction. The experiment result showed that of the 9 dataset subjects, the accuracy of each subject is 94.25%, 87.36%, 94.25%, 90.80%, 85.06%, 89.66%, 95.40%, 95.40%, and 93.10, with an average accuracy of 91.69%. If all subject data is combined, the accuracy is 89.97%.