Synchronous Classification of SSVEP-EMG Fusion Signal from Occipital Electrodes Using Convolutional Neural Networks

Combinations of steady-state visual evoked potential (SSVEP) and electromyography (EMG) are one of the most widely used hybrid brain-computer interfaces (BCI). For users who are suffering from severe motor impairments and could only control muscles above their necks, the EMG of jaw clench is commonly used together with SSVEP. Traditional asynchronous signal collecting and serial processing method is time-consuming. This study explored the simultaneous classification of SSVEP and jaw clench-related EMG from the same occipital electrodes. A convolutional neural network (CNN) model was adopted to classify the fusion signal after extracting different time and frequency domain features. Synchronous 12-class identification from 3 jaw clench patterns and 4 stimulate frequencies was tested on ten subjects. The 10-folder cross-validation results showed that using only 4 occipital electrodes, the CNN model with continuous wavelet transform feature achieved the overall best
$89.3\pm4.0\%$
accuracy for sync 12-class. Besides, the parallel acquiring and analyzing method has the lowest consuming time. The proposed method could reduce the signal channels and shorten the collect-process time of the SSVEP-EMG hybrid BCI.