A Joint SSA and CNN Algorithm to Remove EMG Artifacts from Single Channel EEG

There often exist unavoidable artifacts in the electroencephalogram (EEG) acquisition process. Singular spectrum analysis (SSA) has been used as an efficient single channel signal decomposition algorithm in EEG denoising task. Recently, deep neural networks have also been utilized to remove artifacts in an end-to-end manner. Compared with other biomedical signals, very few papers have reported to eliminate electromyography (EMG) from EEG signals. However, the existing deep leaning models still have the potential of improvement. Motivated by these prior works, we propose a novel joint SSA and convolutional neural network (CNN) algorithm to remove EMG from EEG signals. Numerical results obtained based on a stimulated hybrid dataset suggest that the proposed algorithm outperforms traditional approaches when the signal-to-noise ratio (SNR) exceeds -1dB.