Table 1.

Compared methods.

Methods	Algorithm Composition and Processing Flow
CSP	Band-pass filtered EEG signals are spatially filtered by CSP. The logarithm of the variances of spatially filtered signals are extracted as features [10].
Wavelet-CSP	The EEG signals of each channel are decomposed using DWT, the wavelet base is db4. The number of decomposition layers of dataset 1 is 3, and the other datasets are 4. The sub-bands related to motor imagery are used to reconstruct new channels, and then feature extraction is performed using CSP [13].
WPD-CSP	The EEG signals of each channel are decomposed using WPD, the wavelet base is db4. The number of decomposition layers of dataset 1 is 3, and the other datasets are 4. The sub-bands related to motor imagery are used to reconstruct new channels, and then feature extraction is performed using CSP [15].
SFBCSP	The original EEG signals are filtered into 17 sub-bands, and features are extracted for each sub-band using CSP. The filter bandwidth is 4 Hz and the overlap rate is 2 Hz in the range of 4-40 Hz. The sub-band features are selected by LASSO [23].
SBLFB	The original EEG signals are filtered into 17 sub-bands, and features are extracted for each sub-band using CSP. The filter bandwidth is 4 Hz and the overlap rate is 2 Hz in the range of 4–40 Hz. The sub-band features are selected by sparse Bayesian learning [24].
CSP-Wavelet+LASSO	After band-pass filtering, features are extracted using CSP-Wavelet. Features are selected by LASSO. Ensemble learning is used for secondary feature selection and classification model construction.
CSP-WPD+LASSO	After band-pass filtering, features are extracted using CSP-WPD. Features are selected by LASSO. Ensemble learning is used for secondary feature selection and classification model construction.
CSP-FB+LASSO	After band-pass filtering, features are extracted using CSP-FB. Features are selected by LASSO. Ensemble learning is used for secondary feature selection and classification model construction.
CSP-Wavelet+LOG	After band-pass filtering, features are extracted using CSP-Wavelet. Features are selected by LOG. Ensemble learning is used for secondary feature selection and classification model construction.
CSP-WPD+LOG	After band-pass filtering, features are extracted using CSP-WPD. Features are selected by LOG. Ensemble learning is used for secondary feature selection and classification model construction.
CSP-FB+LOG	After band-pass filtering, features are extracted using CSP-FB. Features are selected by LOG. Ensemble learning is used for secondary feature selection and classification model construction.