Successful applications of brain-computer interface (BCI) approaches to motor imagery (MI) are still limited. In this paper, we propose a classification framework for MI electroencephalogram (EEG) signals that combines a convolutional neural network (CNN) architecture with a variational autoencoder (VAE) for classification. The decoder of the VAE generates a Gaussian distribution, so it can be used to fit the Gaussian distribution of EEG signals. A new representation of input was developed by combining the time, frequency, and channel information from the EEG signal, and the CNN-VAE method was designed and optimized accordingly for this form of input. In this network, the classification of the extracted CNN features is performed via the deep network VAE. Our framework, with an average kappa value of 0.564, outperforms the best classification method in the literature for BCI Competition IV dataset 2b with a 3% improvement. Furthermore, using our own dataset, the CNN-VAE framework also yields the best performance for both three-electrode and five-electrode EEGs and achieves the best average kappa values 0.568 and 0.603, respectively. Our results show that the proposed CNN-VAE method raises performance to the current state of the art.

No abstract

The application of wireless sensors in the brain-computer interface (BCI) system provides great convenience for the acquisition of electroencephalography (EEG) signals. However, a large amount of training data is needed to build the classification architectures used in motor imagery (MI) brain-computer interface (BCI), which is time-consuming to generate. To address this issue, transfer learning has gained significant attention in a small sample setting BCI system. The transfer learning methods have shown promising results by leveraging labeled patterns from the source domain to learn robust classifiers for the target domain, which has only a limited number of labeled samples. However, the successful application of such approaches in a motor imagery BCI remains limited. In this paper, we present a novel framework called domain transfer multiple kernel boosting (DTMKB), which extends the DTMKL algorithms by applying boosting techniques for learning kernel-based classifiers with the transfer of multiple kernels. Based on the proposed framework, we examined their empirical performance in comparison to several state-of-theart algorithms on two MI task datasets. DTMKB yields the best performance for all datasets and achieves the best average classification accuracy 87.60%, 76.00%, 74.66%, and 74.13%, respectively. In particular, the proposed framework can be applied successfully in a small sample of EEG motor imagery signals.INDEX TERMS Brain-computer interface EEG, transfer learning, boosting, domain transfer multiple kernel boosting.

No abstract