Towards real-world BCI: CCSPNet, a compact subject-independent motor imagery framework

A conventional brain-computer interface (BCI) requires a complete data gathering, training, and calibration phase for each user before it can be used. In recent years, number of subject-independent (SI) BCIs have been developed. Many these methods yield weaker performance compared to the subject-dependent (SD) approach, some are computationally expensive. potential real-world application would greatly benefit from more accurate, compact, efficient BCI. this work, we propose novel BCI framework, named CCSPNet (Convolutional Common Spatial Pattern Network) that is trained on motor imagery (MI) paradigm large-scale electroencephalography (EEG) signals database consisting 400 trials every 54 subjects who perform two-class hand-movement MI tasks. The proposed framework applies wavelet kernel convolutional neural network (WKCNN) temporal (TCNN) in order represent extract spectral features EEG signals. common spatial pattern (CSP) algorithm implemented feature extraction, CSP reduced by dense network. Finally, class label determined linear discriminant analysis (LDA) classifier. evaluation results show possible compact achieves both SD SI state-of-the-art comparable complex expensive models.