A transplantation of subject-independent model in cross-platform BCI

With the development of wearable technology, portable wireless systems have been used gradually for collecting electroencephalogram (EEG) signals. However, the introduction of portable collection devices always means a descent in signal-to-noise ratio (SNR) of EEG. Subject-independent brain-computer interface (BCI) avoids conventional calibration procedure for new users. However, whether subject-independent model can be used in cross-platform BCI has not been discussed so far. This paper transplanted the subject-independent model from a high-SNR platform to a lower one for recognition in P300-Speller. After comparing their EEG features elicited from diverse collection platforms, a model adjustment strategy was proposed to increase recognition accuracy. By model adjustment, the average accuracy was 85.00% in online spell experiments. The results indicate it is feasible for subject-independent model transplantation, especially after model adjustment strategy. It provides technology supported for further development of cross-platform BCI.