Research On Code Modulation Visual Evoked Potential Brain-Computer Interface Based On Transfer Learning

One of the main problems that a brain-computer interface(BCI)face is that a long training stage is required for acquiring sufficient training data to calibrate its classification model just before every use.Transfer learning is a promising method for addressing the problem.This method can reduce training time or even achieve zero training by transferring the experimental data of previous users to new users as training data to construct a classification model.Due to the randomness and non-stationarity of EEG signals,the data distribution between different subjects is very different.If the data of existing subjects are directly transferred to new subjects as their training data without any processing,the classification model constructed is unstable and not accurate.How to improve the performance of transfer learning is a problem that needs to be solved at present.This dissertation proposes a Riemannian geometry-based transfer learning algorithm for code-modulated visual evoked potential(c-VEP)-based BCI.This is the first study combining Riemannian geometry with transfer learning to reduce training time for cVEP BCI.The algorithm includes the main steps of log-Euclidean data alignment(LEDA),super-trial construction,covariance matrix estimation,training accuracybased subject selection(TSS)and minimum distance to mean(MDM)classification.Sixteen subjects participated in a c-VEP BCI experiment and the recorded data were used in offline analysis.The algorithm proposed in this dissertation is evaluated on this dataset and compared with several other algorithms.Data analysis results show that the algorithm achieves higher classification accuracy under the same number of training trials.Equivalently,the algorithm reduces the training(calibration)time of the BCI at the same performance level and thus facilitates its application in real world.