Research And Implementation Of Noninvasive Multimodal Brain-computer Interface

Non-invasive Brain-computer interface(BCI)is a kind of human-computer interaction technology which can directly establish communication between human brain and external devices.It has the advantages of low cost and convenient operation,therefore,has important application values in the fields of medical rehabilitation,education,military,entertainment and smart home.BCI mixed with multiple EEG models are beneficial to higher recognition rate and accuracy and more stable output.In this dissertation,related research attempts are carried out in two directions in BCI:motor imagery(MI)and steady-state visual evoked potential(SSVEP).The specific research work is summarized as follows:(1)A shallow convolutional neural network(SCNN)is employed to classify MI EEG signals,using a single feature layer to mine the intrinsic features of EEG sequences.The EEG signals are firstly converted into two-dimensional image data and then input to the SCNN for training and testing.(2)A new method to reduce the effect of the "BCI illiteracy" is proposed,which combines a sensitivity-based paradigm selection method and generalized Riemann minimum distance to mean classifier(GRMDM).The combination number formula is used to increase the distinguishability of the combination among the categories of paradigms,and then the sensitivity index is used to select paradigms.A weighting factor is added between the log-Euclidean metric and Riemannian divergence in the GRMDM to select a classifier feasible to the individual.(3)A new control strategy based on recall rate is proposed to select control instruction stimulus frequency,for SSVEP based BCIs.A visual stimulus selection method based on the recall rate is designed.By calculating the recall rate index,the visual stimulus frequency of control instructions is selected and determined,aiming to improve the safety performance and reliability of online experiments.(4)A multi-modal BCI robot online control system based on MI+SSVEP is built.MI-EEG signal is generated independently of external stimuli and used to control the orientation of the robot.The SSVEP-EEG signal detection and processing method is relatively simple and more accurate on classification,which is used to control the robot“forward”,“backward” and “stop” movements.This design significantly relaxes the constraints of "BCI illiteracy " and expands the diversity of control instructions of BCI,which is expected to be applied in more abundant scenarios.