Research On Auxiliary Brain State Monitoring Methods In Brain-computer Interfaces

Brain-computer interface(BCI)refers to the establishment of neural pathways between the brain and external devices to achieve the purpose of controlling external devices by the brain.Auxiliary brain state monitoring refers to monitoring the brain’s state in real time when controlling external devices.State,the process of assisting braincomputer decision-making,can effectively reduce the erroneous decision-making output of BCI and improve the reliability of brain-computer joint decision-making,so it is a key issue in BCI research.Aiming at the problems of low recognition accuracy and few separable patterns in auxiliary brain state monitoring research,this paper uses highprecision recognition of attention states from single-channel EEG signals(EEG)and steady-state visual evoked potential(SSVEP)tasks.Two aspects of multi-type auxiliary brain state monitoring were studied and analyzed in detail.Aiming at the problem of single-channel attention-assisted state monitoring,the classic international 10-20 lead EEG acquisition system was used to obtain EEG signals in the attention-related Fp1 brain region.Aiming at the influence of noise and artifacts,the data is preprocessed through filtering,notch,downsampling and other steps.Feature parameters such as spectrum,entropy and reconstruction network are extracted as classification features.Aiming at the multi-feature brain state judgment problem,a feature fusionmethod is proposed.The method integrates features such as spectrum,entropy and reconstructed network,and uses mutual information to screen out the optimal feature combination.The classification performance was evaluated using the receiver operating characteristic curve and the area under the ROC curve,respectively.Experimental results show that the proposed MIFF method outperforms state-of-the-art methods regardless of the data length on both devices.In particular,the method with a data length of 2.5 s has an average AUC of 0.8545,which is 56.79% higher than spectral features,17.98% higher than entropy features,and 16.02% higher than reconstructed network features.Aiming at the problem of multi-task SSVEP-assisted state monitoring,the joint frequency phase modulation method was used to obtain the whole-brain EEG signals of subjects under different SSVEP tasks.Aiming at the influence of noise and artifacts,the data is preprocessed through filtering,notch,downsampling and other steps.Aiming at the problem of multi-task SSVEP-assisted state monitoring,a complex network algorithm based on individualized frequency bands and common spatial patterns(IFBCSPOCN)is proposed.The EEG does spatial filtering and personalized frequency band search to improve the classification accuracy of multi-task SSVEP-assisted condition monitoring.Finally,the classification performance of OCN method,CSPOCN method and IFBCSPOCN method in multi-task SSVEP-assisted state detection is compared.The results show that the proposed IFBCSPOCN method outperforms other methods on all data length scales as well as in binary classification experiments.Especially when the data length is 1 s,the IFBCSPOCN method achieves the highest recognition accuracy of 0.99 in multi-task SSVEP-assisted state detection.Compared with the CSPOCN method,the classification accuracy is 20.73%;compared with the OCN method,the classification accuracy is increased by 41.43% and 24.24%,respectively.