Application-oriented Steady-state Visual Evoked Potential-based Brain-computer Interface Algorithm And System Research

In the past two decades,thanks to the continuous improvement in the fields of sensors,computers,communications,etc.,Brain-Computer Interface(BCI)technology has also achieved leap-forward development.Steady-State Visual Evoked Potential(SSVEP)BCI is favored by its non-invasiveness,high information transmission rate,easy to train,and relatively low cost.At present,SSVEP-BCI has been able to achieve an information transfer rate(ITR)of up to 325.3 bits/min,but in other respects,the existing SSVEP-BCI system still does not meet the actual application requirements.This thesis mainly focused on SSVEP-BCI,and studied from several aspects,such as signal modeling improvement,training costs reduction,applicable population expansion,and high-performance and reliable asynchronous system implement,which can all improve the practicality of SSVEP-BCI.In terms of signal modeling,this paper constructed a set of mathematical models for the background noise of electroencephalogram(EEG)signals,and proposed a set of equalization optimization methods for its spatio-temporal correlation features.The method could effectively suppress the spatio-temporal correlation of the background noise,thereby improving the detection efficiency of the evoked potentials.Experimental studies showed that the correlation detection and the performance of the xDAWN and Bayesian linear discriminant analysis(BLDA)algorithm significantly improved in the P300 dataset after applying the spatio-temporal equalization preprocessing method.In the SSVEP data set,the average ITR of the canonical correlation analysis(CCA)algorithm was increased from 76.0 bits/min to 87.3 bits/min by applying the spatio-temporal equalization preprocessing method,and the average ITR of the multivariate synchronization index(MSI)algorithm was increased from 76.7 bits/min to 91.3 bits/min.In terms of reducing training costs and expanding the applicable population,this thesis proposed a training-free spatio-temporal equalization dynamic window(STE-DW)synchronization recognition algorithm for SSVEP-BCI.It could adaptively control the stimulation time while maintaining the correct rate of recognition,thereby significantly increasing ITR and enhancing the system's adaptability to different subjects.The offline analysis of a Benchmark SSVEP dataset and an offline dataset containing 16 subjects showed that the STE-DW algorithm was superior to FBCCA,CVARS,CCA and CCA-RV algorithms in terms of accuracy and ITR.At the same time,online experiments also showed that the STE-DW algorithm could effectively expand the applicable population of the SSVEP-BCI system.In terms of asynchronous system,this study designed and implemented an asynchronous SSVEP-BCI system using spatio-temporal equalization,which was based on the statistical test-rejection decision criterion and could adaptively decide whether to issue control commands according to the state of the subject's EEG signal.In an online experiment with 40 targets,14 subjects achieved an average recognition accuracy of 97.2% and an ITR of 106.3 bits/min.In the 240 s resting state test,8 of the 14 subjects achieved 0 false alarm.In the first trial in patients with Amyotrophic Lateral Sclerosis(ALS),268 character inputs achieved an average accuracy of 91.4%,and the equivalent ITR was 47.8 bits/min.The system has the characteristics of training-free,high recognition accuracy,fast recognition speed and large applicable population,and has been able to initially meet the needs of communication between ALS patients and the outside world.