Analysis And Recognition Of Event-related Potential (ERP) Evoked By Covert Attention And Auditory Cross-model Stimulation

Covert attention, a new-style visual stimulation model used in brain computerinterface, has been regarded as possessing high research value and widely discusseddue to its availability for those patients with Amyotrophic Lateral Sclerosis (ALS).Event Related Potential (ERP) is a kind of EEG signals which is closely relevant withhuman feeling and cognitive procedure. P300, the main endogenous components ofERP, has been used in a wide range of BCI systems. However, the recognition andevoking rate of ERP signal based on single-model stimulation have been found to berestricted with each other, which further affected the feature extraction andinformation transfer rate of covert attention BCI signals.On the basis of sufficient preliminary researches on the development of covertattention, auditory stimulation, ERP, P300and their features, some changes have beendone on the traditional covert attention paradigm Hex-o-Spell to solve the problem,which adds color, a kind of feature attention. By adding relevant pure audio stimulifile, which differs in pitch and orientation, to covert attention paradigm, a newcross-model stimulation pattern combined covert attention with auditory stimulationhas been designed successfully. Auditory stimulation experiment, covert attentionexperiment and cross-model stimulation experiment, three tasks in total wascompleted by using Neuroscan4.5digital EEG acquisition system (64channels).EEG signal processing and analyses of ERP signals and P300components featurehave been done on recorded data. The results show that cross-model stimulation ownsshorter latency of P300components compared with single-model stimulation, whichindicates that cross-model stimulation is more appropriate for high speedbrain-computer information exchange system. This study went further into the Fisheranalysis of three experiments and pattern recognition of down sampling EEG data often subjects by using Support Vector Machine and Stepwise Linear DiscriminantAnalysis. Above80%correct recognition rate by folding ten times of cross-modelstimulation have been reached by using two pattern recognition methods of tensubjects which out-perform the traditional single channel stimulation. At last, thisstudy tried to compare the filtering result of all three experiments by using xDAWNalgorithm. Above-mentioned results of this study shows that cross-model stimulationbased on covert attention and auditory stimulation could be widely used in the braincomputer interface with both high recognition rate and high speed information transfer rate.