EEG Signal Feature Analysis Based Working Memory In Sober And Drunk

Drinking has an effect on the brain,in severe cases,which could endanger the lives of persons.Alcohol affects the execution of different physiological functions of brain that working memory is particularly susceptible to the effects of alcohol.The effects of alcohol need to be explored which acts on the physiological function of the brain.There is a lack of research specifically addressing changes in working memory before and after drinking alcohol.Therefore,this study proposes to analyze the differences between working memory in sober and drunk states based on electroencephalogram(EEG)that means studied the effect effectually of alcohol on working memory by EEG features.Firstly,this thesis builds a small sample data set of drunken EEG.In order to better induce working memory and conveniently collect EEG signals,there are two different experimental paradigms for comparison between N-back and MS(Memory Span),and then N-back is chosen which uses digital 0-9 as the stimulus material.A total of 10 participants were recorded EEG data between three different tasks difficulties and two state of drunk.At the same time,the characters of Odd Ball paradigm are incorporated in N-back tasks for better induces of P300 waveform in this thesis.Secondly,the experimental behaviors and EEG data differences are analyzed in sober and drunken state.There are analyzed from three aspects: experimental reaction,P300 waveform,and the ERD/ERS phenomenon.The obvious differences are found both experimental reaction and P300 waveform when comparing data on working memory workload difficulty of different tasks in the awake state and on the state in sober and drunken states for tasks of the same difficulty.Based on the results by comparison,this thesis proposes the hypothesis that alcohol affects brain function,causing brain to increase the allocation of task resources,thereby affecting the participants’ behaviors and EEG signals.Finally,working memory workload levels as well as states between awake and drunk are classified by using support vector machine(SVM)and EEGNet.The SVM and EEGNet models perform well in classification tasks.In the SVM model classification of individual participants,the classification results show that accuracies of time-domain are better than the frequency-domain,and the classification accuracies of frontal lobe are better than parietal lobe.Meanwhile,EEGNet model achieves good classification results in both individual and cross-participants classifications.In this thesis,the trained classification models are used to classify task type of EEG data in the2-back task of drunk state,and the classification results showed that the classification results of task type deviate from actual 2-back task type for most of EEG data samples of drunk state.Therefore,through working memory and drunkenness experiments,this thesis shows from multiple angles that drunkenness leads to changes in EEG signals,and does not completely cause a single decrease or growth trend change in working memory load levels,but is due to a dynamic change process.