基于EEG的警觉度客观检测与通道筛选技术研究

Research on Objective Vigilance Detection and Channel Selection Techniques Based on EEG

在航天领域,警觉度的降低会影响航天员的工效,还可能引发重大安全事故。针对航天领域的警觉度客观检测技术进行研究,设计了多轮次PVT和3⁃back任务结合的警觉度建模实验,并全程采集EEG信号和行为学数据,该实验范式成功诱导了37名志愿者警觉度下降,并伴随着大脑快波成分减少,慢波成分增加,复杂度降低等现象,在额叶、颞叶和枕叶区域变化明显。使用方差分析(ANOVA)结合支持向量机(SVM)对单个特征建模的方法筛选出了6个警觉度敏感特征,在此基础上使用SVM⁃RFECV方法筛选出了12个警觉度敏感导联,最后使用SVM进行警觉度客观检测模型的构建。结果显示:特征筛选前的平均分类正确率为83.10%,在导联筛选后为87.16%,提升了约4%。特征筛选和导联筛选工作有效地提高了模型的综合性能,同时简化了脑电采集流程,减少了后续数据处理的工作量和时间成本。

During spaceflight,the diminished vigilance can impair the efficiency of astronauts and it is also a potential risk factor for major safety accidents.In this paper,the objective vigilance detec⁃tion technologies tailored for application in aerospace contexts were explored.An alertness modeling experiment combining multiple rounds of the Psychomotor Vigilance Task(PVT)and 3⁃back tasks was designed,and the EEG signals and behavioral data were continuously collected.Data analysis indicated that the experimental paradigm effectively induced a reduction in alertness among 37 sub⁃jects,marked by a diminishment in fast wave components and an augmentation in slow wave compo⁃nents of the brain;there was a decrease in complexity,and pronounced alterations were observed in the frontal,temporal,and occipital lobes.Using Analysis of Variance(ANOVA)combined with a Support Vector Machine(SVM)for single⁃feature modeling,6 alertness⁃sensitive features were screened out.Subsequently,the SVM⁃RFECV technique was utilized to select 12 alertness⁃sensitive leads,and an objective alertness detection model was constructed using SVM.The results showed that the average classification accuracy was 83.10%before feature selection and increased to 87.16%after lead selection,with an improvement of about 4%.It indicates that feature selection and lead selection can effectively improve the overall performance of the model,simplify the EEG acquisition process,and reduce the workload and time cost of subsequent data processing.