Vigilance Differentiation from EEG Complexity Attributes

摘要

Vigilance is an ability to maintain concentrated attention on a particular event or target stimulus. Monitoring tasks require certainly high vigilance to properly detect rare occurrence or accurately respond to stimulation. Changes in vigilance can be reflected by EEG signal, so vigilance levels can be classified based on features extracted from EEG. Up to now, power spectral density was commonly employed as features to differentiate between vigilance levels in majority of previous studies. To the best of our knowledge, multifractal attributes for vigilance differentiation have not been exploited, and their feasibility still need to be investigated. In this study, we first extracted multifractal attributes based on wavelet leaders, and then selected statistically significant distinct attributes for the following classification (two vigilance levels). According to the results, classification accuracy was improved with increase of time window used for feature extraction. When time window was increased to 50 s, an averaged accuracy of 91.67% was achieved, and accuracies for all subjects were higher than 85%. Our results suggest that multifractal attributes are promising for vigilance differentiation.