Novel feature extraction of EEG signal for accurate event detection

Abstract

Electroencephalograph (EEG) signal analysis is one of the essential for diagnosis of diseases, detect the various event and psychological problems. The accuracy of event detection depends on the feature vectors used. The feature vectors in literature provide performance only for a specific application and perform poor for other application. This paper presents a new feature vector generation using fusion of energy feature vectors of different types. The energy features of alpha, beta and gamma component is extracted as base feature. A fusion rule is formulated to fuse that base features using Hjorth activity features to improve the accuracy of classification. The performance of the proposed new feature is tested on seizures detection, sleep state detection and emotion detection application. The resulting analysis shows that the proposed new feature outperform with improved accuracy of 26% for emotion detection, 5% for seizures detection and 4% for sleep state detection.