Research About Pattern Classification Of Recognition Tasks Based On EEG Signals

Cognitive science is a sophisticated study subject which studies human feelings,perceptions, mental states and mechanisms of thought and information processing.Research in this field is quite significant to reveal the mystery of human brain.Pattern classification of cognitive tasks is wildly used to build brain-computerinteraction (BCI) systems, study the working mechanism of human brain and thepathogenesis of various brain diseases. In order to study the cognitive mechanism ofhuman brain, two kinds of cognitive tasks–motor imagery classification task anddriving fatigue detection and classification task–were chosen to be researchedbased on EEG signals in this paper.Pattern classification of motor imagery tasks is one important way to build BCIsystems. The key to build BCI systems is extracting features from EEG signalswhich are collected when the subjects are performing different motor imagery tasks,then classifying the features and converting the results into control commands ofexternal devices. For motor imagery tasks, this paper focused on the research aboutfeature extraction algorithms. On one hand, the traditional feature algorithms werestudied, and in order to improve the filter component selection method of thetraditional common spatial pattern (CSP) algorithm, a new method based oncorrelation coefficient was proposed. On the other hand, based on the traditionalconcept of micro-status, a generalized definition and a new feature extractionalgorithm based on the generalized definition were proposed in this paper. Both theimproved CSP algorithm and the new feature extraction algorithm were verifiedusing the international BCI competition datasets and self-collected datasets.Driving is a complex task involving many different cognitive functions, such asvision, auditory, thought, judgment and so on. How to distinguish the alertnessbefore a long driving and the drowsiness after that is another point which this paperfocused on. In order to figure out this question, first, a simulated driving experimentwas designed to collect EEG data of a long driving, then brain effective networksbased on granger causality were built to comparatively study the EEG patternchange before and after the driver’s fatigue. This research has found the mostaffected brain region by the fatigue, and found some properties of brain effectivenetwork could be used as indicators to distinguish alertness and drowsiness. To some extend, methods used here have overcome the shortcoming of most current fatiguedetection algorithms, most of which can’t measure the transmission of informationbetween brain regions. The results could be used as a guideline for the electrodesdisposal and indicator selection in practical fatigue detection system.The mentioned two kinds of cognitive tasks were deeply researched in thispaper. For motor imagery classification, the proposed advice to improve traditionalcommon spatial pattern algorithm is brief and effective. Features, extracted bygeneralized micro-status based algorithm, contain spatial information about the EEGpattern, and could be classified easily. For driver fatigue detection and classification,this paper did the research from a brain effective network view. Brain network couldmeasure both the global and local properties; effective connectivity calculated byusing granger causality could reflect the information flow between different brainregions to some extend. This method is quite novel. Both two kinds of tasks have ahigh scientific value and an important practical significance.