Research On Feature Extraction And Classification Of EEG Signal

Electroencephalograph (EEG) signal is a kind of bioelectricity producedspontaneously by the human brain and contains rich effective information. It isdetermined by nervous system; hence it can objectively and truly reflect human‘sphysiology states. Along with the well-known of Brain Computer Interface (BCI)system, research on EEG recognition signal has also been widely concerned. In thispaper, the recognition research of spontaneous EEG and induced EEG are bothstudied. EEG is a kind of non-stationary signals which is complex and randomness.Hence, it is very important that how to extract the effective information and how toclassify EEG signals more effectively. In this paper, we focus on the featureextraction and classification of spontaneous EEG based on motor imagery and EEGevoked by videos. The study work of this paper is as follows:The EEG recognition based on motor imagery is studied in this paper.Thecombination of PCA (Principle Component Analysis) and LDA (Liner DiscriminateAnalysis) was adopted as the extraction method. It solves the problem that PCAcan‘t take the class information into account which leads to the unsatisfactory ofrecognition accuracy. Then recognition method of voting basedextreme learning machine is firstly used to EEG classification. The dataset we usedin this section is BCI dataset Ia. Experiment results show that the feature extractionand the classification algorithm get the accuracy of93.52%，which is more than thebest previous accuracy of92.15%。Compared with SVM，V-ELM obtain betterperformance in both accuracy and time cost。The recognition of emotional EEG evoked by different kinds of movie clips isalso studied in this paper. Section4introduced design of experiment and process ofEEG signal collection; based on the knowledge of psychology that audiovisualaffects different brain area, we used4electrodes (FP1, FP2, P1, P2) which arerelated to audiovisual from64electrodes to analysis EEG. Due to the theory that theeffective information of emotional EEG signals is contained in frequency range ofand β, original EEG signals are also filtered in processing. Variance of waveletpackage is used to extract feature and use SVM to classify the features. Experimentalresults show that our method is effective to identify the emotion. Average accuraciesof two classes are between71.44%and82.94%, while average accuracy of three classes is66.07%.