Research On Feature Extraction And Classification Algorithm Of EEG Based On Motor Imagery

The brain-computer interface combines many fields such as neuroscience,artificial intelligence and information technology,providing a new way for human-computer interaction.The BCI technology based on the motor imagery can make people directly communicate with the outside world through conscious control without relying on muscles and nerves.Since it was proposed,the research field of the BCI system has been widely concerned and favored by various researchers.Therefore,this thesis mainly studies feature extraction and classification of the EEG signals based on the motor imagery,in which,the key is to extract effective features and optimize the classifier performance to improve classification accuracy.In this thesis,the CSP algorithm and the wavelet analysis method are used to extract spatial,temporal and frequency features and fusion features of the motor imagery EEG signals.Then,the extracted eigenvectors are classified and analyzed by applying different traditional algorithms and the SRC algorithms based on the elastic networks.The main contents of this thesis are as follows:(1)Classification of the motor imagery EEG based on CSP and SRC.The CSP filtering method which can effectively extract the different task components is used to extract two classes of motor imagery ECoG signal feature,and then apply the SRC algorithm for classification.In order to improve the classification accuracy of the ECoG signals,this study introduces the SRC algorithm with elastic network constraint,and we apply APG algorithm and LARS algorithm to sparse coding for ECoG signals.This method can not only avoid over-fitting to a certain extent,but also improve the generalization ability of the model.Finally,the algorithm is compared with traditional algorithms such as SVM,KNN and so on.(2)Classification of the motor imagery EEG based on the time-frequency-space fusion features.Some traditional methods usually extract the frequency or spatial features without considering the related information between different channels that would affect the classification performance.This thesis proposes a new method for feature extraction of EEG signals based on the fusion of time-frequency and spatial features.At the beginning,the CSP algorithm is adopted to extract the special features.Then the DWT and the WPD are used to extract the μ rhythm of the motor imagery EEG signals as the time-frequency features.After that,by combining the spatial and time-frequency features,the timefrequency-spatial feature is formed.SVM and SRC are used to classify fusion features,simple time-frequency features and spatial features.Two points can be proved by analyzing the experimental results:1)Compared with other traditional sparse representation classification algorithm with elastic network,it has better and more stable classification performance when the feature dimension changes;2)It is proved by experiments that the fusion feature based on time-frequency=space domain overcomes the shortcomings of the traditional EEG feature extraction method,fully combines the time-frequency characteristics and spatial characteristics of EEG signals and saves operation time in simpler and more efficient classification algorithm and improves classification and recognition efficiency.