| Brain computer interface system is a technology that uses external auxiliary equipment to complete the communication and control between human brain and computer or other electronic equipment.In the aspect of control instruction set extension brain computer interface system,researchers at home and abroad in the experimental paradigm and a lot of research on signal processing algorithm,but because the EEG signals are non-stationary and nonlinear random strong physiological signals,simple extension movement imagination experimental paradigm task type is easy to cause the participants thinking fatigue,reduces the EEG signal acquisition quality.In order to solve the problem that the control instruction set of brain computer interface system based on EEG signal is less,this paper designed the experimental paradigm of finger motion imagination and motion execution,studied the classification and optimization method of finger motion imagination and execution,and extended the instruction set of brain computer interface system.To solve the problem that EEG signals are susceptible to external and autogenous interference,and the source quality is not high,in the aspect of EEG signal preprocessing,the original EEG signal is removed by using the independent component analysis based on Infomax algorithm,and then the optimal smooth noise reduction method constructed by empirical mode decomposition is adopted to remove other noise artifacts.In terms of feature extraction,optimize the mode of the traditional total space feature extraction algorithm,by Hilbert Huang transform to extract the Hilbert marginal spectrum of time-frequency characteristics and spatial pattern to extract features in spatial domain,and then use the Fisher linear classifier to complete the EEG data classification,the method has obtained the ideal classification accuracy and robustness is higher than the other methods;Because the traditional support vector machine only supports dichotomy,after analyzing the advantages and disadvantages of OVO-SVM and OVR-SVM,a hierarchical support vector machine combining OVO-SVM and OVR-SVM was proposed.Moreover,for the characteristics of EEG signals,which are nonlinear and non-stationary,the gaussian kernel function with good interpolation ability and localextraction ability and the polynomial kernel function with good nonlinear processing ability and global extraction ability are selected and applied to the two layers of HSVM respectively to improve the separability of samples.Experimental results show that the accuracy of the proposed classification and optimization method is higher than other methods.Thus,the classification of finger motor execution and motor imagination expands the set of instructions available for brain-computer interface systems. |
