EEG Signal Analysis And Movement Feature Extraction In Specific Virtual Scenes

Brain Computer Interface(BCI)refers to the technology that establishes a channel from human brain to external auxiliary device to complete human beings' desired tasks by means of this device.In recent years,BCI has been widely studied and applied to medical rehabilitation and many other fields.The main goal of BCI is to assist the patient in their recovering trainings and even to replace some severely damaged limbs.However,the development of BCI does not meet its actual needs,and there're lots of problems to be tackled,such as low signal-to-noise ratio caused by loud background noise,difficulty in signal acquisition,and poor classification results induced by differences between participants.Thus,BCI is mostly used in scientific studies,and it still has a long way to go for being applied in real life.Based on previous studies of electroencephalogram(EEG),the acquisition experiment of motor imagery electroencephalogram(MI-EEG)is designed for the patients with right lower extremity disability in an attempt to achieve the identification of different tasks through signal analysis and processing,thereby outputting relevant control instructions to external device.Major working contents are as follows:(1)Based on previous studies of BCI system,Unity and Kinect technologies are adopted to design and build a virtual-reality-based experimental platform,which consists of training and acquisition parts.The production mechanism and features of MI-EEG are clarified,which is further classified based on amplitude values.(2)The MI-EEG signals acquired are analyzed.Firstly,these EEG signals are pre-processed,and their noises are eliminated through bandpass filtering,Independent Component Analysis(ICA)and other methods to obtain clean signals.Secondly,three-channel data are selected,and the time-frequency characteristics of EEG are analyzed to determine the existence of Event-Related Desynchronization(ERD)in the selected three-channel data,thus proving the analyzability of the selected data.In the meanwhile,due to biological differences in the participants,the best time-frequency bands of different patients are selected as the research data for late-stage studies.Finally,Common Spatial Pattern(CSP)is utilized to extract the characteristics of EEG and analyze the divisibility of characteristic values.(3)To solve the problem of poor classification results of EEG,Particle Swarm Optimization and Support Vector Machine(PSO-SVM)and improved Convolutional Neural Network(CNN)are adopted for the classification of EEG.Firstly,Support Vector Machine(SVM)is utilized to identify the characteristics extracted by CSP,and Particle Swarm Optimization(PSO)is adopted to seek for optimal parameter of SVM.Secondly,ten-layer CNN model is designed,and Batch Normalization(BN)is introduced to eliminate the gradient disappearance problem occurring during the calculation process.The results show that BN-CNN model experiences enhanced convergence speed and classification results.Also,the influence of Dropout parameter on BN-CNN is dissected.Finally,the analyses of experimental results from these two algorithms proposed show that the improved BN-CNN has better identification performance.