Research On Virtual And Real Control Of Time-Frequency Information Extraction And Recognition In Brain-Computer Interface

The Brain-Computer Interface(BCI)decodes the information expressed by human brain signals to control external devices to make corresponding actions.It is widely used in medical rehabilitation,auxiliary training and other fields.As China enters an aging society,stroke and other brain injury diseases have led to more and more elderly people being unable to carry out rehabilitation training independently.Due to their inconvenient limb movements,robots are needed to assist them in driving limb movements to achieve rehabilitation.Therefore,the brain-computer interface virtualreality combination technology has been designed.It uses virtual human stimulation to obtain EEG signals,and then extracts time-frequency information from EEG signals.After obtaining the recognition results,the robot arm is controlled.It lays a foundation for driving patients to exercise and realizing rehabilitation functions.The method of channel selection,feature extraction and feature fusion of EEG signals are mainly studied.The main contents are as follows.(1)To address the problems of large number of EEG signal channels,mutual correlation,mutual interference,large amount of computation and high complexity,the method for selecting channels in EEG signals based on the Gradient Boosting Decision Tree algorithm are proposed.The Gradient Boosting Decision Tree algorithm is used to calculate the importance of different channel features,select important channels for analysis,reduce the complexity of calculation,improve the efficiency of subsequent feature extraction,and verify the effectiveness of the algorithm.(2)To address the problem of subtle differences in local brain signals that can lead to different results in different tasks,more precise features need to be extracted.The Generalized Predictive Control model is used to build model for each frame of EEG signal,and the least square method is used to calculate the parameters as the time domain features of each frame of EEG signal.Then the SE-TCNTA model is used to further extract local features to achieve the purpose of extracting precise features.The Generalized Prediction Control model can not only simplify complex EEG signals,but also extract the features of EEG signals in time and frequency domains,improving the effect of identifying subtle differences.(3)In order to further extract the global feature information of EEG signals and explore the correlation between different features,the sum of power spectral density of the EEG signal and control quantity is taken as the frequency domain feature from a global perspective.The Graph Convolution Neural network is used to further extract the frequency domain features.Pearson correlation coefficient is used to calculate the relationship between channels and establish the graph structure.The global and associated frequency domain features are extracted by aggregating node information.Finally,the time domain features and frequency domain features are fused through full connection to obtain the classification results.(4)The designed virtual human is used to stimulate the brain of the people to obtain EEG signals and driving the robotic arm to perform corresponding actions.The design of the virtual and real combined web system is realized.The virtual human and virtual environment are embedded into the web page using the back-end technology.The virtual human randomly generates actions to obtain the EEG signals from the people.The method proposed is used to identify the acquired signals.The robot performs corresponding actions,achieving the control of the robot through the brain-computer interface virtual-real combination.