Spiking Neural Network For Motor Imagery Brain-computer Interface Signal Classification

Brain-Computer Interface(BCI)technology can help people control machines through brain electrical signals,engage in a variety of jobs,with broad application prospects and huge market potential.In recent years,with the rapid improvement of computer software and hardware,the brain-computer interface technology has also developed rapidly.However,the brain-machine interface technology is still in the development stage,and there are problems such as low signal-to-noise ratio and differences between subjects,resulting in low classification accuracy.Based on this issue,this thesis is mainly covered as follows:(1)Because of the different mental load between different subjects,the quality of Electroencephalograph(EEG)data will be affected.From this point of view,the subjects were evaluated for EEG load,and the EEG data with low EEG load were screened out.In order to induce the high brain load state of the subjects,the paradigm task of Unmanned Aerial Vehicle is designed.The behavioral and event-related potentials of mental load state were analyzed by response time task.The quantitative analysis of the subjects can be effectively evaluated on the degree of mental load.(2)Since the original data contains a lot of artifact noise in detail,this thesis designs a series of noise removal methods for artifact processing.The effective band data is first extracted through the Butterworth Bandpass filter and partially removed by the wavelet transform smoothing data.Then,the Independent Component Analysis(ICA)electrical artifact removal algorithm based on Quality Threshold(QT)adaptation.The algorithm first performs blind source separation of data by ICA,then determines the eye power source,and finally removes the eye artifacts by QT adaptive method.(3)This thesis presents a SW-Neu Cube spiking neural network model.In terms of coding model selection,it encodes and decodes the common temporal spiking coding model.For the coding model by evaluating parameters,the most suitable coding model is selected.In terms of weight update,the spatial distance is considered and the Space-STDP time weight update model is proposed;secondly,the SF coding model and the Weight-de SNN frequency weight update model.Using the open motion imagination dataset,the simulation verifies the SW-Neu Cube model,and through the comparative analysis of the separation and accuracy results,we find that the SW-Neu Cube model can certainly improve a classification accuracy relative to the Neu Cube.In addition,the accuracy of traditional neural network model.Overall,through the evaluation of different mental load is solved between the subject data to some extent.EEG data with higher signal-to-noise ratios can be obtained using artifact removal.The SW-Neu Cube pulsed neural network model partly improves the accuracy of the neural network for the classification of EEG data.