Brain-controlled Auxiliary System Based On Steady-state Visual Evoked Potential

Brain Computer Interface is a technology that can convert EEG signals expressing human intentions into control commands.Its initial and most important application is to assist patients with movement disorders or partial movement disorders to achieve information exchange in the external environment.In the selection of EEG signals,steady-state visual evoked potential EEG signal has been widely studied in the field of brain-computer interface due to their advantages of high classification accuracy and high signal-to-noise ratio.The brain-computer interface system based on steady-state visual evoked potential requires high real-time performance.Therefore,it is very important to improve the performance of accurate analysis based on user-dependent and user-independent EEG signal in a short time window,and to design and implement a brain-controlled auxiliary system on this basis.The main work of this thesis is as follows:(1)Aiming at the problem of poor classification performance caused by insufficient latent feature extraction in a short time window,this thesis proposes a convolutional neural network model that adaptively extracts time domain and frequency domain features.The model considers that EEG harmonics are the main basis for the classification of steady-state visual evoked potential,so the filter bank method is used in both time domain and frequency domain analysis.In the time domain,a multi-scale temporal attention module is designed according to the characteristics of steady-state visual evoked potential under different harmonics.The multi-scale temporal attention module adaptively fuses the time domain features extracted from each group of filters at a specific scale to eliminate noise interference.In the frequency domain,according to the characteristics of the frequency characteristics of the steady-state visual evoked potential under a short time window,an adaptive multi-band convolution module is designed,which adaptively learns the harmonic weight of each group of filters under the consistent network architecture,and obtains enhanced frequency domain information.Comparing the performance of the existing mainstream models under the 1s time window,it is found that the average classification accuracy and information transmission rate of the proposed method are increased by 16.01% and 30.34 bpm under the 10-subject dataset;the average classification accuracy and information transmission rate are improved by 12.73%and 22.01 bpm respectively under the Benchmark dataset,which proves that this model has good classification performance and robustness in a short time window of user-dependent.(2)Given the large individual variation of steady-state visual evoked potential among different subjects,this thesis proposes a model based on attention adaptive extraction of timespace-frequency features.User-dependent accurate analysis of steady-state visual evoked potential EEG signal can be fully extracted in the time-frequency dimension.However,different subjects still have large differences in the spatial characteristics of EEG signal for the same stimulus target or the same subject for the same stimulus target at different times.Therefore,the spatial characteristics should be considered in the user-independent research.In this model,the preprocessed time domain feature map,spatial feature map,and frequency domain filter bank feature map containing harmonic information are used as network input.The channel-based attention mechanism and the time/frequency domain-based attention mechanism are applied successively in the time domain and frequency domain feature extraction to make the feature maps in the time domain and frequency domain contain features that can reflect individual differences.In the spatial feature extraction,the individual differences between channels are focused to obtain enhanced spatial information.Finally,the feature maps extracted from the three dimensions are fused before classification.Experimental verification shows that the proposed method performs well in 10-subjects dataset.Therefore,an accurate analytical model that fully considers the time-space-frequency characteristics of steady-state visual evoked potential EEG signal is more conducive to improving the user-independent classification performance under a short time window.(3)Based on the high real-time performance required by the brain-computer interface system of steady-state visual evoked potential,the brain-control assistance system is designed and implemented in this thesis based on the user-dependent EEG accurate analytical model in a short time window and user-independent EEG accurate analytical model in a short time window proposed above.In the design and implementation of the system,after the subjects generate visual stimuli by staring at the visual stimulus module,the EEG acquisition module will send the collected EEG data into the signal processing module for identification and analysis.Then,based on the preset class-instruction mapping relationship in the control module,the corresponding instructions will be sent to the corresponding devices or applications,and the control results will be fed back to the users.Through online and offline experiments,it is proved that subjects can independently complete the grasping and placing tasks required by the brain control auxiliary system,which can meet the needs of users.