Research On Feature Extraction And Classification Of Asynchronous Brain-Computer Interface Based On Motor Imagery

Brain-computer interface technology based on motor imagery can recognize the movement intention of the human brain and convert it into specific control commands,so as to realize the direct control of human brain to external devices.Because motor imagery is a spontaneous brain computer interface with higher autonomy,it has a broad application prospect.At present,there are many researches on the application of brain-computer interface system for motor imagery,especially in the field of stroke rehabilitation.However,most current systems still use synchronous mode.In the synchronous mode,users can only follow the rhythm of the system and control the device through motor imagery in a specific time window,which greatly destroys the autonomy of the motor imagery paradigm.The more natural way should be the asynchronous mode.In the asynchronous mode,the system will continuously detect whether the user wants to control the device.Only when it is detected that the user is in the control state,will the control command be further issued according to the type of control state.The asynchronous mode has great advantages,but it also brings more challenges,so there are still many issues to be resolved.This article focuses on the feature extraction and classification methods of asynchronous motor imagery.The specific work is as follows:(1)In order to solve the problem of feature instability caused by the complexity of non-control state,this paper optimizes the features by feature selection and multi feature fusion to improve the accuracy of state recognition.Firstly,the feature selection algorithm based on r~2coefficient is used to select the best feature for each subject.This feature can distinguish non control state from control state and distinguish one control states from the other one.Then,the energy feature that reflect amplitude information and PLV feature that reflect phase information are fused by feature splicing and ensemble learning,and several basic classifiers are combined into ensemble classifiers by ensemble learning method.In the task of classifying non control states,imagining left-handed movements,imagining right-handed movements,this method achieved 76.54%average classification accuracy,which is better than the previous research.(2)In order to further enhance the stability and accuracy of asynchronous motor imagery state recognition,this paper has carried out exploration research on multi-brain brain-computer interface technology.the experimental paradigm of asynchronous motor imagery for two person based on Hyperscanning is designed.The cooperative multi-brain brain-computer interface is used to achieve more stable and accurate state recognition through group decision-making and signal superposition enhancement.In this paper,according to the characteristics of energy and PLV,we design a variety of two person fusion classification schemes.The results show that the energy features based on decision level fusion and three PLV features based on feature level fusion can achieve better classification results,among which the average accuracy of the three classification based on feature level fusion is 83.78%,which is better than the results of single person mode.(3)In order to add brain state information to conventional features to improve the stability of the features.The difference between the brain workload index of control state and the non control state was studied,and it is added to the energy and PLV characteristics of the motor area for classification,and better classification result is obtained.