| Brain-computer Interface(BCI)is a means for the brain to interact with the external environment.It can directly control external devices by using EEG signals.At present,it has been applied in many fields such as military,medical,entertainment,etc.The BCI based on motor imagery is a kind of spontaneous BCI,which does not need to receive external stimulation and only requires subjects to generate EEG signals through motor imagery.In this thesis,an EEG signal recognition method based on channel attention mechanism is proposed and applied to brain-controlled robot system.The research focus of this thesis is carried out in the following ways.1.This thesis proposes that the Attention Mechanism(AM)can be used for feature extraction of motor imagery EEG signals.In this thesis,we improve the original AM and form a Channel Attention Mechanism(CAM)suitable for EEG signal processing.In addition,One Versus Rest Filter Bank Common Spatial Pattern(OVRFB-CSP)is introduced to extract the features of different frequency bands in EEG signals,and the multi-task motor imagery feature extraction is implemented.Finally,the Softmax classifier was used to classify the extracted EEG signals.2.The CAM-OVRFB-CSP-Softmax method proposed in this thesis is used to analyze the EEG signals of motor imagery.First,the fourth BCI competition 2A dataset was used to demonstrate the effectiveness of the EEG signal classification for four categories.Then,the algorithm is applied to the offline analysis of the self-collected EEG data,and a good effect is also obtained.In the feedback experiment,the model is constantly updated to further improve its effect,which is ready for the next online experiments.In this thesis,the EEG signal classification results are combined with the subjective fatigue degree of the subjects to analyze the practical application effect of the algorithm.3.The method proposed in this thesis is applied to the online control of mobile robot.The experimental results show that the method proposed in this thesis can effectively extract the motion intention of the subjects and complete the robot control task.Combined with the fatigue evaluation results,it can be seen that the subjects can control the robot freely and effectively in real time through motor imagery in more complex online experiments,which indicates that the EEG recognition algorithm proposed in this thesis has strong adaptive ability and can effectively extract the motion direction features in the EEG signals. |
