Research Of Feature Extraction And Classification Methods Based On Active Imagination EEG Signals

Motor imagery EEG decoding aims to recognize human movement intention,which is an important research topic in the field of brain computer interface(BCI).It has been applied to many fields such as mechanical prosthesis control,neurological rehabilitation therapy,mind games,and has broad application prospects.In the motor imagery classification task,the refined motor imagery is difficult to produce discriminative changes,and the EEG is highly noisy and unstable.Therefore,the public datasets are less involved in the refined motor imagery task,and the existing motor imagery EEG feature extraction and classification methods have poor classification effects for the refined task.Aiming at above challenges,this thesis constructs a two-handed multi-task motor imagery dataset,and proposes three motor imagery EEG feature extraction and classification methods.The main research contents of this thesis are as follows:1.This thesis constructs a multi-task EEG dataset suitable for refined motor imagery classification task.This dataset is collected from 10 subjects for 6 categories of motor imagery brainwaves,including left-and-right-handed motor imagery tasks.Compared with the existing mainstream public motor imagery datasets,this dataset contains more refined motor imagery tasks,and the task duration and task order are more reasonable.2.This thesis focuses on the motor imagery classification algorithm based on deep fusion of multi-scale spatio-temporal features.In order to solve the problem of insufficient time-frequency feature extraction and fusion in motor imagery classification task,this thesis adopts multi-scale temporal convolution to fully extract the features of multiple frequency bands.According to the characteristics of different splicing methods of features,deep detachable convolution and temporal convolution network are used to deeply fuse the features,to fully mining their internal information and improving the accuracy of task classification.3.This thesis focuses on motor imagery classification algorithm based on 3D spatiotemporal convolution and multi-domain high-level feature fusion.Aiming at the problem of the lack of real spatial information and the loss of spatio-temporal correlation information of features in each domain in motor imagery tasks,this thesis stacks EEG according to the spatial relationship of electrodes to supplement spatial information,designs3 D spatio-temporal convolution based on spatio-temporal feature extraction and retains the real distribution of time-frequency spatial features.In this thesis,the capsule network is used to preserve the temporal and spatial association information of features while conducting feature fusion,and the GRU network is used to select the distribution of various features,preserving the integrity of the features corresponding to tasks,so as to improve the performance of the model.4.This thesis focuses on motor imagery classification algorithm based on wavelet transform and self-attention mechanism.Aiming at the problem of low classification performance of existing methods in multi-channel EEG time-frequency maps,this thesis explores and improves an effective time-frequency map convolution method.The selfattention mechanism is used to further fuse the extracted features,and the self-attention EEG features and time-frequency map fusion features are introduced to supplement each other’s information,so as to alleviate the impact of information loss in the conversion of time-frequency maps and improve the classification effect of multi-channel time-frequency maps.