Motion Related Feature Detection And Law Research Based On Brain Electrical Signals

Brain electrical signal is generated by the brain in life activity,which contains abundant information of life activity.Based on brain electrical signals,the brain-computer interface(BCI)is an effective way to analyze the information of brain activity and realize the exchange of information between the brain and external devices.For the motion-oriented brain-computer interface,the motor function reconstruction and recovery can be realized by analyzing and decoding the electrical signals related to motion and combining with external devices such as mechanical prosthesis,which is of significant value to the recovery and treatment of patients with motor dysfunction.In the research of motion-oriented brain-computer interface,the analysis and decoding of motion-related brain electrical signals is crucial.In the paper,the electroencephalogram(EEG)in motor imagery is decoded.Additionally,we analyzed the changes of phase-locking distribution of spike and local field potential(LFP)in continuous movement.The following two aspects are included in the major research:Firstly,a motor imagery EEG signal classification algorithm is proposed based on multibranch 3D dilated convolutional neural network.On the basis of the 3D convolution neural network,the input 2D motor imagery EEG signals were transformed into 3D signals according to the spatial distribution of the sampling electrode and the time dimension information,while retaining both features in time domain and spatial domain.In addition,a dilated convolution is introduced into the 3D convolutional neural network to construct the multi-branch 3D dilated convolutional neural network based on 3D dilated convolution kernel.It is found that not only the temporal and spatial features of EEG signals can be extracted simultaneously,but the receptive field can also be enlarged without enlarging the size of the convolutional kernel by utilizing the multi-branch 3D dilated convolutional neural network,thus ensuring the adequacy of feature extraction.The experimental results showed that the average accuracy and standard deviation in different subjects were 76.98% and 9.51,indicating the excellent classification performance and adaptability.Second,we analyzed the phase-locking relationship between the spike and the LFP delta rhythm under continuous motion.We analyzed the changes of the distribution of phase-locking in Primary motor cortex,Primary somatosensory cortex and Posterior parietal cortex regions in the single directional extension and grip movement paradigm of rhesus monkeys at different movement stages,at the time bins of 100 ms before and after three events and at the time scale of continuous time bins.The experimental results indicated that there were obvious differences of the phase distribution of phase-locking of spike and LFP and the peak position in the different stages.Results of experiments in the time bins of 100 ms before and after the three events showed that the change of the distribution and peak value were obvious differences.By analyzing the phase distribution of phase-locking with the continuous time bins,it is found that the change was most obvious before and after the reaching stage,and the difference value of the change was the largest in the waiting stage and the reaching stage.This is an exploration for further revealing the relationship between EEG signals and physiological electrical activities of neurons.The paper is based on the experimental paradigm related to motion.The electrical signals of brain at different structural levels are classified and analyzed in several domains including time domain,spatial domain and tissue cell level,which further promoting the understanding and application of brain electrical signals in motor cortex.Moreover,a new idea is presented for interpreting the encoding of electroencephalogram signals in the process of motion.