The Analysis And Recognition Of EEG Features During Action Observation

Action observation as a cognitive activity of human brain,the study of EEG signals during action observation is helpful to explore the working mechanism of human brain.In addition,the feature extraction and recognition of EEG signals during action observation has great application value in military investigation and target tracking,and also provides a new idea for the design of brain computer interface system.However,during the process of action observation,the brain does not take the initiative thinking tasks involved,it can not determine whether it is in the effective action observation state through the EEG signals directly,and compared with the motor imagery and the steady-state visual evoked potential,the amplitude of EEG signals during action observation is weaker and more difficult to obtain.The goal of this thesis is to realize and recognize the feature of EEG signals in the process of observing the car turn left or right.Firstly,the SMI eye tracker and Neuroscan EEG equipment are used to collect signals synchronously,the experimental paradigm is designed to observe the car turn left or right.The eye movement trajectory signal analysis is used to determine the effective motion observation task.The active brain regions and the energy spectrum distribution of different frequency bands are analyzed to determine the characteristic frequency bands from the perspective of time-frequency analysis.Because of the interaction between the neurons in the cognitive activity,there is a causal network analysis method that can describe the flow of information between different brain regions.Through the analysis of the network measure of the causal network during the action observation,we find the difference of the network measure,and classify the network measure with obvious difference.Finally,the CSP and the SVM algorithm are used to recognize the features of the EEG signals during action observation.The main work of this thesis is as follows:(1)According to the characteristics of the highly non-stationary and signal to noise ratio(SNR)of EEG during action observation,studying the frequency band ofEEG during action observation as the starting point,and the EEG signals of the effective task is preprocessed to improve the EEG signals to noise ratio;Then,the brain topographic maps of different frequency bands of EEG are analyzed,and the brain regions are activated to determine the critical channel;Finally,the energy spectrum distribution of the critical channel EEG in different frequency bands is analyzed by WPT and power spectrum fusion method,and the characteristic frequency band is determined.The results show that the characteristic frequency band is 0.49-0.98 Hz.(2)Based on the information flow in different brain regions,the causal network measurement difference analysis method is used to study the characteristics of action observation signal.The GC,DTF,PDC three kinds of analysis methods are used to construct the causal network of EEG signals in different frequency bands.By analyzing the network density and global efficiency of causal networks under different thresholds,we select the appropriate threshold,and analyze the difference of the network measures(including degree,clustering coefficient and global efficiency).The results show that the clustering coefficient of GC has significant difference in0-4Hz.(3)According to the feature recognition problem of EEG signals during action observation,the CSP algorithm is used to filter EEG signals,based on the feature of the filtered signal energy,and the SVM is used to recognize feature,compared with the classification and recognition rate of EEG signals in different frequency bands,the highest classification rate is 86.15% on 0-4 Hz.Finally,through the optimization of the channel,we can achieve a higher classification accuracy in the case of fewer channels,and the feature recognition based on the clustering coefficient of GC is realized.