Research On Recognition Method Of Motor Imagery Eeg And Construction Of BCI System For Brain-controlled Vehicle

As a new expandable human-computer interaction method,brain-computer interface(BCI)can directly convert human brain nerve activity into the control instructions of peripheral devices by constructing a pathway from human brain to related equipment.The device control based on BCI system does not require the participation of nerves and muscles,which breaks through the shackles of limbs and has important research significance in the fields of disability assistance,military,entertainment and so on.Electroencephalogram(EEG)signal processing is the core step to realize BCI system.Improving the recognition rate of EEG signal can achieve higher accuracy of device operation.In this paper,the recognition problems for different types of motor imagery EEG are investigated,and the brain-controlled driving is realized by using related EEG recognition algorithms.The main research contents are as follows:(1)A hybrid feature extraction method based on time and frequency is applied to solve the difficulty of feature extraction for few-channel binary motor imagery EEG.First,wavelet packet filter is used to improve the signal-to-noise ratio(SNR)of original EEG signal.Then,the EEG feature in frequency domain is extracted with welch power spectrum.Combining with the energy feature in time domain,a time-frequency hybrid feature vector is constructed.Finally,the classification of feature vector is accomplished by using gaussian mixture model initialized with K-means.The results show that the extracted feature vector contain more information of motor imagery,and the recognition rate of motor imagery EEG can be effectively improved by using this hybrid feature extraction method.(2)To improve the feature extraction effect of regularized common spatial pattern(RCSP)for multi-channel binary motor imagery EEG,a time-frequency-feature selection method based on Fisher discriminant rate and information gain is proposed.RCSP is used to extract EEG features under different time-frequency combinations.The optimal time-frequency combination is selected based on Fisher discriminant rate.The corresponding feature with less motor imagery information is eliminated according to the information gain,and the final feature vector can be obtained.Random forest is applied as classifier to complete the recognition of feature vector.The results show that the time-frequency combination selected by Fisher discriminant rate matches the prior knowledge of motor imagery EEG in time-frequency domain.Using information gain to optimize the feature vector structure can further improve the classification accuracy of motor imagery EEG.(3)To solve the problem of low SNR and low recognition rate for multi-class motor imagery EEG,a method for selecting filtering frequency of each channel based on band power and Fisher distance is proposed.Band power of each channel in different frequencies is calculated,and the Fisher distance of each frequency is obtained according to band power.The frequency corresponding to larger Fisher distance is used as the final filtering frequency.One-versus-one CSP(OVO-CSP)is applied to extract EEG feature vector,and the classification of EEG is completed based on support vector machine(SVM).The results demonstrate that the frequency with larger Fisher distance has better performance to suppress the noise in corresponding channel.The average classification accuracy of the filtered four-class EEG is 86.48%,which is better than that of the traditional filtering method and BCI competition method.(4)To reduce the number of channels for multi-class motor imagery EEG,a channel selection method based on common spatial pattern(CSP)and sequential floating forward selection-sequential floating backward selection(SFFS-SFBS)algorithm is proposed.According to the CSP filtering coefficient,the search space of EEG channel is reduced.Then the SFFS-SFBS algorithm can be used to select the channel in the corresponding small space.It is found that,without reducing the classification accuracy,the number of channels selected by this method is reduced by 51.36% and 47.52%,respectively,and the corresponding search time is shortened by 90.95% and 80%,compared with the traditional SFFS algorithm and the improved SFFS algorithm.Therefore,the channel sequence with high quality can be selected in short time by the proposed method,which leads to a significant improvement of the braincomputer interface applied in practice.(5)To verify the reliability of the related algorithms and the feasibility of brain-controlled driving,a BCI system of brain-controlled vehicle is built.Depending on this system,the lane line detection based on machine vision is completed,and the assistance of brain-controlled driving is realized.By designing an online EEG recognition algorithm,combining with a reasonable vehicle control strategy,an accurate brain-controlled driving is achieved.