Research On Steady-State Visual Evoked Potential-Based Brain-Computer Interface From Ear-EEG

The applications of brain-computer interface(BCI)systems have been developing rapidly in recent years.The steady-state visual evoked potential(SSVEP)-based BCI systems have received a lot of attention due to the high classification accuracy and information transfer rate in a short training time.Since the main response area of SSVEP is located in the visual cortex,the SSVEP-based BCI systems usually place the electrodes on the hair-covered areas above the occipital brain region to acquire electroencephalogram(EEG).This type of signal acquisition requires a cumbersome preparation process,which is not conducive to the practical application of BCI systems.Some studies have implemented the BCI system based on SSVEP measured from non-hair-bearing areas to improve system usability.However,the response amplitudes of SSVEP acquired from non-hair-bearing areas are small and the classification accuracies of the BCI systems are low.In this thesis,the personalized stimulation method and classification algorithms of SSVEP measured from ear-electroencephalogram(ear-EEG)were investigated to address the above problems,and an asynchronous BCI system was implemented by combining the ear-EEG-based SSVEP and electrooculogram(EOG)signals.The main results were presented as follows:(1)A strategy based on personalized stimulation frequencies to induce ear-EEG-based SSVEP was proposed.The stimulation frequencies were selected for subjects based on the response amplitudes and signal-to-noise ratios of ear-EEG-based SSVEP under stimulations with different frequencies in the range of 6 to 40 Hz.Compared with generic stimulation frequencies,using personalized stimulation frequencies to evoke ear-EEG-based SSVEP facilitated BCI system to extract response characteristics,and the average classification accuracy of ear-EEG-based SSVEP by the task-related component analysis algorithm was significantly higher,reaching 83.85% at 3 s time window.(2)A classification algorithm based on canonical correlation analysis using multiple reference signals and multiple trials of signals was proposed.The spatial filters were estimated by combining EEG signals from multiple trials with the sine-cosine based reference and the SSVEP training data based reference,respectively.Then the filters were used for feature extraction of the test signals.In intra-session validation,the proposed method significantly improved the classification accuracy of ear-EEG-based SSVEP to90.89% at 3 s time window.In addition,EEGNet with ensemble learning was implemented to enhance the classification performance by combining features extracted from multiple base models.The ensemble model significantly improved the classification accuracy of earEEG-based SSVEP processed in an asynchronous manner to 89.0% at 3 s time window in cross-session validation.(3)An asynchronous BCI system based on EOG and ear-EEG-based SSVEP was proposed.The BCI system was implemented to control the robotic arm to accomplish the task of grasping and placing the target object.To address the problem that subjects cannot autonomously switch between control and idle states in synchronous BCI systems,the proposed system used the EOG signals of three consecutive blinks as asynchronous switch in combination with the classification results of ear-EEG-based SSVEP as control commands,and identified the EOG signals of a wink as cancellation command.The system had a false activation rate of 0.01 /min for asynchronous switch,a true positive rate of 93.5%for cancellation command,and an accuracy of 86.38% for ear-EEG-based SSVEP.