Research On Error-related Potentials Classification And Its Application In Shared Control

Motor imagery brain-computer interface(BCI)is an innovative human-computer interaction technology that enables people to control external devices through electroencephalogram(EEG)signals.Error-related potentials(Err Ps)are spontaneously generated EEG signals related to wrong behavior detection in the middle area of the brain frontal lobe.Combining Err Ps signals with motor imagery can correct the results after detecting Err Ps to avoid executing wrong instructions and ultimately improve the accuracy of BCI.The current research on Err Ps is mostly based on offline conditions with low detection accuracy and pre-set error rates,which cannot accurately reflect the subject's motor imagery BCI performance and is difficult to apply to online systems.Aiming to assist subjects to complete daily tasks,the feature extraction and classification algorithms of error-related potentials and the shared control of online BCI were studied in depth.Robot auxiliary motion control system is realized based on the BCI of motor imagery and error-related potentials.The main work is as follows:(1)Feature extraction of Err Ps based on Common Spatial Pattern(CSP)and Adaptive Autoregressive Models(AAR).CSP was used to extract the spatial characteristics of all channels.In view of the lack of time domain features of CSP,channels with high correlation with Err Ps was selected,and the AAR model parameters were estimated by the Kalman filter algorithm.The model parameters are selected as the time domain features,and a feature extraction algorithm combining time domain and space domain is proposed.The results show that the proposed method effectively extracts the time domain and space domain features of the EEG signal.Compared with other feature extraction algorithms,the classification accuracy is improved.(2)Research on Spectral Regression Discriminant Analysis(SRDA)classification algorithm based on samples with class-imbalance.Taking into account the stable performance of the subject's motor imagery model and the class-imbalance of the Err Ps,the evaluation index of the classifier is adjusted by introducing a compensation function.The false positive rate is optimized and the best classification threshold is selected based on the confusion matrix and sample distribution.The results show that compared with support vector machines and random forests,the proposed algorithm effectively reduces the false positive rate of the classifier,and further improves the accuracy of classification.(3)Research on shared control based on online BCI.Designed and constructed a brain-controlled robot system.The subject controls the robot's movement through motor imagery,and corrects the instruction by Err Ps online detection.Aiming at the shortcomings of the low accuracy and the few instruction dimensions of BCI,a shared control method based on Bayes filter is implemented.Combining robot sensor measurement,the output of motor imagery and error-related potential to recursively estimate the state of movement,and dynamically allocate brain-computer control ratios.The motion state of robot can also be adjusted automatically during the command interval.Shared control helps patients complete daily tasks,improves the fluency of patient's brain control and accuracy of task completion.