The Key Technique Research Of Lateral Recognition On Imaginary Upper Extremity Movement In Brain-Computer Interface

Brain Computer Interface (BCI) gives their users communication and control channels that do not depend on the brain's normal output channels of peripheral nerves and muscles. Simply to say, the purpose of BCI research is to control computers or other electronic devices only on user's mind. The critical step in BCI is how to using signal processing, feature extraction and pattern recognition techniques to extract mind intents and transform them to output operation commands. Currently, in study of EEG-based BCI, lots of researchers focus on the Event Related Desynchronization/ Synchronization (ERD/ERS) signals evoked by imaginary movement of upper extremity. The occurance and distribution of ERD/ERS signals can provide lateral information of movement, which corresponding to the self-dependent control intent, and consequently, the extraction of this lateral information plays a key role of transforming mind to device operation command.However, electroencephalogram signals are nonstationary in long term measurement and extremely sensitive to noise. For these reasons, the extraction of ERD/ERS always suffers a low efficiency, poor stability and high dependence of user training. Given this condition, how to recognize the lateral information from electroencephalogram signals evoked by imaginary upper extremity movement is one of the critical problems in current BCI research.An evoked electroencephalogram experiment on different lateral imaginary movement of upper extremity was designed and operated in this study. In time frequency analysis, Short Time Fourier Transformation and Fisher Separability Analysis were used to identify the special characters of power spectral density distribution. Then the key patterns for lateral information recognition, such as specific time range and frequency band, were extracted. To enhance the evoked signal, Independent Component Analysis (ICA) was introduced to reject artifact components of EOG and EMG. And Kmeans clustering method coupled with ICA was used to extract components corresponding to evoked response. For further characterizeing the dynamic features of evoked electroencephalogram signals in imaginary movement task, four parameters include Power Spectral Entropy, Wavelet Entropy, Kolmogrov Complexity and Approximate Entropy were measured, then statistical tests were applied to these parameters. In the lateral label recognition step, a SVM based recursive eliminate algorithm was chose to optimize the features combination and ensemble machine learning method was employed to boosting SVM classifier.Result of this study demonstrated that the correct rate and stability in recognition of lateral information were increased by using these techniques described above. More importantly, even for subjects with no training experience, our method also provides a high recognition correct ratio. Generally to say, this study provides a good ground for the development and research of online upper extremity imaginary movement BCI system.