Key Technology Researches In Online BCI Based On SSVEP Signal

The BCI(Brain Computer Interface) system can convert brain awareness into commands of controlling external devices, which covers neuroscience, computer technology, control science, and so on. Among them, as one kind of EEG(Electroencephalogram) signal, the SSVEP(Steady State Visual Evoked Potentials) signal can generate rhythmic signals with distinctive features through external visual stimulus, and has certain degree of stability and persistence. Due to its intensities of power spectrum around the stimulus frequencies, methods of power spectrum can be used to extract features, which has a wide range of applications in BCI systems.The BCI system has a high requirement for the accuracy, real-time and stability of SSVEP signal analysis. One of the key contents is how to analyze the features of online SSVEP signal accurately, and design real-time, stable online BCI system.This thesis focuses on the four classification conscious recognition, aiming at improving the accuracy, real-time and stability of SSVEP signal analysis. It mainly studies the algorithm of artifacts removal, feature extraction and classification, and enhances the real-time and stability of online system by combining underlying data streaming technology. The main results are as follows:(1) SSVEP visual stimulus paradigms are designed and realized. Appropriate stimulus frequencies are chosen to generate relatively stable SSVEP signals, the theoretical knowledge about off-line recognition and online process of SSVEP signal is introduced, and so is the overall design concept of the BCI system.(2) Recognition algorithms of SSVEP signal based on canonical correlation analysis and principal components analysis are designed and realized. Against the obvious EOG(Electrooculogram) artifact, the second order blind identification algorithm is compared with the independent components analysis, and then the second order blind identification algorithm based on canonical correlation analysis is designed to improve the real-time of SSVEP signal preprocessing. About the real-time feature extraction and classification, the canonical correlation analysis algorithm is compared with power spectral density analysis according to processing speed, then against the problems of channels selection, channels number selection, and sample size, the principal components analysis is used to improve the canonical correlation analysis by dimensionality reduction of original SSVEP signal, and enhanced the real-time of canonical correlation analysis without influencing the recognition accuracy.(3) An online BCI system based on SSVEP signal is designed and realized. Against the real-time processing of SSVEP data stream, the multi-thread mechanism is adopted to decompose the whole process into multiple subtasks executing concurrently, and the buffer strategy is used to solve the speed matching problems among threads. Against the flow rate change of SSVEP data streams, adaptive one-sided fuzzy inference is adopted to predict the change of data streams. Against the disorder of data recombination due to thread concurrency, a method of mutual exclusion and synchronization with semaphore is designed to recombine the intermediate data orderly.The experiment results show that the canonical correlation analysis based on principal components analysis is widly suitable to multiple subjects, and has relatively high recognition accuracy and rate. The method of thread concurrency and fuzzy inference solves the problem of real-time and stability in online BCI system, decreases the average delay time of a single trial, and improves the information transmission rates.