Research On The Key Problems Of Coding And Decoding For The Event-related Potential And Its Applications In The Brain-computer Interface

Brain-computer interface(BCI) is a system directly transfering the activity of central neural system(CVS) to the output, which could replace, repair, augment, supplement and assist the normal pathway of CVS, thereby improving the interaction between the CVS and the outer environement. The event-related potential(ERP)-based BCI is one of the most important paradigms. The ERP is an EEG variation caused by a trigger event, which reflects the neural information processing of the inner brain. Therefore, it has been widely used in the fields of neuroscience, cognitive science, psychophysiology and BCI.However, the current ERP-based BCI is confronted with several tremendous challenges. First, the poor understanding of the neural mechanism of ERP prevents us from obtaining a reliable ERP model. Second, the decoding algorithm for the ERP is not good enough. Third, the conventional ERP-based BCI paradigm has a big problem in supporting high speed. Based on the challenges mentioned above, this study conducts overall and deep researches on the problem of ERP encoding and decoding for the BCI application. It includes the researches on the generation model of ERP, the channel selection algorithm for the P300 speller, the ERP recognition algorithm using inter-subject information, the ERP-based hybrid BCI system, the ERP-based gaze-independent BCI, and the divisibility of ERP features caused by the brain predictive code.First, for a further understanding of the ERP generation, this study proposes a novel and effective method to research the ERP evolution, by which it shows a three-period evolution for the visual ERP, the phase resetting mechanism is necessary for the ERP and the outer stimulus is a factor for the ERP generaion model.Second, for better decoding the ERPs, this study proposes two novel algorithms to select and recognize the ERP features. One is a new channel selection algorithm, phase locking and concentrating value-based recursive feature elimination(PLCV-RFE), for the P300 speller, which could get less channels with no loss of accuracy than two other tranditional algorithms. The other one is an effective classification algorithm, weighted ensemble learning generic information(WELGI), which combines both inter- and intra- subject information for ERP recognition and shows a significant improvement.Last, for a more efficient encoding stragegy, this study proposes a new pure hybrid BCI paradigm, the P300+SSVEP-B speller, based on which a visual parallel BCI system has been developed with a high information transfer rate. Moreover, it first finds the interaction feature between the ERP and SSVEP is sensitive to the covert attention level, which could be used for the the gaze-independent BCI. It also first demonstrates that the EEG feature caused by the brain predictive code is as effective as the ERP feature in the P300 speller, which shows a great potential in BCI application.In summary, this study conducts a number of innovative researches on the ERP-based BCI, and obtains many significant findings on the ERP mechanism and important improvements on the BCI system.