A Study On High Speed P300-Based Brain-Computer Interface Speller

Brain-computer interface(BCI)can directly collect brain signals and convert them into corresponding commands to control external devices without depending on the organization of the muscles,so as to complete the process of human-computer interaction.It provides a way to communicate with the outside world for people with motor disabilities.In this thesis,we study the BCI system based on scalp electroencephalography(EEG).Among them,P300-based BCI system has been developed for many years,but there are still a lot of problems in practical application.This thesis mainly focuses on the problem of low information transfer rate in the current P300 BCI,and we work on the following two aspects:1)An interactive P300 speller paradigm is proposed in this thesis.In the traditional P300 BCI speller system,the stimulus sequence is presented to subject for several rounds to improve the signal-to-noise ratio and achieve reliable P300 detection.The number of rounds is usually fixed,and all characters appear in the stimulus sequence,which results in low information transfer rate of the system.In order to solve this problem,an algorithm for shortening the stimulus sequence based on real-time EEG data is first proposed in this paper.According to the EEG data obtained online,the algorithm calculates the posterior probability of the corresponding flashing character to be the target,and deletes the character of lower posterior probability from the stimulus sequence.Hence,the length of the stimulus sequence is dynamically shortened,and the time for the subject to complete the spelling task is reduced.Then,this paper combines the stimulus sequence shortening algorithm with the dynamic stopping criterion(DSC),and proposes a new algorithm,which is named sequence shortening dynamic stopping criterion(SS-DSC)algorithm.The algorithm uses the above posterior probability to determine whether P300 exists in the EEG signal corresponding to the flashing character while shortening the stimulus sequence.If a reliable P300 is detected,the character flashing is immediately stopped,and the corresponding flashing character is directly selected as the target character and output,so as to realize the dynamic output of the system,and further improve the information transfer rate of the system.2)Combining Bayesian linear discriminant analysis(BLDA)with the proposed SS-DSC algorithm,a high speed P300-based BCI speller online experiment system is developed.The proposed SS-DSC algorithm is verified by online experiments,and compared with other stopping criterion algorithms.In order to solve the problem that the target character may be mistaken removed from the stimulus sequence in the SS-DSC algorithm,a method of detecting mistaken deletion is proposed.In addition,we combine convolution neural network(CNN)with SS-DSC algorithm to realize P300 high-speed character input.The results of offline analysis show that the BCI system based on CNN has higher spelling accuracy and information transfer rate.The experimental results show that,compared with the BCI system based on traditional static stopping criterion(SSC)or dynamic stopping criterion(DSC)algorithms,the BCI system based on SS-DSC algorithm can significantly improve the information transfer rate of the system while the spelling accuracy is basically not declined.