Research On Event-related Potentials (ERP) Evoked By Visual P300-Speller Based On Tridimensional Encoding Oddball Paradigm

P300-Speller is a classic human-machine interaction paradigm to realize the character select and input by using the rare event-related potential (ERP) P300 signal feature. It could build a bridge for the subject to communicate with outside environment through EEG and has also been proved to be one of the most efficient and important information input and exchange approaches of brain-computer interface (BCI). The conventional 6*6 P300-Speller, by flashing its rows and columns sequentially, which is described with limitied characters for choosing, low information transfer rate, inconvenient transmission with huge amount of instructions , couldn't meet the applicable demand. Therefore, the key technology needed to be solved is to introduce the new coding stimulate paradigm to improve the traditional P300-Speller paradigm, expand its amount of characters, and increase its information transfer rate as well as keeping the high accuracy.This thesis first proposed the flashing character stimulate paradigm based on tridimensional coding, analyzed the P300 signal feature evoked by the new paradigm in details. and discussed its feasibility obtaining high classification accuracy as well as information transfer rate. On this basis, two inproved P300-Speller was designed based on tridimensional coding stimulation paradigms with 64 and 125 characters, which were compared in details with conventional 6*6 rows-columns paradigm in experimental data preprocessing, feature extraction and pattern recognition to evaluate their classification accuracies and information transfer rates.In the study, the experimental data were preprocessed by filtering and down sampling initially. Then, the EEG feature discriminability was analysed by applying Fisher ratio and r2 ratio method. Coherent averaging method was used for EEG feature extraction. Linear discriminant analysis (LDA) and support vector machine (SVM) were applied for EEG pattern recognition with above 99% of the average classification accuracy for all subjects. Finally, by comparing the information transfer rate of traditional rows-columns paradigm and propoded tridimensional coding paradigm, the results showed that an obviously higher ITR was obtained in tridimensional coding paradigm (53.59 bit/min) than rows-columns paradigm (35.94 bit/min). Chanel optimization was also completed by using adboost support vector machine-based recursive feature elimination (ABSVM-RFE). The results showed that after the downsize of 60 unimportant channels from 64 collection channels, the classification accuracy could maintain above 80% and the remaining important channels were concentrated in top zone of the head which is in accordance with the P300 optimum location indicated in neurophysiology. Above research outputs may provide a key technology basis to develop new applicable P300-Speller BCI with large amount of characters and instructions, high ITR and classification accuracy.