Studies On Rapid Series Visual Presentation System And Related Algorithms

Rapid serial visual presentation system is a visual evoked brain-computer interfacebased on the rapid serial visual presentation paradigm, which may be used for rapidpicture sequence selection. This thesis studies experiment design, signal processing andthe classification algorithm of RSVP system.We design an RSVP system using pictures of nature scene in its paradigm. Toachieve a better result, we discuss various factors that may influence the experimentoutcome, such as attention blink, valid visual field, sparsity of targets, eye blink,proficiency of subjects and task setting, and provide respective solutions. Specifically,we suggest some pictures should be re-checked due to the influence of attention blinkand eye blink. In addition, taking proficiency of subjects into account, we add a trainingsession with feedback prior to the formal experiment.As to the analysis of the recorded EEG signals, we discuss algorithms in both thefrequency and spatial domains. We proposed to determine the range of band-pass filterby the spectrum of averaged signal wave, which achieves a satisfactory result. We studyspatial filters from several aspects. Firstly, we analyze the performance of ICA methodin the underdetermined setting and then evaluate the feasibility of extracting ERPcomponents by ICA. Secondly, we process the recorded EEG signals using the SIMalgorithm and obtain a better result, which demonstrates that the SIM algorithm is moresuitable to extract ERP components than ICA. Thirdly, we proposed two methods tocheck the determinancy of EEG signals. Lastly, we discuss the limitation of spatialfiltering in EEG signal processing.This thesis also discusses classification algorithms. By summarizing the features ofthe ERP signal and looking into the influence factors of the experiment, we enrich theevent set and improve the feature extraction method, which enhances the classificationperformance. We also make comparisons of different classifier. Furthermore, accordingto the limitation of spatial filter in dealing with single-trial EEG signals and theclassification characteristics of ERP signals, we propose a new classification algorithmcombining the traditional algorithm with the local distinctiveness feature of ERP signals,which achieves high classification performance in Multi-person RSVP experiment.