Research On Encoding And Decoding Very Weak Visuospatial Information For Brain-computer Interfaces

Visual brain-computer interface(v-BCI)has attracted plenty of attention of many researchers for its superior information transfer rate and classification accuracy.However,there are several severe problems in traditional interactive mode of v-BCI,such as visual fatigue and excessive visual resources consumption.According to the retino-cortical mapping,the spatial pattern of VEPs is corresponding to the position of visual stimuli in the visual field,and the activated area of visual cortex is positively correlated with the size of visual stimuli.Therefore,designing v-BCI based on retinocortical mapping is an effective way to release visual resources and improve user's comfort.Firstly,the experiment was designed to explore the difference and separability of the VEPs,which were induced by weak visual stimuli in different eccentricities(1°,2°,5°)and different polar angles(0°,45°,90°,180°,-135°,-90°,-45°).The results showed that the VEPs under the same eccentricity but different polar angles have different temporal-spatial characteristics.The VEPs under the same polar angle but different eccentricities shared similar temporal-spatial characteristics,but their amptitude differed a lot.It is preliminarily demonstrated that there is separability among VEPs induced by very weak visual stimuli in different locations.Secondly,a natural and friendly multi-instruction v-BCI paradigm using a set of weak visual stimuli was further designed.Using a set of weak visual stimuli that flickers in a fixed area as a marker,this paradigm can achieve calibration of 16 types of gaze points,which are under 2 kinds of eccentricity and 8 kinds of polar angles.With multidimensional analysis,it was found that the differential characteristics of VEPs evoked by weak visual stimuli in this paradigm are mainly performed in the temporal-spatial dimension,such as the amplitude variation and the increasement phenomenon of Intertrial Coherence.Finally,two algorithms named Multi-class discrimative spatial pattern(Multi-DSP)and Multi-DSP after common component rejection(Multi-DSPCCR)were proposed to decode very weak visuospatial information.The average accuracy of binary classification between VEPs under the same polar angle but different eccentricities was up to 82.69%,which was higher than DSP 2.22%.Multi-DSP and Multi-DSPCCR were used to decode VEPs corresponding to 8 kinds of polar angles at 2°(4°)eccentricity,and average accuracy of the 8-class classifications were 81.47%(74.17%)and 82.37%(74.75%),respectively.The average accuracies of 16-class classification of VEPs under two eccentricities and eight polar angles,were 65.58% and 64.63% for Multi-DSP and Multi-DSPCCR.In summary,this paper verified the feasibility of designing v-BCI based on very weak visuospatial information,designed a multi-instruction paradigm,and proposed two argorithms for decoding very weak visuospatial information.This study lay the foundation for researches aiming to improve user interaction experience and innovate v-BCI paradigm.