Research On Feature Extraction And Recognition Of EEG Markers Related To Visual Attention

Attention is one of the most important cognitive functions,which can optimize the allocation of cognitive resources to the most relevant sensory inputs.However,for most people,it is very difficult to maintain a high level attention state for a long time.The decline in attention would directly cut down the efficiency of both work and study,and even result in serious accidents.Therefore,it's of great significance to detect the attention state objectively.The behavioral and EEG data were recorded from 50 healthy subjects,when they participated in the go/no-go discrimination task based on the features of spatial location.The no-go stimuli,which had an emergency probability of 12%,were extracted for further analysis.We defined the no-go trials without button-press as the high-attentionlevel trials,whereas the trials with wrong button-press as the low-attention-level trials.For the pre-and post-no-go stimulus period,we analyzed the behavior and EEG performance under high and low visual attention levels,the EEG markers closely related to visual attention levels were selected,and effective classification models to detect visual attention levels were established.For the trials before the target no-go stimuli,the low level attention led to significantly shorter reaction time and higher variability of reaction time than the high level attention.The observations indicated that the low level attention reduced the reaction stopping capacity,which confirmed the validity of our experimental design.The neural markers,which can be used for the detection of visual attention level,were analyzed by the event-related potential(ERP)technique,time-frequency analysis,and brain network analysis based on partial orientation coherence.For the pre-stimulus period,the high level attention had significantly lower energy of the ?-band in occipital region but higher clustering coefficient of the brain network in the 0.5?40 Hz than the low level attention.For the post-stimulus period,the high level attention had significantly higher amplitude of N1 and P3 but lower energy in the 40?200 Hz of the forehead area than the low level attention.In addition,there had significantly differences in the energy distribution of the 4?30Hz in the prefrontal area between high and low visual attention levels.The detection models of visual attentional level were established based on the TRCA,DCPM and CSP algorithms,respectively.The results showed that the CSP algorithm was best for the pre-stimulus period,which can achieve an average accuracy up to 73.1%,while the DCPM algorithm was best for the post-stimulus period,which achieved an average accuracy up to 82.9%.The current study designed an effective experimental paradigm to induce two different visual attention states,found six EEG markers which closely related to visual attention level,and established effective classification models to detect visual attention level.This study can provide a new perspective to the early detection of attention state.