Research On EEG Feature Extraction And Recognition Based On Visual Attention

It's very difficult for human beings to sustain attention for a long term.However,attentional lapses can lead to a series of negative results,reducing working efficiency and even causing serious accidents.If the attentional levels can be detected online,we can properly control and reduce the incidents caused by attentional lapses.However,it still remains a challenge to automatically predict the brain attentional states that can be used to alert the attentional lapses in advance.This study focuses on the attentional level detection based on the EEG method.In this study,we designed two kinds of experimental conditions,including no background visual stimulation and steady-state background visual stimulation.The experiment is designed based on go/no-go paradigm.Subjects need to press the corresponding button or not according to the relative position of the color dot in the white background round.30 healthy subjects were recruited in the EEG experiment,and their corresponding keypress were recorded.The statistic result shows that there are significant differences in their response time between high and low attentional levels.In the experiment with no background visual stimulation,we analyzed the eventrelated spectral perturbation(ERSP)based on short-time Fourier transform,finding that the energy of 15 Hz in high attentional level is significantly smaller than that in low level at the 700 ms after no-go stimulus.We also extracted frequency characteristics of EEG signals based on Hilbert transform,and classified these features using SWLDA algorithm.The average accuracy could be up to 85%,which means this kind of algorithm has good classification performance for attentional levels.Besides,the classification accuracy increased along with the growing of signal frequency,conforming to the rules of ? > ? > ? > ?.In the experiment with steady-state background visual stimulation,the P3 component amplitude was significantly higher for the high attentional level than the low level.LDA algorithm,CCA algorithm and improved CCA algorithm were respectively used to classify different attentional states.Results showed that the improved CCA algorithm was better than the other two algorithms,and its classification accuracy of single person could be up to 76%.Furthermore,fundamental frequency and the 80 Hz,90Hz of the SSVEP signal were sensitive to the attentional state and had higher accuracies than other frequencies.In summary,this study explored the relationship between the attentional state and EEG signal characteristics.The methods proposed in this study are promising to monitor the attentional level.