Research On Event Related Potential EEG Signal Analysis Algorithm And Its Application

Brain-Computer Interface(BCI)technology plays an important role in human-computer hybrid intelligence.BCI collects,recognizes,and converts electrical signals from the human brain,thereby establishing a direct information communication channel between the human brain and external devices.Among them,event-related potential(ERP)reflects the direct response of the brain to a specific events,and has extensive research and application in the field of neuroscience.In this study,the Rapid Serial Visual Presentation(RSVP)paradigm for evoking P300 signals and the asymmetric Visual Evoked Potentials(aVEPs)paradigm for evoking visual potentials were taken as the starting point.Deep learning models are used to identify Electroencephalogram(EEG)signals,and two algorithm models were proposed to decode the EEG signals of these two paradigms,which expanded the application of deep learning models in EEG analysis.(1)The RSVP paradigm has been widely used in image recognition based on EEG signals.In order to improve the performance of EEG signal recognition of target images,this thesis proposed an improved EEGNet network model,which added x DAWN filter to the EEGNet model to enhance the signal-to-noise ratio of EEG signal and achieve dimensionality reduction and increase the computational speed.We have validated it on the benckmark offline dataset of Tsinghua University,and the experimental results showed that the recall rate of target images in the binary classification problem with unbalanced samples could reach76.07% ± 11.07% in group A and 78.11% ± 11.87% in group B.We applied the proposed model at the BCI Robot Contest in the World Robot Conference Contest 2021.In the triple classification problem with unbalanced samples,the online recognition recall of the target images could reach 51.56%,which won the second place in this competition.The model achieved good results in both offline and online recognition.And it provides a new and effective method for image recognition based on the RSVP paradigm.(2)The stimulation paradigm of character spelling systems based on aVEPs is a new paradigm in recent years.There is also relatively little character recognition research on this paradigm.Aiming at this paradigm,we proposed a deep convolutional neural network(Deep Conv Net)model based on convolutional block attention module(CBAM)to complete the recognition of characters.To evaluate the performance of the model,we examined the character recognition accuracy and information transfer rate(ITR).The experimental results showed that the model achieved the highest recognition accuracy and ITR when the time length was 300 ms.The character recognition accuracy was 65.57%±5.09% in dataset A and77.63%±3.41% in dataset B.The highest ITR was 38.23±6.00 bits/min in dataset A and66.74±7.98 bits/min in dataset B.This study provided a new solution idea for character recognition based on aVEPs.In summary: This thesis carried out a more systematic study on event-related potential brain-computer interfaces involving P300 EEG signals and aVEPs EEG signals.This thesis had two main contributions: first,an improved EEGNet model was proposed for image recognition in the RSVP paradigm,and second,a Deep Conv Net model based on CBAM was proposed for character recognition in the aVEPs paradigm.Both offline and online experiments had verified the effectiveness of the above algorithms.This research in this thesis provided new methods and ideas for the study of brain-computer interfaces for event related potentials.