Research On Brain-Computer Interface Technology For Group Brain Based On Event-related Potentials

With the deep cross-fertilization of artificial intelligence in the fields of neuroscience and human-computer interaction,brain-computer interface(BCI),as a special human-computer interaction channel,is a new way to generate direct contact with the external world directly through the brain without any physical interaction.Event-related potentials(ERPs)are the changes in potentials caused by specific activities such as cognition and movement.The most common EEG signal in ERPs is the P300 signal,which is one of the hot spots in the current research in the field of brain-computer interface.In this paper,we mainly focus on P300-BCI,which consists of two steps of classification,the first step is to detect the presence of P300 potentials in EEG signals,and the second step is to identify the target characters.To solve the typical problems in existing P300 classification,such as low accuracy,complexity and time consuming,this paper proposes a collaborative brain-computer interface(c BCI)system based on P300 to achieve fast and high accuracy classification of P300 signals.The P300 collaborative brain-computer interface is also called group brain-computer interface,which means that the EEG signals of multiple subjects are collected and processed simultaneously in the same brain-computer interface system to establish communication and control channels with external devices.It can effectively solve the problem that the P300 signal cannot be effectively evoked or the P300 signal characteristics are not prominent due to the unstable physical or mental state of the subjects during the "Oddball" experiment.In this paper,we combine deep learning,Support Vector Machine(SVM),Linear Discriminant Analysis(LDA)and Random Forest(RF)to construct a group brain algorithm model,which can realize a stable and efficient group brain joint system.In this paper,we investigate the P300 swarm brain-computer interface technology,and design two methods for centralized data fusion from multiple single-person data in the data preprocessing stage,and analyze the feasibility of such methods in terms of characteristics,as well as design algorithmic models for deep learning,SVM,LDA and RF for single-person as well as swarm brain,and construct the data acquisition part under the P300 swarm brain paradigm.In this paper,classification recognition of multi-trial P300 and single-trial P300 was performed using deep learning models.By adding Dropout layer,maximum pooling layer and Flatten layer in constructing CNN,it can effectively mitigate the gradient disappearance and overfitting problem and extract the main features and dimensionality reduction out of P300 to reduce the computational effort.The use of Batch Normalization for training data with small batches of data enables the network to generalize well,where the evaluation metrics are Accuracy,Precision,Recall and F1-socre,proving that it can be faster and more accurate for P300 signals for classification.The algorithmic models based on deep learning,SVM,LDA and RF of P300 designed in this paper have taken better results.Among them,the group brain brain-computer interface based on SVM+xdawn+filtl achieves an average classification accuracy of 97.8% for P300 classification detection and 95.5% for character recognition after 2-person parallel data fusion.