EEG Signal Recognition Based On Brain Network And Parallel Convolutional Neural Network

With the rapid development of brain science,brain-computer interface(BCI)has been widely applied in various fields as a frontier technology of multidisciplinary integration.The electroencephalogram(EEG)signal is considered as the main input signal in the brain-computer interface system by virtue of its easy acquisition.In order to improve the practicality of braincomputer interface,it is necessary to optimize the modules of information processing,feature extraction and classifier in BCI system.In this paper,the construction of a brain network is implemented for analyzing the interdependencies between different brain regions in the consciousness task.Based on a full analysis of the characteristic differences of brain networks in different rhythms,a method combining a split-frequency multi-feature brain network with a parallel convolutional neural network is proposed in this paper to classify and identify EEG signals.The classification accuracy is also significantly improved.We first introduce the dataset used in this paper and analyze the correlation between EEG signals under different rhythms.Through the traditional method combining filter bank common spatial pattern(FBCSP)and support vector machine(SVM)to extract features and classify EEG signals,the average classification accuracy of 10 subjects is 71.93%.Based on the above experiments,in order to obtain more abundant EEG signal characteristics,a split-frequency multi-feature brain network under different consciousness tasks by calculating the correlation coefficient and phase locking value between each channel is constructed in this paper.Compared with the traditional single vector extraction of EEG signal characteristics,this algorithm maps the statistical dependencies between brain regions during the imagination task to the network framework.Thus,the EEG signals can be analyzed more clearly from the network.At the same time,in order to avoid the possible overfitting problem caused by the small data set of EEG signals,a data augmentation algorithm on the time domain is used in this paper to expand the data of EEG signals according to the characteristics of brain network construction.The average accuracy was increased from 70.87% to 77.82% by using the data enhanced frequency division brain network as the input of the convolutional neural network.Secondly,in order to avoid the inability of single-size convolutional neural network structure to comprehensively learn the features of multi-band brain network.In this paper,Parallel Convolutional Neural Network based on Multi-Band Brain Networks(MBBN-PCNN)is proposed to better learn the features in the topology of multi-feature brain network.The average accuracy of single-size input increased from 77.82% to 85.76%.It is further verified that the model can significantly enhance the performance of the classifier in decoding EEG signals based brain networks.Finally,the improved Shallow Conv Net and Deep Conv Net parallel multi-input structures are compared with the MBBN-PCNN model.The experimental results were also subjected to paired samples t-test,which proved that the classification accuracy of the model proposed in this paper was significantly improved.It is also verified that the split-frequency multi-feature brain network has a better effect in EEG signal recognition,which lays the foundation for the practical application of brain-computer interface system.