Pattern Recognition Of P300 EEG Signal Based On ICA

Diseases such as Amyotrophic Lateral Sclerosis(ALS)will cause patients to gradually lose the ability to control muscle movement,eventually leading to functional and cognitive impairment.Brain-Computer Interface(BCI)technology can build an artificial information transmission channel between the brain and the surrounding environment,and is a new type of interaction method that can replace traditional neuromuscular pathways.The main purpose of developing brain-computer interface technology is to help people with motor dysfunction regain the ability to interact with the outside world.The P300 character spelling system based on electroencephalogram(EEG)studied in this paper is one of the most popular application directions in the field of brain-computer interface research.In this paper,the electrode channel for recording EEG data is firstly selected adaptively for different test conditions.Secondly,the data under the electrode channel dimension is projected to the independent component level and correlation analysis is carried out.An integrated Support Vector Machine(SVM)classifier based on the Bagging strategy is constructed for the P300 EEG signal pattern recognition task.Finally,the P300 sensitive components in the signal were enhanced,and then the P300 EEG signal was extracted and classified and recognized by spatio-temporal features using Convolutional Neural Network(CNN).The specific research content is as follows:(1)An adaptive channel selection algorithm based on Principal Component Analysis(PCA)is proposed.Due to the large redundancy in the information recorded by adjacent electrode channels,the computational burden in the subsequent feature extraction and model construction process is aggravated.This article first uses the character matrix blinking sequence recorded during the experiment to superimpose and average the EEG samples containing the P300 potential response to restore the P300 EEG waveform generated by a specific user.Secondly,the PCA algorithm is used to rotate the electrode channel dimension,the purpose is to keep the P300 waveform information obtained after superposition and average in a certain dimension to the greatest extent.The EEG sequence in this dimension is called the P300 principal component.After the data is subjected to the PCA operation,the feature vector corresponding to the principal component can be obtained,and its value in each dimension represents the information content of the principal component in the original channel dimension.This article uses this as the basis for selecting adaptive sensitive channels to sort the channels.(2)At the independent component level,the feature extraction method of P300 EEG signal is discussed.Because the signal-to-noise ratio of the EEG signal is too low,the P300 potential components are easily submerged in other background signals.This article first uses the ICA blind source separation method to decompose the collected mixed information into independent source components,so as to increase the proportion of P300 potential in some independent components.Secondly,perform the same demixing operation on the EEG sample containing the P300 potential response,and superimpose and average each independent component to construct the standard independent component of the P300 potential response mode in this demixing mode.Then,perform the unmixing operation on all sample data,and perform correlation analysis on the independent components of each sample and the independent components of each standard.The correlation analysis methods used in this article include Pearson correlation analysis in the time domain and frequency domain.Coherence analysis(Coherence)and time-frequency cross mutual information analysis in the time-frequency domain(Time-frequence cross mutual information).(3)A SVM ensemble classifier based on Bagging strategy is constructed.Due to the uneven number of positive and negative samples in the P300 data set.This article first uses random sampling and superimposed averaging to expand the minority sample class.In addition,in order to further improve the recognition performance of the classifier for binary classification tasks,this paper introduces the Bagging ensemble strategy to construct 20 training subsets with replacement sampling to train the homogeneous weak classifier SVM,the final sample type The judgment is decided by voting.(4)A P300 EEG signal enhancement algorithm based on ICA is proposed.Inspired by the ICA removal process of electrooculogram artifacts,this article first superimposes and averages the independent components after the sample is unmixed,and then performs the correlation between all the independent components and the above-mentioned P300 principal components in the time domain,frequency domain,and time-frequency domain.Analysis,and use the sum of correlation coefficients as the independent component elimination index.Finally,use the remaining independent components to restore the data to its original size.Experimental results show that the signal enhancement algorithm is effective in improving the accuracy of model recognition.(5)A new type of convolutional neural network is designed for pattern recognition of P300 EEG signals.In view of the excellent performance of the convolutional neural network on the EEG pattern recognition task,an iterative scheme is used to design a convolutional neural network with a 10-layer structure.The convolution kernel of the network is used to analyze the spatio-temporal information in the EEG signal.Perform extraction.In order to alleviate the over-fitting problem,a dropout strategy is added to the network,and batch-normalization is used to accelerate its training process.Experimental results show that the single-sample recognition accuracy of the model proposed in this paper is up to 98.3%,and the character recognition accuracy rate is up to 97%.