| Electroencephalogram(EEG)contains rich information related to people’s mental and health status.But eye blink artifact usually has a negative impact on EEG analysis because of its inevitability in EEG signal acquisition.In addition,the mainstream eye blink artifact detection methods often ignore the screening of epileptiform discharges,resulting in signal misjudgment,which seriously affects the diagnosis of epilepsy.Moreover,EEG signals have significant individual differences,but most of the current detection methods lack individual difference processing,and the multi-channel EEG representation ability is poor,which leads to a significant decline in the performance of these models on real data.For the problems described above,the following parts are the main research work of this dissertation:1.Aiming at the problem that the current mainstream eye blink artifact detection methods lack epileptiform discharge signal screening and misjudge it as eye blink artifact,a frontal epileptiform discharge signal screening method(MIS-EDS)based on monotonic increasing slope is proposed.This method is based on the physiological characteristics of epileptiform discharge signal and smooth nonlinear energy operator(SNEO)filtering method,the maximum value of monotonically increasing slope of prefrontal channel and the statistical characteristics beyond the threshold count are specially designed to screen epileptiform discharge signals,Finally,K-means unsupervised clustering is performed on the eight dimensional features of the prefrontal channel.On the CHZU dataset,the accuracy of this method is 96.51%,which solves the problem that the current mainstream methods lack epileptiform discharge signal detection and misjudge it as eye blink artifact.It is of great significance for EEG signal analysis and disease diagnosis in patients with epilepsy.2.An eye blink artifact detection method based on multi-channel and multi-dimensional feature optimization(MMF-BAD)is proposed to solve the problems of ignoring the relevant information between channels,weak signal representation ability and lack of individual differentiation processing of EEG in traditional methods.This method extracts 16 key features to distinguish eye blink artifact from normal EEG signals.These features cover the time-domain,Frequency domain and spatial distribution characteristics of multi-channel EEG signals,and combined with support vector machine(SVM)algorithm to achieve high-accuracy eye blink artifact detection.In addition,through the comparison and analysis with Relief algorithm,this dissertation proposes to use variance filtering as a differentiated feature selection algorithm to solve the problem of individual differences in EEG,and can effectively remove the redundant features in the feature extraction step,and finally improve the accuracy of eye blink artifact detection model.MIS-EDS and MMF-BAD can be combined to form eye blink artifact detection method under the background of epileptiform discharge(EDS-BAD).Compared with ICA-SVM,WICA-SVM,SVM-AE and GMM,the average accuracy of proposed method is improved by 28.48%,23.19%,11.94% and 7.23% respectively.3.Aiming at the disadvantage that MIS-EDS is not suitable for epileptiform discharge signal with high-frequency fluctuation and the failure of eye blink artifact detection in VMEDWT method,an eye blink artifact detection method(IVME-BAD)based on improved variational mode extraction(VME)is proposed.VME can smooth the EEG and reduce the local high-frequency fluctuation.The combination of VME and monotonic increasing sequence feature extraction method can further optimize the features and improve the screening accuracy in the screening of epileptiform discharge signals.In addition,the threshold based detection algorithm proposed by VME-DWT method will be seriously invalid when the EEG has an overall offset and multiple eye blink artifacts exist continuously.The proposed improved method optimizes the threshold calculation method by using the difference signal,and takes it as the one-dimensional feature for eye blink artifact detection based on SVM.In terms of accuracy,the proposed method is greatly improved compared with VME-DWT method,and the average accuracy is also improved by 1.71% compared with EDS-BAD method.Finally,from the perspective of engineering practice and clinical application,a set of EEG eye blink artifact detection system based on multi-channel feature optimization is developed by using the GUI environment of MATLAB.Combined with EEG multi-dimensional feature representation,different feature selection algorithms and classifier models,the system realizes the automatic detection of eye blink artifacts,which is very helpful to assist doctors and researchers in the follow-up EEG signal processing and analysis,and makes the research content more practical value. |
