Research On Automatic Seizure Detection Algorithm Of Electroencephalogram Signals Based On Flexible Analytic Wavelet Transform

Epilepsy is one of nervous system disease that cause troubles to human beings,whose occurrence regularity is difficult to predict.Electroencephalography(EEG)is one of the most effective techniques for diagnosing epilepsy.When epileptic seizure occurs,the brain is periodically in high frequency discharge,and the characteristic waves such as spinous wave,spike wave,slow wave,and slow compound wave can all be reflected by EEG.At present,the detection of epilepsy is based on observation and comparison of EEG in different stages to determine the type of epileptic disease,the incubation period,the treatment plan or surgical site from an intuitive perspective,which usually requires long-term monitoring of brain electrical signals from epileptic patients.This process produces an enormous amount of EEG data.It will sonsume a lot of time and energy to diagnose based on the medical workers' visual acuity.Meanwhile,depending on the work experience,diagnostic habits and criteria of judgment,deviation or misdiagnosis will occur.So,it is of great significance to develop a method which can be applied to automatic detection of EEG.For the complex and detailed characteristics of EEG signals,a large number of details are lost when extract EEG signals.In this paper,an automatic detection method of time-frequency analysis is proposed in the process of EEG feature extraction,and the effectiveness of this method is verified by experiments.The process of the EEG is as follows: firstly,the Flexible Analytic Wavelet Transform(FAWT)is applied in the processing of data sets to decompose the signal and reconstruct them.Multiscale transformation makes the reconstruction signal more subtle than the original signal.Secondly,the Short-Time Fourier Transform(STFT)is used to analyze and reconstruct the signal.The width of window is set to 256 points,and the overlap number is 200 points.Then,the Non-negative Matrix Factorization(NMF)is used to reduce the time-frequency characteristic matrix of the samples.Finally,the new time-frequency characteristics are put into the Support Vector Machine(SVM)to classify.The mean accuracy of the recognition classification,as the main evaluation index,is 99%.The basic steps of the automatic detection method are just like as we detail.Basedon the above works,multigroup contrast experiments are conducted,including:1.Find the optimal number of decomposition layers for FAWT.Due to the large difference in the degree of decomposition of EEG in different decomposition layers,In the experiment,we change the number of layers J from 1 to 16,and compare the classification accuracy of each J value.Then,we can find that the classification accuracy is highest when J is 6.2.Determine the way which FAWT is refactored.With the different frequency bands of EEG refined in different reconstructions,the results of accuracy are very different.In the experiment,four reconstruction methods are compared.And the results show that the classification accuracy is the highest in the first reconstruction method.3.Compare the multi-scale time-frequency analytic performance among FAWT,Tunable Q-factor Wavelet Transform(TQWT)and Discrete Wavelet Transform(DWT).The results show that the accuracy rate of FAWT is higher than that of TQWT and DWT when it is reconstructed in the first way.4.Compare the time-frequency characteristics of STFT and Fast Fourier Transform(FFT).In order to make the experimental results more objective,the time-frequency window size of STFT is adjusted in the experiment,so that the number of time-frequency characteristics of STFT and FFT are consistent.The results of the accuracy show that the accuracy of STFT is higher than that of FFT.5.Compare the performance of dimension reduction about NMF,Factor Analysis(FA)and Singular Value Decomposition(SVD).Analysis the performance of dimension reduction from the accuracy and time when the dimension reduction is changed from 1 to 30.The results show that the dimension reduction performance of NMF is higher than that of FA and SVD.6.Verify the effectiveness of this method.Compare the accuracy of this paper with the literature accuracy in this field,and the results show that the accuracy of this paper is higher than other literatures.