Automatic Classification Of Epileptic EEG Signals Based On Local Pattern

Electroencephalogram(EEG)signals represent the electrical activity of the human brain.Because of its high accuracy,safety,non-invasiveness and cheapness,it has become the mainstream technology for epilepsy diagnosis.However,the classified study of epilepsy EEG signals still has problems such as high algorithm complexity,unstable performance caused by imbalanced data categories,and poor classification effects due to too few sample sizes.In order to effectively detect epilepsy signals in EEG,this paper focuses on the classification of EEG signals in local mode,and introduces local modes,multi-granularity scanning,and extreme learning machines into the detection of epileptic EEG signals.The innovations and main contributions proposed in this article are as follows:(1)An improved Multi-Granularity Local Binary Pattern(M-LBP)operator in local mode is proposed.Multi-granularity scanning is used to process multiple different types of EEG signals to obtain multiple granularity corresponding high-dimensional EEG signals with more detailed information;by calculating the mean and standard deviation of the local binary transformation codes of the high-dimensional EEG signals at each granularity,the final feature vectors are fed to five different classifiers for classification.Experiments were performed on the epilepsy dataset of the University of Bonn.The experimental results show that the method has a good classification effect on different data sets of two and three classifications,especially the two-class Z-S,Z-F classification accuracy reaches 99.5%.This operator analyzes and calculates the node information of the signals under different coarse and fine granularity,so as to prevent the omission of detailed information and improve the classification accuracy.(2)An epileptic EEG signal classification model based on 1D-Local Ternary Pattern(1D-LTP)and Extreme Learning Machine(ELM)was implemented.Firstly,use the 1DLTP operator to extract the features of the top mode and the bottom mode of the EEG signal;secondly,use the Principal Component Analysis(PCA)to reduce the feature code to use a limit learning machine to classify the feature code;and finally use 10 x cross validation to assess classification performance.The experimental data show that the proposed algorithm has better classified accuracy on the experimental data set,reducing the number of redundant features and the recognition accuracy on Z-S can reach 99.79%,showing that the classification model has a strong classification ability on epileptic EEG signals.Through this experiments,the effects of various parameters of the multi-granular local binary pattern operator and the one-dimensional local ternary pattern operator on the classification are simulated and analyzed.The results show that the proposed algorithm improves the classification accuracy of epilepsy EEG signals to varying degrees,and achieves the expected classification goal.It not only provides a solution for rapid and accurate automatic diagnosis of epilepsy diseases,but also an effective reference for other classification studies.