Classification Of Epileptic Eeg Signals Based On Sparse Representation

Epilepsy is one of the most common chronic nervous system diseases in the world,and its seizure will cause central nervous system dysfunction,with repeated,irregular,sudden characteristics,which seriously harms to the patient’s physical and mental health and normal life.Electroencephalogram(EEG)is an important tool for detecting epileptic seizures,but it is a difficult and time-consuming task to identify seizures by human eyes.Therefore,this paper studies the classification of epileptic EEG signals using the method of sparse representation,which provides an effective tool for automatic detection of epileptic EEG signals.Firstly,since the performance of automatic detection of epileptic EEG signals is influenced by feature extraction methods and classifiers,linear and nonlinear features of epileptic EEG signals were analyzed,and the performance of classifiers such as support vector machine,decision tree,linear discriminant analysis,K-nearest neighbor,and Bayesian were compared.The characteristics of epileptic EEG signals and the mechanism of seizures were explored,and the classification methods for epileptic seizure detection were summarized,laying the foundation for subsequent experiments on automatic epileptic seizure detection and classification.Secondly,a detection method based on sparse representation and Bayesian classifier was implemented to classify epileptic EEG signals.By performing sparse representation on the original EEG signals,the better atoms and coefficients in the sparse representation of each training sample were determined as classification features,and the obtained feature vectors were input into the Bayesian classifier for classification.The performance of the classification method was tested using EEG data from epileptic patients at the University of Bonn,and the results showed that the method had high accuracy in binary classification tasks and strong robustness to noise.Finally,in order to avoid tedious and complex feature extraction steps and reduce processing time,an online dictionary learning and elastic net constraint method was used to further improve the sparse representation classification system.The category of the test sample was determined by comparing the reconstruction error of the normal period and the seizure period,and the performance of the method was tested using the epilepsy dataset from the University of Bonn.The influence of different dictionary atom numbers on the classification effect was analyzed,and the results showed that the accuracy of binary classification reached 100% under a certain number of atoms,indicating that the method has application value in automatic epileptic seizure classification.