Epilepsy is a transient and dysfunctional brain disease,which is caused by sudden,hyperynchronous and abnormal discharge of neurons.Due to the different parts of abnormally discharged neurons in the brain,the clinical manifestations are also more complex,and the disability rate is high and the disease process is long,which causes great trouble to doctors and patients.The emergence and development of machine learning promote the research of automatic epilepsy detection system,breakthroughs have been made in the research on automatic epilepsy detection systems,and the threat of epilepsy to life is gradually reduced.Research in this field involves subjects such as nervous system,signal processing,dynamics,etc.How to obtain useful information from signals and build models to identify different epileptic states is a hot and difficult research topic.Firstly,this thesis introduces the current status of epileptic EEG classification,summarizes the current signal preprocessing,feature extraction,and classification methods,and compares the advantages and disadvantages of different methods.Then,the basic knowledge about EEG signals is introduced,from EEG,EEG signals,experimental data and evaluation criteria.Secondly,based on the similarity between epileptic EEG signals and wavelet basis function,an epileptic EEG signals detection algorithm based on discrete wavelet transform and mixed features is proposed.The original EEG signal is decomposed into four layers by using Db4 wavelet,and the obtained sub-bands are reconstructed with the original signals.In order to distinguish the features of EEG signals in different states of epilepsy to the greatest extent,the nonlinear features and linear features are mixed,and the epileptic EEG signals are classified by support vector machine to form a complete epilepsy detection system.The experimental results confirmed the effectiveness of the preprocessing step and the feature extraction method,and the accuracy in the four classification cases are more than 98%.Finally,considering the advantages and disadvantages of adaptive frequency decomposition method in completeness and reconstruction error,this thesis proposes a epileptic EEG signals classification method based on complete ensemble empirical mode decomposition with adaptive noise.Using complete ensemble empirical mode decomposition with adaptive noise,the epileptic EEG signals is decomposed into IMF components of different scales,and the information that can not be fully obtained by traditional methods under a single time scale is mined.By calculating the correlation between the original signals and each order components,the appropriate modal components are selected,and then the result of feature extraction is input into support vector machine.Whether it is two classification experiment or three classification experiment,the experimental results show that our preprocessing method has a positive impact. |