Feature Extraction And Classification Of Epileptic EEG Signals

Electroencephalogram(EEG)is a physiological signal from the complex system of the brain.It is the spontaneous or induced electrical activity of nerve cell groups in specific parts of the brain during physiological processes.It is a typical field to study the brain function with the method of biological cybernetics.EEG signal has a strong randomness,a variety of rhythms,including rich physiologica l and pathological information.From traditional visua l observation,simple time and frequency domain analysis,to modern time-frequency analysis and nonlinear dynamics research,with the continuous development of EEG signal research methods,the understanding of EEG is becoming more and more profound,and it s application is becoming more and more extensive.In clinical medicine,it is an irreplaceable and effective method to detect some brain diseases(such as epilepsy,depression,encephalitis,schizophrenia,etc.)by EEG,especially in the field of diagnosi s and treatment of epilepsy.At present,the research on epileptic EEG automatic classification mainly includes two parts: feature extraction and classification.Feature extraction is a key step in EEG analysis and research,good quality of features has an important impact on the classification results;a suitable classification algorithm plays an important role in training and recognition speed,classification accuracy and stability.In this paper,research is carried out in two aspects: feature extraction and classification.The main work is as follows:Feature extraction and classification are studied from the perspective of machine learning.In the feature extraction stage,a feature extraction method based on wavelet packet-based time-frequency analysis combined with nonlinear and statistical features is proposed,which can deeply mine EEG signal features from multiple perspectives,and then fuse to form feature vectors.In the research of classification,the idea of integrated learning is used,combined with the grid search optimized XGBoost algorithm for classification,and a random forest model is established for comparison.The results show that the proposed classification system can fully extract the EEG features of epilepsy,and accurately classify different types of EEG signals,which has good robustness.the overall system has higher classification accuracy,better universality,and realizes the automatic extraction of epilepsy EEG features.Feature extraction is more sufficient,and the classifiable samples have a shorter duration,which provides a new idea for automatic detection of epilepsy EEG.Further research is carried out with the method of machine learn ing combined with deep learning,the neural network method is used to extract features automatically,and combined with machine learning methods to further improve model classification performance.Firstly,a model of epilep tic EEG classification based on one-dimensional convolutional neural network was studied;then a hybrid classification model based on one-dimensional convolutional neural network and XGBoost algorithm was proposed.Through experimental comparison,the results show that the proposed hybrid model can further improve the performance of the one-dimensional convolution model,the overall system has higher classification accuracy,has better universality,and realizes the automatic extraction of epilep tic EEG features.Feature extraction is more sufficient,and the classifiable samples have a shorter duration,which provides a new idea for automatic detection of epilep tic EEG.