The Research On Epilepsy Prediction Method Based On EEG Time Series Data And Deep Learning

Epilepsy is sudden and random,which often poses a serious threat to the life safety of patients.It is also an abnormal discharge disease of the brain,so EEG can be used to study epilepsy prediction.In view of the problems existing in the current research,this paper proposes to study the method of epilepsy prediction by using deep learning network on the basis of EEG time series signal data.The main work contents are as follows:1.At present,almost all researches on epilepsy prediction based on EEG signals convert one-dimensional time series data into two-dimensional image data,and then carry out feature extraction and classification prediction,resulting in complicated data preprocessing and high network training load.In this paper,a research idea of epilepsy prediction based on one-dimensional EEG time series data is proposed,and the corresponding preprocessing process,data screening methods and standards are studied and formulated,and the initial data set is established through this process.2.In order to solve the problem that the sample size of epileptic EEG data set is too small and unbalanced,this paper puts forward the idea of building a data enhancement network based on the structure of Generative Adversarial Network(GAN),and constructs two initial data enhancement networks,one is based on long-short-term memory network,LSTM-gan and DCGAN based on convolution,and designed experiments to verify the performance of the two networks.The verification results show that DCGAN is more excellent in performance;After that,the initial DCGAN is improved and optimized,and finally a W-c DCGAN is constructed.The quality of generated data of this network is verified by using the original data set.The verification results show that the generated data quality is basically up to standard,which can achieve the effect of data enhancement.3.Aiming at the problem that the network training time for epilepsy prediction based on deep learning is too long due to the use of image processing methods,this paper puts forward an epilepsy prediction network based on one-dimensional convolution(CNN classification prediction network),and then makes further improvement and optimization on this basis,and finally constructs a combination of Global Average Pooling,GAP)method and Support Vector Machine(SVM)classifier(CNN-GAP-SVM),and designed experiments to verify the performance of the epilepsy prediction network before and after improvement and the epilepsy prediction network(CNN-GAP-SVM)before and after data enhancement.The verification results show that the improvement of the initial CNN classification prediction network and data enhancement have improved the network performance to some extent.