Research On Classification Of Epileptic EEG Based On Deep Learning

Epilepsy is one of modern common diseases,heavily threatening public health.According to the produce of the special brain wave such as spike wave and sharp wave,EEG monitoring is frequently-used in clinical treatment to detect epileptic attack.For the amount of EEG signal generated in this process is huge,manual recognition is inefficient and not accurate enough.Therefore,it is of great significance to develop a method of automatically identifying epileptic EEG signals.This paper presents an assortment method of epileptic EEG signal based on depth learning,focusing on the EEG database of the Epilepsy Research Center in Bonn University.This method includes the following steps:extracting the characteristics of epileptic EEG based on sparse automatic coding,analysing the further characteristic gotten from SAE based on Long-Short-Term-Memory Recurrent Network Networks and signal classification based on softmax classifier.The specific work of this paper has the following aspects:Firstly,get the sample database needed for deep learning network training.The original epilepsy EEG data is huge number of continuous real brain signals.In order to not lose the signal information with depth learning network structure training,the data were normalized,segmented stored and label processed,etc.Secondly,extract characteristic epilepsy EEG signals based on SAE.After the segmentation,a total of 20,000 data samples with a dimension of 100 are obtained.Putted into the SAE,then ascertain the corresponding weight matrix in the process of constant parameter adjustment to determine the network structure.According to the network structure,the output of SAE will become the input of the LSTM-RNN.Thirdly,further characteristic gotten from SAE analyzed based on the LSTM-RNN.In order to incorporate the timing characteristics of EEG signals into analysis,this paper adopts the LSTM-RNN which is more adept in timing analysis.The LSTM-RNN has a total of 10 memory modules.The network sample data dimension is set to 1000,and the large sample which has the dimension of 1000 is divided into 10 small samples which have the dimension of 100,orderly.10 small samples are inputted into the SAE corresponding to each memory modules,to optimize the network structure in the constant parameter adjustment process.The output of the LSTM-RNN will be used as the input of the softmax classifier.Finally,complete the signal classification according to the softmax classifier.After the experiment,the accuracy of the two classification was 65.3%,and the accuracy of the three classification was 56.7%.The results show that the method of this paper has the ability to recognize and classify epileptic EEG signals.