The EEG Recognition Combined With Convolution Neural Network And Recurrent Neural Network

Brain computer interface(BCI)is defined as a combination of software and hardware that allows brain to control computer or other devices without using the peripheral nerves and muscles,which is a new way of interaction.The main goal is to help seriously disabled persons live as ordinary people and even recover in medical treatment.However,many other fields have become interested in the application of BCI technology,such as entertainment,life,games,military and so on.Since the EEG signal is high-dimensional multi-channel and the signal-to-noise ratio is very low,there is very little useful information that can be used for analysis in the original data.It is difficult to extract the essential features in the EEG signal,and it is a random sequence signal.Convolution extracts abstract features with great ability to enhance the signal-to-noise ratio of EEG signals.In addition,the timing of EEG signals is often neglected.Recurrent Neural Network(RNN)is used to implement EEG,which can make full use of the signal.Therefore,it is determined to study the feature extraction and classification of EEG signals based on Convolutional Neural Network(CNN)and RNN.For the problem that the RNN easily leads to the disappearance of the gradient,by introducing a gated recursive unit(GRU)of the gated structure,it has a longer "memory";In addition,the problem of over-fitting of the network,we use dropout,a certain percentage of neurons are randomly dropped during training to prevent over-fitting and enhance the ability of generalization.At the same time,Batch Normalization(BN)is also introduced to keep the input distribution of data consistent,speed up network training,and prevent over-fitting to some extent.And it is applied to the recognition of motor image EEG signals.Compared with traditional and common algorithms,experiments showed that the proposed algorithm had higher accuracy.For the depth of the network is increasing,the depth of the extracted features will be deeper,and the shallow features will be ignored.At the same time,the parameters will increase exponentially,and the amount of calculation will become larger and larger.A cross-connected continuous convolutional long-term memory networks algorithm is proposed for EEG signal recognition.The idea of cross-layer connection is introduced to connect the shallow features nd deep features into the LSTM network,and then using the Merge layer to complete the feature fusion,it can make full use of the features;the traditional fully connected layer is replaced by a global average pooling layer,which can reduce the training parameters of the network and speed up the network training.Finally,experiments showed that the proposed algorithm can achieve higher recognition rate in the recognition of EEG signals.