Research On The Classification Of Attentional States Based On EEG

EEG signals have become a hot research topic in the field of deep learning because they can reflect human mental and emotional states without the control of human subjective consciousness.Among the studies based on EEG signals,the detection of attentional states has received much interest because of its high relevance to people’s lives.Nowadays,most of the attentional states classification models based on deep learning and EEG signals are studied and analyzed using a single structure.The deep-level information included in EEG signals cannot be effectively extracted by these single models.And as the number of neural network layers increases,more losses are generated.This leads to the final classification results are not accurate enough.Therefore,this thesis takes attentional states as the main research object and explores multiple deep learning network models.Combining the Attention mechanism,convolutional neural network,and bi-directional long short-term memory network.According to the characteristics of different models,the corresponding improvement methods are proposed,and the specific research is as follows:(1)A student attentional states classification model integrating Attention mechanism and Bi LSTM was designed.In order to improve the accuracy of deep learning models on the attentional states classification problem and further explore the effective features of EEG signals,this thesis designs FF-Bi ALSTM,a student attentional states classification model based on Attention mechanism and Bi LSTM,to realize the binary classification of attentional states.By combining the Attention mechanism with the bi-directional long short-term memory network.The global features of EEG signals are captured using the Attention mechanism,and the time-domain features of EEG data are extracted by two consecutive layers of bi-directional long short-term memory network.The experimental results show that the FF-Bi ALSTM model can better extract EEG signal features,and this model is improved in several aspects compared with traditional neural network models,which can effectively improve the classification performance of attentional states.(2)An attentional states multi-classification model with 1DCNN and Bi ALSTM in parallel was designed.In this thesis,based on the study of the attentional states binary classification problem,a further exploration of the attentional states multi-classification problem based on EEG signals is presented.The attentional states multi-classification model with the parallel integration of 1DCNN and Bi ALSTM is designed.Firstly,the EEG data collected from multiple channels are analyzed for the time-frequency features using power spectrum estimation and short-time Fourier transform to explore the most suitable feature extraction method for EEG signals.Then,the local features and the global features of EEG signals are learned separately using a modified 1DCNN model in parallel with the Bi ALSTM model.Finally,the features learned by the two models are combined and put into the full connection layer to accomplish the multi-classification of attentional states.The experimental results show that the 1DCNN-Bi ALSTM model has better classification results on the attentional multi-classification problem compared with the baseline method support vector machine and FF-Bi ALSTM model.