Research On EEG-based Visual And Auditory Cognitive Task State Recognition And Analysis Methods

The brain,as the most advanced information processing center for human beings,is the material basis for humans to perform cognitive tasks.Cognitive science and neuroscience research results show that in different stages or states of cognitive tasks or different task,brain activity has different characteristics.Therefore,it is possible to identify current task state by analyzing the activity of the brain.The states in a cognitive task and the current task itself,are called the cognitive task states.In recent years,researches on the neural mechanism,influencing factors,and recognition and detection methods of the human brain cognitive task state have been obtained high attention from experts and researchers in related fields.As a highly accurate,responsive,low-cost and easy-to-apply technique for observing brain activity,EEG signal acquisition is well suited for cognitive task state detection.Visual and auditory cognitive tasks are the most common cognitive tasks,therefore this thesis focuses on how to efficiently and accurately analyze EEG signals for cognitive task state identification when performing audio and visual cognitive tasks.In the research and application of cognitive tasks based on EEG,the number of EEG acquisition channels and the number of acquisitions directly affect the equipment cost and use effect of EEG applications.The fewer the number of channels,the lower the cost of the device,and the more comfortable people wear the device;the fewer the number of acquisitions,the faster the response speed of the entire system.So,how to use fewer channels and fewer acquisitions of EEG data for cognitive task state identification and how to more accurately identify the cognitive task state and the current cognitive task has been an important issue in cognitive neuroscience and cognitive computation.To address these two issues,the main research of this thesis includes the following points.(1)This thesis proposes an EEG classification feature extraction method for cognitive tasks based on sparse dictionary learning.This method uses fewer channels of EEG data for real-time cognitive task state recognition,based on limited data and fully extracts the cognitive task state EEG classification characteristics.This method uses self-organizes mapping on sparse dictionary atoms,and utilizes the correlation between the weight vector in the self-organizing nnetwork and the EEG signal waveform as the classification features.This method used single-channel EEG data to acquire better accuracy as traditional method with 64-channel EEG data,achieving the goal of less EEG data for recognition.(2)This thesis proposes a brain activity pattern analysis method based on group sparse decomposition,which can perform cognitive task states on single-trial EEG data when traditional method requires multiple trials to obtain analysis results.This method greatly reduces the amount of EEG data required for analysis.The EEG signal generated by the human brain during cognitive tasks contains waveforms that are closely related to the cognitive task state,which are collectively referred to as Brain Activity Pattern(BAP)in this thesis.This method proposes to extract cognitive task-related BAPs from EEG signals using a group sparse decomposition technique to realize the analysis of eventrelated potentials in a single trial,to accurately and quickly extract the waveform related to cognitive tasks in EEG signals,and to obtain refined components of event-related potentials.(3)Cross Spectral Pattern(CCSP)features and its extraction technique are proposed to solve the problem that the existing methods do not closely integrate the frequency domain characteristics and spatial distribution of EEG signals.The method makes full use of the regions of the brain involved in different The method takes advantage of the differences in the regions of the brain involved in different cognitive tasks and the frequency domain characteristics of the EEG signals to identify the current cognitive task more accurately by the EEG signals.In the CCSP extraction method,the spatial distribution of EEG signals in different frequency bands corresponding to each cognitive task is first extracted using the band power matrix,then the band power matrix is diagonalized using the orthogonal transform to obtain a joint spatial and frequency domain filter matrix,and finally the logarithmic power matrix of each frequency band is obtained as the classification feature using the filter matrix,and higher recognition accuracy and response speed are obtained using this method..(4)This thesis proposes the use of convolutional neural network-long and short-term memory network(Convolutional Neural Network-Long Short-Term Memory network,CNN-LSTM)to extract time domain multi-scale features and multi-domain features for EEG signal recognition.For single-channel EEG signals,a time-domain multi-scale CNN-LSTM network model is proposed.By adjusting the size of the convolution kernel in CNN,multi-layer CNN is used to extract the features of EEG signals on multiple time scales,and combined with LSTM for EEG signals Recognition;for multi-channel EEG data,multi-domain CNN-LSTM(Multiple Domain CNN-LSTM,MD-CNN-LSTM)network is proposed to extract the Spatial-Spectral domain joint features that change over time,for EEG signal classification.The CNN-LSTM network model proposed in this thesis focuses on the problem that traditional methods do not make full use of the properties of EEG signal in the temporal domain,Spectral domain,and Spatial domain.The proposed network models improves the classification accuracy of cognitive task states and expands the The availability of EEG-based cognitive task state recognition technology.Finally,this thesis integrates the aforementioned methods into an application validation system.On this system EEG-based cognitive task state recognition experiments were conducted to verify the effectiveness of the methods in this thesis.The experimental results show that the EEG can be used for cognitive task state recognition accurately and efficiently using proposed methods.This thesis presents an in-depth improvement study on the key issues of EEG signalbased cognitive task state analysis,focusing on making the EEG signal analysis and recognition methods more accurate and simple,improving the ability of EEG to identify cognitive task states,which contributes to the advancement of research in this field and also inspires other related fields...