Research On Auditory Attentional Brain Mechanism And Application

Auditory attention is the key to the normal operation of the auditory system.The study of auditory attention has made people have a better understanding of the auditory system.Understanding the mechanism of auditory attention is of great significance in practical applications.The Electroencephalograph(EEG)signal contains abundant physiological information and has the characteristics of low cost and high time resolution,and it has been relatively common to apply EEG technology to brain analysis.Based on EEG,this thesis studies the classification method of auditory tracking attention state,explores the spatial-temporal mechanism of auditory attention,and constructs a braincomputer interface based on auditory attention.The content of the detailed study includes the following three aspects:First,the classification research of auditory tracking attention state.The correlation between neural response and speech stimuli is directly related to the quality of auditory attention.Based on this,a new convolutional neural networks(CNN)model with source spatial feature images as input was proposed to decode the state of auditory tracking attention in a cocktail party environment.First,based on weighted minimum norm estimation and rhythm entropy,a source spatial feature image was constructed and used as a classification feature.Then,a CNN classification model with 3 convolutional layers was built,and the classification accuracy of the model reached 80.4%.Compared with other feature input methods,the source spatial feature image obtains the largest classification result,and the result is significant.Second,the exploration of the spatial-temporal mechanism of auditory attention.Attention plays a key role in the process of eliminating interference and processing specific information in the auditory system.The exploration of the activation area and activation time during auditory attention is a hot issue.Deep neural networks have the ability to automatically learn features from data,and visualizing the learned features helps to understand the cognitive information of the corresponding task.Therefore,by constructing a deep neural network classification model based on auditory attention and non-attention states,and visualizing the learned features through saliency maps,it is found that the most discriminating brain regions are in the temporal lobe and frontal lobe,and the most discriminating time period is mainly within 250 ms after the end of the target speech stimulation.The results show that the temporal lobe and frontal lobe are the key activation areas of auditory attention,and 250 ms is the key period of auditory attention.Third,the design of the “hearing aid” brain-computer interface system.A braincomputer interface system for auditory attention state detection was built based on the Windows platform.Data acquisition and data processing modules were integrated through application software.The stimulation module interacts with the data processing module based on the user datagram protocol.The system built a classification model based on the idea of deep transfer learning,that is,a small amount of data was used to fine-tune the 3-layer CNN model,and the fine-tuned model was used to perform classification tasks.The classification accuracy of the model reached 81.2%,which was 6.6% higher than the classification without fine-tuning,namely the direct classification method.During the online experiment,the classification model detects the attention state in real time,and feeds back the attention state in time through the stimulation module.Through feedback,the correct rate and response time of the subjects’ behavioral response were significantly improved,indicating that the system improved the hearing efficiency and achieved the purpose of hearing aid.