The Research On Brainprint Recognition Based On Deep Learning

In today’s society where personal information security is important,biometricbased identification methods are playing an important role in various industries.Unlike traditional biometric identification,which has various shortcomings,EEG signals have the advantages of non-stealing,non-coercion and in live detection,therefore the field of EEG signal identification is receiving more and more attention from researchers.Most of the previous studies on EEG identity recognition are based on a single feature in temporal domain,frequency domain or spatial domain,without considering learning the temporal domain information,frequency domain information and spatial domain information of EEG signals at the same time,and therefore only based on the limited identity eigeninformation in EEG signals for recognition.Aiming at the characteristics of EEG signals with temporal domain,frequency domain and spatial domain,this paper proposes the study based on deep learning identification of individuals using EEG signals,and the main work includes.(1)To address the problem that EEG signals cannot characterize temporal domain information,frequency domain information and spatial domain information simultaneously,this paper proposes a method for EEG signal identity recognition based on multispectral brain topography and joint attention.In this paper,we use azimuthal equidistant projection to project EEG electrodes in 3D space to 2D plane,and calculate the power spectral density features in different frequency bands of each channel,and then use interpolation algorithm to fill the features in the remaining positions to complete the construction of multispectral brain topography map.For the multispectral brain topography sequences with both temporal domain features,frequency domain features and spatial domain features,this paper proposes a joint attention network to construct frequency attention,spatial attention and temporal attention to make the network focus on the important parts of the input features while reducing the interference of the unimportant parts.Experiments are conducted on three datasets,and the results show that the joint attention network is able to learn multispectral brain topography effectively and has better identity recognition results compared with other models.(2)To address the problems of temporal dependence,frequency band differences and left-right half-brain asymmetry in EEG signals,as well as the need to build large models when deep learning is used for identity recognition,an identity recognition method for EEG signals based on multi-scale convolution and identity subspace is proposed.First,the EEG signal is divided into multiple frames of data,and the shorttime Fourier transform is used to convert the EEG signal to the time-frequency domain to construct features with time-domain information,frequency domain information and spatial domain information.Then the multi-scale convolution module is used to obtain the temporal dependence of the features,the differences between different frequency bands and the left and right half-brain spatial asymmetry of the subjects,and the output of the multi-scale module is downscaled using subspace learning to construct the identity subspace of the subjects,and finally the cosine similarity between the test samples and each category of the identity subspace is used for recognition.The experimental results show that the EEG identity recognition method based on multiscale convolution and identity subspace has good recognition performance on multi-task data and also has good recognition accuracy in cross-task recognition scenarios.(3)In order to apply the EEG signal identity recognition algorithm to practical scenarios,an EEG signal identity recognition system is designed and implemented in this paper.The system mainly includes four modules: stimulus display module,signal acquisition module,pre-processing module and registration and login module.The stimulus display module is designed and implemented for the motion imagery task and the fast sequence visual presentation task;the signal acquisition module is responsible for receiving data from the EEG acquisition device and providing it to the front-end page for display;the pre-processing module is mainly used to remove the noise from the EEG signal;the registration module is used to save the data locally and train the model for identity login verification.The experimental tests show that the system can effectively identify legitimate visitors and trespassers,providing ideas for the application of EEG identity recognition system in practical scenarios.Based on EEG signals and oriented to identity recognition tasks,this paper have conducted research on deep learning-based EEG signal identity recognition analysis,and these research results both complement the current EEG signal identity recognition technology and provide a reference solution for the application of EEG identity recognition.