Study Of Music Induced Autobiographical Memory EEG In Identity Verification

With the development of computer and information security technologies,EEG signals have become a popular direction in the field of identity verification,and a variety of experimental protocols for EEG acquisition have been developed.As the first step of most systems,authentication must be highly secure and reliable,while the process is as convenient as possible.EEG signals are obviously different from person to person due to their biological characteristics,and are difficult to copy,meeting the needs of security and reliability.Convenience is a major reason for limiting EEG signals as authentication keys.A convenient,user-friendly,and signal-reliable stimulation modality is necessary.Autobiographical memory refers to the mixed memory of personal complex life events and emotions.It has the characteristics of long-term stability,accessibility and uniqueness,while its accessing is simple and easy,and will not increase mental burden.Based on the above research foundation,we believe that the autobiographical memory of adolescence offers great information for identity verification.The main work of this paper is as follows.First,a new identity verification protocol based on autobiographical memory EEG signals is proposed to improve convenience and friendliness of the system,which is used in the signal acquisition and authentication stages of the identity verification system.Specifically,a stable autobiographical memory is induced by specific music,and the EEG signals of the recall process are collected as the key for authentication.Specific music is closely related to an individual’s long-term memory,which can not only play an evoking role,but also limit the range of memory content.This method can obtain a signal with a high signal-to-noise ratio and make the subjects feel comfortable.The EEG signal was preprocessed by independent component analysis to remove EEG artifacts for the construction of the identity verification EEG data set.On the other hand,the realization of complex functions such as memory,reasoning,and movement requires multiple brain regions to complete different internal processes to jointly realize the characteristics.In feature extraction,multi-dimensional and large-scale are emphasized.In this paper,signals’ feature extracted from several dimension such as time domain,frequency domain,time-frequency domain,nonlinearity and functional connectivity.The influence of different feature selection on the results is explored through the comparison between single-class features.The results show that the optimal features are different when building models with different subjects as the target,which verifies the validity of the signal features,that is,for different individuals,the effective features vary from person to person.Finally,in order to improve the accuracy and stability of the identity recognition model and maximize the use of effective information in different types of features,according to the characteristics of the dataset,this study introduces the discussion and verification of the class imbalance problem.Then a multi-modal weak classifier ensemble learning algorithm based on the idea of Adaboost to construct weak classifiers with different features is proposed.Through comparative analysis with other classifiers,the improved algorithm has significantly improved the verification accuracy and model stability for both target and non-target objects.At the same time,the robustness of the authentication system is further analyzed.