CN114638252A - Electroencephalogram-based identity recognition method - Google Patents

Abstract

The invention provides an electroencephalogram-based identity recognition method, which comprises the following steps: inducing the testee to perform stable self-transmission memory, and collecting electroencephalogram signals of the testee in the process of performing stable self-transmission memory; the electroencephalogram signals of the testee in the process of stable self-memory are used for identity recognition. The invention can improve the safety and convenience of identity identification. The invention induces stable self-memory by setting specific music, collects the brain electrical signals in the memory process and uses the brain electrical signals for identification. Through the preprocessing such as denoising to the initial signal, carry out feature extraction from a plurality of dimensions, the music induces the brain electrical signal from the memory of the transmitter and has excellent effect on authentication.

Description

Electroencephalogram-based identity recognition method

Technical Field

The invention relates to an electroencephalogram-based identity recognition method, and belongs to the technical field of identity recognition.

Background

The increasingly updated information technology enables personal identity authentication to be widely applied to the fields of information security, access control, security monitoring and the like, and biological characteristics occupy a place and are continuously expanded in identity authentication due to uniqueness of the biological characteristics. The biological features include voice, signature and gait based on human behavior biological features, and focus on human physiological features such as face, iris and fingerprint. Currently, facial recognition and fingerprint recognition are widely used in daily life, and become one of the most common authentication methods. Nevertheless, it is undeniable that these reproducibly and irreproducibility based on human biological features pose a great risk to information security.

On the other hand, in view of the development of portable medical devices including EEG headset-type acquisition devices, electroencephalogram signals are also more and more frequently appearing in the field of authentication and biometrics as very potential and special biometrics. Compared with other characteristics, the electroencephalogram signal has two visual advantages, the first is that the electroencephalogram signal is very high in privacy, and cannot be intuitively detected and difficultly copied as an electric signal acquired on cerebral cortex when a human senses or thinks; secondly, the acquisition of the brain electrical signals must require that the user be alive or even conscious.

Currently, there are three main paradigms for applying brain wave EEG to biometric identification. The first is to collect the brain electrical signals in a resting state, including opening or closing the eyes. Except that the electroencephalogram signal acquisition equipment does not use other equipment, the acquisition protocol generally requires quiet and comfortable environment, a testee completely relaxes, and equipment except the signal acquisition equipment is not used, so that the method is simple and convenient. However, the resting brain electrical signals are susceptible to environmental noise and other electrical signal artifacts, the signal-to-noise ratio is low, and the result may be affected if the subject thinks or moves mentally. The second is a signal acquisition protocol based on external stimuli, which mainly includes visual stimuli and auditory stimuli. Visual stimuli, also known as Visual Evoked Potentials (VEPs), refer to a set of ERPs (event-related potentials) that are evoked by external visual stimuli. Compared with the signal to noise ratio in a resting state, the electroencephalogram signal acquisition device has the advantages that the electroencephalogram signal acquisition device acquires the electroencephalogram signals by external stimulation, the signal to noise ratio is higher, the characteristics are more obvious, in addition to the electroencephalogram acquisition device, an additional preparation is needed for providing stimulated objects or equipment, and the convenience is more complicated and reduced. The third experimental paradigm refers to a mental task protocol in which participants perform various types of cognitive or mental activities and collect the brain electrical signals during the process. Compared with visual stimulation, the mental task does not need special stimulation equipment, is simpler and more convenient, and is more reliable than signals acquired in a resting state. Until now, many mental tasks are to let the testees learn to memorize new things, think about problems, etc., and a person's thinking ability, memory energy and attention level are limited to some extent by environment, emotion and personal state, etc., and few studies prove that these factors have no influence on the final result, and on the other hand, some mental tasks increase the mental fatigue of the user. The prior art is difficult to apply to identity recognition due to various reasons, and the accuracy, operability and convenience are insufficient.

Disclosure of Invention

The invention aims to overcome the defects in the prior art, provides an identification method based on electroencephalogram, and can improve the safety and convenience of identification.

In order to achieve the purpose, the invention is realized by adopting the following technical scheme:

in a first aspect, the invention provides an electroencephalogram-based identity recognition method, which comprises the following steps:

inducing the testee to perform stable self-transmission memory, and collecting electroencephalogram signals of the testee in the process of performing stable self-transmission memory;

the electroencephalogram signals in the process of stable self-memory of the testee are used for identity recognition.

Further, the method for inducing the subject to perform stable self-memory and collecting the electroencephalogram signals of the subject during the process of performing stable self-memory comprises the following steps:

recording the subjective fatigue degree of the testee in the implementation interval of each experiment; placing brain area collecting electrodes of multiple brain areas, and acquiring corresponding electroencephalogram signals by using the brain area collecting electrodes placed on the scalp of the brain;

enabling the testee to close eyes to follow the music for association and recall, recording electroencephalogram signals of the testee for 3 minutes in the recall period, and recording 5 related keywords for assisting the music to define the recall content of repeated experiments;

and (4) amplifying and performing analog-to-digital conversion on the electroencephalogram signals, and storing the electroencephalogram signals as digital signals.

Furthermore, the method for placing the brain area collecting electrodes of the multiple brain areas and acquiring the corresponding electroencephalogram signals by utilizing the brain area collecting electrodes placed on the brain scalp comprises the following steps:

selecting a forehead control brain area and an occipital-temporal sensation brain area related to music, memory and vision, recording leads FP1, FP2, F1, F2, T3, T4, C1, C2, P1, P2, O1, O2, Cz, Pz and Oz by adopting a 10-20 electrode lead positioning standard calibrated by the International electroencephalogram society and using a binaural plumbing method, wherein the reference electrodes are M1 and M2, the sampling frequency is 512HZ, and the impedance of each channel lead is less than 5k omega;

the method for amplifying and performing analog-to-digital conversion on the electroencephalogram signal and storing the electroencephalogram signal as a digital signal comprises the following steps: electroencephalogram signals are collected by a Neuroscan64 device, amplified, subjected to analog-to-digital conversion and input into a computer.

Further, the method for performing identity recognition by using the electroencephalogram signals of the testee in the process of performing stable self-transmission memory comprises the following steps:

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the characteristic vector subjected to dimension reduction into an identity recognition model, and outputting the category of the characteristic vector, namely the identity information of the tested person;

the identity recognition model is obtained by training an integrated classifier based on the RUSBoost algorithm.

Further, the training method of the identity recognition model comprises the following steps:

designing an experiment for inducing the memory of the self-transmission body by the music;

collecting brain waves in the experimental process;

the electroencephalogram signals are amplified and subjected to analog-to-digital conversion, and are stored as digital signals;

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the feature vector subjected to dimension reduction into a training database;

and training the integrated classifier based on the RUSBoost algorithm by using a training database to obtain a trained identity recognition model.

Further, the experimental method for designing music-induced memory comprises:

selecting specific music familiar to the testee as an anchor point, considering both the experience of the testee and the number of samples, and adopting a mode of repeated experiments at intervals;

the first time of the test is relaxed and sits on a chair, after the eyes are closed and the rest is calm, the user can listen to music and simultaneously carry out association and recall, and the memory electroencephalogram signals with the duration of 3 minutes are recorded;

after the rest of the testee is finished, 5 keywords about memory are recorded, wherein the keywords are mainly used for assisting music to define memory contents, deepen prints and facilitate subsequent repeated experiments;

the experiment was repeated after full rest.

Further, the method for training the ensemble classifier based on the RUSBoost algorithm by using the training database comprises the following steps:

setting normalized sample weight W (i) for all samples;

② defining the iteration number as T, and for T being 1,2, …, T:

randomly extracting a certain number of majority samples, forming a training data set S by all minority samples, and obtaining the weight S of the samples in S_wNormalizing it;

training a weak classifier h (t) by using the data set S according to the weight, wherein the output of h (t) is the probability of judgment, the probability of correct judgment of the ith sample by h (t) is p1(i), and the probability of wrong judgment is p2 (i);

calculating an error e, e ═ Σ w (i) [1-p1(i) + p2(i) ] of the misclassification;

updating the weight:

and normalizing;

output integrated classifier

W (i) is the updated weight, e is the error of the erroneous classification, h (t) is the weak classifier learned at each iteration, p1(i) is h (t) the probability of correct decision on the ith sample, and p2(i) is the probability of incorrect decision.

Further, the method for preprocessing the electroencephalogram signal of the digital signal comprises the following steps: filtering the acquired electroencephalogram signals, reducing interference of power frequency signals, environmental noise, myoelectricity and the like, removing baseline drift, and removing ocular artifacts by using Independent Component Analysis (ICA);

the method for filtering the acquired electroencephalogram signals comprises the following steps: and performing an FIR filter with the band pass of 0.5-40Hz on the acquired original electroencephalogram signals to remove baseline drift and high-frequency interference components.

Further, the method for respectively extracting the features of all the samples comprises the following steps:

extracting frequency domain characteristics of the sample to obtain an AR model coefficient, power spectral density and power difference of a left hemisphere and a right hemisphere of a frequency domain;

extracting the time-frequency domain characteristics of the sample to obtain a discrete wavelet transform coefficient of a time-frequency domain;

carrying out nonlinear feature extraction on the sample to obtain a nonlinear feature sample entropy;

extracting functional connectivity characteristics of the sample to obtain a functional connectivity characteristic phase lag index;

and (3) forming the characteristic vector of the sample by the AR model coefficient of the frequency domain, the power spectral density, the power difference of the left hemisphere and the right hemisphere, the discrete wavelet transform coefficient of the time-frequency domain, the nonlinear characteristic sample entropy and the functional connectivity characteristic phase lag index.

Further, the method further comprises:

inputting the feature vectors of a test sample into a classifier for classification, wherein the test sample comprises the feature vectors of the sample and corresponding classes;

obtaining the number (TP) correctly identified in all the positive samples, the number (FN) incorrectly identified in the positive samples, the probability (TN) correctly identified in the negative samples and the probability (FP) incorrectly identified in the negative samples according to whether the feature vector of the test sample can be correctly judged as the category of the corresponding sample;

calculating a comprehensive False Rejection Rate (FRR), a False Acceptance Rate (FAR) and a F1 score, and evaluating the identity recognition model by adopting three evaluation indexes of the False Rejection Rate (FRR), the False Acceptance Rate (FAR) and the F1 score, wherein the three indexes are as follows:

in the formula, TP is the number of correctly recognized positive samples, FN is the number of incorrectly recognized positive samples, TN is the number of correctly recognized negative samples, and FP is the number of incorrectly recognized negative samples.

Compared with the prior art, the invention has the following beneficial effects:

1. the invention starts from an experimental paradigm, utilizes the natural advantages of stability and easy accessibility of self-memory in the brain, and is based on simple head-mounted trial signal acquisition equipment, so that the acquisition of the personal electroencephalogram signal can be completed, the electroencephalogram signal used as the key can be acquired, and the operation difficulty is extremely low. Meanwhile, complex work such as calculation, memory and the like is not required to be completed by the test, the increase of mental burden and psychological pressure of the test is avoided, the convenience and the user friendliness are greatly improved, accidents caused by complex acquisition processes are avoided, and the data robustness is enhanced.

2. The invention considers the time span of the identity recognition system in a use scene, increases specific music as a recall anchor point, combines the special function of the music in memory, limits the approximate content and range of recall through the music, reduces the influence of time on data and lays a foundation for the accuracy of the whole system.

3. In the aspect of construction of the identification system, the signal characteristics are extracted from multiple dimensions, and the identification accuracy is improved by focusing on the problem of class imbalance in the identity identification system.

4. The method and the device ensure the accuracy of the identity recognition model, reduce the complexity of user operation in the verification process, improve the experience of the user and enhance the user friendliness. Meanwhile, the class imbalance of the training samples is improved in a targeted manner, an integrated classifier based on the RUSBoost algorithm is used, and the evaluation index F1 score more suitable for the class imbalance samples is used, so that the accuracy of the model is further improved.

Drawings

Fig. 1 is an overall frame diagram of brain electricity for music-induced self-memory for identity recognition.

FIG. 2 is a schematic diagram of channel leads of an electroencephalogram signal.

Fig. 3 is a schematic diagram of discrete wavelet transform and coefficient selection.

Fig. 4 is a diagram showing a feature extraction process and a feature vector composition.

Fig. 5 is a graph illustrating FAR value comparison based on three classifiers.

Fig. 6 is a graph showing FRR value comparison based on three classifiers.

Fig. 7 is a graph showing comparison of F1 scores based on three classifiers.

Detailed Description

The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.

The first embodiment is as follows:

the embodiment provides an identification method based on electroencephalogram, which comprises the following steps:

inducing the testee to perform stable self-transmission memory, and collecting electroencephalogram signals of the testee in the process of performing stable self-transmission memory;

the electroencephalogram signals in the process of stable self-memory of the testee are used for identity recognition.

Self-memory refers to the mixed memory of the emotion of a person's complex life events, including both the recall of the person's past events and the factual knowledge about himself, which is present in all persons with normal memory. Studies have shown that humans can retain a great deal of detail in personal memory. Also, the amount and quality of memory varies greatly from individual to individual, and some long-term self-transmitting memories in adolescence and early adulthood remain highly accessible throughout life. In other studies, music is inseparable from emotion, memory, according to brain imaging studies, music-induced emotion, emotional arousal is associated with activation of central reward circuits and dopaminergic mechanisms, which in turn affect cognitive performance and memory formation. In general, the self-transmission memory of human adolescence and early adulthood has strong stability and accessibility, and the natural existence in consciousness can be easily induced without increasing mental burden. In fact, human memory, as a material stored in the brain, is not only simple and convenient to recall, but also formed according to the actual personal experience and perception of each person, and has very strong personal characteristics.

Specifically, the method for inducing the subject to perform stable self-memory and collecting the electroencephalogram signals of the subject during the process of performing stable self-memory comprises the following steps:

recording the subjective fatigue degree of the testee in the implementation interval of each experiment; placing brain area collecting electrodes of multiple brain areas, and acquiring corresponding electroencephalogram signals by using the brain area collecting electrodes placed on the scalp of the brain;

enabling the testee to close eyes to follow the music for association and recall, recording electroencephalogram signals of the testee for 3 minutes in the recall period, and recording 5 related keywords for assisting the music to define the recall content of repeated experiments;

and (4) amplifying and performing analog-to-digital conversion on the electroencephalogram signals, and storing the electroencephalogram signals as digital signals.

Specifically, the method for acquiring corresponding electroencephalogram signals by using the brain area acquisition electrodes arranged on the brain scalp comprises the following steps:

selecting a forehead control brain area and an occipital-temporal sensation brain area related to music, memory and vision, recording leads FP1, FP2, F1, F2, T3, T4, C1, C2, P1, P2, O1, O2, Cz, Pz and Oz by using a 10-20 electrode lead positioning standard calibrated by the International electroencephalogram society and using a binaural plumbing method, wherein the reference electrodes are M1 and M2, the sampling frequency is 512HZ, and the impedance of each channel lead is less than 5k omega.

The method for amplifying and performing analog-to-digital conversion on the electroencephalogram signal and storing the electroencephalogram signal as a digital signal comprises the following steps: electroencephalogram signals are collected by a Neuroscan64 device, amplified, subjected to analog-to-digital conversion and input into a computer.

Specifically, the method for performing identity recognition by using the electroencephalogram signals of the testee in the process of performing stable self-transmission memory comprises the following steps:

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the characteristic vector subjected to dimension reduction into an identity recognition model, and outputting the category of the characteristic vector, namely the identity information of the tested person;

the identity recognition model is obtained by training an integrated classifier based on a RUSBoost algorithm. The training samples input by the classifier include the features of the samples and the corresponding labels (i.e. positive or negative samples representing the tested identity), and the label output by the classifier for judging the feature vector is also called as a class. The classification result is that the classifier judges which class the sample belongs to according to the input sample characteristics, and whether the judgment is correct or not is known by comparing the judged class with the real class. In the identity recognition, one type is a positive sample, namely a sample of a designated user, a negative sample is a sample of other users, the classifier classifies the feature vector into the positive sample or the negative sample, and if the classification is correct, the recognition is correct.

Specifically, the training method of the identity recognition model includes:

designing an experiment for inducing the memory of the self-transmission body by the music;

collecting brain waves in the experimental process;

amplifying and performing analog-to-digital conversion on the electroencephalogram signals, and storing the electroencephalogram signals as digital signals;

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the feature vector subjected to dimension reduction into a training database;

and training the integrated classifier based on the RUSBoost algorithm by using a training database to obtain a trained identity recognition model.

Specifically, the experimental method for designing music-induced memory includes:

selecting specific music familiar to the testee as an anchor point, considering both the experience of the testee and the number of samples, and adopting a mode of repeated experiments at intervals;

the first time of the relaxation of the testee is performed, the testee sits on a chair, after the eyes are closed and the rest is calm, the testee can listen to music and simultaneously perform association and recall, and the recall electroencephalogram signals with the duration of 3 minutes are recorded;

after the rest of the testee is finished, 5 keywords about memories are recorded, and the keywords are mainly used for assisting music in defining memory contents, deepening the print and facilitating subsequent repeated experiments;

the experiment was repeated after full rest.

Specifically, the method for training the ensemble classifier based on the RUSBoost algorithm by using the training database comprises the following steps of:

setting normalized sample weight W (i) for all samples;

② defining the iteration number as T, and for T being 1,2, …, T:

randomly extracting a certain number of majority samples, forming a training data set S by all minority samples, and obtaining the weight S of the samples in S_wNormalizing it;

training a weak classifier h (t) by using the data set S according to the weight, wherein the output of h (t) is the probability of judgment, the probability of correct judgment of the ith sample of h (t) is p1(i), and the probability of wrong judgment is p2 (i);

calculating error e, e ═ Σ w (i) ([ 1-p1(i) + p2(i) ];

updating the weight:

and normalizing;

fifthly, output an integrated classifier

W (i) is the updated weight, e is the error of the erroneous classification, h (t) is the weak classifier learned at each iteration, p1(i) is h (t) the probability of correct decision on the ith sample, and p2(i) is the probability of incorrect decision.

Specifically, the method for preprocessing the electroencephalogram signal of the digital signal comprises the following steps: filtering the acquired electroencephalogram signals, reducing interference of power frequency signals, environmental noise, myoelectricity and the like, removing baseline drift, and removing ocular artifacts by using Independent Component Analysis (ICA);

the method for filtering the acquired electroencephalogram signals comprises the following steps: and performing an FIR filter with the band pass of 0.5-40Hz on the acquired original electroencephalogram signals to remove baseline drift and high-frequency interference components.

Specifically, the method for respectively extracting the features of all samples comprises the following steps:

extracting frequency domain characteristics of the sample to obtain an AR model coefficient, power spectral density and power difference of a left hemisphere and a right hemisphere of a frequency domain;

extracting the time-frequency domain characteristics of the sample to obtain a discrete wavelet transform coefficient of a time-frequency domain;

carrying out nonlinear feature extraction on the sample to obtain a nonlinear feature sample entropy;

extracting functional connectivity characteristics of the sample to obtain a functional connectivity characteristic phase lag index;

and (3) forming the characteristic vector of the sample by the AR model coefficient of the frequency domain, the power spectral density, the power difference of the left hemisphere and the right hemisphere, the discrete wavelet transform coefficient of the time-frequency domain, the nonlinear characteristic sample entropy and the functional connectivity characteristic phase lag index.

The characteristic extraction of multiple dimensions is required because the change of the electroencephalogram signal is random and irregular, and the memory process involves the synergistic effect of multiple brain areas, so that the characteristic extraction is carried out from the multiple dimensions as much as possible to identify

Specifically, the feature extraction method is as follows:

(1) extracting frequency domain features: realizing an AR model by adopting a parameter model method to obtain an AR coefficient, wherein the parameter model method generally assumes that a signal x (n) is output by exciting a linear system H (z) by a white noise sequence u (n), observes obtained data x (n) or an autocorrelation function thereof to estimate parameters of H (z), and estimates a power spectrum of x (n) by the parameters of H (z); namely:

calculating power spectral density by using an improved periodogram algorithm Welch method, and extracting power spectral density characteristics of 15 channels in 4 frequency bands, namely theta (Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30Hz) respectively;

after obtaining the power spectral density of each lead, the power difference PD of 6 pairs of signal channels symmetrically distributed in the left and right hemispheres is calculated:

wherein P is_leftIs the power of the left hemisphere channel, P_rightIs the power of the right hemispherical channel;

(2) extracting time-frequency domain features: decomposing the signal into a weighted sum of wavelet bases using a DB4 wavelet base in the Daubechies wavelet family for Discrete Wavelet Transform (DWT); the wavelet transform is a time-frequency two-dimensional domain analysis method of signals, and shows the spectrum distribution condition of the signals at any time or in any short time; performing 6-order wavelet decomposition based on the sampling rate of the original signal at 512Hz to obtain wavelet coefficients of corresponding frequency bands; meanwhile, in order to reduce the dimensionality of the feature vector, the average value and the mean square error of an absolute value and the Shannon entropy of the wavelet coefficient are further extracted from the wavelet coefficient to be used as the feature vector of wavelet transformation;

(3) and (3) carrying out nonlinear feature extraction: the electroencephalogram signals are complex and chaotic, and besides linear analysis on EEG, nonlinear research is also indispensable; the sample entropy measures the complexity of a time sequence by measuring the probability of generating a new mode in a signal, and the larger the probability of generating the new mode is, the larger the complexity of the sequence is, and the larger the chaos degree of the signal is;

(4) and (3) performing functional connectivity feature extraction: any behavior or perception is produced by multiple regions of the brain, each function involving a network of elements that may be located in different lobes of the cerebral cortex and sub-cortical structures, thereby developing a network of functional connections; calculating a phase lag index PLI to study the relationship between different channels of the electroencephalogram signal, wherein the PLI is a functional connection method based on the phase relationship and is used for measuring the phase synchronization degree of signals of the two channels:

where N represents the length of the time series,

representing two-channel signals at time t_nA phase difference of (d); PLI has a value between 0 and 1, with a larger value indicating a stronger phase synchronization between the two signals.

In addition, the method of the present invention further comprises:

inputting the feature vectors of a test sample into a classifier for classification, wherein the test sample comprises the feature vectors of the sample and corresponding classes;

according to whether the feature vector of the test sample can be correctly judged as the category of the corresponding sample, obtaining the number (TP) correctly identified in all the positive samples, the number (FN) incorrectly identified in the positive samples, the probability (TN) correctly identified in the negative samples and the probability (FP) incorrectly identified in the negative samples;

calculating a comprehensive False Rejection Rate (FRR), a False Acceptance Rate (FAR) and a F1 score, and evaluating the identity recognition model by adopting three evaluation indexes of the False Rejection Rate (FRR), the False Acceptance Rate (FAR) and the F1 score, wherein the three indexes are as follows:

in the formula, TP is the number of correctly recognized positive samples, FN is the number of incorrectly recognized positive samples, TN is the number of correctly recognized negative samples, and FP is the number of incorrectly recognized negative samples.

In order to test the feasibility of the identity recognition method of the present embodiment, the present embodiment provides a research method for using electroencephalogram based on music-induced self-memory for identity recognition, which includes the following steps:

(1) recording the subjective fatigue degree of the testee in the implementation interval of each experiment; scalp electrode placement for multiple brain regions: the memory behavior relates to a plurality of brain areas, and the brain area acquisition electrodes on the brain scalp are used for acquiring corresponding brain electrical signals, amplifying the brain electrical signals, performing analog-to-digital conversion on the brain electrical signals and storing the brain electrical signals as digital signals.

(2) The tested eye is closed to follow the music for association and recall, electroencephalogram signals of 3 minutes in the period of the tested recall are recorded, and 5 related keywords are recorded to assist the music in defining the recall content of repeated experiments.

(3) Preprocessing the electroencephalogram signals, filtering the acquired electroencephalogram signals, reducing interference of power frequency signals, environmental noise, myoelectricity and the like, removing baseline drift, and removing ocular artifacts by Independent Component Analysis (ICA).

(4) And carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals in the time length of 1s, wherein each segment of signals is used as a sample.

(5) And (3) respectively extracting the characteristics of all samples from multiple dimensions such as frequency domain, nonlinear characteristics, functional connectivity characteristics and the like, and performing dimension reduction processing on the total characteristic vector by using Principal Component Analysis (PCA).

(6) Classifying the feature vectors, and observing classification results by integrating three evaluation indexes of the False Rejection Rate (FRR), the False Acceptance Rate (FAR) and the F1 score.

The experimental design strategy for inducing memory of the memory by the music in the step (2) is as follows: selecting the specific music familiar to the testee as an anchor point, considering both the experience of the testee and the number of samples, and adopting a mode of repeated experiments at intervals. Sitting on a chair for the first time in a relaxed way, listening to music and simultaneously performing association and recall after resting calmly by closing eyes, and recording a recall electroencephalogram signal with the duration of 3 minutes. When the rest is finished, 5 keywords about memory are recorded, wherein the keywords are mainly used for assisting music to define memory contents, deepen prints and facilitate subsequent repeated experiments. The experiment was repeated after sufficient rest to expand the database.

Extracting the frequency domain, nonlinear characteristics, functional connectivity characteristics and other multi-dimensional characteristics of the brain electrical signals preprocessed in the step (5), wherein a specific decomposition algorithm is as follows:

(1) and (3) extracting frequency domain features: the method comprises the steps of realizing an AR model by adopting a parameter model method, obtaining an AR coefficient, wherein the parameter model method generally assumes that a signal x (n) is output by exciting a linear system H (z) by a white noise sequence u (n), observing obtained data x (n) or an autocorrelation function thereof to estimate parameters of H (z), and estimating a power spectrum of x (n) by the parameters of H (z). Namely:

calculating power spectral density by using a modified periodogram algorithm (Welch method), and extracting power spectral density characteristics of 15 channels in 4 frequency bands, namely theta (Hz), alpha (8-13 Hz), beta (13-30 Hz) and gamma (30Hz) respectively;

after obtaining the power spectral density of each lead, the power difference PD of 6 pairs of signal channels symmetrically distributed in the left and right hemispheres is calculated:

wherein P is_leftIs the power of the left hemisphere channel, P_rightIs the power of the right hemisphere channel.

(2) And (3) extracting time-frequency domain features: discrete Wavelet Transform (DWT) is performed using DB4 wavelet basis in Daubechies wavelet family to decompose the signal into a weighted sum of wavelet basis. Wavelet transform is a time-frequency two-dimensional domain analysis method of signals, and shows the spectrum distribution of the signals at any time or in any short time. Based on the sampling rate of the original signal 512Hz, wavelet decomposition of 6 th order is carried out to obtain the wavelet coefficient of the corresponding frequency band. Meanwhile, in order to reduce the dimensionality of the feature vector, the average value and the mean square error of the absolute value of the wavelet coefficient and the Shannon entropy of the wavelet coefficient are further extracted to be used as the feature vector of the wavelet transformation.

(3) And (3) carrying out nonlinear feature extraction: electroencephalogram signals are complex and chaotic, and in addition to linear analysis of EEG, nonlinear research is also indispensable. The sample entropy measures the time sequence complexity by measuring the probability of generating a new pattern in the signal, and the larger the probability of generating the new pattern is, the greater the sequence complexity is, and the greater the signal chaos degree is.

(4) And (3) performing functional connectivity feature extraction: any behavior or perception is produced by multiple regions of the brain, and each function involves a network of elements that may be located in different lobes of the cerebral cortex and sub-cortical structures, thereby developing a network of functional connections. Calculating a phase lag index PLI to study the relationship between different channels of the electroencephalogram signal, wherein the PLI is a functional connection method based on the phase relationship and is used for measuring the phase synchronization degree of two channel signals:

where N represents the length of the time series,

representing two-channel signals at time t_nThe phase difference of (a). PLI has a value between 0 and 1, with a larger value indicating a stronger phase synchronization between the two signals.

And (3) classifying the feature vectors in the step (6), wherein due to the imbalance of the distribution of positive and negative samples, an integrated classifier based on an RUSBoost algorithm is adopted for feature classification, and meanwhile, the classification results of a Logistic Regression (LR) and a Support Vector Machine (SVM) are used as comparison, the comparison is set because the RUSBoost classifier is a sample more suitable for class imbalance, and the LR and the SVM are common ordinary classifiers. The specific decomposition algorithm of the RUSBoost is as follows:

setting normalized sample weight W (i) for all samples;

② defining the iteration number as T, and for T equal to 1,2, …, T:

randomly extracting a certain number of majority samples, forming a training data set S by all minority samples, and obtaining the weight S of the samples in S_wNormalizing it;

training a weak classifier h (t) by using the data set S according to the weight, wherein the output of h (t) is the probability of judgment, the probability of correct judgment of the ith sample of h (t) is p1(i), and the probability of wrong judgment is p2 (i);

calculating error e, e ═ Σ w (i) ([ 1-p1(i) + p2(i) ];

updating the weight:

and normalizing;

output integrated classifier

The input samples include the features of the samples and the corresponding labels (i.e. positive or negative samples, representing the tested identity), and the output of the classifier is the label that it judges on the feature vector, which can also be called the class. The classification result is that the classifier judges which class the sample belongs to according to the input sample characteristics, and whether the judgment is correct or not is known by comparing the judged class with the real class. In the identity recognition, one type is a positive sample, namely a sample of a designated user, a negative sample is a sample of other users, the classifier classifies the feature vector into the positive sample or the negative sample, and if the classification is correct, the recognition is correct.

The error rejection rate (FRR), the error acceptance rate (FAR) and the F1 score in the step (6) are three evaluation indexes: due to the unbalanced experimental data class, the conventional classifier evaluation index such as classification accuracy is no longer applicable, and the F1 score is one of the indexes for measuring the accuracy of the classification model, and it considers the accuracy and recall of the model. The False Rejection Rate (FRR) and the False Acceptance Rate (FAR) are important evaluations of the identity recognition system, and the three indexes are as follows:

in the existing research, the identity recognition experiment paradigm based on other mental tasks, such as calculation, memory, dictation and the like, is complex to operate, poor in popularization, even easy for people to fatigue, and poor in user experience, and has the beneficial effects that: the accuracy of the identity recognition model is guaranteed, meanwhile, the complexity of user operation in the verification process is reduced, the experience of a user is improved, and the user friendliness is enhanced. Meanwhile, the class imbalance of the training samples is improved in a targeted manner, the accuracy of the model is further improved, namely an integrated classifier based on the RUSBoost algorithm is used, and the evaluation index F1 score more suitable for the class imbalance samples is used.

As shown in FIG. 1, the overall scheme of the experiment is that the experiment process is tried to be reminisced by listening to specific music, and electroencephalograms are collected and stored by a Neuroscan64 device for subsequent analysis. In the experimental data analysis part, the original brain electrical signals are led into an EEGLAB tool package of Matlab software for pretreatment: the band-pass filter removes baseline drift and power frequency interference, and then an independent component analysis algorithm is used for removing ocular artifacts. Extracting the characteristics of the preprocessed signals from multiple dimensions, further reducing the dimensions of the characteristic vectors by using a principal component analysis algorithm, classifying by using an integrated classifier based on a RUSBoost algorithm, taking the results of a support vector machine classifier and a logistic regression classifier as reference, and calculating the FAR value, the FRR value and the F1 score value of the final result.

As shown in fig. 2, this figure is a brain scalp potential placement method. The invention adopts the 10-20 electrode lead positioning standard calibrated by the international electroencephalogram society, uses a binaural plumbing method, injects conductive paste to enhance the conductivity of the electrode, and correctly wears the electroencephalogram cap. Fifteen channels of FP1, FP2, F1, F2, T3, T4, C1, C2, P1, P2, O1, O2, Cz, Pz and Oz of different brain areas are selected to place electrodes so as to acquire brain electrical signals of different brain areas, and the channels M1 and M2 are taken as reference electrodes.

Fig. 3 is a schematic diagram of discrete wavelet transform and coefficient selection, where the wavelet transform is a time-frequency two-dimensional domain analysis method of signals, different from the short-time fourier transform, which always uses windows with the same width for different frequencies, and the wavelet transform uses a time-frequency local analysis method in which the window size (i.e., window area) is fixed but the shape can be changed, so that the wavelet transform has a lower time resolution and a higher frequency resolution in the low frequency part and a higher time resolution and a lower frequency resolution in the high frequency part, and is very suitable for extracting local features of electroencephalogram signals. In the present invention, Discrete Wavelet Transform (DWT) is performed using DB4 wavelet basis in the Daubechies wavelet family to decompose the signal into a weighted sum of wavelet basis. Since the original signal sampling rate is 512Hz, a wavelet decomposition of order 6 is performed to obtain wavelet coefficients of corresponding frequency bands, as shown in fig. 2, and in order to reduce the dimension of the feature vector, the obtained wavelet coefficients are not directly used, but the average value and the mean square error of absolute values and the shannon entropy of the wavelet coefficients are further extracted from the wavelet coefficients as the feature vector of the wavelet transform.

As shown in fig. 4, after the electroencephalogram signal is segmented, feature extraction is performed on fifteen signal channels of each segment of the electroencephalogram signal, wherein the feature extraction includes an AR model coefficient in a frequency domain, a Power Spectral Density (PSD), a left-right hemisphere Power Difference (PD), a Discrete Wavelet Transform (DWT) coefficient in a time-frequency domain, a nonlinear feature sample entropy (SampEn), and a functional connectivity feature Phase Lag Index (PLI), and a final feature vector is formed by the coefficients.

Fig. 5, 6 and 7 show the classification results of 12 experimental data tested on three control classifiers, respectively, fig. 5 is the FAR value, fig. 6 is the FRR value, and fig. 7 is the F1 score. From the figure, it can be seen that the experimental data can achieve excellent identification effect on three classifiers, the total F1 value of 12 subjects is mostly above 0.9, the self-passing memory is proved to contain clear personal characteristics, and the electroencephalogram signal in the process can be used as a personal identification technology. Comparing the results of the three classifiers at the same time, the F1 value of the RUSBoost is generally higher than that of the other two classifiers, and for data with unbalanced classes, the integrated classifier based on the RUSBoost algorithm is more friendly than the other two classifiers.

As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (10)

1. An identification method based on electroencephalogram is characterized by comprising the following steps:

inducing the testee to perform stable self-transmission memory, and collecting electroencephalogram signals of the testee in the process of performing stable self-transmission memory;

the electroencephalogram signals of the testee in the process of stable self-memory are used for identity recognition.

2. The electroencephalogram-based identity recognition method according to claim 1, wherein the method for inducing the subject to perform stable self-memory and collecting the electroencephalogram signals of the subject during the process of performing stable self-memory comprises the following steps:

recording the subjective fatigue degree of the testee in the implementation interval of each experiment; placing brain area collecting electrodes of multiple brain areas, and acquiring corresponding electroencephalogram signals by using the brain area collecting electrodes placed on the scalp of the brain;

enabling the testee to close eyes to follow the music for association and recall, recording electroencephalogram signals of the testee for 3 minutes in the recall period, and recording 5 related keywords for assisting the music to define the recall content of repeated experiments;

and (4) amplifying and performing analog-to-digital conversion on the electroencephalogram signals, and storing the electroencephalogram signals as digital signals.

3. The electroencephalogram-based identification method according to claim 2, wherein brain region acquisition electrodes for a plurality of brain regions are placed, and the method for acquiring corresponding electroencephalogram signals by using the brain region acquisition electrodes placed on the scalp of the brain comprises the following steps:

selecting a forehead control brain area and an occipital temporal sensation brain area related to music, memory and vision, recording leads FP1, FP2, F1, F2, T3, T4, C1, C2, P1, P2, O1, O2, Cz, Pz and Oz by adopting a 10-20 electrode lead positioning standard calibrated by the International electroencephalogram (EEG) society and using a binaural lappinging method, wherein the reference electrode selects M1 and M2, the sampling frequency is 512HZ, and the impedance of each channel lead is less than 5k omega;

the method for amplifying and performing analog-to-digital conversion on the electroencephalogram signal and storing the electroencephalogram signal as a digital signal comprises the following steps: electroencephalogram signals are collected by a Neuroscan64 device, amplified, subjected to analog-to-digital conversion and input into a computer.

4. The electroencephalogram-based identity recognition method according to claim 1, wherein the method for performing identity recognition by using electroencephalogram signals in a process of a subject performing stable self-memory comprises the following steps:

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the characteristic vector subjected to dimension reduction into an identity recognition model, and outputting the category of the characteristic vector, namely the identity information of the tested person;

the identity recognition model is obtained by training an integrated classifier based on the RUSBoost algorithm.

5. The electroencephalogram-based identity recognition method according to claim 4, wherein the training method of the identity recognition model comprises the following steps:

designing an experiment for inducing the memory of the self-transmission body by the music;

collecting brain waves in the experimental process;

amplifying and performing analog-to-digital conversion on the electroencephalogram signals, and storing the electroencephalogram signals as digital signals;

preprocessing the electroencephalogram signal of the digital signal;

carrying out non-overlapping segmentation on the preprocessed electroencephalogram signals for 1s, wherein each segment of signals is used as a sample;

respectively extracting features of all samples to obtain feature vectors of all samples, and performing dimensionality reduction on the feature vectors of all samples by using Principal Component Analysis (PCA);

inputting the feature vector subjected to dimension reduction into a training database;

and training the integrated classifier based on the RUSBoost algorithm by using a training database to obtain a trained identity recognition model.

6. The brain electricity-based identity recognition method of claim 5, wherein designing an experimental method for music-induced self-memory comprises:

selecting specific music familiar to the testee as an anchor point, considering both the experience of the testee and the number of samples, and adopting a mode of repeated experiments at intervals;

the first time of the test is relaxed and sits on a chair, after the eyes are closed and the rest is calm, the user can listen to music and simultaneously carry out association and recall, and the memory electroencephalogram signals with the duration of 3 minutes are recorded;

after the rest of the testee is finished, 5 keywords about memory are recorded, wherein the keywords are mainly used for assisting music to define memory contents, deepen prints and facilitate subsequent repeated experiments;

the experiment was repeated after full rest.

7. The electroencephalogram-based identification method according to claim 5, wherein the method for training the RUSBoost algorithm-based ensemble classifier using the training database comprises:

setting a normalized sample weight w (i) for all samples;

defining the number of iterations as T, for T ═ 1,2, …, T:

randomly extracting a certain number of majority samples, forming a training data set S by all minority samples, and obtaining the weight S of the samples in S_wNormalizing it;

training a weak classifier h (t) by using the data set S according to the weight, wherein the output of h (t) is the probability of judgment, the probability of correct judgment of the ith sample of h (t) is p1(i), and the probability of wrong judgment is p2 (i);

calculating an error e, e ═ Σ w (i) ([ 1-p1(i) + p2(i) ];

updating the weight:

and normalizing;

output integration classifier

W (i) is the updated weight, e is the error of the erroneous classification, h (t) is the weak classifier learned at each iteration, p1(i) is h (t) the probability of correct decision on the ith sample, and p2(i) is the probability of incorrect decision.

8. The electroencephalogram-based identity recognition method of claim 5, wherein the method for preprocessing the electroencephalogram signal of the digital signal comprises: filtering the acquired electroencephalogram signals, reducing interference of power frequency signals, environmental noise, myoelectricity and the like, removing baseline drift, and removing ocular artifacts by using Independent Component Analysis (ICA);

the method for filtering the acquired electroencephalogram signals comprises the following steps: and performing an FIR filter with the band pass of 0.5-40Hz on the acquired original electroencephalogram signals to remove baseline drift and high-frequency interference components.

9. The electroencephalogram-based identification method according to claim 5, wherein the method for respectively extracting the features of all samples comprises the following steps:

extracting frequency domain characteristics of the sample to obtain an AR model coefficient, power spectral density and power difference of a left hemisphere and a right hemisphere of a frequency domain;

extracting the time-frequency domain characteristics of the sample to obtain a discrete wavelet transform coefficient of a time-frequency domain;

carrying out nonlinear feature extraction on the sample to obtain a nonlinear feature sample entropy;

extracting functional connectivity characteristics of the sample to obtain a functional connectivity characteristic phase lag index;

and (3) forming the characteristic vector of the sample by the AR model coefficient of the frequency domain, the power spectral density, the power difference of the left hemisphere and the right hemisphere, the discrete wavelet transform coefficient of the time-frequency domain, the nonlinear characteristic sample entropy and the functional connectivity characteristic phase lag index.

10. The brain electricity based identity recognition method of claim 1, further comprising:

inputting the feature vectors of a test sample into a classifier for classification, wherein the test sample comprises the feature vectors of the sample and corresponding classes;

obtaining the number (TP) correctly identified in all the positive samples, the number (FN) incorrectly identified in the positive samples, the probability (TN) correctly identified in the negative samples and the probability (FP) incorrectly identified in the negative samples according to whether the feature vector of the test sample can be correctly judged as the category of the corresponding sample;

calculating a comprehensive False Rejection Rate (FRR), a False Acceptance Rate (FAR) and a F1 score, and evaluating the identity recognition model by adopting three evaluation indexes of the False Rejection Rate (FRR), the False Acceptance Rate (FAR) and the F1 score, wherein the three indexes are as follows:

in the formula, TP is the number of correctly recognized positive samples, FN is the number of incorrectly recognized positive samples, TN is the number of correctly recognized negative samples, and FP is the number of incorrectly recognized negative samples.

Images