CN107122643A - Personal identification method based on PPG signals and breath signal Fusion Features - Google Patents

Abstract

The invention discloses a kind of personal identification method based on PPG signals and breath signal Fusion Features, the problem of existing method discrimination is relatively low is mainly solved.Implementation step：1) training PPG signals and breath signal are pre-processed；2) carry out crest detection respectively to the signal after processing, obtain the waveform sample of two signals；3) the smooth non-negative matrix factorization method of utilization index sparse constraint extracts feature to two sample of signal respectively, obtains eigenmatrix；4) eigenmatrix of two signals is merged, obtains training ATL；5) test PPG samples and breath sample are projected respectively, obtains test feature；6) Fusion Features are carried out, fusion test sample is obtained；7) to test sample class prediction, it is identified result.The present invention, which can be generated, stablizes effective fusion feature sample.Simulation result shows that its discrimination reaches 100%, can be applied to medical, safe defence etc. and requires higher application field to identification accuracy rate.

Description

Identity recognition method based on feature fusion of PPG signal and respiratory signal

Technical Field

The invention belongs to the technical field of information processing, and particularly relates to an identity identification method based on feature fusion, which can be applied to the application fields of medical treatment, safety defense and the like.

Background

With the development of wireless networks, the wireless network applications such as remote medical treatment and electronic commerce play more and more important roles in the life of people. Security issues are particularly important as these applications involve important information about people, property, etc. While the traditional identity authentication carries out identification authorization through an identity card or a password, the security protection is not enough. The identity card or password information is easy to steal or forget, and the biological identification system has uniqueness, reliability, confidentiality and the like, and is widely applied. Some biological characteristic identification methods still have certain potential safety hazards, for example, fingerprints can be extracted by latex, face identification can be deceived by forged photos, sounds can also be simulated, and electroencephalogram or electrocardio signal methods cannot be widely used due to the fact that various electrodes are needed for collection.

The photoplethysmography (PPG) signal is acquired by a non-invasive means, is convenient and simple to acquire, reflects rich microcirculation physiological information of a human body, is a unique physiological characteristic of the human body, is difficult to copy and imitate, and has high safety. In addition, the respiratory signal contains various information such as the movement of internal organs of the human body and the like, and is mutually complemented with the human body information contained in the PPG signal. The identification rate of the existing identity recognition technology based on PPG signals is low, and the application occasions with high accuracy requirements are difficult to meet.

The identity recognition method based on the PPG signal is proposed as follows:

a method for extracting PPG Signal characteristics by using a kernel principal component analysis KPCA method to identify is provided in a publication of 'the fourth International Conference on Signal and Image Processing' in 2012 by N.S.Girish Rao Salanke, N.Maherswis and Andrews Samraj et al. The article analyzes the identification performance of the PPG signals in a stressed state and a relaxed state, but the identification rate of an individual is not specifically given, and the number of experimental subjects is small, so that the identification of the method cannot be fully displayed.

NI Mohammed Nadzr, M Sulaimi, LF Umadi, KA Sidek et al, 2016 in "Indian journal of Science and Technology" journal by "Photophysogram based biometric Recognition for Twins" using a PPG signal single cycle waveform for identification. The method comprises the steps of denoising an original PPG signal by using a low-pass filter, segmenting a PPG signal waveform, extracting a monocycle waveform, and identifying and classifying the monocycle waveform by using a radial basis function network and a naive Bayes classifier, wherein the final identity correct identification rate is more than 97%. The method verifies the effectiveness of the single-cycle waveform characteristics of the PPG signals on individual identification, but the characteristics of the single-cycle waveforms of the PPG signals are not fully mined and utilized, and the identification rate still has a rising space.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides an identity recognition method based on feature fusion of a PPG signal and a respiratory signal so as to improve the correct recognition rate of the identity.

The technical scheme for realizing the aim of the invention is that effective information of the respiratory signal and the PPG signal is mutually supplemented by respectively extracting the stable characteristics of the respiratory signal and the PPG signal for fusion, so as to generate the identity information fusion characteristic which contains the full and stable individual, and then identity recognition is carried out, and the implementation steps are as follows:

(1) separately for PPG training data sets X ═ X₁,X₂,…,X_nAnd a set of respiratory training data Y ═ Y₁,Y₂,…,Y_nDenoising and normalizing to obtain a normalized PPG signal set Z ═ Z₁,Z₂,…,Z_nAnd a set of normalized post-respiration signals R ═ R₁,R₂,…,R_nWhere n represents the total number of people;

(2) performing peak detection on the PPG signal set Z to obtain a PPG signal peak position set Loc ═ Loc of all persons₁,Loc₂,…,Loc_n}; performing peak detection on the respiratory signal set R to obtain a respiratory signal peak position set L ═ { L ═ of all people₁,L₂,…,L_n}；

(3) Taking all elements in the PPG signal peak position set Loc as datum points, and extracting a waveform sample of the PPG signal; taking all elements in the respiratory signal peak position set L as datum points, and extracting a waveform sample of the respiratory signal;

(4) removing samples with large difference in all human PPG waveform samples to obtain PPG training set M_p(ii) a Removing samples with large difference in all human respiratory signal waveform samples to obtain a respiratory training set M_R；

(5) Respectively carrying out smoothing nonnegative matrix decomposition on PPG training set M by using exponential sparse constraint_pAnd respiratory training set M_RExtracting the features to obtain a PPG feature matrix C_PPPG projection matrix W_PRespiratory characteristic matrix C_RAnd the respiratory projection matrix W_R；

(6) Using PPG feature matrix C_PEach column in (1) and a respiratory feature matrix C_RThe corresponding columns are connected in series to generate a fusion training template, and a training template library F is formed by all the fusion training templates;

(7) using PPG projection matrix W_PAnd the respiratory projection matrix W_RObtaining PPG test feature matrix T_PAnd respiratory characteristics test matrix T_R；

(8) Testing the PPG characteristic matrix T_PEach column and breath test feature matrix T_RConnecting corresponding columns in series to obtain a fusion characteristic column, taking the fusion characteristic column as a fusion test sample, and forming a test sample library Lb by all the fusion test samples;

(9) and performing class prediction on all the fusion test samples in the sample library Lb to obtain an identity recognition result of the identified person.

Compared with the prior art, the invention has the following advantages:

firstly, the invention utilizes the exponential sparse constraint smooth nonnegative matrix decomposition method to extract the characteristics of the PPG signal waveform sample and the respiratory signal waveform sample, fully obtains the effective information of the PPG signal and the effective information of the respiratory signal, and improves the accuracy of identity identification.

Secondly, the invention extracts the main characteristics of the PPG signal and the respiration signal waveform sample of the individual and fuses the PPG signal characteristics and the respiration signal characteristics, so that the effective information of the PPG signal and the effective information of the respiration signal are mutually supplemented, the fusion characteristics are obtained, the identity recognition is carried out by utilizing the fusion characteristics, and the accuracy of the identity recognition is improved.

Drawings

FIG. 1 is a flow chart of an implementation of the present invention;

FIG. 2 is a diagram of the identification rate results of the MIMIC database;

fig. 3 is a graph of the identification rate results of the MIMIC2 database.

Detailed Description

The following describes the embodiments and effects of the present invention in further detail with reference to the accompanying drawings.

Referring to fig. 1, the implementation steps of the invention are as follows:

step 1, preprocessing training data.

The commonly disclosed PPG signal database comprises a MIMIC database, a MIMIC2 database and the like, the example selects PPG signals and respiratory signals of 50 persons in the MIMIC database, wherein the PPG signals with the length of 400 seconds and the respiratory signals with the length of 400 seconds of each person are taken as training data, the sampling frequency f of the PPG signals and the respiratory signals is 125Hz, the length of the PPG signals data of each person is 400 × 125 — 50000, the length of the respiratory signals data is 400 × 125 — 50000, and a PPG training data set X is formed by the PPG signals of 50 persons { X ═ 50000₁,X₂,…,X_i,…,X_nAnd forming a respiratory training data set Y (Y) by respiratory signals of 50 persons₁,Y₂,…,Y_i,…,Y_nIn which X_iPPG signal data, Y, representing the ith person_iThe PPG training data set X and the respiratory training data set Y are preprocessed, the respiratory signal data representing the ith individual, i ═ 1,2, …, n, n represents headcount:

respectively carrying out denoising and normalization processing on the PPG training data set X and the respiratory training set Y in sequence to enable normalizationAll subsequent sampling points take values in the interval [0,1]Obtaining a normalized PPG signal set Z ═ { Z ═ Z₁,Z₂,…,Z_nAnd a set of normalized breathing signals R ═ R₁,R₂,…,R_n}；

Common methods for denoising the PPG signal and the respiratory signal include low-pass filtering, wavelet transform denoising, adaptive morphological filtering, and the like, and the present example adopts a method without limitation to low-pass filtering.

Step 2. normalizing PPG signal Z for ith person_iPerforming peak detection to obtain a PPG signal peak position sequence Loc of the ith person_iAnd obtaining a PPG signal peak position set Loc formed by the peak position sequence of all person PPG signals₁,Loc₂,…,Loc_i,…,Loc_n}; normalized respiration signal R for the ith person_iPerforming peak detection to obtain the peak position sequence L of the ith person's respiratory signal_iAnd obtaining a respiratory signal peak position set L ═ { L ═ L consisting of all human respiratory signal peak position sequences₁,L₂,…,L_i,…,L_n}。

The peak detection of the PPG signal and the respiration signal adopts a dynamic differential threshold peak detection method, and introduces a text of 'pulse signal peak detection algorithm based on dynamic differential threshold' published in the engineering edition of the university of Jilin university school in the year 2014 of Youn-new of Zhang Aiwa, Wang and Chou.

And 3, acquiring a PPG signal waveform sample and a respiratory signal waveform sample.

(3a) With the peak position sequence Loc of PPG signal of the ith person_iIs a reference point, in the PPG signal Z of the ith person_iBefore a of each reference point₁A sampling point and a₂A sampling point and a reference point consisting of₁+a₂+1) sampling points constitute one PPG signal waveform sample, and the acquisition and sequence Loc_iAnd (b) obtaining PPG signal waveform samples with the same number of middle elements, and further obtaining PPG signal waveform samples of all people, wherein a₁，a₂Is two positive integers;

(3b) with the sequence L of the ith individual's breathing signal positions_iIs a reference point, the respiration signal R of the ith person_iBefore b is extracted from each reference point₁A sampling point and b₂A sampling point and a reference point consisting of₁+b₂+1) sampling points constitute a respiratory signal waveform sample, and the obtained respiratory signal waveform sample and sequence L_iAnd the respiratory signal waveform samples with the same number of the medium elements are obtained, and then the respiratory signal waveform samples of all people are obtained, wherein b₁，b₂Are two positive integers.

And 4, removing the waveform samples with large differences to obtain a training set.

(4a) Performing waveform averaging on all PPG signal waveform samples of the ith person to obtain an average waveform sample, calculating Euclidean distances between each waveform sample of the ith person and the average waveform sample, arranging all Euclidean distances from small to large, and selecting the first m_iWaveform samples corresponding to the Euclidean distances take the selected waveform samples as column vectors, and all the selected human PPG signal waveform samples form a PPG training set M_p＝{SP₁,SP₂,…,SP_gnIn which m is_iIs a positive integer and is a non-zero integer,is shown (a)₁+a₂Real vector space of +1) dimension, gn representing PPG training set M_pThe number of the middle samples;

(4b) performing waveform averaging on all respiratory signal waveform samples of the ith person to obtain an average waveform sample, calculating Euclidean distances between each waveform sample of the ith person and the average waveform sample, arranging all Euclidean distances from small to large, and selecting the first m_iThe waveform samples corresponding to the Euclidean distances take the selected waveform samples as column vectors, and the selected waveform samples of the respiratory signals of all people form a respiratory training set M_R＝{SR₁,SR₂,…,SR_gn}，Wherein,is represented by (b)₁+b₂A real number vector space of +1) dimension.

Step 5, respectively aligning PPG training set M_pAnd carrying out feature extraction on the respiratory training set MR to obtain a PPG feature matrix C_PPPG projection matrix W_PRespiratory characteristic matrix C_RAnd the respiratory projection matrix W_R。

(5a) In order to effectively improve the sparsity of the decomposition result of the non-smooth non-negative matrix decomposition method and the interpretability of local features and reduce decomposition errors, index sparsity constraint is added into a target function of the non-smooth non-negative matrix decomposition method to obtain an index sparsity constraint target function of the smooth non-negative matrix decomposition method:

the constraint conditions need to be satisfied:

W^TW＝I,S^TS＝I，

wherein V represents a data matrix to be decomposed, W represents a base matrix, S represents a smoothing matrix, H represents a coefficient matrix, and V_qsThe element representing the qth row and the s column of the matrix V to be decomposed, W_κDenotes the k-th column of the matrix W, S_vRepresents the v-th column of the matrix S, | · |. non-woven phosphor₂The L2 norm, β, λ representing the vector represent two constraint parameters, (. cndot.) respectively^TRepresents a transpose of a matrix, I represents an identity matrix;

(5b) method for decomposing PPG training set M by using exponential sparse constraint smooth nonnegative matrix_pDecomposing to obtain PPG characteristic matrix C_PAnd PPG projection matrix W_PThe method comprises the following specific operation steps:

(5b1) randomly initializing a base matrix W⁽⁰⁾Smoothing matrix S⁽⁰⁾Sum coefficient matrix H⁽⁰⁾All elements are within the interval (0,1), wherein,andrespectively represent (a)₁+a₂+1)×r₁Vitamin, vitamin R₁×r₁A sum of₁× gn-dimension real number matrix space, r₁A decomposition dimension representing the PPG signal;

(5b2) according to the following iterative formula, the base matrix W is processed^(t)Of (2) element(s)Updating:

firstly, the intermediate variable value is obtained by updating according to the following formula

Then, the intermediate variable value is centeredPerforming column normalization to obtain

Wherein,representing the base after t iterationsMatrix W^(t)To (1) aLine θ column element, t represents the number of iterations, t ∈ [1, iter]Iter is a predefined maximum number of iterations,θ＝1,2,…,r₁，represents the basis matrix W after t-1 iterations^(t-1)To (1) aLine theta column element, S^(t-1)Represents the smoothing matrix after t-1 iterations, H^(t-1)Representing a coefficient matrix after t-1 times of iteration, and e representing a natural constant;

(5b3) according to the following iterative formula, for the coefficient matrix H^(t)Of (2) element(s)Updating:

wherein,representing the coefficient matrix H after t iterations^(t)Line theta ofThe elements of the column are, in turn,representing the coefficient matrix H after t-1 iterations^(t-1)Line theta ofA column element;

(5b4) for the smooth matrix S according to the following formula^(t)Of (2) element(s)Updating:

firstly, the intermediate variable value is obtained by updating according to the following formula

Then, for the intermediate variable valueColumn normalization processing to obtain

Wherein,representing the smoothing matrix S after t iterations^(t)Row u, column v, u, v ═ 1,2, …, r₁，Smoothing matrix S after iterating t-1 times^(t-1)Row u and column v elements of (1);

(5b5) repeating the steps (5b2) - (5b4), stopping iteration when the maximum iteration number iter is reached, and outputting the base matrix W^(iter)Sum coefficient matrix H^(iter)(ii) a Will radicalMatrix W^(iter)As PPG projection matrix W_PCoefficient matrix H^(iter)As PPG feature matrix C_PWherein

(5b) according to step (5a), training set M of breaths_RDecomposing to obtain a respiratory characteristic matrix C_RAnd the respiratory projection matrix W_RWhereinandrespectively represent r₂× gn and (b)₁+b₂+1)×r₂Real matrix space of dimension, r₂Representing the respiratory signal decomposition dimension.

Step 6, PPG feature matrix C_PEach column of (2) and a respiratory feature matrix C_RThe corresponding columns in the system are connected in series one to generate a fusion training template; forming a training template library F by all the fusion training templates;

step 7, PPG test matrix T is obtained_PAnd a breath test matrix T_R。

(7a) Respectively carrying out the steps 1-4 processing on the PPG signal test data and the respiratory signal test data of the person to be identified to obtain a PPG test set P and a respiratory test set Q, wherein,andrespectively represent (a)₁+a₂+1) × G and (b)₁+b₂+1) × G-dimensional real matrix space, G representing the number of PPG signal test samples;

(7b) using PPG projection matrix W_PFor PPG test set PExtracting the characteristics to obtain a PPG test characteristic matrix T_P＝inv((W_P)^T×W_P)×(W_P)^T× P, wherein,is represented by r₁× G-dimensional real matrix space, inv (·) represents matrix inversion operation;

(7c) using a breathing projection matrix W_RExtracting the characteristics of the breath test set Q to obtain a breath test characteristic matrix T_R＝inv((W_R)^T×W_R)×(W_R)^T× Q, whereinIs represented by r₂× G-dimensional real matrix space.

Step 8, testing the PPG characteristic matrix T_PEach column of (1) and breath test feature matrix T_RAnd connecting the corresponding columns in series to obtain a fusion characteristic column, taking the fusion characteristic column as a fusion test sample, and forming a test sample library Lb by all the fusion test samples.

And 9, carrying out identity recognition by using the K neighbor classifier.

(9a) Calculating Euclidean distances between each fusion test sample in the test sample library Lb and all training templates in the training template library, arranging all Euclidean distances from small to large, selecting the training templates corresponding to the first K Euclidean distances, then counting the occurrence times of each category in the K training templates, and selecting the category with the maximum occurrence frequency as a prediction category of the fusion test sample, wherein K is a positive integer;

(9b) according to the step (9a), performing class prediction on all the fusion test samples in the test sample library Lb, counting the occurrence frequency of each prediction class in all the fusion test samples, and taking the class with the highest occurrence frequency as the identity of the identified person.

The classifier for performing class prediction on the fusion test sample is not limited to a K-nearest neighbor classifier, and a Support Vector Machine (SVM) classifier, a Bayesian classifier and the like can be selected.

The effects of the present invention can be further explained by the following simulations.

1. Simulation conditions

The simulation experiment of the invention respectively uses PPG signals and respiratory signals in two public databases MIMIC and MIMIC2 as experimental data to simulate PPG signals and respiratory signals collected from a human body, and is carried out on a computer with Intel Pentium E58003.2GHz CPU and 2GB memory.

2. Emulated content

Firstly, randomly selecting 50 persons of PPG signals and respiratory signals from databases MIMIC and MIMIC2, respectively, using the method of the present invention to identify each person in the two databases, respectively, calculating the identification rate of each person, and obtaining the identification rate curve chart of each database, as shown in fig. 2 and fig. 3, wherein the identification rate calculation method of each person is as follows:

the identity recognition rate is the number of fusion test samples with correct category prediction/the total number of test samples of the appraised person;

then, the average value of the identification rates of all persons in each database is taken as the identification rate of the database.

As can be seen from FIGS. 2 and 3, the average identification rate of each library reaches 100%, which fully illustrates the effectiveness and high correct identification rate of the present invention.

The foregoing description is only an example of the present invention and should not be construed as limiting the invention, as it will be apparent to those skilled in the art that various changes and modifications in form and detail may be made therein without departing from the principles and structures of the invention, but such changes and modifications are within the scope of the invention as defined by the appended claims.

Claims (8)

1. An identity recognition method based on feature fusion of PPG signals and respiratory signals comprises the following steps:

(1) separately for PPG training data sets X ═ X₁,X₂,…,X_nAnd a set of respiratory training data Y ═ Y₁,Y₂,…,Y_nDenoising and normalizing to obtain a normalized PPG signal set Z ═ Z₁,Z₂,…,Z_nAnd a set of normalized post-respiration signals R ═ R₁,R₂,…,R_nWhere n represents the total number of people;

(2) performing peak detection on the PPG signal set Z to obtain a PPG signal peak position set Loc ═ Loc of all persons₁,Loc₂,…,Loc_n}; performing peak detection on the respiratory signal set R to obtain a respiratory signal peak position set L ═ { L ═ of all people₁,L₂,…,L_n}；

(3) Taking all elements in the PPG signal peak position set Loc as datum points, and extracting a waveform sample of the PPG signal; taking all elements in the respiratory signal peak position set L as datum points, and extracting a waveform sample of the respiratory signal;

(4) removing samples with large difference in all human PPG waveform samples to obtain PPG training set M_p(ii) a Removing samples with large difference in all human respiratory signal waveform samples to obtain a respiratory training set M_R；

(5) Respectively carrying out smoothing nonnegative matrix decomposition on PPG training set M by using exponential sparse constraint_pAnd respiratory training set M_RExtracting the features to obtain a PPG feature matrix C_PPPG projection matrix W_PRespiratory characteristic matrix C_RAnd the respiratory projection matrix W_R；

(6) Using PPG feature matrix C_PEach column in (1) and a respiratory feature matrix C_RThe corresponding columns are connected in series to generate a fusion training template, and a training template library F is formed by all the fusion training templates;

(7) using PPG projection matrix W_PAnd the respiratory projection matrix W_RObtaining PPG test feature matrix T_PAnd respiratory characteristics test matrix T_R；

(8) Testing the PPG characteristic matrix T_PEach column and breath test feature matrix T_RConnecting corresponding columns in series to obtain a fusion characteristic column, taking the fusion characteristic column as a fusion test sample, and forming a test sample library Lb by all the fusion test samples;

(9) and performing class prediction on all the fusion test samples in the sample library Lb to obtain an identity recognition result of the identified person.

2. Root of herbaceous plantThe method of claim 1, wherein the extracting of the PPG signal waveform samples in step (3) is performed as a sequence of PPG signal peak positions Loc of the ith person_iIs a reference point, in the PPG signal Z of the ith person_iBefore a of each reference point₁A sampling point and a₂A sampling point and a reference point, from which (a)₁+a₂+1) sampling points constitute one PPG signal waveform sample, and the acquisition and sequence Loc_iPPG signal waveform samples with the same number of middle elements are obtained, and then PPG signal waveform samples of all people are obtained; wherein, a₁，a₂Is two positive integers, i is 1,2, …, n, n represents the total number of people.

3. The method of claim 1, wherein the step (3) of extracting the waveform samples of the respiration signal is a sequence L of respiration signal positions of the ith person_iIs a reference point, the respiration signal R of the ith person_iBefore b is extracted from each reference point₁A sampling point and b₂A sampling point and a reference point, from which (b)₁+b₂+1) sampling points constitute a respiratory signal waveform sample, and the obtained respiratory signal waveform sample and sequence L_iThe respiratory signal waveform samples with the same number of the middle elements are obtained, and then the respiratory signal waveform samples of all people are obtained; wherein, b₁，b₂Are two positive integers.

4. The method of claim 1, wherein step (4) acquires a PPG training set M_pThe method comprises the following specific operations:

performing waveform averaging on all PPG signal waveform samples of the ith person to obtain an average waveform sample, calculating Euclidean distances between each waveform sample of the ith person and the average waveform sample, arranging all Euclidean distances from small to large, and selecting the first m_iWaveform samples corresponding to the Euclidean distances take the selected waveform samples as column vectors, and PPG training sets M are formed by the selected PPG signal waveform samples of all persons_p＝{SP₁,SP₂,…,SP_gnIn which m is_iIs a positive integer and is a non-zero integer,is shown (a)₁+a₂Real vector space of +1) dimension, gn representing PPG training set M_pThe number of samples in the sample.

5. The method of claim 1, wherein step (4) acquires a respiratory training set M_RThe method comprises the following specific operations:

performing waveform averaging on all respiratory signal waveform samples of the ith person to obtain an average waveform sample, calculating Euclidean distances between each waveform sample of the ith person and the average waveform sample, arranging all Euclidean distances from small to large, and selecting the first m_iThe waveform samples corresponding to the Euclidean distances take the selected waveform samples as column vectors, and the selected waveform samples of the respiratory signals of all people form a respiratory training set M_R＝{SR₁,SR₂,…,SR_gnIn which m is_iIs a positive integer and is a non-zero integer,is represented by (b)₁+b₂A real number vector space of +1) dimension.

6. The method according to claim 1, wherein step (5) separately applies the exponential sparsity constrained smooth non-negative matrix factorization method to the PPG training set M_pAnd respiratory training set M_RPerforming feature extraction according to the following steps:

(5a) randomly initializing a base matrix W⁽⁰⁾Smoothing matrix S⁽⁰⁾Sum coefficient matrix H⁽⁰⁾；

(5b) According to the following iterative formula, the base matrix W is processed^(t)Of (2) element(s)Updating:

firstly, the intermediate variable value is obtained by updating according to the following formula

Then, the intermediate variable value is centeredNormalizing the row to obtain

Wherein,representing the basis matrix W after t iterations^(t)To (1) aLine θ column element, t represents the number of iterations, t ∈ [1, iter]Iter is a predefined maximum number of iterations,θ＝1,2,…,r，r＝r₁or r₂，r₁Representing the PPG signal decomposition dimension, r₂Representing the respiratory signal decomposition dimension, V representing the matrix to be decomposed, V ═ M_POr M_R，Represents the basis matrix W after t-1 iterations^(t-1)First, theColumn theta of rowsElement of (1), S^(t-1)Represents the smoothing matrix after t-1 iterations, H^(t-1)Coefficient matrix after t-1 iterations is represented, β represents constraint parameters, (-)^TDenotes the transpose of the matrix, e denotes the natural constant;

(5c) according to the following formula, for coefficient matrix H^(t)Of (2) element(s)Updating:

wherein,representing the coefficient matrix H after t iterations^(t)Line thetaThe elements of the column are, in turn,representing the coefficient matrix H after t-1 iterations^(t-1)Line theta ofColumn elements, λ represents a constraint parameter;

(5d) for the smooth matrix S according to the following formula^(t)Of (2) element(s)Updating:

firstly, the intermediate variable value is obtained by updating according to the following formula

<mrow> <msubsup> <mi>S</mi> <mrow> <mi>u</mi> <mi>v</mi> </mrow> <mrow> <mo>&amp;prime;</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </msubsup> <mo>=</mo> <msubsup> <mi>S</mi> <mrow> <mi>u</mi> <mi>v</mi> </mrow> <mrow> <mo>(</mo> <mi>t</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </msubsup> <mfrac> <msub> <mrow> <mo>(</mo> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mi>T</mi> </mrow> </msup> <msup> <mi>VH</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mi>T</mi> </mrow> </msup> <mo>)</mo> </mrow> <mi>w</mi> </msub> <mrow> <msub> <mrow> <mo>(</mo> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mi>T</mi> </mrow> </msup> <msup> <mi>W</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </msup> <msup> <mi>S</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> </msup> <msup> <mi>H</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </msup> <msup> <mi>H</mi> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> <mi>T</mi> </mrow> </msup> <mo>)</mo> </mrow> <mrow> <mi>u</mi> <mi>v</mi> </mrow> </msub> <mo>+</mo> <msubsup> <mi>&amp;lambda;eS</mi> <mrow> <mi>u</mi> <mi>v</mi> </mrow> <mrow> <mo>(</mo> <mi>t</mi> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </msubsup> </mrow> </mfrac> <mo>,</mo> </mrow>

Then, for the intermediate variable valueColumn normalization to obtain

<mrow> <msubsup> <mi>S</mi> <mrow> <mi>u</mi> <mi>v</mi> </mrow> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </msubsup> <mo>=</mo> <mfrac> <msubsup> <mi>S</mi> <mrow> <mi>u</mi> <mi>v</mi> </mrow> <mrow> <mo>&amp;prime;</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </msubsup> <mrow> <munder> <mo>&amp;Sigma;</mo> <mi>&amp;rho;</mi> </munder> <msubsup> <mi>S</mi> <mrow> <mi>&amp;rho;</mi> <mi>v</mi> </mrow> <mrow> <mo>&amp;prime;</mo> <mrow> <mo>(</mo> <mi>t</mi> <mo>)</mo> </mrow> </mrow> </msubsup> </mrow> </mfrac> <mo>,</mo> </mrow>2

Wherein,representing the smoothing matrix S after t iterations^(t)Row u, column v element u, v 1,2, …,r₁，Smoothing matrix S after iteration t-1 times^(t-1)Row u and column v elements of (1);

(5e) repeating the steps (5b) to (5d), stopping iteration when the maximum iteration number iter is reached, and outputting the base matrix W^(iter)Sum coefficient matrix H^(iter)(ii) a When the matrix V to be decomposed is PPG training set M_PThen, the base matrix W^(iter)As PPG projection matrix W_PCoefficient matrix H^(iter)As PPG feature matrix C_P(ii) a When the matrix V to be decomposed is a breathing training set M_RThen, the base matrix W^(iter)As a breathing characteristic matrix C_RCoefficient matrix H^(iter)As a breathing projection matrix W_R。

7. The method according to claim 1, wherein step (7) acquires the PPG test feature matrix T_PAnd breath test feature matrix T_RThe method comprises the following specific operations:

(7a) using PPG projection matrix W_PObtaining PPG test feature matrix T_P＝inv((W_P)^T×W_P)×(W_P)^T× P, where inv (·) represents a matrix inversion operation and P represents a PPG test set;

(7b) using a breathing projection matrix W_RObtaining a breath test feature matrix T_R＝inv((W_R)^T×W_R)×(W_R)^T× Q, where Q represents a breath test set.

8. The method of claim 1, wherein the step (9) of obtaining the identification result of the authenticated person is performed according to the following steps:

(9a) calculating Euclidean distances between each fusion test sample in the test sample library Lb and all training templates in the training template library, arranging all Euclidean distances from small to large, selecting the training templates corresponding to the first K Euclidean distances, then counting the occurrence times of each category in the K training templates, and selecting the category with the maximum occurrence frequency as a prediction category of the fusion test sample, wherein K is a positive integer;

(9b) and (4) according to the step (9a), performing class prediction on all fused test samples in the test sample library Lb, counting the occurrence frequency of each prediction class in all test samples, and taking the class with the highest occurrence frequency as the identity of the identified person.