PPG signal identity recognition method and system based on singular value decomposition

Abstract

The present invention belongs to the technical field of PPG signal biometric identification, and provides a PPG signal identity identification method and system based on singular value decomposition. The present invention uses the singular value decomposition method to obtain the corresponding feature vector of the PPG signal, reduces signal noise, uses the constructed multiple features and their grouping to obtain highly discriminative features in the PPG signal, and makes full use of the correlation information between different PPG signals to further enhance the ability of the PPG signal identity identification method, thereby improving the robustness and performance of the PPG signal identity identification system; by calculating the Euclidean distance between the feature vector and each template in a preset user PPG vector template library, the template with the smallest distance value and reaching a certain threshold is used as the user to be identified, thereby achieving the purpose of PPG signal identity identification.

Description

Technical Field

[0001] The present invention belongs to the field of PPG signal biometric identification technology, and in particular relates to a PPG signal identity recognition method and system based on singular value decomposition. Background Art

[0002] In recent years, biometrics such as face, voice, iris, electroencephalogram (EEG), electrocardiogram (ECG), lip prints, gait, photoplethysmography (PPG), fingerprints, and palm and finger veins have gained increasing attention and achieved promising results. However, face recognition can be disguised through artificial disguise, voice can be easily imitated, fingerprints can be modified with latex, and EEG and ECG-based methods are very complex to acquire biometric signals. PPG, on the other hand, is a non-invasive photoelectric signal that can be acquired at low cost from locations such as the fingertip, wrist, or earlobe. As a new biometric recognition technology, PPG's universality, uniqueness, robustness, and adaptability have been verified by research, and PPG signals have been shown to have the ability to distinguish individuals. Therefore, PPG biometric feature extraction and recognition are gaining increasing attention.

[0003] The inventors have discovered that feature extraction methods for PPG biometric identification have been extensively researched, with some promising results. However, some of these PPG biometric extraction methods lack in-depth analysis of local features of PPG signals, are time-consuming, and require complex hyperparameter adjustments, making them unsuitable for training with small-scale data. Furthermore, existing PPG signal identification methods consider only a limited range of internal structural characteristics and fail to fully utilize the correlation information between different PPG signals. Due to the intra-individual variability of PPG signals, the accuracy of existing PPG signal identification methods is relatively low. Summary of the Invention

[0004] In order to solve the above problems, the present invention proposes a PPG signal identity identification method and system based on singular value decomposition, which is a PPG signal identity identification method based on singular value decomposition and multi-feature discriminant analysis. The singular value decomposition method is used to reduce signal noise, and the constructed multiple features and their grouping are used to obtain highly discriminative features in the PPG signal. The correlation information between different PPG signals is fully utilized to further enhance the capability of the PPG signal identity identification method, thereby improving the robustness and performance of the PPG signal identity identification system.

[0005] In order to achieve the above object, the present invention is implemented through the following technical solutions:

[0006] In a first aspect, the present invention provides a PPG signal identity recognition method based on singular value decomposition, comprising:

[0007] Get PPG signal;

[0008] Using the projection matrix, the corresponding eigenvector of the PPG signal is obtained;

[0009] Calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and take the template with the smallest distance value that reaches a certain threshold as the user to be identified;

[0010] Among them, the construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; based on the single cycle segmentation results, constructing multiple features of the PPG signals in the training set and grouping them to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library.

[0011] Furthermore, the PPG signal is pre-processed including:

[0012] Get PPG signal;

[0013] Construct the Hankel matrix based on the number of samples in the acquired PPG signal;

[0014] Perform singular value decomposition on the constructed Hankel matrix to obtain a set of singular values;

[0015] Process the singular value set, calculate the mean of the singular values, retain the singular values ​​greater than or equal to the mean, and set the remaining singular values ​​to zero, and obtain;

[0016] According to the processed singular value set, perform the inverse operation of singular value decomposition to obtain a new Hankel matrix;

[0017] The anti-diagonal elements of the new Hankel matrix are averaged to obtain the preprocessed PPG signal.

[0018] Further, a peak point in the preprocessed PPG signal is determined as a reference point;

[0019] Determine the nearest valley point before the benchmark point and the nearest valley point after the benchmark point;

[0020] The length of each single-cycle waveform is the sum of the sampling points between the two valley points and the reference point plus 1;

[0021] All single-cycle waveforms are normalized.

[0022] Furthermore, the multiple features include time domain features, three-layer sparse representation features and local mean features of PPG signals.

[0023] Furthermore, the time domain features, three-layer sparse representation features, and local mean features of the single-cycle PPG signal are serially fused, and the templates of all individuals after the fusion of multiple features are calculated using the average method;

[0024] Use Euclidean distance to calculate the difference between multiple features and templates, and sort the corresponding PPG signals according to the size of the difference;

[0025] Divide the sorted PPG signals into several groups in sequence.

[0026] Furthermore, the projection matrix is ​​used to project the test samples of each group, and the multiple sets of features of each sample are combined to obtain the test set projection matrix.

[0027] Furthermore, the corresponding feature submatrix is ​​obtained for each individual's training sample and test sample through the projection matrix, and the distance between each eigenvector in the feature submatrix and other eigenvectors is calculated through the Euclidean distance to obtain the homology similarity matrix; the Euclidean distance between the eigenvector of the test sample of the PPG signal and each template in the user PPG vector template library is calculated, and the template with the smallest distance value that reaches the threshold is used as the user to be identified.

[0028] In a second aspect, the present invention further provides a PPG signal identity recognition system based on singular value decomposition, comprising:

[0029] The data acquisition module is configured to: acquire PPG signals;

[0030] The eigenvector construction module is configured to: obtain the eigenvector corresponding to the PPG signal using the projection matrix;

[0031] The identity recognition module is configured to: calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and select the template with the smallest distance value that reaches a certain threshold as the user to be identified;

[0032] Among them, the construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; based on the single cycle segmentation results, constructing multiple features of the PPG signals in the training set and grouping them to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library.

[0033] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the PPG signal identity identification method based on singular value decomposition described in the first aspect.

[0034] In a fourth aspect, the present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the PPG signal identity recognition method based on singular value decomposition described in the first aspect are implemented.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. The present invention utilizes the singular value decomposition method to obtain the corresponding feature vector of the PPG signal, thereby reducing signal noise. The constructed multiple features and their grouping are used to obtain highly discriminative features in the PPG signal. Furthermore, the correlation information between different PPG signals is fully utilized to further enhance the ability of the PPG signal identity recognition method, thereby improving the robustness and performance of the PPG signal identity recognition system. By calculating the Euclidean distance between the feature vector and each template in a preset user PPG vector template library, the template with the smallest distance value that reaches a certain threshold is selected as the user to be identified, thereby achieving the purpose of PPG signal identity recognition.

[0037] 2. The present invention uses the inter-individual scatter matrix and the intra-individual scatter matrix to obtain a linear discriminant analysis model in a multi-feature group shared space. By solving the optimal projection vector, multiple PPG signal multi-feature group data with differences are uniformly projected into a common discriminant subspace, so that the PPG signal samples of the same individual are clustered together after projection, and the samples of different individuals are dispersed after projection. In this way, the final solved projection vector is more distinguishable and discriminative, which is also conducive to extracting the main features of the PPG signal, thereby improving the recognition rate of the PPG signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of the specification of this embodiment are used to provide a further understanding of this embodiment. The schematic embodiments and descriptions of this embodiment are used to explain this embodiment and do not constitute an improper limitation on this embodiment.

[0039] Figure 1 This is a block diagram of the identity recognition process of Example 1 of the present invention;

[0040] Figure 2 1 is a typical two-cycle PPG signal waveform diagram of Example 1 of the present invention. DETAILED DESCRIPTION

[0041] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0042] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present application belongs.

[0043] Example 1:

[0044] This embodiment provides a PPG signal identity recognition method based on singular value decomposition, including:

[0045] Acquire a PPG signal; in this embodiment, it can be understood that the PPG signal acquired at this time is a signal to be identified or a PPG signal in the test set in this embodiment;

[0046] Using the projection matrix, the corresponding eigenvector of the PPG signal is obtained;

[0047] Calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and take the template with the smallest distance value that reaches a certain threshold as the user to be identified;

[0048] The construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; constructing multiple features of the PPG signals in the training set and grouping them according to the single cycle segmentation results to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library. Figure 1 As shown, the PPG signal identity recognition method based on singular value decomposition in this embodiment specifically includes:

[0049] S1. Preprocessing the PPG signal using the singular value decomposition method;

[0050] S2, segmenting the pre-processed PPG signal into single cycles;

[0051] S3. Based on the single-cycle segmentation results, construct multiple features of the PPG signal and group them to obtain a multi-feature group;

[0052] S4. Perform discriminant analysis on the multi-feature group of the PPG signal, calculate the projection matrix of all training samples, and construct a multi-feature module library of the training set;

[0053] S5. Use the projection matrix to perform corresponding projection processing on the test samples and construct multiple features of the test set;

[0054] S6. Classify according to the Euclidean distance and use the equal error rate (EER) to measure the recognition effect of the method.

[0055] The original PPG signal contains noise interference, the main noises include motion artifacts, electromyographic interference, high-frequency interference, and baseline drift. Since the presence of these noises will inevitably affect the accuracy of identity recognition, in step S1 of the embodiment, the PPG signal is preprocessed using the following technical solution:

[0056] S1.1, obtain PPG signal, such as represents the kth sample of the nth class individual, N is the number of sample points;

[0057] S1.2. Construct the following Hankel matrix H m×n : A Hankel Matrix is ​​a matrix in which all elements on the subdiagonal are equal.

[0058]

[0059] S1.3, according to the singular value decomposition theorem, H m×n Singular value decomposition can be expressed as U∑V T , the elements on the diagonal of Σλ i (i=1,2,3,...,k) is H m×n The singular value of can be recorded as λ={λ i |i=1,2,3,...,k};U and V are orthogonal matrices, UU T =U T U=I m , VV T =V T V=I n ;

[0060] S1.4. Use the following formula (2) to calculate the mean of the singular values ​​ω, and use ω as λ i The threshold for keeping or discarding;

[0061]

[0062] That is, the singular values ​​in λ that are greater than or equal to ω are retained, and the other singular values ​​are set to zero. At this time, ∑ can be replace,

[0063] S1.5, according to the following formula (3) use Do the inverse singular value decomposition operation with U and V to get

[0064]

[0065] S1.6, yes The average value of the anti-diagonal elements is obtained The preprocessed signal is recorded as:

[0066]

[0067] PPG signal segmentation after preprocessing: Typical PPG signal waveform is as follows Figure 2 As shown in FIG2 , in this embodiment, in step S2, the reference point P for PPG signal segmentation is first determined; then all single-cycle PPG signal waveforms are segmented with point P as the center; finally, all single-cycle PPG signal waveforms are normalized and a single-cycle PPG signal waveform template is calculated. The specific steps of PPG signal segmentation are as follows:

[0068] S2.1. Determine the peak point P in the PPG signal as the reference point. The reference point P may be determined using the pan_tompkin algorithm.

[0069] S2.2. Segment the waveform into a single-cycle PPG signal. Determine the nearest valley point V before point P and the nearest valley point V after point P. Assuming there are m sampling points between the nearest valley point V before point P and point P, and n sampling points between the nearest valley point V after point P and point P, the length of each single-cycle waveform segment is m+n+1.

[0070] S2.3. Normalize all single-cycle PPG signal waveforms.

[0071] Construction and grouping of multiple features: The time domain features, three-layer sparse representation features, and local mean features of PPG signals all have relatively good PPG biometric recognition performance. However, these single features are often biased and cannot fully reflect the essential characteristics of PPG signals in terms of biometric recognition. Therefore, this embodiment constructs multiple features including time domain features, three-layer sparse representation features, and local mean features to achieve better recognition results.

[0072] S3.1. Extracting time domain features. This embodiment extracts 10 time domain features including the maximum value, minimum value, peak value, peak-to-peak value, mean value, square root amplitude, variance, standard deviation, skewness, and margin of the PPG signal. The calculation formulas are as follows:

[0073]

[0074]

[0075]

[0076]

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] S3.2. Extracting three-layer sparse representation features. Sparse representation exhibits strong robustness to small changes in input data, and therefore sparse representation feature extraction is widely used. The process of extracting three-layer sparse representation feature vectors in this embodiment is as follows:

[0084] S3.2.1. Obtain the sparse representation vector, which can be obtained by the following formula (14):

[0085]

[0086] Among them, y is the test sample, X k is the sample set of individuals of the kth category, W k is the correlation coefficient of the kth individual. For the convenience of description, the above formula (14) can be simplified as:

[0087] S V =SSV X (y,w,K) (15)

[0088] S3.2.2. Extract three-layer sparse representation features:

[0089] The first layer sparsely represents the feature vector:

[0090] The second-level sparse representation feature vector is first divided into four equal-length blocks, denoted as y1, y2, y3, and y4. Correspondingly, the sample set X is also divided into four blocks, denoted as X1, X2, X3, and X4. The second-level sparse representation feature vector is then obtained:

[0091] The third layer of sparse representation feature vector is to divide the sum X into 8 blocks.

[0092] S3.2.3, the extracted three-layer sparse representation feature vector is:

[0093] S3.3. Extract local mean features. The local mean features are designed to obtain local information in the PPG signal that can reflect the essential characteristics of the individual, so as to make up for the shortcomings of the time domain features and the three-layer sparse representation features. First, the local mean method is used to obtain the PF (Product Function) component of the PPG signal. The PF component of the kth sample of the nth category individual can be recorded as The energy ratio is then extracted by RMS Kurtosis Waveform Index and sample entropy

[0094]

[0095]

[0096]

[0097]

[0098] Start with Continuously take the m-dimensional data in the i-th row of the matrix to form a vector and the m-dimensional data in the j-th row form a vector Note that i≠j; statistics (r is a threshold) and is recorded as Aim(r); Similarly, the number of two m+1 dimensional vectors that meet the conditions can be counted and recorded as Finally, the sample entropy is calculated using the following formula (20):

[0099]

[0100] Finally, the extracted local mean feature is expressed as

[0101] S3.4. Construct multiple features and grouping:

[0102] S3.4.1. Serially fuse the time domain features, three-layer sparse representation features, and local mean features of the single-cycle PPG signal, and calculate the templates of all individuals using the average method.

[0103] S3.4.2. Calculate the difference between the constructed multi-features and the template using Euclidean distance, and sort the corresponding PPG signals according to the difference.

[0104] S3.4.3. Divide the sorted PPG signals into several groups, such as G groups, in sequence according to actual needs.

[0105] Multi-feature discriminant analysis: In step S4, multi-feature discriminant analysis is performed based on the constructed multi-feature group. The G group data can be projected into the common subspace of the discriminant analysis through linear transformation, so that the projected samples in this space have the properties of minimal intra-individual variation and maximum inter-individual variation. The projection transformation and constructed objective function of the multi-feature discriminant analysis are as follows:

[0106]

[0107]

[0108]

[0109]

[0110]

[0111] in, represents the jth training sample with individual k in group i, m k represents the number of samples of the kth individual in all groups, m represents the number of all samples in all groups, K represents the total number of individuals, m kj represents the number of samples of the kth type of individuals in the jth group, represents the intra-individual scatter matrix, represents the inter-individual scatter matrix, represents the mean of the kth sample in all groups, W i represents the projection matrix of the i-th group,

[0112] The process of multi-feature discriminant analysis is as follows:

[0113] S4.1. Obtaining multiple features for training

[0114]

[0115] Where K represents the number of categories;

[0116] S4.2. Obtain the intra-individual scatter matrix according to formula (21) According to formula (23), the inter-individual scatter matrix is ​​obtained

[0117] S4.3. Obtain the projection matrix according to formula (25) Where G represents the number of groups;

[0118] S4.4. Obtaining a single-cycle PPG signal template for the i-th individual

[0119] S4.5. Obtaining samples from test individuals

[0120] S4.6. Use the projection matrix to project the test samples of each group accordingly, and then combine the multiple sets of features of each sample to obtain the test set projection matrix.

[0121] Identification and classification: Through multi-feature discriminant analysis, each individual training sample and test sample can obtain the corresponding feature sub-matrix through the projection matrix. The distance between each eigenvector and other eigenvectors in the feature sub-matrix can be calculated by Euclidean distance to obtain the homology similarity matrix. In this embodiment, by calculating the Euclidean distance between the eigenvector of a test sample of a PPG signal and each template in the user PPG vector template library, the template with the smallest distance value and reaching a certain threshold is used as the user to be identified. The threshold here can be understood as a constant. For example, the eigenvector of a test sample (SF∈R 1 ×M ) and the template library matrix (TF∈R M×N The Euclidean distance calculation formula for an individual of ) is as follows (26):

[0122]

[0123] Then, according to formula (26), the Euclidean distance between the test sample and all individuals in the template library can be calculated.

[0124] Finally, the false acceptance rate (FAR), false rejection rate (FRR) and equal error rate (EER) are calculated to measure the effectiveness of the PPG signal identification method based on multi-feature discriminant analysis. The specific formula is as follows:

[0125]

[0126]

[0127]

[0128] Where NFA is the number of false acceptances, NIRA is the total number of inter-individual tests; NFR is the number of false rejections, NGRA is the total number of intra-individual tests.

[0129] This embodiment constructs multiple features and groups of PPG signals by performing singular value decomposition denoising and segmentation on the PPG signals, performs discriminant analysis on them, calculates the projection matrix of the multi-feature groups of all training samples, and then uses the projection matrix to perform corresponding projection on each multi-feature group of the test sample, and finally fuses the obtained multi-features of each sample; this method uses the inter-individual scatter matrix and the intra-individual scatter matrix to obtain a linear discriminant analysis model in the multi-feature group shared space, and by solving the optimal projection vector, multiple PPG signal multi-feature group data with differences are uniformly projected into a common discriminant subspace, so that the PPG signal samples of the same individual are clustered together after projection, and the samples of different individuals are dispersed after projection. In this way, the projection vector finally solved is more distinguishable and discriminative, which is also conducive to extracting the main features of the PPG signal, thereby improving the recognition rate of the PPG signal.

[0130] Example 2:

[0131] This embodiment provides a PPG signal identity recognition system based on singular value decomposition, including:

[0132] The data acquisition module is configured to: acquire PPG signals;

[0133] The eigenvector construction module is configured to: obtain the eigenvector corresponding to the PPG signal using the projection matrix;

[0134] The identity recognition module is configured to: calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and select the template with the smallest distance value that reaches a certain threshold as the user to be identified;

[0135] Among them, the construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; based on the single cycle segmentation results, constructing multiple features of the PPG signals in the training set and grouping them to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library.

[0136] The working method of the system is the same as the PPG signal identity recognition method based on singular value decomposition in Example 1, and will not be repeated here.

[0137] Example 3:

[0138] This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the steps of the PPG signal identity identification method based on singular value decomposition described in Example 1 are implemented.

[0139] Example 4:

[0140] This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps of the PPG signal identity recognition method based on singular value decomposition described in Example 1 are implemented.

[0141] The above description is merely a preferred embodiment of this embodiment and is not intended to limit this embodiment. Those skilled in the art will readily appreciate that this embodiment may be modified and varied in various ways. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this embodiment shall be within the scope of protection of this embodiment.

Claims

1. A PPG signal identification method based on singular value decomposition, characterized in that: include: Get PPG signal; Using the projection matrix, the corresponding eigenvector of the PPG signal is obtained; Calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and take the template with the smallest distance value that reaches a certain threshold as the user to be identified; The construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; constructing multiple features of the PPG signals in the training set based on the single cycle segmentation results and grouping them to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library; Multiple features include time domain features of PPG signals, three-layer sparse representation features, and local mean features; The time domain features, three-layer sparse representation features, and local mean features of the single-cycle PPG signal are serially fused, and the templates of all individuals after the fusion of multiple features are calculated using the average method; Use Euclidean distance to calculate the difference between multiple features and templates, and sort the corresponding PPG signals according to the size of the difference; Divide the sorted PPG signals into several groups in sequence; Multi-feature discriminant analysis is performed on the basis of the constructed multi-feature group. The G group data can be projected into the common subspace of the discriminant analysis through linear transformation, so that the projected samples in this space have the properties of minimal intra-individual variation and maximum inter-individual variation. The inter-individual scatter matrix and the intra-individual scatter matrix are used to obtain the linear discriminant analysis model in the multi-feature group shared space, so that the PPG signal samples of the same individual are clustered together after projection, and the samples of different individuals are dispersed after projection. The final solved projection vector is more distinguishable and discriminative. The intra-individual scatter matrix is ​​specifically: ； The inter-individual scatter matrix is ​​specifically: ； Where, represents the intra-individual scatter matrix, represents the scatter matrix between individuals, and G represents the number of groups.

2. A PPG signal identity recognition method based on singular value decomposition according to claim 1, characterized in that: Preprocessing of PPG signals includes: Get PPG signal; Construct the Hankel matrix based on the number of samples in the acquired PPG signal; Perform singular value decomposition on the constructed Hankel matrix to obtain a set of singular values; Process the singular value set, calculate the mean of the singular values, retain the singular values ​​greater than or equal to the mean, and set the remaining singular values ​​to zero, and obtain; According to the processed singular value set, perform the inverse operation of singular value decomposition to obtain a new Hankel matrix; The anti-diagonal elements of the new Hankel matrix are averaged to obtain the preprocessed PPG signal.

3. The PPG signal identity recognition method based on singular value decomposition according to claim 1, characterized in that: Determine the peak point in the preprocessed PPG signal as the reference point; Determine the nearest valley point before the benchmark point and the nearest valley point after the benchmark point; The length of each single-cycle waveform is the sum of the sampling points between the two valley points and the reference point plus 1; All single-cycle waveforms are normalized.

4. The PPG signal identity recognition method based on singular value decomposition according to claim 1, characterized in that: The projection matrix is ​​used to project the test samples of each group, and the multiple sets of features of each sample are combined to obtain the test set projection matrix.

5. The PPG signal identity recognition method based on singular value decomposition according to claim 1, characterized in that: The training and test samples of each individual are projected to obtain the corresponding feature submatrix. The distance between each eigenvector in the feature submatrix and other eigenvectors is calculated using the Euclidean distance to obtain a homology similarity matrix. The Euclidean distance between the eigenvector of the test sample of the PPG signal and each template in the user PPG vector template library is calculated, and the template with the smallest distance value that reaches the threshold is used as the user to be identified.

6. A PPG signal identity recognition system based on singular value decomposition using the method of claim 1, characterized in that: include: The data acquisition module is configured to: acquire PPG signals; The eigenvector construction module is configured to: obtain the eigenvector corresponding to the PPG signal using the projection matrix; The identity recognition module is configured to: calculate the Euclidean distance between the feature vector and each template in the preset user PPG vector template library, and select the template with the smallest distance value that reaches a certain threshold as the user to be identified; Among them, the construction of the user PPG vector template library includes: preprocessing the PPG signals in the training set using the singular value decomposition method; segmenting the preprocessed PPG signals in the training set into single cycles; based on the single cycle segmentation results, constructing multiple features of the PPG signals in the training set and grouping them to obtain multi-feature groups; performing discriminant analysis on the multi-feature groups, calculating the projection matrix of all training samples, and using the projection matrix to construct a multi-feature module library of the training set as the user PPG vector template library.

7. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the PPG signal identity recognition method based on singular value decomposition as described in any one of claims 1 to 5 are implemented.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the steps of the PPG signal identity identification method based on singular value decomposition as described in any one of claims 1 to 5 are implemented.

Images