CN114098691A

CN114098691A - Pulse wave identity authentication method, device and medium based on Gaussian mixture model

Info

Publication number: CN114098691A
Application number: CN202210090684.3A
Authority: CN
Inventors: 魏日令; 徐晓刚; 王军; 何鹏飞
Original assignee: Zhejiang Gongshang University; Zhejiang Lab
Current assignee: Zhejiang Gongshang University; Zhejiang Lab
Priority date: 2022-01-26
Filing date: 2022-01-26
Publication date: 2022-03-01

Abstract

The invention relates to the field of non-contact physiological signal detection, in particular to a pulse wave identity authentication method, a device and a medium based on a Gaussian mixture model, wherein the method comprises the steps of firstly utilizing a user finger pulse wave signal acquired by an oximeter, inputting the pulse wave signal into a manual feature extraction module to extract time domain and frequency domain features, respectively processing the time domain and frequency domain features into dynamic features, and respectively inputting the obtained dynamic features of the time domain and the frequency domain into a feature extraction network to extract deep level features of the frequency domain and the time domain; further, performing feature screening on the two deep-level features obtained in the last step through a probability linear discriminant analysis algorithm; and finally, carrying out identity recognition on the screened features by utilizing a Gaussian mixture model algorithm. The method can be effectively applied to a biological recognition system based on the pulse wave.

Description

Pulse wave identity authentication method, device and medium based on Gaussian mixture model

Technical Field

The invention relates to the field of non-contact physiological signal detection, in particular to a pulse wave identity authentication method, a device and a medium based on a Gaussian mixture model.

Background

The photoplethysmography signals are that the light absorption degrees of blood and other tissue components to different frequency bands are different, and the blood volume amount of the blood in a blood vessel changes along with the pulsation of the heart, so that the absorption amount of the blood to light also shows periodic pulse fluctuation along with the heart contraction in the processes of the heart contraction and relaxation, and the fluctuation reflects the change of signals received by a video sensor, namely, PPG signals.

In theory, the PPG signals of different people are different, each person has own unique PPG signal characteristics, the identity of the person can be identified through the characteristics, and the method can be applied to the field of biological identification. In the aspect of feature extraction, the identities of different persons can be distinguished by extracting time-domain and frequency-domain features of the signal.

A Gaussian Mixture Model (GMM) can approximate any continuous probability distribution, and can be regarded as a universal approximator of continuous probability distribution. The basic idea is to use known samples to extrapolate the parameter values that are most likely to lead to the result. The GMM can generalize well the features in the PPG signal and perform matching between features in the inference phase to complete identity authentication.

In the process of using PPG to perform biometric identification, because the states of people are different, the PPG signals of the same person are sometimes different, and how to extract effective features of different people and how to summarize the features to complete identity authentication is also a major problem in the field at present.

Disclosure of Invention

In order to solve the technical problems in the prior art, the invention provides a pulse wave identity authentication method, a device and a medium based on a Gaussian mixture model, and the specific technical scheme is as follows:

a pulse wave identity authentication method based on a Gaussian mixture model comprises the following steps:

step one, collecting a pulse wave signal PPG of a fingertip of a user by using an oximeter, and performing framing processing to obtain M sections of signal segments, wherein M is a positive integer and is greater than 1;

step two, arranging the M sections of signal segments according to a time sequence, extracting frequency domain and time domain characteristics, and respectively calculating to obtain dynamic characteristics of the frequency domain and the time domain;

inputting the dynamic characteristics of the frequency domain and the time domain into a trained characteristic extraction network, extracting to obtain deep level characteristics of the frequency domain and the time domain, and fusing;

step four, screening the fused deep level features through a probability linear discriminant analysis algorithm;

and fifthly, identifying the identity of the screened features by using a Gaussian mixture model algorithm.

Further, the step one specifically includes:

firstly, acquiring a pulse wave signal PPG of a fingertip of a user by using an oximeter, wherein the sampling frequency is N Hz, and the acquisition duration is T s;

then, performing framing operation on pulse wave signals PPG acquired by an oximeter, and performing signal interception on the pulse wave signals with the duration of T s by using a sliding window mode, wherein the moving step length of the sliding window is fixed, the length of the sliding window is a random numerical value, and M sections of signal segments are extracted, and users corresponding to the M sections of signal segments are the same user.

Further, the second step specifically includes:

arranging the signal segments according to a time sequence and then extracting frequency domain and time domain waveform characteristics, wherein the frequency domain waveform characteristics of the signals are extracted through a spectrogram after Fourier transform, amplitudes and phases corresponding to various frequencies are extracted, W frequency bands are extracted in total, and 2W characteristics are extracted; in the extraction of time domain waveform characteristics, diastolic period time, systolic period time, corresponding amplitude and slope are extracted, K characteristics are counted, and 2W + K manual characteristics are extracted, wherein W, K are positive integers;

respectively calculating dynamic characteristics of a frequency domain and a time domain, wherein the dynamic characteristics are obtained by calculating the characteristics of different frames, and when the number of frames is less than the number of Gaussian models, the dynamic characteristics are calculated by the difference value of a previous frame and a next frame of a current frame; if the current frame is larger than the number of the characteristic quantity minus the number of the Gaussian models, calculating by the difference value of the current frame and the previous frame; in other cases, the difference between the k frame after the current frame and the k frame before the current frame is accumulated.

Further, the feature extraction network specifically includes:

in the time domain feature extraction, a structure of taking a depth residual error shrinkage network as a backbone network and adding a TDNN network is adopted; in the frequency domain feature extraction, a depth residual error shrinkage network and an LSTM network structure are adopted; pre-training in a backbone network: in the pre-training stage, a combined data set comprises a common video and an infrared video of the same user, two different videos are randomly combined, if the videos are of the same user, a label is 1, otherwise, the labels are 0, a main network is pre-trained by using a cross entropy loss function and a gradient descent method, and model parameters are stored after training is finished.

Further, the third step specifically includes:

loading the trained feature extraction network;

respectively inputting the dynamic characteristics of the frequency domain and the time domain into a corresponding frequency domain characteristic extraction network and a corresponding time domain characteristic extraction network according to the frequency domain characteristics and the time domain characteristics to obtain deep level characteristics of the frequency domain and the time domain, and outputting a characteristic vector of the time domain and a characteristic vector of the frequency domain;

and the obtained time domain feature vector and the frequency domain feature vector are fused into a segment of feature vector, and the vector length is as follows: the sum of the length of the feature vector in the frequency domain and the length of the feature vector in the time domain.

Further, the fourth step specifically includes:

the effective features are screened out through a probability linear discriminant analysis mode, and the specific mode is that the effective features are solved by adding all data mean values, expressions of different users in the identity space of the users and error spaces.

Further, the fifth step is specifically:

grouping signals belonging to different users, wherein the signals of each group belong to the same user, and establishing respective Gaussian mixture models for each user;

establishing respective Gaussian mixture model parameters for each user by using an EM algorithm: estimating the mean value, the variance and the weight of all training data, updating the model parameters through continuous iteration, and when the model parameters meet the following conditions: stopping updating and storing the model parameters after the norm of the difference between the current iteration updated model parameters and the last model parameters is less than 0.001;

the effective characteristics are calculated in all user parameter models, and the highest value is taken as an authentication result.

The device comprises a memory and one or more processors, wherein the memory stores executable codes, and the one or more processors are used for realizing the pulse wave identity authentication method based on the Gaussian mixture model when executing the executable codes.

A computer-readable storage medium having stored thereon a program for implementing the method for identity authentication of a pulse wave based on a gaussian mixture model when the program is executed by a processor.

The invention has the advantages that:

firstly, the invention reduces the noise interference by extracting the dynamic characteristics of time domain and frequency domain as input; secondly, the time characteristic is fully considered in the feature extraction network to extract the deep-level features of the signals from the manual features; effective features are selected by utilizing a PLDA algorithm in the feature screening stage, so that the length of a feature vector is reduced, and the reasoning speed of the model is improved; in the final identity authentication, a Gaussian mixture model belonging to each person is respectively established for identity authentication.

Drawings

FIG. 1 is a schematic flow diagram of the process of the present invention;

FIG. 2 is a schematic diagram of a feature extraction network of the method of the present invention;

FIG. 3 is a schematic diagram of an extraction module in the feature extraction network of the method of the present invention;

FIG. 4 is a schematic diagram of a TDNN network in a feature extraction network of the method of the present invention;

fig. 5 is a block diagram of a pulse wave identity authentication apparatus based on a gaussian mixture model according to the present invention.

Detailed Description

In order to make the objects, technical solutions and technical effects of the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and examples.

As shown in fig. 1, a method for authenticating identity of pulse wave based on gaussian mixture model includes the following steps:

step one, collecting pulse wave signals PPG of fingertips of a user by using an oximeter, and performing framing processing to obtain M signal segments.

Specifically, firstly, a pulse wave signal PPG of a fingertip of a user is acquired by using an oximeter, the sampling frequency is N Hz, and the acquisition duration is T s;

then, performing framing operation on a PPG signal acquired by an oximeter, and performing signal interception on a pulse wave signal with the duration of T s by using a sliding window mode, wherein the moving step of the sliding window is fixed and is step _ time s, the length of the sliding window is a random numerical value, the operation is to extract characteristics of signals with different lengths and extract effective characteristics with time invariance in a subsequent process, and the variation range of the sliding window is [ window _ size/2, window _ size/2 ]; by the method, M signal segments are extracted, and the users corresponding to the signal segments are the same user.

And step two, arranging the M sections of signal segments according to a time sequence, extracting frequency domain and time domain characteristics, and respectively calculating to obtain dynamic characteristics of the frequency domain and the time domain.

Specifically, the divided signal segments are arranged according to a time sequence, and then frequency domain and time domain waveform characteristics are extracted, wherein the frequency domain characteristics of the signals are mainly extracted through a spectrogram after Fourier transform, amplitudes and phases corresponding to various frequencies are extracted, W frequency bands are extracted in total, and 2W characteristics are calculated; in the extraction of time domain waveform characteristics, extracting diastolic period time, systolic period time, corresponding amplitude and slope, and totaling K characteristics; in the stage, 2W + K manual features are extracted, wherein W, K are positive integers;

the extracted manual features are easily interfered by noise, the noise mainly comes from noise introduced by acquisition equipment during acquisition and noise caused by movement of a user finger during the acquisition, dynamic features of a frequency domain and a time domain are respectively calculated for reducing noise interference, the dynamic features are obtained by calculating features of different frames, and when the number of frames is less than the number of Gaussian models, the dynamic features are calculated by the difference value of a previous frame and a next frame of a current frame; if the current frame is larger than the number of the characteristic quantity minus the number of the Gaussian models, calculating by the difference value of the current frame and the previous frame; in other cases, the calculation is performed by accumulating the difference between the k frame after the current frame and the k frame before the current frame, and the expression is:

in the above formula, L is the number of features, C is the features of different frames, t represents different times, Ct is the feature of different time periods, and k represents the kth gaussian model.

Repeating the step two, and extracting dynamic characteristics from all the signal segments; and repeating the first step and the second step to extract the dynamic characteristics of the signals of all the users.

And thirdly, inputting the dynamic characteristics of the frequency domain and the time domain into the trained characteristic extraction network, extracting to obtain deep level characteristics of the frequency domain and the time domain, and fusing the deep level characteristics.

The pre-training and reasoning of the feature extraction network specifically comprises the following steps:

as shown in fig. 2 to 4, in the feature extraction network, two networks with different structures are used for extracting time domain features and frequency domain features, and in the time domain feature extraction, a structure in which a trunk network with a deep residual shrinkage network as an extraction module and a TDNN network (time delay neural network) are used as a time extraction module is used; in the frequency domain feature extraction, a deep residual error shrinkage network is adopted as a network structure of a main network of an extraction module and an LSTM (long-short term memory network) to extract features; as shown in fig. 3, the depth residual shrinkage network assigns different attention ranges to different features in a multi-scale manner, and outputs a feature map through a 3 × 3 convolution layer; then, generating a first branch, wherein the first branch is composed of 3 × 3 convolution layers and 1 × 1 convolution layers to further extract features, and the 1 × 1 convolution layers are mainly used for reducing the output size of the feature map; the main branch generates a new feature map after passing through a 3 x 3 convolution layer, the new feature map generates a second branch, wherein the feature map is kept unchanged by the main branch, the current feature map is further extracted by the second branch through two 3 x 3 convolution layers and a 1 x 1 convolution layer, finally, the feature maps of the three branches are fused, and the fused feature map is output by an output layer; wherein their backbone networks are pre-trained; in the pre-training stage, a common video and an infrared video of the same user are included through a combined data set; by randomly combining two different videos, if the videos are the same user, the label is 1, otherwise, the label is 0; pre-training a backbone network by using a cross entropy loss function and a gradient descent method, wherein the training times are 50 times; after training, storing model parameters;

loading the trained feature extraction network;

inputting all the feature vectors obtained in the step two into a corresponding frequency domain feature extraction network and a corresponding time domain feature extraction network according to the frequency domain features and the time domain features to obtain frequency domain and time domain deep level features, and outputting the time domain feature vectors and the frequency domain feature vectors, wherein the length of each feature vector is the frequency domain length L _ freq and the time domain length L _ time;

and the obtained time domain feature vector and the frequency domain feature vector are fused into a segment of feature vector, and the vector length is as follows: l _ freq + L _ time.

And step four, screening the fused deep level features through a probability linear discriminant analysis algorithm.

Specifically, the obtained fused features are screened, and effective features are screened out in a mode of Probability Linear Discriminant Analysis (PLDA), wherein the effective features are solved by adding all data mean values, expressions of different users in the identity space of the users and error spaces, and the expressions are as follows:

the parameters of the formula are

And mu represents the mean value of the whole training data, F can be considered as an identity space and comprises bases of different user identities,

representing the identity of a user, G can be considered as an error space, comprising bases that can be used to represent different variations of the same identity,

indicating the position in the space, last

To represent what has not been explained, the term is a gaussian distribution with a mean of 0 and a variance of Σ.

Continuously iterating and updating the EM algorithm (maximum expectation algorithm) to obtain a parameter theta, and calculating the screened effective characteristics

；

And repeating the third step and the fourth step to extract features of all the signals.

And fifthly, carrying out identity recognition on the screened effective characteristics by using a Gaussian mixture model algorithm.

Specifically, signals belonging to different users are grouped, and the signals of each group belong to the same user; establishing a Gaussian mixture model for each user, wherein the formula of the Gaussian mixture model is as follows:

is the probability density function of the kth gaussian model,

is the weight of the kth model, wherein

。

In the above formula, the parameters of the model are

Using the EM algorithm to obtain the gaussian mixture model parameters for each user: estimating the mean value, the variance and the weight of all training data, updating the model parameters through continuous iteration, and stopping updating when the model parameters meet the norm of the difference between the updated model parameters of the current iteration and the last model parameters, wherein the expression is as follows:

the formula indicates that updating of parameters is stopped when the model parameters change very little, wherein,

for the updated model parameters of the current iteration,

is the parameter of the last model,

taking 0.001;

the model parameters of each person are then saved.

In the inference stage, after a section of new PPG signals are input into the model, the screened effective characteristics are respectively calculated in all current user parameter models, and finally the user with the highest score is obtained as an authentication result.

The following table 1 shows that indexes of the processed remote pulse wave signals in the identity authentication are better than those of the remote pulse wave signals obtained by other methods.

Table 1: the application of the invention in biological identification is compared with the performance of the existing biological identification method

Method	Rate of accuracy
		Fuzzy Logic	82.3%
KNN	74.0%
		Method of the invention	98.0%

Corresponding to the embodiment of the pulse wave identity authentication method based on the Gaussian mixture model, the invention also provides an embodiment of a pulse wave identity authentication device based on the Gaussian mixture model.

Referring to fig. 5, the apparatus for authenticating identity of pulse wave based on a gaussian mixture model according to the embodiment of the present invention includes one or more processors, and is configured to implement a method for authenticating identity of pulse wave based on a gaussian mixture model in the foregoing embodiment.

The embodiment of the pulse wave identity authentication device based on the Gaussian mixture model can be applied to any equipment with data processing capability, such as computers and other equipment or devices. The device embodiments may be implemented by software, or by hardware, or by a combination of hardware and software. The software implementation is taken as an example, and as a logical device, the device is formed by reading corresponding computer program instructions in the nonvolatile memory into the memory for running through the processor of any device with data processing capability. In terms of hardware, as shown in fig. 5, a hardware structure diagram of an arbitrary device with data processing capability where a pulse wave identity authentication device based on a gaussian mixture model is located according to the present invention is shown, except for the processor, the memory, the network interface, and the nonvolatile memory shown in fig. 5, in an embodiment, the arbitrary device with data processing capability where the device is located may also include other hardware according to the actual function of the arbitrary device with data processing capability, which is not described again.

The implementation process of the functions and actions of each unit in the above device is specifically described in the implementation process of the corresponding step in the above method, and is not described herein again.

For the device embodiments, since they substantially correspond to the method embodiments, reference may be made to the partial description of the method embodiments for relevant points. The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of the invention. One of ordinary skill in the art can understand and implement it without inventive effort.

The embodiment of the present invention further provides a computer-readable storage medium, on which a program is stored, and when the program is executed by a processor, the method for authenticating identity of pulse wave based on a gaussian mixture model in the above embodiments is implemented.

The computer readable storage medium may be an internal storage unit, such as a hard disk or a memory, of any data processing capability device described in any of the foregoing embodiments. The computer readable storage medium may also be an external storage device of the wind turbine, such as a plug-in hard disk, a Smart Media Card (SMC), an SD Card, a Flash memory Card (Flash Card), and the like, provided on the device. Further, the computer readable storage medium may include both an internal storage unit and an external storage device of any data processing capable device. The computer-readable storage medium is used for storing the computer program and other programs and data required by the arbitrary data processing-capable device, and may also be used for temporarily storing data that has been output or is to be output.

The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way. Although the foregoing has described the practice of the present invention in detail, it will be apparent to those skilled in the art that modifications may be made to the practice of the invention as described in the foregoing examples, or that certain features may be substituted in the practice of the invention. All changes, equivalents and modifications which come within the spirit and scope of the invention are desired to be protected.

Claims

1. A pulse wave identity authentication method based on a Gaussian mixture model is characterized by comprising the following steps:

2. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 1, wherein the first step is specifically as follows:

3. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 1, wherein the second step is specifically:

4. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 1, wherein the feature extraction network specifically comprises:

5. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 4, wherein the third step is specifically:

loading the trained feature extraction network;

6. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 5, wherein the fourth step is specifically:

7. The method for authenticating identity of pulse wave based on Gaussian mixture model as claimed in claim 6, wherein the fifth step is specifically as follows:

8. A pulse wave identity authentication device based on a gaussian mixture model, comprising a memory and one or more processors, wherein the memory stores executable codes, and the one or more processors are used for implementing the pulse wave identity authentication method based on the gaussian mixture model according to any one of claims 1 to 7 when executing the executable codes.

9. A computer-readable storage medium, having stored thereon a program which, when being executed by a processor, is configured to implement the method for identity authentication of a pulse wave based on a gaussian mixture model according to any one of claims 1 to 7.