CN109919050A - Identity recognition method and device - Google Patents
Identity recognition method and device Download PDFInfo
- Publication number
- CN109919050A CN109919050A CN201910130479.3A CN201910130479A CN109919050A CN 109919050 A CN109919050 A CN 109919050A CN 201910130479 A CN201910130479 A CN 201910130479A CN 109919050 A CN109919050 A CN 109919050A
- Authority
- CN
- China
- Prior art keywords
- signal
- cardiac cycle
- cycle signal
- identity
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses an identity recognition method and an identity recognition device, wherein the identity recognition method comprises the following steps: dividing the acquired PPG signal of the current user into a plurality of cardiac cycle signals, and performing feature extraction on each cardiac cycle signal to obtain a first identity attribute feature corresponding to each cardiac cycle; comparing each first identity attribute characteristic with a second identity attribute characteristic of a standard PPG signal of at least one pre-stored registered user respectively, and judging whether the first identity attribute characteristics and the second identity attribute characteristics come from the same user or not; and if the judgment result is from the same user, the pre-stored identity information of the corresponding user is used as the identity information of the current user.
Description
Technical field
The present invention relates to living things feature recognition field, intelligent medical and intelligent wearable devices, in particular to a kind of body
Part recognition methods and device.
Background technique
Living things feature recognition is also more and more with its application demand of the development of artificial intelligence and deep learning, increasingly
Extensively.Living things feature recognition known at present includes fingerprint, iris, refers to vein, vocal print, face, person's handwriting, gait, auricle etc.,
It is respectively applied to different occasions.
The detection of PPG signal and the research and development of the relevant technologies are that measuring of human health brings many conveniences, are based on PPG signal
The detections of the physiological signals such as BOLD contrast, heart rate, blood pressure also widely come into average family, accordingly do not increasing it is additional
Carrying out authentication and identification to user under conditions of equipment is the demand occurred therewith.Pass through the certification to user identity
It can be realized the monitoring long-term to multiple kinsfolks, lasting, independent simultaneously of same equipment, realize the automatic respective tube of data
Reason.
Living things feature recognition has always research in voice and image domains.Currently, carrying out authentication based on PPG signal
For method mostly from traditional mode identification method, basic procedure is Signal Pretreatment, characteristic Design, Feature Selection, classification
Deng.The accuracy of authentication extremely relies on extracted feature, and the extracted feature of method traditional at present is mostly based on PPG
Relationship between some key points of signal, such as main wave wave crest point, dicrotic wave wave crest point in PPG signal, and between the two
The triangle area of this 3 points compositions in valley point.The maximum defect of artificial design features is it is impossible to ensure that obtaining optimal to authenticating
Feature, cause the robustness of authentication result poor.
Summary of the invention
The present invention provides a kind of personal identification method and device, to overcome it is existing in the prior art at least one ask
Topic.
According to a first aspect of the embodiments of the present invention, a kind of personal identification method is provided, comprising the following steps:
The PPG signal of acquired active user is divided into multiple cardiac cycle signals, to each cardiac cycle
Signal carries out feature extraction, obtains corresponding first identity attribute feature of each cardiac cycle；
By each first identity attribute feature respectively with the standard PPG of at least one pre-stored registered users
Whether the second identity attribute feature of signal compares, judge the two from same user；
If judging result is from same user, using the identity information of pre-stored corresponding user as active user
Identity information.
Optionally, described that the PPG signal of acquired active user is divided into multiple cardiac cycle signals, to each institute
It states cardiac cycle signal and carries out feature extraction, obtaining corresponding first identity attribute feature of each cardiac cycle includes:
The PPG signal of acquired active user is split, multiple cardiac cycle signals are obtained；
Detect the key point of each cardiac cycle signal；
Based on the key point, each cardiac cycle signal is aligned with reference cycle signal, is obtained each
The corresponding normalized signal of the cardiac cycle signal；
Feature extraction is carried out to each normalized signal using CNN network, it is corresponding to obtain each cardiac cycle signal
The first identity attribute feature.
Optionally, described to be split the PPG signal of acquired active user, obtain multiple cardiac cycle signal packets
It includes:
The wave crest and trough of the PPG signal of active user are obtained using Time-Domain algorithm；
Acquired wave crest and trough are analyzed based on heuristic criterion, obtain multiple cardiac cycle signals and every
The starting point and terminating point of a cardiac cycle signal.
Optionally, the key point of each cardiac cycle signal includes starting point, terminating point and main wave wave crest point, the inspection
The key point for surveying each cardiac cycle signal includes:
The starting point and terminating point obtained when each cardiac cycle signal is divided is as its starting point and terminating point；
The judgment criterion for being zero by amplitude maximum and gradient detects the time-domain signal of each cardiac cycle
Obtain the main wave wave crest point of each cardiac cycle signal.
Optionally, described to be based on the key point, each cardiac cycle signal and reference cycle signal are carried out pair
Together, obtaining the corresponding normalized signal of each cardiac cycle signal includes:
The signal amplitude of each cardiac cycle signal is normalized；
The key point of cardiac cycle signal after the normalization of each signal amplitude is corresponding with reference cycle signal respectively
Crucial point alignment obtains the mapping relations between key point；
The remaining data points of each cardiac cycle signal are aligned according to the mapping relations, by each letter
The length of cardiac cycle signal after number amplitude normalization is normalized.
Optionally, the signal amplitude to each cardiac cycle signal, which is normalized, includes:
Assuming that the main wave crest height of each cardiac cycle signal is H, the main wave crest height after normalization is NH,
Then for each cardiac cycle signal X, signal amplitude is normalized as follows:
Xn=X* (NH/H),
Wherein, Xn is the signal after signal amplitude normalization.
Optionally, described to be carried out the remaining data points of each cardiac cycle signal pair according to the mapping relations
Together, the length of the cardiac cycle signal after the normalization of each signal amplitude is normalized and includes:
Wave crest point full-length Lf according to preset full-length L and main wave wave crest point relative to starting point, to each letter
Cardiac cycle signal after number amplitude normalization carries out length normalization method processing according to following formula:
Wherein, W is the length of each cardiac cycle signal, and w1 is the main wave wave crest of each cardiac cycle signal
Length of the position relative to its starting point, XiFor the remaining data points in each cardiac cycle signal in addition to key point.
Optionally, described that feature extraction is carried out to each normalized signal using CNN network, obtain each week aroused in interest
The corresponding first identity attribute feature of phase signal includes:
Adopt using the input picture that two paths of data point sequence of the convolution sum pond to each normalized signal is formed
Sample obtains the first down-sampling feature vector of the input picture；
The input picture is up-sampled using deconvolution, obtain the input picture first up-sampling feature to
Amount；
The first up-sampling feature vector and the first down-sampling feature vector are subjected to group by way of superposition
It closes, obtains multiple feature vectors of each cardiac cycle signal；
The corresponding multiple feature vectors of each cardiac cycle signal are compressed by way of convolution, will be obtained
N dimension identity attribute feature vector as the corresponding first identity attribute feature of each cardiac cycle signal, wherein n is positive
Integer.
Optionally, it is described by each first identity attribute feature respectively at least one pre-stored registered use
Second identity attribute feature of the standard PPG signal at family compares, and judges whether the two comes from same user and include:
Each first identity attribute feature and at least one pre-stored registered use are calculated according to following formula
Similarity sim (X1, X2) between second identity attribute feature of the standard PPG signal at family:
Wherein, X1 is each first identity attribute feature, and X2 is pre-stored one of registered users
Second identity attribute feature of standard PPG signal；
The size relation for judging each similarity and threshold value Th then determines when the similarity is greater than threshold value Th
The corresponding cardiac cycle signal of X1, X2 comes from same user, otherwise, then determines X1, the corresponding cardiac cycle signal of X2 is not from
Same user；
It votes according to following formula the corresponding identity information of multiple cardiac cycle signals, obtains active user
The corresponding identity information C of PPPG signal*:
Wherein fijPass through the posterior probability that certification is i-th of registration user for j-th of cardiac cycle signal, m is multiple institutes
The number of cardiac cycle signal is stated, m, i, j are positive integer.
Optionally, if being to believe the identity of pre-stored corresponding user from same user in the judging result
After breath is as the identity information of active user further include:
Pre-stored, registered users PPG signal the second identity attribute feature vector is carried out according to following formula
It updates:
Xnew=(1- β) * X+ β * Xcur
Wherein, X is the second identity attribute feature vector of user i registration before updating, and Xcur is that current detection mentions in the process
The first identity attribute feature vector with X from same registration user i is obtained and is identified as, Xnew is user after updating
Second identity attribute feature vector of i registration；β is to update coefficient.
Optionally, the used loss function when constructing the CNN network are as follows:
Wherein, W, W※Weight vectors after respectively normalizing and before normalization, x, x※Be divided into for normalization after and normalizing
Feature vector before change, θjBe weight vectors WjWith feature vector xiAngle.
cos(θj, i) and=Wj Txi,
Wherein, W, W※Weight vectors after respectively normalizing and before normalization, x, x※Be divided into for normalization after and normalizing
Feature vector before change, θjBe weight vectors WjWith feature vector xiAngle.
According to the second aspect of this specification embodiment, a kind of identity recognition device is also provided, comprising:
Divide extraction module, is configured as the PPG signal of acquired active user being divided into multiple cardiac cycle letters
Number, feature extraction is carried out to each cardiac cycle signal, obtains corresponding first identity attribute of each cardiac cycle
Feature；
Contrast judgement module is configured as each first identity attribute feature respectively with pre-stored at least one
Whether the second identity attribute feature of the standard PPG signal of a registered users compares, judge the two from same user；
Authentication module, if being configured as judging result is from same user, by pre-stored corresponding user
Identity information of the identity information as active user.
Optionally, the segmentation extraction module includes:
Cutting unit is configured as the PPG signal of acquired active user being split, obtains multiple cardiac cycles
Signal；
Detection unit is configured as detecting the key point of each cardiac cycle signal；
Normalization unit is configured as believing each cardiac cycle signal and reference period based on the key point
It number is aligned, obtains the corresponding normalized signal of each cardiac cycle signal；
Feature extraction unit is configured as carrying out feature extraction to each normalized signal using CNN network, obtain each
The corresponding first identity attribute feature of the cardiac cycle signal.
Optionally, the cutting unit includes:
Peak valley obtains subelement, is configured as obtaining the wave crest and trough of the PPG signal of active user using Time-Domain algorithm；
Start-stop point obtains subelement, is configured as analyzing acquired wave crest and trough based on heuristic criterion,
Obtain the starting point and terminating point of multiple cardiac cycle signals and each cardiac cycle signal.
Optionally, the key point of each cardiac cycle signal includes starting point, terminating point and main wave wave crest point, the inspection
Unit is surveyed to be configured to:
The starting point and terminating point obtained when each cardiac cycle signal is divided passes through as its starting point and terminating point
The judgment criterion that amplitude maximum and gradient are zero is detected to obtain each described to the time-domain signal of each cardiac cycle
The main wave wave crest point of cardiac cycle signal.
Optionally, the normalization unit includes:
Amplitude normalization subelement is configured as that the signal amplitude of each cardiac cycle signal is normalized；
Be aligned subelement, be configured as by each signal amplitude normalize after cardiac cycle signal key point respectively with
The corresponding crucial point alignment of reference cycle signal, obtains the mapping relations between key point；
Length normalization method subelement, be configured as according to the mapping relations by each cardiac cycle signal remaining
Data point is aligned, and the length of the cardiac cycle signal after the normalization of each signal amplitude is normalized.
Optionally, the amplitude normalization subelement is configured to:
Assuming that the main wave crest height of each cardiac cycle signal is H, the main wave crest height after normalization is NH,
Then for each cardiac cycle signal X, signal amplitude is normalized as follows:
Xn=X* (NH/H),
Wherein, Xn is the signal after signal amplitude normalization.
Optionally, the length normalization method subelement is configured to:
Wave crest point full-length Lf according to preset full-length L and main wave wave crest point relative to starting point, to each letter
Cardiac cycle signal after number amplitude normalization carries out length normalization method processing according to following formula:
Wherein, W is the length of each cardiac cycle signal, and w1 is the main wave wave crest of each cardiac cycle signal
Length of the position relative to its starting point, XiFor the remaining data points in each cardiac cycle signal in addition to key point.
Optionally, the feature extraction unit includes:
Down-sampling subelement is configured as the two paths of data point sequence shape using convolution sum pond to each normalized signal
At input picture carry out down-sampling, obtain the first down-sampling feature vector of the input picture；
Subelement is up-sampled, is configured as up-sampling the input picture using deconvolution, obtains the input
First up-sampling feature vector of image；
Stack combinations subelement, be configured as by it is described first up-sampling feature vector and the first down-sampling feature to
Amount is combined by way of superposition, obtains multiple feature vectors of each cardiac cycle signal；
Subelement is compressed, is configured as passing through convolution to the corresponding multiple feature vectors of each cardiac cycle signal
Mode is compressed, using obtained n dimension identity attribute feature vector as corresponding first body of each cardiac cycle signal
Part attributive character, wherein n is positive integer.
Optionally, the contrast judgement module includes:
Similarity calculated is configured as calculating each first identity attribute feature and in advance according to following formula
Storage at least one registered users standard PPG signal the second identity attribute feature between similarity sim (X1,
X2):
Wherein, X1 is each first identity attribute feature, and X2 is pre-stored one of registered users
Second identity attribute feature of standard PPG signal；
Threshold decision unit is configured as judging the size relation of each similarity and threshold value Th, when described similar
When degree is greater than threshold value Th, then X1 is determined, the corresponding cardiac cycle signal of X2 comes from same user, otherwise, then determines that X1, X2 are corresponding
Cardiac cycle signal come from different user；
It votes unit, is configured as carrying out the corresponding identity information of multiple cardiac cycle signals according to following formula
Ballot, obtains the corresponding identity information C of PPPG signal of active user*:
Wherein fijPass through the posterior probability that certification is i-th of registration user for j-th of cardiac cycle signal, m is multiple institutes
The number of cardiac cycle signal is stated, m, i, j are positive integer.
Optionally, the identification authentication system further include:
Update module is configured as the second identity according to following formula to pre-stored registered users PPG signal
Attribute feature vector is updated:
Xnew=(1- β) * X+ β * Xcur
Wherein, X is the second identity attribute feature vector of user i registration before updating, and Xcur is that current detection mentions in the process
The first identity attribute feature vector with X from same registration user i is obtained and is identified as, Xnew is user after updating
Second identity attribute feature vector of i registration；β is to update coefficient.
Optionally, the used loss function when constructing the CNN network are as follows:
Wherein,
cos(θj, i) and=Wj Txi,
Wherein, W, W※Weight vectors after respectively normalizing and before normalization, x, x※Be divided into for normalization after and normalizing
Feature vector before change, θjBe weight vectors WjWith feature vector xiAngle.
This specification embodiment carries out identification and certification using PPG signal, multiple aroused in interest by the way that PPG signal to be divided into
Periodic signal, and then the identity attribute feature that cardiac cycle signal is objective, can reflect different user physiological characteristic is extracted, it will
Extracted identity attribute feature and the identity attribute feature of pre-stored registered users PPG signal compare, judgement
Whether two signals are from the same person, to identify which in database be current PPG signal correspondence have already registered with
One ID realizes automated validation and the identification of the user identity based on PPG signal；Simultaneously as this specification embodiment institute
Using the physiological attribute feature for the different user that PPG signal is reflected, rather than the feature artificially designed, so that this
The personal identification method of specification embodiment is available to certification preferably feature, has preferable robustness.
The inventive point of this specification embodiment includes at least:
1, CNN network uses symmetrical U-net network, and bottom convolutional layer is merged and can be had with high-rise convolutional layer
The ability to express of the raising feature of effect reduces loss of the feature in convolution process, is one of inventive point of the embodiment of the present invention.
2, the key point of the cardiac cycle signal after segmentation is aligned, when remaining data points are according to crucial point alignment
Mapping ruler is mapped, and (these differences may be because electric signal is strong to the difference between the PPG signal to eliminate the same ID
Caused by weak transformation, movement etc.), it is one of the inventive point of this specification embodiment；
3, self-adaptive features study and online updating method, by by the identity attribute feature of user's current matching to pre-
The identity attribute of the user's PPG signal first stored is updated, and is this specification to adapt to the slow transformation of human body PPG signal
One of inventive point of embodiment.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the personal identification method flow chart of one embodiment of the invention；
Fig. 2 is the personal identification method flow chart of another embodiment of the present invention；
Fig. 3 is the cardiac cycle signal schematic diagram of one embodiment of the invention；
Fig. 4 is the CNN schematic network structure of one embodiment of the invention；
Fig. 5 is the identity recognition device module map of one embodiment of the invention.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other under that premise of not paying creative labor
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the personal identification method flow chart of one embodiment of the invention；As shown in Figure 1, the personal identification method packet
Include following steps:
The PPG signal of acquired active user is divided into multiple cardiac cycle signals, to each described aroused in interest by S110
Periodic signal carries out feature extraction, obtains corresponding first identity attribute feature of each cardiac cycle.
Wherein, PPG signal collected includes infrared signal and red signal light two-way in the present embodiment.A kind of realization side
In formula, can by the method for sliding window, by the PPG signal sampling of acquired original at the signals of several setting length, for example,
Signal duration is set as 32s.
In a kind of implementation, before step S110 further include:
Noise reduction process is carried out to the PPG signal of active user collected, it specifically, can be using bandpass filter to going through
History PPG signal carries out noise reduction process to eliminate baseline drift and partial noise.
In specific implementation, can using kaise (Keyes) FIR, (Finite Impulse Response has limit for length single
Position impulse response) bandpass filter.In one example, it can be produced using the tool box matlab according to following parameter setting
Filter, design parameter are as follows:
Frequency parameter [Fstop1Fpass1Fpass2Fstop2]=[0.1 0.5 8 10]；
Magnitude parameters [Astop1Apass Astop2]=[30 0.01 30]；
Wherein, Fpass1 and Fpass2 is the cutoff frequency of passband；Fstop1 and Fstop2 is the cutoff frequency of stopband.
Astop1 refers to the amplitude attenuation of cutoff frequency Fstop1；Astop2 refers to the amplitude attenuation of cutoff frequency Fstop2；Apass is
Amplitude attenuation in free transmission range, also referred to as passband fluctuation.After can be obtained denoising by filter progress signal convolution
PPG signal.
Kaise FIR bandpass filter is one of this specification example exemplary method, is not limited in this specification
Link of making an uproar uses kaise FIR bandpass filter, and other any filters that can be used for PPG signal denoising can use.
It is described that the PPG signal of acquired active user is divided into multiple cardiac cycle signals in a kind of embodiment,
Feature extraction is carried out to each cardiac cycle signal, obtains corresponding first identity attribute feature of each cardiac cycle
Include:
The PPG signal of acquired active user is split, multiple cardiac cycle signals are obtained；
Detect the key point of each cardiac cycle signal；
Based on the key point, each cardiac cycle signal is aligned with reference cycle signal, is obtained each
The corresponding normalized signal of the cardiac cycle signal；
Feature extraction is carried out to each normalized signal using CNN network, it is corresponding to obtain each cardiac cycle signal
The first identity attribute feature.
In specific implementation, described to be split the PPG signal of acquired active user, obtain multiple cardiac cycles
Signal may include:
The wave crest and trough of the PPG signal of active user are obtained using Time-Domain algorithm；
Acquired wave crest and trough are analyzed based on heuristic criterion, obtain multiple cardiac cycle signals and every
The starting point and terminating point of a cardiac cycle signal.
In specific implementation, the key point of each cardiac cycle signal includes starting point, terminating point and main wave wave crest point,
The key point of each cardiac cycle signal of the detection may include:
The starting point and terminating point obtained when each cardiac cycle signal is divided is as its starting point and terminating point；
The judgment criterion for being zero by amplitude maximum and gradient detects the time-domain signal of each cardiac cycle
Obtain the main wave wave crest point of each cardiac cycle signal.
In specific implementation, described to be based on the key point, by each cardiac cycle signal and reference cycle signal
It is aligned, obtaining the corresponding normalized signal of each cardiac cycle signal includes:
The signal amplitude of each cardiac cycle signal is normalized；
The key point of cardiac cycle signal after the normalization of each signal amplitude is corresponding with reference cycle signal respectively
Crucial point alignment obtains the mapping relations between key point；
The remaining data points of each cardiac cycle signal are aligned according to the mapping relations, by each letter
The length of cardiac cycle signal after number amplitude normalization is normalized.
In specific implementation, the signal amplitude to each cardiac cycle signal, which is normalized, includes:
Assuming that the main wave crest height of each cardiac cycle signal is H, the main wave crest height after normalization is NH,
Then for each cardiac cycle signal X, signal amplitude is normalized as follows:
Xn=X* (NH/H),
Wherein, Xn is the signal after signal amplitude normalization.
In specific implementation, described to be clicked through the remainder data of each cardiac cycle signal according to the mapping relations
Row alignment, the length of the cardiac cycle signal after each signal amplitude is normalized, which is normalized, includes:
Wave crest point full-length Lf according to preset full-length L and main wave wave crest point relative to starting point, to each letter
Cardiac cycle signal after number amplitude normalization carries out length normalization method processing according to following formula:
Wherein, W is the length of each cardiac cycle signal, and w1 is the main wave wave crest of each cardiac cycle signal
Length of the position relative to its starting point, XiFor the remaining data points in each cardiac cycle signal in addition to key point.
In specific implementation, described that feature extraction is carried out to each normalized signal using CNN network, it obtains each described
The corresponding first identity attribute feature of cardiac cycle signal includes:
Adopt using the input picture that two paths of data point sequence of the convolution sum pond to each normalized signal is formed
Sample obtains the first down-sampling feature vector of the input picture；
The input picture is up-sampled using deconvolution, obtain the input picture first up-sampling feature to
Amount；
The first up-sampling feature vector and the first down-sampling feature vector are subjected to group by way of superposition
It closes, obtains multiple feature vectors of each cardiac cycle signal；
The corresponding multiple feature vectors of each cardiac cycle signal are compressed by way of convolution, will be obtained
N dimension identity attribute feature vector as the corresponding first identity attribute feature of each cardiac cycle signal, wherein n is positive
Integer.
In order to enable extracting identity attribute feature using CNN network has preferable robustness, the CNN network is being constructed
When used loss function can be with are as follows:
Wherein,
cos(θj, i) and=Wj Txi,
Wherein,
S120, by each first identity attribute feature mark at least one pre-stored registered users respectively
Whether the second identity attribute feature of quasi- PPG signal compares, judge the two from same user.
In a kind of implementation, it is described by each first identity attribute feature respectively with it is pre-stored at least one
Second identity attribute feature of the standard PPG signal of registered users compares, and judges whether the two wraps from same user
It includes:
Each first identity attribute feature and at least one pre-stored registered use are calculated according to following formula
Similarity sim (X1, X2) between second identity attribute feature of the standard PPG signal at family:
Wherein, X1 is each first identity attribute feature, and X2 is pre-stored one of registered users
Second identity attribute feature of standard PPG signal；
The size relation for judging each similarity and threshold value Th then determines when the similarity is greater than threshold value Th
The corresponding cardiac cycle signal of X1, X2 comes from same user, otherwise, then determines X1, the corresponding cardiac cycle signal of X2 is not from
Same user；
It votes according to following formula the corresponding identity information of multiple cardiac cycle signals, obtains active user
The corresponding identity information C* of PPPG signal:
Wherein fijPass through the posterior probability (posteriority here that certification is i-th of registration user for j-th of cardiac cycle signal
Probability is the cosine similarity sim that front calculates), m is the number of multiple cardiac cycle signals, and m, i, j are positive whole
Number.
S130, if judging result is from same user, using the identity information of pre-stored corresponding user as working as
The identity information of preceding user.
This specification embodiment carries out identification and certification using PPG signal, multiple aroused in interest by the way that PPG signal to be divided into
Periodic signal, and then the identity attribute feature that cardiac cycle signal is objective, can reflect different user physiological characteristic is extracted, it will
Extracted identity attribute feature and the identity attribute feature of pre-stored registered users PPG signal compare, judgement
Whether two signals are from the same person, to identify which in database be current PPG signal correspondence have already registered with
One ID realizes automated validation and the identification of the user identity based on PPG signal；Simultaneously as this specification embodiment institute
Using the physiological attribute feature for the different user that PPG signal is reflected, rather than the feature artificially designed, so that this
The personal identification method of specification embodiment is available to certification preferably feature, has preferable robustness.
In a kind of implementation, if being from same user, by pre-stored corresponding user in the judging result
Identity information as the identity information of active user after further include:
Pre-stored, registered users PPG signal the second identity attribute feature vector is carried out according to following formula
It updates:
Xnew=(1- β) * X+ β * Xcur
Wherein, X is the second identity attribute feature vector of user i registration before updating, and Xcur is that current detection mentions in the process
The first identity attribute feature vector with X from same registration user i is obtained and is identified as, Xnew is user after updating
Second identity attribute feature vector of i registration；β is to update coefficient, can also be considered as learning rate.Renewal rate when β value is larger
Fastly, renewal speed is slow when β is smaller, rule of thumb generally can be set to 10e-4To 10e-6Between.
Fig. 2 is the identity identifying method flow chart of another embodiment of this specification；As shown in Fig. 2, the identity identifying method
Include:
1, Signal Pretreatment
Noise reduction process is carried out to the PPG signal of active user collected, it specifically, can be using bandpass filter to going through
History PPG signal carries out noise reduction process to eliminate baseline drift and partial noise.
2, signal is divided
The effect of signal segmentation is that continuous P PG signal is divided into the set of complete cardiac cycle signal.When can use
Domain algorithm is split PPG signal, such as the method based on gradient, after seeking gradient to signal, obtains wave crest and wave
Valley point is then based on heuristic criterion and analyzes to obtain the start-stop point position in each period.
3, critical point detection
Key in this example refers to starting point, terminating point and the main wave wave crest of a complete cardiac cycle PPG signal
Point.As shown in figure 3, S is the starting point of a complete cardiac cycle signal, E is terminating point, and A is main wave wave crest point.
Its starting point and terminating point are the starting point and terminating point of signal after the completion of periodic signal segmentation.Main wave wave crest point then leads to
It crosses amplitude maximum and judgment criterion that gradient is zero is detected to obtain to the time-domain signal of a cycle.
4, signal alignment
Signal alignment refers to that the key point in one that will enter into this system complete periodic signal carries out in this example
Alignment, mapping ruler when remaining data points are according to crucial point alignment are mapped.The effect of signal alignment is primarily to disappear
Except the difference between the PPG signal of the same ID, these differences may be because transformation, the movement etc. of electric signal power cause
's.
Signal alignment is divided into the progress of two steps:
Step 1: signal amplitude normalizes
Assuming that main wave crest height is H, the height after normalization is NH, then for the signal X of a complete cycle, width
Degree is normalized as follows
Xn=X* (NH/H)
Wherein, Xn is the signal after amplitude normalization.
Step 2: main wave wave crest point alignment and signal length normalization
Why select main wave wave crest as key point because of key feature points such as dicrotic waves often by signal
Interference or gathered person's self reason lead to not obviously observe, and main wave wave crest point is usually more stable and is easy to pass through
Algorithm detects.The position of main wave wave crest point is 42 after normalized signal length is set as 128 and normalizes.As shown in figure 3,
Assuming that being W1 to normalized signal main wave wave crest point position, then the planning of signal in the longitudinal direction carries out according to the following formula:
Wherein i is signaling point XiAbscissa.
5, feature extraction
Feature extracting method in the present embodiment uses the neural network based on depth convolution, is situated between by taking u-net as an example
It continues, this method can extend to the general depth convolutional network such as GoogLenet, resnet.
Here input signal is that two-way passes through normalized feux rouges and infrared signal, and the length of signal is 128.
In a kind of implementation, feature extraction is carried out to each normalized signal using CNN network, obtains each heart
Moving the corresponding first identity attribute feature of periodic signal includes:
Adopt using the input picture that two paths of data point sequence of the convolution sum pond to each normalized signal is formed
Sample obtains the first down-sampling feature vector of the input picture；
The input picture is up-sampled using deconvolution, obtain the input picture first up-sampling feature to
Amount；
The first up-sampling feature vector and the first down-sampling feature vector are subjected to group by way of superposition
It closes, obtains multiple feature vectors of each cardiac cycle signal；
The corresponding multiple feature vectors of each cardiac cycle signal are compressed by way of convolution, will be obtained
N dimension identity attribute feature vector as the corresponding first identity attribute feature of each cardiac cycle signal, wherein n is positive
Integer.
In specific implementation, port number is stepped up during down-sampling to avoid information excessive loss, is adopted upper
Port number is gradually reduced using convolution during sample.
Fig. 4 is the U-net network structure of this specification one embodiment；As shown in figure 4, CNN network uses U-net net
Network, what the convolution in the present embodiment was all made of is the convolution filter of 3x3, and nonlinear activation function uses Relu (Rectified
Linear Unit, line rectification function) function.Number in Fig. 4 in U-net network structure is characteristic layer in the present embodiment
Quantity.Downward arrow indicates down-sampling (method in max-pooling (maximum pond) is used in this example)；Upward arrow
Head indicates to use the deconvolution of 2x2.
U-Net makees down-sampling to input picture first with convolution sum pond in the present embodiment, during down-sampling gradually
Increase port number to avoid information excessive loss；Then deconvolution (transposition convolution) is recycled to up-sample image, it will be upper
Obtained feature vector is sampled to combine with shallow-layer feature vector (i.e. the feature vector of down-sampling).In the interval of up-sampling, benefit
Port number is gradually reduced with convolution.Different from the method summed pixel-by-pixel that FCN is used, U-Net middle-shallow layer Fusion Features use
The mode of superposition directly combines shallow-layer feature with up-sampling feature vector, generates the spy that port number is the sum of the two port number
Levy vector.It is T by the characteristic that U-net is finally extracted, T can empirically take the numerical value between 128~1024.The value
The classification number for needing to identify in general viewing system, i.e., when scene huge applied to number, the spy of CNN extraction in training process
Sign number can be set larger accordingly.
Certification and identification:
Use in the training process the most common loss function of Verification System based on depth convolutional neural networks for
Softmax loss function, i.e.,
Wherein N is sample number, and pi is the posterior probability of identification, and yi is to identify correct label, and C is the class in identifying system
Other quantity.
Since the scale of weight vectors Wj and feature vector x are related to signal, when signal normalization effect is not achieved
It is expected that when, it is very big to will cause the fj difference that different training samples are calculated, and eventually leads to model and is not easy to restrain, therefore this Shen
Please in example using to weight vectors and the normalized method of feature vector and using the softmax loss after normalization into
Row training.Softmax loss after normalization is
Wherein,
cos(θj, i) and=Wj Txi,
Certification mainly calculates the distance between the feature of two signal extraction, i.e. cosine similarity using cosine function
Wherein, X1, X2 are to extract obtained feature.When similarity is greater than threshold value Th, i.e. when sim (X1, X2) > Th, then recognize
It is same people for X1, X2；Conversely, then X1, X2 come from different people.
The method that recognition methods in present application example uses authentication result Nearest Neighbor with Weighted Voting.Assuming that system acquires m altogether
The signal in period simultaneously, shares C classification, namely have registered C user, then recognition result is in identifying system
Wherein fijPass through the posterior probability that certification is i-th of classification for j-th of periodic signal, before posterior probability here is
The cosine similarity sim that face calculates.
Self-adaptive features study and update:
Since with advancing age, certain variation can occur for the position of dicrotic wave in PPG signal, and this variation is slow
Accuracy that is slow and will affect the certification based on PPG signal, present application example propose a kind of study of self-adaptive features and
The method of online updating adapts to the slow transformation of human body PPG signal, this is also one of the inventive point of this specification embodiment.
Xnew=(1- β) * X+ β * Xcur
Wherein, Xcur extracts obtained feature for current detection in the process, and is to come from feature X by system identification
The same registration user；Xnew is updated registration feature；β is to update coefficient.
It is adapted with above method embodiment, Fig. 5 is the identity recognition device module map of one embodiment of this specification.Such as figure
Shown in 5, which includes:
Divide extraction module 510, is configured as the PPG signal of acquired active user being divided into multiple cardiac cycles
Signal carries out feature extraction to each cardiac cycle signal, obtains corresponding first identity category of each cardiac cycle
Property feature；
Contrast judgement module 520, be configured as by each first identity attribute feature respectively with it is pre-stored extremely
Whether the second identity attribute feature of the standard PPG signal of few registered users compares, judge the two from same
User；
Authentication module 530, if being configured as judging result is from same user, by pre-stored to application
Identity information of the identity information at family as active user.
Optionally, the segmentation extraction module includes:
Cutting unit is configured as the PPG signal of acquired active user being split, obtains multiple cardiac cycles
Signal；
Detection unit is configured as detecting the key point of each cardiac cycle signal；
Normalization unit is configured as believing each cardiac cycle signal and reference period based on the key point
It number is aligned, obtains the corresponding normalized signal of each cardiac cycle signal；
Feature extraction unit is configured as carrying out feature extraction to each normalized signal using CNN network, obtain each
The corresponding first identity attribute feature of the cardiac cycle signal.
Optionally, the cutting unit includes:
Peak valley obtains subelement, is configured as obtaining the wave crest and trough of the PPG signal of active user using Time-Domain algorithm；
Start-stop point obtains subelement, is configured as analyzing acquired wave crest and trough based on heuristic criterion,
Obtain the starting point and terminating point of multiple cardiac cycle signals and each cardiac cycle signal.
Optionally, the key point of each cardiac cycle signal includes starting point, terminating point and main wave wave crest point, the inspection
Unit is surveyed to be configured to:
The starting point and terminating point obtained when each cardiac cycle signal is divided passes through as its starting point and terminating point
The judgment criterion that amplitude maximum and gradient are zero is detected to obtain each described to the time-domain signal of each cardiac cycle
The main wave wave crest point of cardiac cycle signal.
Optionally, the normalization unit includes:
Amplitude normalization subelement is configured as that the signal amplitude of each cardiac cycle signal is normalized；
Be aligned subelement, be configured as by each signal amplitude normalize after cardiac cycle signal key point respectively with
The corresponding crucial point alignment of reference cycle signal, obtains the mapping relations between key point；
Length normalization method subelement, be configured as according to the mapping relations by each cardiac cycle signal remaining
Data point is aligned, and the length of the cardiac cycle signal after the normalization of each signal amplitude is normalized.
Optionally, the amplitude normalization subelement be configured to include:
Assuming that the main wave crest height of each cardiac cycle signal is H, the main wave crest height after normalization is NH,
Then for each cardiac cycle signal X, signal amplitude is normalized as follows:
Xn=X* (NH/H),
Wherein, Xn is the signal after signal amplitude normalization.
Optionally, the length normalization method subelement is configured to:
Wave crest point full-length Lf according to preset full-length L and main wave wave crest point relative to starting point, to each letter
Cardiac cycle signal after number amplitude normalization carries out length normalization method processing according to following formula:
Wherein, W is the length of each cardiac cycle signal, and w1 is the main wave wave crest of each cardiac cycle signal
Length of the position relative to its starting point, XiFor the remaining data points in each cardiac cycle signal in addition to key point.
Optionally, the feature extraction unit includes:
Down-sampling subelement is configured as the two paths of data point sequence shape using convolution sum pond to each normalized signal
At input picture carry out down-sampling, obtain the first down-sampling feature vector of the input picture；
Subelement is up-sampled, is configured as up-sampling the input picture using deconvolution, obtains the input
First up-sampling feature vector of image；
Stack combinations subelement, be configured as by it is described first up-sampling feature vector and the first down-sampling feature to
Amount is combined by way of superposition, obtains multiple feature vectors of each cardiac cycle signal；
Subelement is compressed, is configured as passing through convolution to the corresponding multiple feature vectors of each cardiac cycle signal
Mode is compressed, using obtained n dimension identity attribute feature vector as corresponding first body of each cardiac cycle signal
Part attributive character, wherein n is positive integer.
Optionally, the contrast judgement module includes:
Similarity calculated is configured as calculating each first identity attribute feature and in advance according to following formula
Storage at least one registered users standard PPG signal the second identity attribute feature between similarity sim (X1,
X2):
Wherein, X1 is each first identity attribute feature, and X2 is pre-stored one of registered users
Second identity attribute feature of standard PPG signal；
Threshold decision unit is configured as judging the size relation of each similarity and threshold value Th, when described similar
When degree is greater than threshold value Th, then X1 is determined, the corresponding cardiac cycle signal of X2 comes from same user, otherwise, then determines that X1, X2 are corresponding
Cardiac cycle signal come from different user；
It votes unit, is configured as carrying out the corresponding identity information of multiple cardiac cycle signals according to following formula
Ballot, obtains the corresponding identity information C of PPPG signal of active user*:
Wherein fijPass through the posterior probability that certification is i-th of registration user for j-th of cardiac cycle signal, m is multiple institutes
The number of cardiac cycle signal is stated, m, i, j are positive integer.
Optionally, the identification authentication system further include:
Update module is configured as the second identity according to following formula to pre-stored registered users PPG signal
Attribute feature vector is updated:
Xnew=(1- β) * X+ β * Xcur
Wherein, X is the second identity attribute feature vector of user i registration before updating, and Xcur is that current detection mentions in the process
The first identity attribute feature vector with X from same registration user i is obtained and is identified as, Xnew is user after updating
Second identity attribute feature vector of i registration；β is to update coefficient.
Optionally, the used loss function when constructing the CNN network are as follows:
Wherein,
cos(θj, i) and=Wj Txi,
Wherein, W, W※Weight vectors after respectively normalizing and before normalization, x, x※Be divided into for normalization after and normalizing
Feature vector before change, θjBe weight vectors WjWith feature vector xiAngle.
In conclusion this specification embodiment realize it is following the utility model has the advantages that
This specification embodiment carries out identification and certification using PPG signal, multiple aroused in interest by the way that PPG signal to be divided into
Periodic signal, and then the identity attribute feature that cardiac cycle signal is objective, can reflect different user physiological characteristic is extracted, it will
Extracted identity attribute feature and the identity attribute feature of pre-stored registered users PPG signal compare, judgement
Whether two signals are from the same person, to identify which in database be current PPG signal correspondence have already registered with
One ID realizes automated validation and the identification of the user identity based on PPG signal；Simultaneously as this specification embodiment institute
Using the physiological attribute feature for the different user that PPG signal is reflected, rather than the feature artificially designed, so that this
The personal identification method of specification embodiment is available to certification preferably feature, has preferable robustness.
This specification embodiment is based on CNN network and carries out the feature extraction of PPG signal attribute, in one embodiment convolution
Neural network is using U-Net, and unconventional CNN network.Because PPG Signal-to-Noise is lower, before entering CNN network
Straight line drift and a part of noise can be removed by carrying out noise reduction, alleviate signal amplitude variation caused by equipment dims, but can not reach
To ideal effect, i.e. PPG signal after conventional method noise reduction still contains a large amount of noise.Currently used AlexNet,
The typical case such as GoolgeNet and resNet CNN extracts feature on the PPG signal directly after noise reduction, network performance also can be by
Large effect.Noise can make the between class distance of pattern feature become smaller, and network performance is caused to be deteriorated, and U-Net passes through convolution
Down-sampling and up-sampling also achieve the denoising and reconstruction process of signal while extracting feature, therefore the feature extracted is more
Add robust.
Those of ordinary skill in the art will appreciate that: attached drawing is the schematic diagram of one embodiment, module in attached drawing or
Process is not necessarily implemented necessary to the present invention.
Those of ordinary skill in the art will appreciate that: the module in device in embodiment can describe to divide according to embodiment
It is distributed in the device of embodiment, corresponding change can also be carried out and be located in one or more devices different from the present embodiment.On
The module for stating embodiment can be merged into a module, can also be further split into multiple submodule.
Finally, it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: it still may be used
To modify to technical solution documented by previous embodiment or equivalent replacement of some of the technical features；And
These are modified or replaceed, the spirit and model of technical solution of the embodiment of the present invention that it does not separate the essence of the corresponding technical solution
It encloses.
Claims (10)
1. a kind of personal identification method, which comprises the following steps:
The PPG signal of acquired active user is divided into multiple cardiac cycle signals, to each cardiac cycle signal
Feature extraction is carried out, corresponding first identity attribute feature of each cardiac cycle is obtained；
By each first identity attribute feature respectively with the standard PPG signal of at least one pre-stored registered users
The second identity attribute feature compare, whether both judge from same user；
If judging result is from same user, using the identity information of pre-stored corresponding user as the body of active user
Part information.
2. personal identification method according to claim 1, which is characterized in that the PPG by acquired active user
Signal is divided into multiple cardiac cycle signals, carries out feature extraction to each cardiac cycle signal, obtains each heart
Moving period corresponding first identity attribute feature includes:
The PPG signal of acquired active user is split, multiple cardiac cycle signals are obtained；
Detect the key point of each cardiac cycle signal；
Based on the key point, each cardiac cycle signal is aligned with reference cycle signal, is obtained each described
The corresponding normalized signal of cardiac cycle signal；
Feature extraction is carried out to each normalized signal using CNN network, obtains each cardiac cycle signal corresponding the
One identity attributive character.
3. identity identifying method described in any one of -2 according to claim 1, which is characterized in that it is described will be acquired current
The PPG signal of user is split, and is obtained multiple cardiac cycle signals and is included:
The wave crest and trough of the PPG signal of active user are obtained using Time-Domain algorithm；
Acquired wave crest and trough are analyzed based on heuristic criterion, obtain multiple cardiac cycle signals and each institute
State the starting point and terminating point of cardiac cycle signal.
4. identity identifying method according to any one of claim 1-3, which is characterized in that each cardiac cycle letter
Number key point include starting point, terminating point and main wave wave crest point, the key point packet of each cardiac cycle signal of detection
It includes:
The starting point and terminating point obtained when each cardiac cycle signal is divided is as its starting point and terminating point；
The judgment criterion for being zero by amplitude maximum and gradient is detected to obtain to the time-domain signal of each cardiac cycle
The main wave wave crest point of each cardiac cycle signal.
5. identity identifying method described in any one of -4 according to claim 1, which is characterized in that described to be based on the key
Each cardiac cycle signal is aligned by point with reference cycle signal, and it is corresponding to obtain each cardiac cycle signal
Normalized signal include:
The signal amplitude of each cardiac cycle signal is normalized；
The key point of cardiac cycle signal after the normalization of each signal amplitude is corresponding with reference cycle signal crucial respectively
Point alignment obtains the mapping relations between key point；
The remaining data points of each cardiac cycle signal are aligned according to the mapping relations, by each signal width
The length of cardiac cycle signal after degree normalization is normalized.
6. identity identifying method according to any one of claims 1-5, which is characterized in that described to each described aroused in interest
The signal amplitude of periodic signal, which is normalized, includes:
Assuming that the main wave crest height of each cardiac cycle signal is H, the main wave crest height after normalization is NH, then right
In each cardiac cycle signal X, signal amplitude is normalized as follows:
Xn=X* (NH/H),
Wherein, Xn is the signal after signal amplitude normalization.
7. identity identifying method according to claim 1 to 6, which is characterized in that described to be closed according to the mapping
The remaining data points of each cardiac cycle signal are aligned by system, by the week aroused in interest after the normalization of each signal amplitude
The length of phase signal, which is normalized, includes:
Wave crest point full-length Lf according to preset full-length L and main wave wave crest point relative to starting point, to each signal width
Cardiac cycle signal after degree normalization carries out length normalization method processing according to following formula:
Wherein, W is the length of each cardiac cycle signal, and w1 is the position of the main wave wave crest of each cardiac cycle signal
Set the length relative to its starting point, XiFor the remaining data points in each cardiac cycle signal in addition to key point.
8. identity identifying method described in any one of -7 according to claim 1, which is characterized in that described to use CNN network pair
Each normalized signal carries out feature extraction, obtains the corresponding first identity attribute feature packet of each cardiac cycle signal
It includes:
Down-sampling is carried out using the input picture that two paths of data point sequence of the convolution sum pond to each normalized signal is formed, is obtained
To the first down-sampling feature vector of the input picture；
The input picture is up-sampled using deconvolution, obtains the first up-sampling feature vector of the input picture；
The first up-sampling feature vector and the first down-sampling feature vector are combined by way of superposition, obtained
To multiple feature vectors of each cardiac cycle signal；
The corresponding multiple feature vectors of each cardiac cycle signal are compressed by way of convolution, the n that will be obtained
Identity attribute feature vector is tieed up as the corresponding first identity attribute feature of each cardiac cycle signal, wherein n is positive whole
Number.
9. identity identifying method according to claim 1 to 8, which is characterized in that described by each described first
Identity attribute feature the second identity attribute feature with the standard PPG signal of at least one pre-stored registered users respectively
It compares, judges whether the two comes from same user and include:
Each first identity attribute feature and at least one pre-stored registered users are calculated according to following formula
Similarity sim (X1, X2) between second identity attribute feature of standard PPG signal:
Wherein, X1 is each first identity attribute feature, and X2 is the standard of pre-stored one of registered users
Second identity attribute feature of PPG signal；
The size relation for judging each similarity and threshold value Th then determines X1, X2 when the similarity is greater than threshold value Th
Corresponding cardiac cycle signal comes from same user, otherwise, then determines X1, the corresponding cardiac cycle signal of X2 comes from different use
Family；
It votes according to following formula the corresponding identity information of multiple cardiac cycle signals, obtains active user's
The corresponding identity information C of PPPG signal*:
Wherein fijPass through the posterior probability that certification is i-th of registration user for j-th of cardiac cycle signal, m is multiple hearts
The number of dynamic periodic signal, m, i, j are positive integer.
10. a kind of identity recognition device characterized by comprising
Divide extraction module, is configured as the PPG signal of acquired active user being divided into multiple cardiac cycle signals, it is right
Each cardiac cycle signal carries out feature extraction, obtains corresponding first identity attribute feature of each cardiac cycle；
Contrast judgement module, be configured as by each first identity attribute feature respectively with it is pre-stored at least one
Whether the second identity attribute feature for registering the standard PPG signal of user compares, judge the two from same user；
Authentication module, if being configured as judging result is from same user, by the body of pre-stored corresponding user
Identity information of part information as active user.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130479.3A CN109919050B (en) | 2019-02-21 | 2019-02-21 | Identity recognition method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910130479.3A CN109919050B (en) | 2019-02-21 | 2019-02-21 | Identity recognition method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109919050A true CN109919050A (en) | 2019-06-21 |
| CN109919050B CN109919050B (en) | 2021-07-20 |
Family
ID=66962093
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910130479.3A Active CN109919050B (en) | 2019-02-21 | 2019-02-21 | Identity recognition method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109919050B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458197A (en) * | 2019-07-11 | 2019-11-15 | 启东市知微电子科技有限公司 | Personal identification method and its system based on photoplethysmographic |
| CN111046738A (en) * | 2019-11-14 | 2020-04-21 | 五邑大学 | Precision improvement method of light u-net for finger vein segmentation |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140148709A1 (en) * | 2012-11-23 | 2014-05-29 | Pixart Imaging Incorporation | System and method for integrating heart rate measurement and identity recognition |
| CN104102915A (en) * | 2014-07-01 | 2014-10-15 | 清华大学深圳研究生院 | Multiple-template matching identity recognition method based on ECG (Electrocardiogram) under electrocardiogram abnormality state |
| CN104573458A (en) * | 2014-12-30 | 2015-04-29 | 深圳先进技术研究院 | Identity recognition method, device and system based on electrocardiogram signals |
| CN106473750A (en) * | 2016-10-08 | 2017-03-08 | 西安电子科技大学 | Personal identification method based on photoplethysmographic optimal period waveform |
| CN107088069A (en) * | 2017-03-29 | 2017-08-25 | 西安电子科技大学 | Personal identification method based on human body PPG signal subsections |
-
2019
- 2019-02-21 CN CN201910130479.3A patent/CN109919050B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140148709A1 (en) * | 2012-11-23 | 2014-05-29 | Pixart Imaging Incorporation | System and method for integrating heart rate measurement and identity recognition |
| CN104102915A (en) * | 2014-07-01 | 2014-10-15 | 清华大学深圳研究生院 | Multiple-template matching identity recognition method based on ECG (Electrocardiogram) under electrocardiogram abnormality state |
| CN104573458A (en) * | 2014-12-30 | 2015-04-29 | 深圳先进技术研究院 | Identity recognition method, device and system based on electrocardiogram signals |
| CN106473750A (en) * | 2016-10-08 | 2017-03-08 | 西安电子科技大学 | Personal identification method based on photoplethysmographic optimal period waveform |
| CN107088069A (en) * | 2017-03-29 | 2017-08-25 | 西安电子科技大学 | Personal identification method based on human body PPG signal subsections |
Non-Patent Citations (3)
| Title |
|---|
| LUKE EVERSON ET AL.: "BiometricNet: Deep Learning based Biometric Identification using Wrist-Worn PPG", 《ISCAS》 * |
| 滕毅: "基于光电容积脉搏波的生物识别研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 翁剑枫 等: "有效保留血液容积图(PPG)波形信息的新方法", 《航天医学与医学工程》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110458197A (en) * | 2019-07-11 | 2019-11-15 | 启东市知微电子科技有限公司 | Personal identification method and its system based on photoplethysmographic |
| CN111046738A (en) * | 2019-11-14 | 2020-04-21 | 五邑大学 | Precision improvement method of light u-net for finger vein segmentation |
| CN111046738B (en) * | 2019-11-14 | 2023-03-17 | 五邑大学 | Precision improvement method of light u-net for finger vein segmentation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109919050B (en) | 2021-07-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104573458B (en) | A kind of personal identification method based on electrocardiosignal, apparatus and system | |
| CN102542281B (en) | Non-contact biometric feature identification method and system | |
| CN106473750B (en) | Personal identification method based on photoplethysmographic optimal period waveform | |
| CN109543526B (en) | True and false facial paralysis recognition system based on depth difference characteristics | |
| CN106951753B (en) | Electrocardiosignal authentication method and device | |
| CN106980852B (en) | Based on Corner Detection and the medicine identifying system matched and its recognition methods | |
| CN111461176A (en) | Multi-mode fusion method, device, medium and equipment based on normalized mutual information | |
| CN106778664A (en) | The dividing method and its device of iris region in a kind of iris image | |
| CN103793690A (en) | Human body biotic living body detection method based on subcutaneous bloodstream detection and application | |
| CN106557723A (en) | A kind of system for face identity authentication with interactive In vivo detection and its method | |
| CN111062021B (en) | Method and device for identity authentication based on wearable equipment | |
| CN110786849B (en) | Electrocardiosignal identity recognition method and system based on multi-view discriminant analysis | |
| CN107293301B (en) | Recognition methods and system based on dental articulation sound | |
| CN107169479A (en) | Intelligent mobile equipment sensitive data means of defence based on fingerprint authentication | |
| CN109472247A (en) | Face identification method based on the non-formula of deep learning | |
| CN109063643A (en) | A kind of facial expression pain degree recognition methods under the hidden conditional for facial information part | |
| CN109919050A (en) | Identity recognition method and device | |
| Wang et al. | Stable EEG Biometrics Using Convolutional Neural Networks and Functional Connectivity. | |
| CN114578963A (en) | Electroencephalogram identity recognition method based on feature visualization and multi-mode fusion | |
| CN116881853B (en) | Attention assessment method, system, equipment and medium based on multi-mode fusion | |
| CN107977622B (en) | Eye state detection method based on pupil characteristics | |
| CN110547806B (en) | Gesture action online recognition method and system based on surface electromyographic signals | |
| CN115240241A (en) | Emotion recognition method and device for road and bridge engineering scene | |
| CN106548112A (en) | A kind of personal identification method and system | |
| CN109255318A (en) | Based on multiple dimensioned and multireel lamination Fusion Features fingerprint activity test methods |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |