CN110490245A - Authentication model training method and device, storage medium, electronic equipment - Google Patents
Authentication model training method and device, storage medium, electronic equipment Download PDFInfo
- Publication number
- CN110490245A CN110490245A CN201910750225.1A CN201910750225A CN110490245A CN 110490245 A CN110490245 A CN 110490245A CN 201910750225 A CN201910750225 A CN 201910750225A CN 110490245 A CN110490245 A CN 110490245A
- Authority
- CN
- China
- Prior art keywords
- identity
- branching networks
- network
- authentication model
- information
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Security & Cryptography (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Collating Specific Patterns (AREA)
Abstract
The disclosure provides a kind of authentication model training method and device, electronic equipment, storage medium;It is related to field of artificial intelligence.The authentication model training method includes: to obtain sample identity characteristic and the corresponding identity label of the sample identity characteristic;The first study processing is carried out to be trained to the identity branching networks to the identity branching networks in pre-established authentication model according to the sample identity characteristic;The second study processing is carried out to be trained to the field branching networks to the field branching networks in the authentication model according to the sample identity characteristic;Third study processing is carried out to construct the authentication model according to the identity branching networks after training to the identity branching networks by the field branching networks after the identity label and training.The disclosure can eliminate the field difference in the identity characteristic data of same people when carrying out authentication.
Description
Technical field
This disclosure relates to field of artificial intelligence, in particular to a kind of authentication model training method, identity
Verify model training apparatus, electronic equipment and computer readable storage medium.
Background technique
With the development of internet technology, identity validation technology achieves huge progress, including fingerprint authentication, face are tested
Card, vocal print/iris verification, pedestrian identify again.Particularly, with continuous universal, the body based on intelligent terminal of intelligent terminal
Part verification technique also develops rapidly.
However, identity identifying method in the prior art often faces the deviation in data, such as field difference, i.e., one
A model in one field but verified in another field by training.Such as in pedestrian again identification field, work as season
Cause people to wear the clothes when changing or when the relative angle of people and video camera changes, verifying may all be affected, and cause
It is verified that rate is lower, influences the usage experience of user.
It should be noted that information is only used for reinforcing the reason to the background of the disclosure disclosed in above-mentioned background technology part
Solution, therefore may include the information not constituted to the prior art known to persons of ordinary skill in the art.
Summary of the invention
The disclosure is designed to provide a kind of authentication model training method, authentication model training device, electricity
Sub- equipment and computer readable storage medium, and then overcome the limitation and defect due to the relevant technologies to a certain extent and lead
The problem of causing, field difference in identity characteristic data can not be eliminated when carrying out authentication.
According to the disclosure in a first aspect, providing a kind of authentication model training method, comprising:
Obtain sample identity characteristic and the corresponding identity label of the sample identity characteristic;
The is carried out to the identity branching networks in pre-established authentication model according to the sample identity characteristic
One study processing is to be trained the identity branching networks;
Second is carried out to the field branching networks in the authentication model according to the sample identity characteristic to learn
Processing is practised to be trained to the field branching networks;
The is carried out to the identity branching networks by the field branching networks after the identity label and training
Three study processing are to construct the authentication model according to the identity branching networks after training.
In a kind of exemplary embodiment of the disclosure, the authentication model further includes Feature Conversion network, in root
The first study processing is carried out to the identity branching networks in pre-established authentication model according to the sample identity characteristic
Before being trained to the identity branching networks, the method also includes:
Conversion process is carried out to generate the sample to the sample identity characteristic according to the Feature Conversion network
The corresponding hidden layer feature vector of identity characteristic data.
In a kind of exemplary embodiment of the disclosure, the identity branching networks include that identity information extracts network, body
Part information differentiates network and target loss function;
It is described according to the sample identity characteristic to the identity branching networks in pre-established authentication model into
The study of row first, which is handled to be trained to the identity branching networks, includes:
By the identity information differentiate network and the target loss function to the identity information extract network into
Row unsupervised learning is handled so that the identity information, which extracts network, generates the first attribute vector according to the hidden layer feature vector;
Wherein first attribute vector includes identity characteristic information and does not include domain features information.
In a kind of exemplary embodiment of the disclosure, the field branching networks include that realm information extracts network, neck
Domain information differentiates network and Target Countermeasure loss function;
It is described that the is carried out to the field branching networks in the authentication model according to the sample identity characteristic
Two study, which are handled to be trained to the field branching networks, includes:
Differentiate that network and the Target Countermeasure loss function extract net to the realm information by the realm information
Network carries out unsupervised learning processing so that the realm information, which extracts network, generates the second attribute according to the hidden layer feature vector
Vector;Wherein second attribute vector includes domain features information and does not include identity characteristic information.
In a kind of exemplary embodiment of the disclosure, pass through the field branch after the identity label and training
Network carries out third study processing according to the identity branching networks building after training to the identity branching networks
Authentication model includes:
It is exercised supervision by the identity label to identity information extraction network after learning processing and passing through training
The field branching networks to the identity information extract network carry out confrontation study handle so that the identity information extract
Network generates identity characteristic vector according to the sample identity characteristic.
In a kind of exemplary embodiment of the disclosure, the target loss function includes depth measure study loss function
Or cross entropy loss function;And calculation processing is carried out to the target loss function and generates the Target Countermeasure loss letter
Number.
According to the second aspect of the disclosure, a kind of auth method is provided, comprising:
Current identity characteristic data are obtained, and the current identity characteristic data are input to the authentication mould of pre-training
With the corresponding current signature vector of the determination current identity characteristic data in type;
Pre-stored original identity characteristic data are obtained, and the original identity characteristic data are input to the identity and are tested
With the corresponding original feature vector of the determination original identity characteristic data in model of a syndrome;
If the difference of the current signature vector and the original feature vector is less than preset threshold, it is determined that described current
Identity characteristic data pass through authentication.
According to the third aspect of the disclosure, a kind of authentication model training device is provided, comprising:
Sample data acquiring unit, it is corresponding for obtaining sample identity characteristic and the sample identity characteristic
Identity label;
Identity branching networks training unit, for according to the sample identity characteristic to pre-established authentication mould
Identity branching networks in type carry out the first study processing to be trained to the identity branching networks;
Field branching networks training unit is used for according to the sample identity characteristic in the authentication model
Field branching networks carry out second study processing to be trained to the field branching networks;
Authentication model construction unit, for passing through the field branching networks after the identity label and training
Third study processing is carried out to construct the identity according to the identity branching networks after training to the identity branching networks
Verify model.
In a kind of exemplary embodiment of the disclosure, the authentication model training device further include hidden layer feature to
Generation unit is measured, the hidden layer feature vector generation unit is configured as:
Conversion process is carried out to generate the sample to the sample identity characteristic according to the Feature Conversion network
The corresponding hidden layer feature vector of identity characteristic data.
In a kind of exemplary embodiment of the disclosure, the identity branching networks include that identity information extracts network, body
Part information differentiates network and target loss function;
The identity branching networks training unit is configured as:
By the identity information differentiate network and the target loss function to the identity information extract network into
Row unsupervised learning is handled so that the identity information, which extracts network, generates the first attribute vector according to the hidden layer feature vector;
Wherein first attribute vector includes identity characteristic information and does not include domain features information.
In a kind of exemplary embodiment of the disclosure, the field branching networks include that realm information extracts network, neck
Domain information differentiates network and Target Countermeasure loss function;
The field branching networks training unit is configured as:
Differentiate that network and the Target Countermeasure loss function extract net to the realm information by the realm information
Network carries out unsupervised learning processing so that the realm information, which extracts network, generates the second attribute according to the hidden layer feature vector
Vector;Wherein second attribute vector includes domain features information and does not include identity characteristic information.
In a kind of exemplary embodiment of the disclosure, the authentication model construction unit is configured as:
It is exercised supervision by the identity label to identity information extraction network after learning processing and passing through training
The field branching networks to the identity information extract network carry out confrontation study handle so that the identity information extract
Network generates identity characteristic vector according to the sample identity characteristic.
In a kind of exemplary embodiment of the disclosure, the target loss function includes depth measure study loss function
Or cross entropy loss function;And calculation processing is carried out to the target loss function and generates the Target Countermeasure loss letter
Number.
According to the fourth aspect of the disclosure, a kind of authentication means are provided, comprising:
Current signature vector determination unit, for obtaining current identity characteristic data, and by the current identity characteristic number
According to being input in the authentication model of pre-training with the corresponding current signature vector of the determination current identity characteristic data;
Original feature vector determination unit, for obtaining pre-stored original identity characteristic data, and by the original body
Part characteristic be input in the authentication model with the corresponding primitive character of the determination original identity characteristic data to
Amount;
Authentication is by unit, if being less than for the difference of the current signature vector and the original feature vector pre-
If threshold value, it is determined that the current identity characteristic data pass through authentication.
According to the 5th of the disclosure the aspect, a kind of electronic equipment is provided, comprising: processor;And memory, for storing
The executable instruction of the processor;Wherein, the processor is configured to above-mentioned to execute via the executable instruction is executed
Authentication model training method or auth method described in any one.
According to the 6th of the disclosure the aspect, a kind of computer readable storage medium is provided, computer program is stored thereon with,
The computer program realizes authentication model training method or body described in above-mentioned any one when being executed by processor
Part verification method.
Disclosure exemplary embodiment can have it is following partly or entirely the utility model has the advantages that
In the authentication model training method provided by an example embodiment of the disclosure, according to sample identity spy
It levies data and the first study processing is carried out so that identity branching networks are distinguished in sample identity characteristic to identity branching networks
Identity information and field difference, at the same according to sample identity characteristic to field branching networks carry out second study processing with
So that field branching networks is distinguished the identity information in sample identity characteristic and field difference and eliminates field difference;Finally
Third study processing is carried out according to training to identity branching networks by the field branching networks after identity label and training
Identity branching networks afterwards construct the authentication model.On the one hand, according to field branching networks secondary identities branching networks
It is trained, so that identity branching networks can accurately extract the identity characteristic vector in sample identity characteristic, eliminates neck
Domain difference promotes the accuracy of authentication, promotes the usage experience of user;On the other hand, by identity branching networks and
Field branching networks carry out confrontation learning training authentication model, and required sample data is few, reduce mark, time, money
The cost in source promotes the training effectiveness of authentication model.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.It should be evident that the accompanying drawings in the following description is only the disclosure
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.
Fig. 1 show can using the embodiment of the present disclosure a kind of authentication model training method and device it is exemplary
The schematic diagram of system architecture;
Fig. 2 shows the structural schematic diagrams of the computer system of the electronic equipment suitable for being used to realize the embodiment of the present disclosure;
Fig. 3 diagrammatically illustrates the schematic diagram of the elimination field differences method according to one embodiment of the disclosure;
Fig. 4 is diagrammatically illustrated to be illustrated according to the process of the authentication model training method of one embodiment of the disclosure
Figure;
Fig. 5 diagrammatically illustrates the structural schematic diagram of the authentication model of one embodiment according to the disclosure;
Fig. 6, which is diagrammatically illustrated, eliminates showing for systematical difference in the authentication process itself according to one embodiment of the disclosure
It is intended to;
Fig. 7, which is diagrammatically illustrated, eliminates showing for difference in thickness in the authentication process itself according to one embodiment of the disclosure
It is intended to;
Fig. 8 diagrammatically illustrates the flow diagram of auth method according to another embodiment of the present disclosure;
Fig. 9, which is diagrammatically illustrated, verifies model training method and body according to the application identity of one embodiment of the disclosure
The flow diagram of part verification method;
Figure 10 diagrammatically illustrates the schematic block of the authentication model training device according to one embodiment of the disclosure
Figure;
Figure 11 diagrammatically illustrates the schematic block diagram of the authentication means of one embodiment according to the disclosure.
Specific embodiment
Example embodiment is described more fully with reference to the drawings.However, example embodiment can be with a variety of shapes
Formula is implemented, and is not understood as limited to example set forth herein;On the contrary, thesing embodiments are provided so that the disclosure will more
Fully and completely, and by the design of example embodiment comprehensively it is communicated to those skilled in the art.Described feature, knot
Structure or characteristic can be incorporated in any suitable manner in one or more embodiments.In the following description, it provides perhaps
More details fully understand embodiment of the present disclosure to provide.It will be appreciated, however, by one skilled in the art that can
It is omitted with technical solution of the disclosure one or more in the specific detail, or others side can be used
Method, constituent element, device, step etc..In other cases, be not shown in detail or describe known solution to avoid a presumptuous guest usurps the role of the host and
So that all aspects of this disclosure thicken.
In addition, attached drawing is only the schematic illustrations of the disclosure, it is not necessarily drawn to scale.Identical attached drawing mark in figure
Note indicates same or similar part, thus will omit repetition thereof.Some block diagrams shown in the drawings are function
Energy entity, not necessarily must be corresponding with physically or logically independent entity.These function can be realized using software form
Energy entity, or these functional entitys are realized in one or more hardware modules or integrated circuit, or at heterogeneous networks and/or place
These functional entitys are realized in reason device device and/or microcontroller device.
Fig. 1 show can using the embodiment of the present disclosure a kind of authentication model training method and device it is exemplary
The schematic diagram of the system architecture of application environment.
As shown in Figure 1, system architecture 100 may include one or more of terminal device 101,102,103, network
104 and server 105.Network 104 between terminal device 101,102,103 and server 105 to provide communication link
Medium.Network 104 may include various connection types, such as wired, wireless communication link or fiber optic cables etc..Terminal is set
Standby 101,102,103 can be the various electronic equipments with display screen, including but not limited to desktop computer, portable computing
Machine, smart phone and tablet computer etc..It should be understood that the number of terminal device, network and server in Fig. 1 is only to show
Meaning property.According to needs are realized, any number of terminal device, network and server can have.For example server 105 can be with
It is the server cluster etc. of multiple server compositions.
Authentication model training method provided by the embodiment of the present disclosure is generally executed by server 105, correspondingly, body
Part verifying model training apparatus is generally positioned in server 105.But it will be readily appreciated by those skilled in the art that the disclosure is real
Applying authentication model training method provided by example can also be executed by terminal device 101,102,103, correspondingly, identity is tested
Model of a syndrome training device also can be set in terminal device 101,102,103, and it is special not do in the present exemplary embodiment to this
It limits.For example, in a kind of exemplary embodiment, can be user by terminal device 101,102,103 will work as predecessor
Part characteristic is uploaded to server 105, and server passes through authentication model training method provided by the embodiment of the present disclosure
It determines the corresponding current signature vector of current identity characteristic data and is compared with original feature vector to determine verification result,
And verification result is transferred to terminal device 101,102,103 etc..
Fig. 2 shows the structural schematic diagrams of the computer system of the electronic equipment suitable for being used to realize the embodiment of the present disclosure.
It should be noted that Fig. 2 shows the computer system 200 of electronic equipment be only an example, should not be to this public affairs
The function and use scope for opening embodiment bring any restrictions.
As shown in Fig. 2, computer system 200 includes central processing unit (CPU) 201, it can be read-only according to being stored in
Program in memory (ROM) 202 or be loaded into the program in random access storage device (RAM) 203 from storage section 208 and
Execute various movements appropriate and processing.In RAM 203, it is also stored with various programs and data needed for system operatio.CPU
201, ROM 202 and RAM 203 is connected with each other by bus 204.Input/output (I/O) interface 205 is also connected to bus
204。
I/O interface 205 is connected to lower component: the importation 206 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 207 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.;Storage section 208 including hard disk etc.;
And the communications portion 209 of the network interface card including LAN card, modem etc..Communications portion 209 via such as because
The network of spy's net executes communication process.Driver 210 is also connected to I/O interface 205 as needed.Detachable media 211, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 210, in order to read from thereon
Computer program be mounted into storage section 208 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer below with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communications portion 209, and/or from detachable media
211 are mounted.When the computer program is executed by central processing unit (CPU) 201, execute in the present processes and device
The various functions of limiting.In some embodiments, computer system 200 can also include AI (Artificial
Intelligence, artificial intelligence) processor, the AI processor is for handling the calculating operation in relation to machine learning.
It should be noted that computer-readable medium shown in the disclosure can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to: have the electrical connection, just of one or more conducting wires
Taking formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In the disclosure, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.And at this
In open, computer-readable signal media may include in a base band or as the data-signal that carrier wave a part is propagated,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By the use of instruction execution system, device or device or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. are above-mentioned
Any appropriate combination.
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the disclosure, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of above-mentioned module, program segment or code include one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it wants
It is noted that the combination of each box in block diagram or flow chart and the box in block diagram or flow chart, can use and execute rule
The dedicated hardware based systems of fixed functions or operations is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
Being described in unit involved in the embodiment of the present disclosure can be realized by way of software, can also be by hard
The mode of part realizes that described unit also can be set in the processor.Wherein, the title of these units is in certain situation
Under do not constitute restriction to the unit itself.
As on the other hand, present invention also provides a kind of computer-readable medium, which be can be
Included in electronic equipment described in above-described embodiment;It is also possible to individualism, and without in the supplying electronic equipment.
Above-mentioned computer-readable medium carries one or more program, when the electronics is set by one for said one or multiple programs
When standby execution, so that method described in electronic equipment realization as the following examples.For example, the electronic equipment can be real
Now such as Fig. 4~each step shown in Fig. 9.
The technical solution of the embodiment of the present disclosure is described in detail below:
In recent years, identity validation technology achieves huge progress, including fingerprint authentication, face verification, vocal print/iris are tested
Card, pedestrian identify again.However, the authentication procedures of data-driven often face the deviation in data, such as field is poor
Different, i.e. a model in one field but verified in another field by training.Such as in pedestrian again identification field,
When season, which causes people to wear the clothes, to change or when the relative angle of people and video camera changes, verifying all may be by shadow
It rings.Further, due to cost and operation possibility the problem of, obtains the labeled data of clothes dress or the mark of shooting angle
It is more difficult to infuse data.
Particularly, as the continuous of intelligent terminal is popularized, the identity validation technology based on intelligent terminal is also developed rapidly.And
Above situation in the authentication based on intelligent terminal there is also.For example, for some particular user (by a virtual account
Association), device type (being determined by hardware feature or Software Coding) may change, and operation mode is (left hand, the right hand, double
Hand) may change, physical condition (sit, crouch, lying) may change, local environment (office, bus,
Iron, taxi, train, aircraft) it can also happen that variation.It is used to when using the model of training in old data distribution in new data
When being predicted in distribution, it is possible to the authentication result of output error.Further, asking due to cost and operation possibility
Topic, the labeled data of above-mentioned many types can not obtain (such as the labeled data such as operation mode, physical condition).
In a kind of authentication scheme, unknown different information is realized by directly eliminating edge distribution difference between not same area
Study, such methods include but is not limited to: migration constituent analysis (Transfer Component Analysis, TCA), deep
Spend adaptive network (Deep Adaptation Network, DAN), reverse gradient (Reversing Gradient,
RevGrad), confrontation differentiate domain-adaptive (Aversarial Discriminative Domain Adaptation,
ADDA).Such methods eliminate the domain difference of learned feature while learning Main classification task (such as authentication).Fig. 3 signal
Property show the schematic diagram of elimination field differences method according to one embodiment of the disclosure, refering to what is shown in Fig. 3, primitive character
301 are learnt by generating network (Generator network) 302, and the feature vector of output will do 303 He of identification simultaneously
Type differentiates 304.Type differentiates 304 corresponding networks by confrontation study, and elimination generates network 302 and exports machine in feature vector
Type differentiates 304 information.But such methods need to know field label, and can only learn specific field difference.And this
The network structure of class method cannot carry out decoupling study well.
In another authentication scheme, by realizing that unknown difference is believed using the sample for generating the combinations of attributes that do not see
The study of breath, such methods include but is not limited to: ELEGANT, DNA-GAN, multilayer variation are from code machine (Multi-Level
Variational Autoencoder, ML-VAE), CausalGAN, ResGAN, SaGAN.Such methods first generate in training set
The corresponding sample of the combinations of attributes that do not see, for example (ID=001, system=IOS) was met in training set, but do not see (ID=
001, system=Android), then learnt using this part sample as aid sample input authentication system.It is this kind of
In method, ML-VAE can unsupervised ground learning areas label, while carrying out the study of authentication, but identity information and neck
Do not have to be decoupled with confrontation study between domain information.Remaining method in this kind of scheme is unable to identity-based label
It practises, so the estimated performance of authentication is very limited, and not can be carried out decoupling study;Also, their learning process needs are known
Road field label can only learn specific field difference.
In another authentication scheme, the generation of combinations of attributes sample and edge distribution difference are had no by carrying out simultaneously
Eliminate the study for realizing unknown different information, such methods include but is not limited to: crossing domain indicates decoupler (Cross-
Domain Representation Disentangler, CDRD), the synthesis Sample Method of zero sample learning of broad sense
(Synthesized Examples for Generalized Zero-Shot Learning, SE-GZSL), domain-adaptive
Decoupling synthesize (Disentangled Synthesis for Domain Adaptation, DiDA), the synthesis based on attribute
Network (Attribute-Based Synthetic Network, ABS-Net), the unbiased identity based on additivity confrontation study are tested
Card method (Additive Adversarial Learning for Unbiased Authentication, AAL-UA).AAL-
For UA structure by feature to be processed by multiple network mappings to multiple hidden layer spaces, each hidden layer space corresponds to an attribute, but
The training of AAL-UA network needs to know field label, can only learn specific field difference.Moreover, the generation of AAL-UA network
The quantity of network is directly proportional to number of attributes, differentiates the quantity of network and square directly proportional, the training time cost of number of attributes
It is relatively high with required resource.Remaining method of this kind of scheme weight also requires to know field label, can only learn specifically to lead
Domain difference.
In above-mentioned one or more technical solutions, not only train at high cost, discrimination is lower, transmission speed is slow, safety
Property it is poor, and most method requires to know field label, can only learn specific field difference;Next is related to unsupervised
The method of learning areas information carries out decoupling study not over confrontation study, and the coupling of realm information and identity information is still
Do not solve effectively.
Based on said one or multiple problems, this example embodiment provides a kind of authentication model training method.
The authentication model training method can be applied to above-mentioned server 105, also can be applied to above-mentioned terminal device 101,
102, one or more of 103, particular determination is not done to this in the present exemplary embodiment.Refering to what is shown in Fig. 4, the authentication
Model training method may comprise steps of S410 to step S440:
Step S410, sample identity characteristic and the corresponding identity label of the sample identity characteristic are obtained;
Step S420, according to the sample identity characteristic to the identity branched network in pre-established authentication model
Network carries out the first study processing to complete the training to the identity branching networks;
Step S430, according to the sample identity characteristic to the field branching networks in the authentication model into
The study processing of row second is to complete the training to the field branching networks;
Step S440, by the field branching networks after the identity label and training to the identity branched network
Network carries out third study processing to construct the authentication model according to the identity branching networks after training.
In the authentication model training method provided by this example embodiment, on the one hand, according to field branched network
Network secondary identities branching networks are trained, so that identity branching networks can accurately extract the body in sample identity characteristic
Part feature vector, eliminates field difference, promotes the accuracy of authentication, promote the usage experience of user;On the other hand, pass through
Identity branching networks and field branching networks carry out confrontation learning training authentication model, and required sample data is few,
Reduce mark, the time, resource cost, promote the training effectiveness of authentication model.
In the following, the above-mentioned steps for this example embodiment are described in more details.
In step S410, sample identity characteristic and the corresponding identity mark of the sample identity characteristic are obtained
Label.
In an example embodiment of the disclosure, sample identity characteristic can refer to pre-stored for training
The sample data of authentication model, such as sample identity characteristic can be stored in different user in sample database
Various dimensions identity characteristic data, certainly, sample identity characteristic can also be stored in other positions for training identity
The sample data of model is verified, this example embodiment does not do particular determination to this.Identity label can refer to sample identity feature
The corresponding sample label data of data, such as identity label can refer to the corresponding identity of sample identity characteristic, when
So, this example embodiment is not limited.The sample identity characteristic and corresponding identity label can be stored in advance in
In server, also it can store in the terminal, naturally it is also possible to imported into server or terminal, originally show by other means
Example embodiment does not do particular determination to this.
In the step s 420, according to the sample identity characteristic to the identity in pre-established authentication model point
Branch network carries out the first study processing to be trained to the identity branching networks.
In an example embodiment of the disclosure, pre-established authentication model can refer to that developer creates in advance
The authentication model built, the authentication model may include Feature Conversion network, identity branching networks and field branch
Network.Identity branching networks can refer to the branched network for carrying out identity characteristic information extraction in authentication model and being differentiated
Network.First study processing, which can refer to, carries out unsupervised learning (Unsupervised Learning) to identity branching networks
Treatment process.The corresponding hidden layer feature vector of sample identity characteristic is input to identity branching networks and to identity branched network
Network carries out unsupervised learning, so that identity branching networks can distinguish identity characteristic information and domain features difference.
Specifically, according to sample identity characteristic to the identity branching networks in pre-established authentication model into
Before the study processing of row first is to be trained identity branching networks, according to Feature Conversion network to sample identity characteristic
Conversion process is carried out to generate the corresponding hidden layer feature vector of sample identity characteristic.Feature Conversion network can refer to identity
For the sample identity characteristic of various dimensions to be converted into the coding network of the characteristic of vector expression, example in verifying model
The sample identity characteristic of various dimensions is converted into the encoder network of vector as Feature Conversion network can be
(Encoder), certainly, this example embodiment is not limited.Conversion process can refer to Feature Conversion network to sample identity
The process that characteristic is encoded, hidden layer feature vector can refer to the corresponding feature vector of sample identity characteristic.It is logical
It crosses Feature Conversion network and the sample identity characteristic of various dimensions is converted into feature vector, convenient for other in authentication model
Network handles the sample identity characteristic of various dimensions, improves data-handling efficiency in authentication model, promotes body
The working efficiency of part verifying model.
Further, differentiate that network and target loss function extract network to identity information and carry out nothing by identity information
Supervised learning is handled so that identity information, which extracts network, generates the first attribute vector according to hidden layer feature vector;Wherein the first attribute
Vector may include identity characteristic information and not include domain features information.Identity branching networks in authentication model are specific
It may include that identity information extracts network, identity information differentiates network and target loss function, wherein identity information extracts net
Network can refer to the generator net for extracting identity characteristic vector in the corresponding hidden layer feature vector of sample identity characteristic
Network (Generator), identity information differentiate that network can refer to the differentiation network that network associate is extracted with identity information
(Discriminator), target loss function can refer to the loss letter for identity branching networks to be carried out with unsupervised learning
Number, such as target loss function can be depth measure study (deep metric learning, DML) loss function, can also
To be cross entropy loss function, certainly, this example embodiment is not limited.First attribute vector can refer to that identity information mentions
The feature vector for taking network to be generated according to hidden layer feature vector.Differentiate network and target loss function to body by identity information
Part information extraction network carries out unsupervised learning processing, so that identity information extracts that network is generated according to hidden layer feature vector
One attribute vector only includes identity characteristic information without including domain features information.
In step S430, according to the sample identity characteristic to the field branched network in the authentication model
Network carries out the second study processing to be trained to the field branching networks.
In an example embodiment of the disclosure, field branching networks can refer in authentication model for extracting
Domain features information and the branching networks differentiated.Second study processing can refer to unsupervised to the progress of field branching networks
The treatment process of study.The corresponding hidden layer feature vector of sample identity characteristic is input to field branching networks and to field
Branching networks carry out unsupervised learning, so that field branching networks can distinguish domain features difference and identity characteristic information.
Specifically, differentiating that network and Target Countermeasure loss function extract network to realm information and carry out by realm information
Unsupervised learning is handled so that realm information, which extracts network, generates the second attribute vector according to hidden layer feature vector;Wherein second belongs to
Property vector may include domain features information and not include identity characteristic information.Field branching networks in authentication model can
To include that realm information extracts network, realm information differentiates network and Target Countermeasure loss function;Wherein realm information extracts
Network can refer to the generator for extracting domain features vector in the corresponding hidden layer feature vector of sample identity characteristic
Network;Realm information differentiates that network can refer to the differentiation network that network associate is extracted with realm information;Target Countermeasure loses letter
Number can refer to the loss function for field branching networks to be carried out with unsupervised learning, such as Target Countermeasure loss function can be with
It is depth measure study loss function, is also possible to cross entropy loss function, certainly, this example embodiment is not limited.Mesh
Mark confrontation loss function can be obtained by carrying out calculation processing to target loss function, such as Target Countermeasure loss function can be with
It is to carry out that negative processing is taken to obtain to target loss function, can be and carried out by comparison loss function (contrastive loss)
Adjustment obtains, this example embodiment does not do particular determination to this.Second attribute vector can refer to that realm information extracts network root
Make this feature by extracting the unsupervised learning training of network to realm information according to the feature vector that hidden layer feature vector generates
Vector only includes domain features information and does not include identity characteristic information.
In step S440, by the field branching networks after the identity label and training to the identity point
Branch network carries out third study processing to construct the authentication model according to the identity branching networks after training.
In an example embodiment of the disclosure, third study processing can refer to by after identity label and training
Field branching networks secondary identities branching networks carry out learning training process.Example third study processing can be according to identity
Label exercises supervision study so that identity branching networks can correctly extract sample identity characteristic pair to identity branching networks
The identity characteristic vector in hidden layer feature vector answered, and according to the field branching networks after training to passing through unsupervised learning
Identity branching networks carry out confrontation study treatment process.Turned by the identity branching networks and feature of completing learning training
Switching network constructs authentication model to complete the training of authentication model.
Specifically, by identity label to identity information extract network exercise supervision study processing and by training after
Field branching networks extract network to identity information and carry out confrontation study processing so that identity information extracts network according to sample body
Part characteristic generates identity characteristic vector.Unsupervised learning is carried out by extracting network to identity information, identity can be made to believe
Breath extracts Network Recognition and goes out identity feature vector and corresponding domain features difference;Identity information is extracted by identity label
Network exercises supervision study, and identity information can be allow to extract network, and correctly to extract sample identity characteristic corresponding hidden
Identity characteristic vector in layer feature vector;Network is extracted to identity information by the field branching networks after training to fight
Study can make identity information extract network and eliminate the domain features difference learnt, identity information is made to extract what network extracted
Feature vector only includes identity characteristic vector, excludes the influence of field difference, promotes the accuracy rate of authentication.
Fig. 5 diagrammatically illustrates the structural schematic diagram of the authentication model of one embodiment according to the disclosure.
Refering to what is shown in Fig. 5, by one Feature Conversion network 502 of building by 501 turns of sample identity characteristic of input
It is changed to hidden layer feature vector 503, and then network 505 and field branch are extracted by the identity information in identity branching networks 504
Realm information in network 509 extracts network 510 and is transformed to the first attribute vector 506 and the second attribute vector 511 respectively.It should
The decoupling of symmetrical expression confrontation study learns so that the first attribute vector 506 and the second attribute vector 511 only include and the attribute
Associated discriminant information.That is only include identity characteristic information in the first attribute vector 506, and do not include potential field difference and believe
Breath.Symmetrically, only include potential field different information in the second attribute vector 511, and do not include identity characteristic information.
The learning process is a kind of confrontation learning process, and identity information extracts network 505 and realm information extracts network
510 carry out characteristic extraction procedure, and identity information differentiates that network 508 and realm information differentiate that network 513 carries out potential neck
The judgement of domain difference.Wherein depth measure study (deep metric learning, DML) loss function 507 can make the
One attribute vector 506 can distinguish identity characteristic information, and identity information differentiates that network 508 is used for so that the first attribute vector 506
Potential field different information cannot be distinguished;(adversarial) the depth measure study loss function 512 of confrontation is for making
Identity characteristic information cannot be distinguished by obtaining the second attribute vector 511;Realm information differentiate network 513 be used for so that the second attribute to
Amount 511 can distinguish potential field different information.After the training for completing each network, it is only necessary to sequential concatenation Feature Conversion net
Network 502 and identity information, which extract network 505, can construct authentication model.
Fig. 6, which is diagrammatically illustrated, eliminates showing for systematical difference in the authentication process itself according to one embodiment of the disclosure
It is intended to.
Refering to what is shown in Fig. 6, in certain iteration of training process, when the difference of key data distribution current in sample data
Different when being the not influence of homologous ray, model unsupervised learning is exactly that system differentiates, and eliminates systematical difference bring data
Distributional difference.Specifically, primitive character 601 is generated into hidden layer feature vector, feature extraction network by Feature Conversion network 602
603 pairs of identifications exercise supervision study, confrontation study are differentiated to system, so that the feature that feature extraction network 603 extracts is only
Only include the information of identification, and does not include system and sentence another characteristic.Similarly, the feature that feature extraction network 604 extracts is only
Another characteristic only is sentenced comprising system, and does not include the feature of identification.
After the difference that system differentiates in sample data is eliminated, most important data distribution difference may be come in data
The difference caused by the mobile phone thickness.So what model progress unsupervised learning was acquired is exactly the difference of thickness, then proceedes to disappear
It, specifically can be with reference to shown in Fig. 7 except thickness bring difference.
Fig. 7, which is diagrammatically illustrated, eliminates showing for difference in thickness in the authentication process itself according to one embodiment of the disclosure
It is intended to.
Refering to what is shown in Fig. 7, detecting key data current in sample data point after the difference for eliminating system differentiation
The difference of cloth is the influence of thickness, and model unsupervised learning is exactly difference in thickness, and eliminates difference in thickness bring data
Distributional difference.Specifically, primitive character 601 is generated into hidden layer feature vector, feature extraction network by Feature Conversion network 602
703 pairs of identifications exercise supervision study, fight and learn to difference in thickness, so that the feature that feature extraction network 703 extracts is only
Only include the information of identification, and does not include the feature of difference in thickness.Similarly, the feature that feature extraction network 704 extracts is only
Only include the feature of difference in thickness, and does not include the feature of identification.
After the difference of thickness in sample data is eliminated, the difference that key data is distributed in sample data is continued to test,
And the difference detected is eliminated.So iteration continues, and authentication model can find the residue of sample data step by step
It most important difference and is eliminated in difference, it is final to eliminate difference all in sample data, it is corresponding to obtain sample data
Identity characteristic vector.
This example embodiment additionally provides a kind of auth method.The auth method can be applied to above-mentioned clothes
Business device 105, also can be applied to one or more of above-mentioned terminal device 101,102,103, right in the present exemplary embodiment
This does not do particular determination.Refering to what is shown in Fig. 8, the auth method may comprise steps of S810 to step S840:
In step S810, current identity characteristic data are obtained, and the current identity characteristic data are input to pre- instruction
With the corresponding current signature vector of the determination current identity characteristic data in experienced authentication model.
In an example embodiment of the disclosure, current identity characteristic data can refer to that user is carrying out authentication
When terminal acquisition, characterization user identity various dimensions characteristic, such as current identity characteristic data can be terminal acquisition
The password of user's input, the facial photo of user and the mode of operation (such as left-handed operation or the right hand operation) of user,
Certainly, it only schematically illustrates herein, does not cope with this example embodiment and cause any particular determination.Authentication model can be with
Refer to the deep learning model obtained by the authentication model training method training in this example embodiment, is tested by identity
Model of a syndrome can verify the identity of user.Current signature vector, which can refer to, handles current identity characteristic by authentication model
The feature vector for the characterization subscriber identity information that data obtain.
In step S820, pre-stored original identity characteristic data are obtained, and the original identity characteristic data are defeated
Enter into the authentication model with the corresponding original feature vector of the determination original identity characteristic data.
In an example embodiment of the disclosure, original identity characteristic data can refer to that user first passes through terminal in advance and adopts
Collect and is stored in target position (storage region of such as sample database or terminal, it is special that this example embodiment does not do this
Limit) verifying characteristic, the correlation data as authentication.Original feature vector can refer to through authentication mould
Type handles the feature vector for the characterization subscriber identity information that original identity characteristic data obtain, when as authentication to bit
Levy vector.
In step S830, if the difference of the current signature vector and the original feature vector is less than preset threshold,
Then determine that the current identity characteristic data pass through authentication.
In an example embodiment of the disclosure, preset threshold can refer to it is pre-set, for judging current spy
Levy the threshold value of the corresponding user identity of vector.Specifically, being preset if the difference of current signature vector and original feature vector is less than
Threshold value, it is determined that current identity characteristic data pass through authentication;If the difference of current signature vector and original feature vector is big
In preset threshold, it is determined that current identity characteristic data and original identity characteristic data mismatch, and current identity characteristic data are not
Authentication can be passed through.
Fig. 9, which is diagrammatically illustrated, verifies model training method and body according to the application identity of one embodiment of the disclosure
The flow diagram of part verification method.
Refering to what is shown in Fig. 9, the authentication model training method provided in this example embodiment can be applied to shown in Fig. 9
Training set 901 in, the auth method provided in this example embodiment can be applied in test set 902 shown in Fig. 9.
Specific step is as follows:
Step S901, input sample identity characteristic data and the corresponding identity label of sample identity characteristic, and root
Sample identity characteristic is converted into hidden layer feature vector according to Feature Conversion network;
Step S902 carries out unsupervised field difference learning to identity branching networks so that identity branching networks are according to hidden
Layer feature vector obtains the first attribute vector;
Step S903, identity branching networks eliminate the field different information learnt, so that identity branching networks generated
First attribute vector only includes identity characteristic information and does not include field different information;
Step S904 exercises supervision study so that identity branching networks can be just to identity branching networks according to identity label
Really extract the identity characteristic vector in hidden layer feature vector;
Step S905 carries out unsupervised field difference learning to field branching networks so that field branching networks are according to hidden
Layer feature vector obtains the second attribute vector;
Step S906, field branching networks eliminate identity information, so that the second attribute vector that identity branching networks generate
Only identity characteristic information is not included comprising field different information;
Whether step S907, the corresponding hidden layer feature vector of detection sample identity characteristic also include other field difference
Information, if in hidden layer feature vector including other field different information, then it is assumed that iteration needs to continue and executes step S902, no
Then according to the authentication model in Feature Conversion network and the identity branching networks generation step S909 of training completion;
Step S908 acquires the identity characteristic data of user by terminal device;
Identity characteristic data are input in the authentication model obtained according to step S901- step 907 by step S909
Obtain identity characteristic vector;
Step S910 is carried out according to the identity characteristic vector and pre-stored identity label (original identity characteristic vector)
It compares to determine verification result.
For example, the specific Optimization Steps of authentication model training method in this example embodiment can following institute
Show:
Step 1: identity branching networks carry out unsupervised field difference learning can be by shown in relational expression (1):
Wherein λ is weight coefficient.D12It is to differentiate network, the last layer uses softmax activation primitive.The full articulamentum
Export the number k of the specified potential field difference of dimension, that is, priori.Form such as relational expression (2) shown in:
Wherein α1, α2For weight coefficient.B is the size of batch size, pikFor D12(G1(P(xi))) r-th output,
And
Step 2: the field difference that the elimination of identity branching networks learns can be by shown in relational expression (3):
WhereinIt is MSE (mean square error) loss function, 1 is complete 1 vector.
Step 3: the study that identity branching networks carry out identification can be by shown in relational expression (4):
Wherein, S1f, SyThe set of feature vector and the set of identity label inside a respectively batch, and
S1f={ f11..., f1B, Sy={ y1..., yB, f1i=G1(P(xi))
It can learn loss function, such as contrastive loss (comparison damage using arbitrary depth measure
Lose function), triplet loss, npair loss, angular loss etc..
Step 4: field branching networks carry out unsupervised field difference learning can be by shown in relational expression (5):
Step 5: field branching networks elimination identity information can be by shown in relational expression (6):
Wherein, S2f, SyThe set of feature vector and the set of identity label inside a respectively batch, and
S2f={ f21..., f2B, Sy={ y1..., yB, f2i=G2(P(xi))
Purpose be eliminate identity information, i.e., so that the differences between samples of different people be not more than the same person sample
This difference.Can directly take-It can also be by above-mentioned logic to existing arbitrary depth measure study loss
Function is transformed.It can use contrastive loss in this example embodiment to be transformed, such as can be such as relational expression
(7) shown in:
It should be noted that the loss function in above-mentioned formula, which can be, intersects entropy loss (cross entropy
Loss), be also possible to logic this spy loss (logistic loss), mean square loss (mean square loss), square
It loses (square loss), l2Norm loss and l1Norm loss one of or it is a variety of, this example embodiment not as
Limit.And complete 1 vector in formula (3) is also possible to full 0 vector, full 0 .5 vector etc., it is special that this example embodiment does not do this
It limits.
It should be noted that although describing each step of method in the disclosure in the accompanying drawings with particular order, this is simultaneously
Undesired or hint must execute these steps in this particular order, or have to carry out the ability of step shown in whole
Realize desired result.Additional or alternative, it is convenient to omit multiple steps are merged into a step and executed by certain steps,
And/or a step is decomposed into execution of multiple steps etc..
Further, in this example embodiment, a kind of authentication model training device is additionally provided.The authentication
Model training apparatus can be applied to a server or terminal device.Refering to what is shown in Fig. 10, the authentication model training device
1000 may include sample data acquiring unit 1010, identity branching networks training unit 1020, field branching networks training list
Member 1030 and authentication model construction unit 1040.Wherein:
Sample data acquiring unit 1010 is for obtaining sample identity characteristic and the sample identity characteristic
Corresponding identity label;
Identity branching networks training unit 1020 is for testing pre-established identity according to the sample identity characteristic
Identity branching networks in model of a syndrome carry out the first study processing to complete the training to the identity branching networks;
Field branching networks training unit 1030 is used for according to the sample identity characteristic to the authentication mould
Field branching networks in type carry out the second study processing to complete the training to the field branching networks;
Authentication model construction unit 1040 is used for through the field branch after the identity label and training
Network carries out third study processing according to the identity branching networks building after training to the identity branching networks
Authentication model.
In a kind of exemplary embodiment of the disclosure, the authentication model training device 1000 further includes hidden layer spy
Vector generation unit is levied, the hidden layer feature vector generation unit is configured as:
Conversion process is carried out to generate the sample to the sample identity characteristic according to the Feature Conversion network
The corresponding hidden layer feature vector of identity characteristic data.
In a kind of exemplary embodiment of the disclosure, the identity branching networks include that identity information extracts network, body
Part information differentiates network and target loss function;
The identity branching networks training unit 1020 is configured as:
By the identity information differentiate network and the target loss function to the identity information extract network into
Row unsupervised learning is handled so that the identity information, which extracts network, generates the first attribute vector according to the hidden layer feature vector;
Wherein first attribute vector includes identity characteristic information and does not include domain features information.
In a kind of exemplary embodiment of the disclosure, the field branching networks include that realm information extracts network, neck
Domain information differentiates network and Target Countermeasure loss function;
The field branching networks training unit 1030 is configured as:
Differentiate that network and the Target Countermeasure loss function extract net to the realm information by the realm information
Network carries out unsupervised learning processing so that the realm information, which extracts network, generates the second attribute according to the hidden layer feature vector
Vector;Wherein second attribute vector includes domain features information and does not include identity characteristic information.
In a kind of exemplary embodiment of the disclosure, the authentication model construction unit 1040 is configured as:
It is exercised supervision by the identity label to identity information extraction network after learning processing and passing through training
The field branching networks to the identity information extract network carry out confrontation study handle so that the identity information extract
Network generates identity characteristic vector according to the sample identity characteristic.
In a kind of exemplary embodiment of the disclosure, the target loss function includes depth measure study loss function
Or cross entropy loss function;And calculation processing is carried out to the target loss function and generates the Target Countermeasure loss letter
Number.
Each module or the detail of unit are in corresponding authentication in above-mentioned authentication model training device
It is described in detail in model training method, therefore details are not described herein again.
In this example embodiment, a kind of authentication means are additionally provided.The authentication means can be applied to one
Server or terminal device.With reference to shown in Figure 11, which may include current signature vector determination unit
1110, original feature vector determination unit 1120 and authentication pass through unit 1130, in which:
Current signature vector determination unit 1100 is for obtaining current identity characteristic data, and by the current identity characteristic
Data are input in the authentication model of pre-training with the corresponding current signature vector of the determination current identity characteristic data;
Original feature vector determination unit 1120 is used to obtain pre-stored original identity characteristic data, and will be described original
Identity characteristic data are input in the authentication model with the corresponding primitive character of the determination original identity characteristic data
Vector;
If authentication is small for the current signature vector and the difference of the original feature vector by unit 1130
In preset threshold, it is determined that the current identity characteristic data pass through authentication.
In above-mentioned authentication means each module or the detail of unit in corresponding auth method into
Detailed description is gone, therefore details are not described herein again.
It should be noted that although being referred to several modules or list for acting the equipment executed in the above detailed description
Member, but this division is not enforceable.In fact, according to embodiment of the present disclosure, it is above-described two or more
Module or the feature and function of unit can embody in a module or unit.Conversely, an above-described mould
The feature and function of block or unit can be to be embodied by multiple modules or unit with further division.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to its of the disclosure
Its embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Person's adaptive change follows the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure
Or conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by following
Claim is pointed out.
It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by the accompanying claims.
Claims (10)
1. a kind of authentication model training method characterized by comprising
Obtain sample identity characteristic and the corresponding identity label of the sample identity characteristic;
First is carried out to the identity branching networks in pre-established authentication model according to the sample identity characteristic to learn
Processing is practised to be trained to the identity branching networks;
The field branching networks in the authentication model are carried out at the second study according to the sample identity characteristic
Reason is to be trained the field branching networks;
Third is carried out to the identity branching networks by the field branching networks after the identity label and training
Processing is practised to construct the authentication model according to the identity branching networks after training.
2. authentication model training method according to claim 1, which is characterized in that the authentication model also wraps
Include Feature Conversion network, according to the sample identity characteristic to the identity branched network in pre-established authentication model
Before network carries out the first study processing to be trained to the identity branching networks, the method also includes:
Conversion process is carried out to generate the sample identity to the sample identity characteristic according to the Feature Conversion network
The corresponding hidden layer feature vector of characteristic.
3. authentication model training method according to claim 2, which is characterized in that the identity branching networks include
Identity information extracts network, identity information differentiates network and target loss function;
It is described that the is carried out to the identity branching networks in pre-established authentication model according to the sample identity characteristic
One study, which is handled to be trained to the identity branching networks, includes:
Differentiate that network and the target loss function extract network to the identity information and carry out nothing by the identity information
Supervised learning is handled so that the identity information, which extracts network, generates the first attribute vector according to the hidden layer feature vector;Wherein
First attribute vector includes identity characteristic information and does not include domain features information.
4. authentication model training method according to claim 3, which is characterized in that the field branching networks include
Realm information extracts network, realm information differentiates network and Target Countermeasure loss function;
It is described that second is carried out to the field branching networks in the authentication model according to the sample identity characteristic
Processing, which is practised, to be trained to the field branching networks includes:
By the realm information differentiate network and the Target Countermeasure loss function to the realm information extract network into
Row unsupervised learning is handled so that the realm information, which extracts network, generates the second attribute vector according to the hidden layer feature vector;
Wherein second attribute vector includes domain features information and does not include identity characteristic information.
5. authentication model training method according to claim 3, which is characterized in that by the identity label and
The field branching networks after training carry out third study processing according to after training to the identity branching networks
Identity branching networks construct the authentication model
Is extracted by network and is exercised supervision for the identity information by the identity label and learns processing and by the institute after training
It states field branching networks and is extracted by network and fight for the identity information and learn to handle so that the identity information extracts network
Identity characteristic vector is generated according to the sample identity characteristic.
6. according to authentication model training method described in claim 3 or 4 any one, which is characterized in that the target
Loss function includes depth measure study loss function or cross entropy loss function;And the target loss function is carried out
Calculation processing generates the Target Countermeasure loss function.
7. a kind of auth method characterized by comprising
Current identity characteristic data are obtained, and the current identity characteristic data are input in the authentication model of pre-training
With the corresponding current signature vector of the determination current identity characteristic data;
Pre-stored original identity characteristic data are obtained, and the original identity characteristic data are input to the authentication mould
With the corresponding original feature vector of the determination original identity characteristic data in type;
If the difference of the current signature vector and the original feature vector is less than preset threshold, it is determined that the current identity
Characteristic passes through authentication.
8. a kind of authentication model training device characterized by comprising
Sample data acquiring unit, for obtaining sample identity characteristic and the corresponding body of the sample identity characteristic
Part label;
Identity branching networks training unit, for according to the sample identity characteristic in pre-established authentication model
Identity branching networks carry out first study processing to be trained to the identity branching networks;
Field branching networks training unit, for according to the sample identity characteristic to the neck in the authentication model
Domain branching networks carry out the second study processing to be trained to the field branching networks;
Authentication model construction unit, for the field branching networks by the identity label and after training to institute
It states identity branching networks and carries out third study processing to construct the authentication according to the identity branching networks after training
Model.
9. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program quilt
It realizes when processor executes such as authentication model training method described in any one of claims 1 to 6 or claim 7 institute
The auth method stated.
10. a kind of electronic equipment characterized by comprising
Processor;And
Memory, for storing the executable instruction of the processor;
Wherein, the processor is configured to execute via the executable instruction is executed such as any one of claims 1 to 6 institute
The authentication model training method stated or auth method as claimed in claim 7.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910750225.1A CN110490245A (en) | 2019-08-14 | 2019-08-14 | Authentication model training method and device, storage medium, electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910750225.1A CN110490245A (en) | 2019-08-14 | 2019-08-14 | Authentication model training method and device, storage medium, electronic equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110490245A true CN110490245A (en) | 2019-11-22 |
Family
ID=68551033
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910750225.1A Pending CN110490245A (en) | 2019-08-14 | 2019-08-14 | Authentication model training method and device, storage medium, electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110490245A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112232241A (en) * | 2020-10-22 | 2021-01-15 | 华中科技大学 | Pedestrian re-identification method and device, electronic equipment and readable storage medium |
-
2019
- 2019-08-14 CN CN201910750225.1A patent/CN110490245A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112232241A (en) * | 2020-10-22 | 2021-01-15 | 华中科技大学 | Pedestrian re-identification method and device, electronic equipment and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
de Araujo-Filho et al. | Intrusion detection for cyber–physical systems using generative adversarial networks in fog environment | |
EP3961441B1 (en) | Identity verification method and apparatus, computer device and storage medium | |
Zhang et al. | Wi-Fi sensing for joint gesture recognition and human identification from few samples in human-computer interaction | |
CN108696331B (en) | Signal reconstruction method based on generation countermeasure network | |
CN109922032A (en) | Method and apparatus for determining the risk of logon account | |
CN106022317A (en) | Face identification method and apparatus | |
CN110232373A (en) | Face cluster method, apparatus, equipment and storage medium | |
Tinsley et al. | This face does not exist... but it might be yours! identity leakage in generative models | |
CN102890776A (en) | Method for searching emoticons through facial expression | |
CN106022241B (en) | A kind of face identification method based on wavelet transformation and rarefaction representation | |
CN109685104B (en) | Determination method and device for recognition model | |
CN110378095A (en) | Validation-cross method, apparatus, medium and calculating equipment based on identifying code | |
CN110197224A (en) | Aerial hand-written character track restoration methods based on the confrontation study of feature space depth | |
CN108806718A (en) | Based on the audio authentication method to ENF phase spectrums and instantaneous frequency spectrum analysis | |
CN108509994A (en) | character image clustering method and device | |
CN108766464A (en) | Digital audio based on mains frequency fluctuation super vector distorts automatic testing method | |
CN109979439A (en) | Audio recognition method, device, medium and electronic equipment based on block chain | |
CN111382403A (en) | Training method, device, equipment and storage medium of user behavior recognition model | |
CN110490245A (en) | Authentication model training method and device, storage medium, electronic equipment | |
CN109242007A (en) | A kind of cross-domain isomery big data multi-angle of view clustering method and device based on tensor | |
CN109147446A (en) | Electric examination system | |
CN111767949A (en) | Multi-task learning method and system based on feature and sample confrontation symbiosis | |
Qiu et al. | Semantic-visual guided transformer for few-shot class-incremental learning | |
CN106650685A (en) | Electrocardiogram signal-based identity identification method and apparatus | |
Yu et al. | A multi-scale feature selection method for steganalytic feature GFR |
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 |