A data security control method based on biometrics

Abstract

The present invention discloses a data security control method based on biometrics, which belongs to the field of data management technology. An identity information recognition model is constructed by adopting an artificial intelligence algorithm, and the identity information recognition model is trained by adopting the biometric information corresponding to the user to be authorized and the corresponding authorization category label. When the user accesses a data platform, the biometric information corresponding to the current accessing user is collected, and the trained identity information recognition model is used to recognize the biometric information corresponding to the current accessing user. When the identity recognition result is that the identity authentication is passed, the current accessing user is allowed to access the data platform, and the access process is controlled according to the data management authority. This not only effectively improves data security, but also can effectively control the user data management authority, avoiding the problem of easy loss and low security of traditional passwords.

Description

Technical Field

[0001] The present invention belongs to the technical field of data management, and in particular relates to a data security control method based on biometrics. Background Art

[0002] Data security refers to a series of measures and strategies to protect digital data from unauthorized access, leakage, tampering, destruction and loss. In the information age, data has become an important asset for enterprises and individuals, so ensuring data security is crucial. With the rapid development of information technology, data security has become a focus of attention in my country and even the world. Traditional data security control methods mainly rely on passwords, digital certificates and other means, but these methods have problems such as easy leakage and easy cracking. Summary of the invention

[0003] The present invention provides a data security control method based on biometrics, which is used to solve the problems of easy leakage and low security performance in the process of managing data security by traditional passwords and digital certificates.

[0004] A data security control method based on biometrics, comprising:

[0005] Obtain the biometric information of the user to be authorized and the data authority management key entered by the administrator, verify the data authority management key, and obtain the authority verification result;

[0006] When the permission verification result is verification passed, an authorization category label corresponding to the biometric information of the user to be authorized is generated;

[0007] An identity information recognition model is constructed using an artificial intelligence algorithm, and the identity information recognition model is trained using the biometric information corresponding to the user to be authorized and the corresponding authorization category label to obtain a trained identity information recognition model;

[0008] Deploy the trained identity information recognition model on the data platform, and simultaneously obtain the data management permissions corresponding to each user to be authorized entered by the administrator;

[0009] When a user accesses the data platform, the biometric information corresponding to the current access user is collected, and the deployed identity information recognition model is used to recognize the biometric information corresponding to the current access user to obtain the identity recognition result;

[0010] When the identity recognition result is that the identity authentication is passed, the current access user is allowed to access the data platform, and the access process is controlled according to the data management authority.

[0011] Furthermore, when a new user to be authorized is added, a transfer learning algorithm is used to re-update and deploy the identity information recognition model.

[0012] Furthermore, the biometric information of the user to be authorized and the data authority management key input by the administrator are obtained, and the data authority management key is verified to obtain the authority verification result, including:

[0013] Obtain a facial image of the user to be authorized and obtain the biometric information corresponding to the user to be authorized;

[0014] Obtaining a data rights management key input by an administrator; wherein the data rights management key is a digital key or a key stored in a physical key device;

[0015] When the data permission management key input by the administrator is the same as the permission management key pre-stored in the data platform, the permission verification result is determined to be verification passed, otherwise the permission verification result is determined to be verification failed.

[0016] Furthermore, when the authority verification result is a verification failure, the administrator is denied from entering the information of the authorized user.

[0017] Furthermore, an artificial intelligence algorithm is used to construct an identity information recognition model, and the identity information recognition model is trained using the biometric information corresponding to the user to be authorized and the corresponding authorization category label, to obtain the trained identity information recognition model, including:

[0018] An identity information recognition model is constructed by using an artificial intelligence algorithm, and model parameters of the identity information recognition model are initialized to determine a plurality of model parameter individuals; wherein each model parameter individual includes all parameters to be trained of the identity information recognition model;

[0019] For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the biometric information corresponding to the user to be authorized is used as the input of the identity information recognition model, and the authorization category label corresponding to the biometric information is used as the expected output of the identity information recognition model to determine the fitness corresponding to the model parameter individual;

[0020] According to the fitness corresponding to each model parameter individual, the optimal model parameter individual, the elite model parameter individual and the common model parameter individual are determined among all the model parameter individuals;

[0021] For elite model parameter individuals, according to the optimal model parameter individuals, the elite model parameter individuals are searched by using a spiral adaptive local optimal search method to obtain elite model parameter individuals after local search;

[0022] For common model parameter individuals, an adaptive collaborative search method is used to perform interval search on the common model parameter individuals to obtain common model parameter individuals after interval search;

[0023] For the elite model parameter individuals after local search and the common model parameter individuals after interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after global search;

[0024] Determine whether the training end conditions are met. If so, re-obtain the optimal model parameter individual based on the elite model parameter individual and the ordinary model parameter individual after the global search, and use the re-obtained optimal model parameter individual as the final parameter of the identity information recognition model to obtain the identity information recognition model after training. Otherwise, return to the step of obtaining fitness.

[0025] Furthermore, for any model parameter individual, after applying the model parameter corresponding to the model parameter individual to the identity information recognition model, using the biometric information corresponding to the user to be authorized as the input of the identity information recognition model, and using the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, the fitness corresponding to the model parameter individual is determined, including:

[0026] For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the identity information recognition model after the parameters are applied is obtained;

[0027] Using the authorization category label corresponding to the biometric information as the input of the identity information recognition model after applying the parameters, and obtaining the actual output of the identity information recognition model;

[0028] Taking the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, and obtaining the cross entropy loss function of the identity information recognition model according to the actual output and the expected output of the identity information recognition model;

[0029] After adding the cross entropy loss function of the identity information recognition model to the preset constant term, the intermediate numerical term is obtained, and the reciprocal of the intermediate numerical term is taken to obtain the fitness corresponding to the individual model parameter.

[0030] Furthermore, according to the fitness corresponding to each model parameter individual, the optimal model parameter individual, the elite model parameter individual and the common model parameter individual are determined among all the model parameter individuals, including:

[0031] According to the fitness corresponding to each model parameter individual, the model parameter individual with the largest fitness is determined as the optimal model parameter individual;

[0032] For other model parameter individuals except the optimal model parameter individual, the first N model parameter individuals are taken as elite model parameter individuals in descending order of fitness, and the remaining model parameter individuals are taken as ordinary model parameter individuals; where N is less than M / 3, M represents the total number of all model parameter individuals, and both M and N are positive integers.

[0033] Furthermore, for the elite model parameter individuals, according to the optimal model parameter individuals, the elite model parameter individuals are searched by using a spiral adaptive local optimal search method to obtain elite model parameter individuals after local search, including:

[0034] Based on the current training times and the maximum training times, the adaptive inertia weight is obtained by combining the sine and cosine functions;

[0035] For any elite model parameter individual, randomly match another elite model parameter individual to obtain the first model parameter individual to be learned corresponding to the elite model parameter individual;

[0036] According to the first model parameter individual to be learned corresponding to the elite model parameter individual, a spiral function is used to obtain the first spiral learning information corresponding to the elite model parameter individual;

[0037] Determine the historical optimal state corresponding to each elite model parameter individual, and obtain the first current update speed by combining the adaptive inertia weight, the historical optimal state and the optimal model parameter individual;

[0038] Based on the first current update speed and the first spiral learning information, the elite model parameter individuals are searched to obtain the elite model parameter individuals after the local search.

[0039] Furthermore, for the common model parameter individuals, an adaptive collaborative search method is used to perform interval search on the common model parameter individuals, and the common model parameter individuals after the interval search are obtained, including:

[0040] For any common model parameter individual, a model parameter individual is randomly matched to the common model parameter individual to obtain a second model parameter individual to be learned corresponding to the common model parameter individual;

[0041] According to the second model parameter individuals to be learned corresponding to the common model parameter individuals, a spiral function is used to obtain the second spiral learning information corresponding to the common model parameter individuals;

[0042] According to the adaptive inertia weight and the second model parameter individual to be learned, the second current update speed is obtained, and according to the second current update speed and the second spiral learning information, the common model parameter individual is interval searched to obtain the common model parameter individual after the interval search.

[0043] Furthermore, for the elite model parameter individuals after the local search and the common model parameter individuals after the interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after the global search, including:

[0044] According to the upper limit individuals and lower limit individuals corresponding to the model parameter individuals, determine the midpoint information of the feasible domain;

[0045] A perturbation factor is randomly generated between (0, 1), and the midpoint information of the feasible domain is perturbed by the perturbation factor to obtain the midpoint information of the feasible domain after perturbation;

[0046] For any target model parameter individual, the elite model parameter individual or the common model parameter individual is perturbed by a perturbation factor to obtain the target model parameter individual after perturbation; wherein the target model parameter individual represents the elite model parameter individual after local search or the common model parameter individual after interval search;

[0047] According to the midpoint information of the feasible domain, the midpoint information of the feasible domain after disturbance and the target model parameter individual after disturbance, a global search is performed on the target model parameter individual to obtain the target model parameter individual after the global search, that is, the elite model parameter individual or the ordinary model parameter individual after the global search is obtained.

[0048] The present invention provides a data security control method based on biometrics. An identity information recognition model is constructed by adopting an artificial intelligence algorithm, and the identity information recognition model is trained by adopting the biometric information corresponding to the user to be authorized and the corresponding authorization category label. When the user accesses the data platform, the biometric information corresponding to the current accessing user is collected, and the trained identity information recognition model is used to recognize the biometric information corresponding to the current accessing user. When the identity recognition result is that the identity authentication is passed, the current accessing user is allowed to access the data platform, and the access process is controlled according to the data management authority. This not only effectively improves data security, but also can effectively control the user's data management authority, avoiding the problem of easy loss and low security of traditional passwords. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] Figure 1 A flowchart of a data security control method based on biometrics provided in an embodiment of the present invention.

[0051] Figure 2A flowchart for training an identity information recognition model provided in an embodiment of the present invention.

[0052] The above drawings have shown clear embodiments of the present invention, which will be described in more detail below. These drawings and text descriptions are not intended to limit the scope of the present invention in any way, but to illustrate the concept of the present invention for those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0053] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of devices and methods consistent with some aspects of the present invention as detailed in the appended claims.

[0054] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0055] like Figure 1 As shown, an embodiment of the present invention provides a data security control method based on biometrics, including:

[0056] S101, obtaining the biometric information of the user to be authorized and the data authority management key input by the administrator, verifying the data authority management key, and obtaining an authority verification result;

[0057] The data permission management key entered by the administrator can be a physical and software combination key such as a U-shield, which makes data management more secure, or it can be a traditional digital key. Therefore, the administrator only needs to save the data permission management key to effectively implement data security control.

[0058] S102, when the authority verification result is verification passed, generating an authorization category label corresponding to the biometric information corresponding to the user to be authorized;

[0059] Optionally, since illegal operations may exist during the data access process, the biometric information of some unauthorized persons may be collected and labeled with unauthorized category labels, so that identity recognition can be performed in subsequent processes. A different authorization category label may be assigned to each user to be authorized, so that after identifying the user's identity information, data management can be performed according to the set data management permissions.

[0060] S103, using an artificial intelligence algorithm to construct an identity information recognition model, and using the biometric information corresponding to the user to be authorized and the corresponding authorization category label to train the identity information recognition model, to obtain a trained identity information recognition model;

[0061] Using artificial intelligence algorithms to build an identity information recognition model may include: using a CNN (convolutional neural network) algorithm to build an identity information recognition model. After building the identity information recognition model, the identity information recognition model can be trained using the biometric information of the user to be authorized and the corresponding authorization category label. It is also necessary to use the biometric information of unauthorized personnel and unauthorized category labels for training, so that identity recognition can be achieved.

[0062] The particle swarm algorithm can be used to train the identity information recognition model, so as to obtain the trained identity information recognition model.

[0063] S104, deploying the trained identity information recognition model on the data platform, and simultaneously obtaining the data management authority corresponding to each user to be authorized input by the administrator;

[0064] Data management authority can be the management authority of certain data or the management authority of some data space. After the administrator assigns management authority and determines the user's identity, the user is only required to manage data within the scope of his or her management authority.

[0065] After the administrator inputs the data management authority corresponding to each user to be authorized, the authorization category label corresponding to the user to be authorized and the data management authority are associated and stored.

[0066] S105. When a user accesses the data platform, the biometric information corresponding to the current accessing user is collected, and the deployed identity information recognition model is used to recognize the biometric information corresponding to the current accessing user to obtain an identity recognition result;

[0067] The biometric information can be set as a face image or a fingerprint image. The face image is easier to collect. Therefore, in the embodiment of the present invention, the biometric information is preferably set as a face image.

[0068] After using the deployed identity information recognition model to identify the biometric information corresponding to the current access user, the authorization category label corresponding to the current access user can be obtained, thereby determining the data management authority.

[0069] S106. When the identity recognition result is that the identity authentication is passed, the current access user is allowed to access the data platform, and the access process is controlled according to the data management authority.

[0070] The present invention provides a biometric-based data security control method, which uses biometric information to control data security, which not only avoids the problems of easy loss and insecurity in the prior art, but also can realize data operation traceability, further improving data security.

[0071] In the embodiment of the present invention, when a new user to be authorized is added, a transfer learning algorithm is used to re-update and deploy the identity information recognition model.

[0072] By re-updating and deploying the identity information recognition model through the transfer learning algorithm, not only the complexity of retraining the identity information recognition model can be reduced, but also the accurate recognition of the identity information recognition model can be ensured.

[0073] In the embodiment of the present invention, the biometric information of the user to be authorized and the data authority management key input by the administrator are obtained, the data authority management key is verified, and the authority verification result is obtained, including:

[0074] Obtain a facial image of the user to be authorized and obtain the biometric information corresponding to the user to be authorized;

[0075] Obtaining a data rights management key input by an administrator; wherein the data rights management key is a digital key or a key stored in a physical key device;

[0076] When the data permission management key input by the administrator is the same as the permission management key pre-stored in the data platform, the permission verification result is determined to be verification passed, otherwise the permission verification result is determined to be verification failed.

[0077] It is worth noting that biometric information can also be set as other identity feature information, which can also realize user identity identification.

[0078] In the embodiment of the present invention, when the authority verification result is verification failure, the administrator is denied from entering the information of the authorized user.

[0079] like Figure 2 As shown, an artificial intelligence algorithm is used to construct an identity information recognition model, and the biometric information corresponding to the user to be authorized and the corresponding authorization category label are used to train the identity information recognition model to obtain the trained identity information recognition model, including:

[0080] S201, constructing an identity information recognition model using an artificial intelligence algorithm, and initializing model parameters of the identity information recognition model to determine a plurality of model parameter individuals; wherein each model parameter individual includes all parameters to be trained of the identity information recognition model;

[0081] Although the embodiment of the present invention preferably uses the CNN algorithm to construct the identity information recognition model, other artificial intelligence algorithms can also be used to construct the identity information recognition model.

[0082] The model parameters of the identity information recognition model can be initialized by using a random initialization method or a chaotic mapping initialization method.

[0083] When the CNN algorithm is used to construct an identity information recognition model, all parameters to be trained of the identity information recognition model can be set as weight parameters.

[0084] S202, for any model parameter individual, after applying the model parameter corresponding to the model parameter individual to the identity information recognition model, using the biometric information corresponding to the user to be authorized as the input of the identity information recognition model, using the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, and determining the fitness corresponding to the model parameter individual;

[0085] S203, according to the fitness corresponding to each model parameter individual, determining the optimal model parameter individual, the elite model parameter individual and the common model parameter individual among all the model parameter individuals;

[0086] S204, searching the elite model parameter individuals according to the optimal model parameter individuals and using a spiral adaptive local optimal search method to obtain elite model parameter individuals after local search;

[0087] S205, for common model parameter individuals, using an adaptive collaborative search method to perform interval search on the common model parameter individuals to obtain common model parameter individuals after the interval search;

[0088] S206, for the elite model parameter individuals after the local search and the common model parameter individuals after the interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after the global search;

[0089] S207, determine whether the training end conditions are met. If so, re-obtain the optimal model parameter individual based on the elite model parameter individual and the ordinary model parameter individual after the global search, and use the re-obtained optimal model parameter individual as the final parameter of the identity information recognition model to obtain the identity information recognition model after training. Otherwise, return to the step of obtaining fitness.

[0090] The training end condition may be set as follows: when the current training times are greater than or equal to a preset maximum training times, it may be determined that the training end condition is met.

[0091] In the prior art, particle swarm optimization is often used to optimize the hyperparameters of artificial intelligence algorithms, but there is a problem of being easily trapped in local optimality, which ultimately leads to poor identity recognition results. Therefore, an embodiment of the present invention provides a model parameter optimization method to solve the problems existing in the prior art and improve the accuracy of identity recognition.

[0092] In the embodiment of the present invention, for any model parameter individual, after applying the model parameter corresponding to the model parameter individual to the identity information recognition model, using the biometric information corresponding to the user to be authorized as the input of the identity information recognition model, using the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, and determining the fitness corresponding to the model parameter individual, including:

[0093] For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the identity information recognition model after the parameters are applied is obtained;

[0094] Using the authorization category label corresponding to the biometric information as the input of the identity information recognition model after applying the parameters, and obtaining the actual output of the identity information recognition model;

[0095] Taking the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, and obtaining the cross entropy loss function of the identity information recognition model according to the actual output and the expected output of the identity information recognition model;

[0096] After adding the cross entropy loss function of the identity information recognition model to the preset constant term, the intermediate numerical term is obtained, and the reciprocal of the intermediate numerical term is taken to obtain the fitness corresponding to the individual model parameter.

[0097] In the embodiment of the present invention, according to the fitness corresponding to each model parameter individual, the optimal model parameter individual, the elite model parameter individual and the common model parameter individual are determined among all the model parameter individuals, including:

[0098] According to the fitness corresponding to each model parameter individual, the model parameter individual with the largest fitness is determined as the optimal model parameter individual;

[0099] For other model parameter individuals except the optimal model parameter individual, the first N model parameter individuals are taken as elite model parameter individuals in descending order of fitness, and the remaining model parameter individuals are taken as ordinary model parameter individuals; where N is less than M / 3, M represents the total number of all model parameter individuals, and both M and N are positive integers.

[0100] In the embodiment of the present invention, for the elite model parameter individuals, according to the optimal model parameter individuals, the elite model parameter individuals are searched by using a spiral adaptive local optimal search method to obtain the elite model parameter individuals after the local search, including:

[0101] Based on the current number of training times and the maximum number of training times, the adaptive inertia weight is obtained by combining the sine and cosine functions:

[0102]

[0103] in, represents the adaptive inertia weight, t represents the current number of training times, T represents the maximum number of training times, represents pi, sin represents sine function, cos represents cosine function, k represents the inertia weight control constant, and k Set to 0.2.

[0104] For any elite model parameter individual, randomly match another elite model parameter individual to obtain the first model parameter individual to be learned corresponding to the elite model parameter individual;

[0105] According to the first model parameter individual to be learned corresponding to the elite model parameter individual, the first spiral learning information corresponding to the elite model parameter individual is obtained by using the spiral function:

[0106]

[0107] in, Indicates i The first spiral learning information corresponding to the elite model parameter individual d Dimension parameters, Indicates t During the training i The elite model parameter individual d Dimension parameters, represents the first model parameter individual to be learned d Dimension parameters, i =1,2,…,N, where N represents the total number of elite model parameter individuals. d =1,2,…,D, where D represents the total dimension of the parameters of the elite model parameter individuals. Represents a random number between [-1,1], e represents the natural constant, b represents the helical morphology constant, and b is set to 1.2.

[0108] Determine the historical optimal state corresponding to each elite model parameter individual, and combine the adaptive inertia weight, the historical optimal state and the optimal model parameter individual to obtain the first current update speed:

[0109]

[0110] in, Indicates t During the training i The elite model parameter individual d The speed corresponding to the dimension parameter, Indicates t +1 training session i The elite model parameter individual d The speed corresponding to the dimension parameter, that is, the first current update speed d Dimension parameters, represents the first learning coefficient, represents the second learning coefficient, e represents the natural constant, represents the search range control coefficient, Indicates i The historical optimal state corresponding to the individual parameter of the elite model d Dimension parameters, The first individual representing the optimal model parameter d Dimension parameters, Indicates i The Euclidean distance between an elite model parameter individual and its historical optimal state, represents the optimal model parameter individual and i The Euclidean distance between the elite model parameters.

[0111] Based on the first current update speed and the first spiral learning information, the elite model parameter individuals are searched, and the elite model parameter individuals after the local search are obtained as follows:

[0112]

[0113] in, Indicates i The elite model parameter individual after local search d Dimension parameters.

[0114] The embodiment of the present invention uses a spiral adaptive local optimal search method to search for elite model parameter individuals, which can not only enable the elite model parameter individuals to search in a spiral route, but also learn better location information, and improve the algorithm's search speed in the process of exploring local areas. In the early stage of training, the initial value of the adaptive inertia weight is large. Due to the nonlinear attenuation characteristics of the adaptive inertia weight, the algorithm achieves higher accuracy at a faster convergence speed and has a stronger global optimization capability; in the later stage of iteration, in the spiral update stage, the adaptive inertia weight decreases with a nonlinear trend, and the algorithm's local optimization capability is improved, so that the elite model parameter individuals continue to approach the target individual to achieve the optimal solution.

[0115] In the embodiment of the present invention, for common model parameter individuals, an adaptive collaborative search method is used to perform interval search on the common model parameter individuals to obtain common model parameter individuals after the interval search, including:

[0116] For any common model parameter individual, a model parameter individual is randomly matched to the common model parameter individual to obtain a second model parameter individual to be learned corresponding to the common model parameter individual;

[0117] According to the second model parameter individual to be learned corresponding to the common model parameter individual, the second spiral learning information corresponding to the common model parameter individual is obtained by using the spiral function:

[0118]

[0119] in, Indicates t During the training m The common model parameter individual d Dimension parameters, The first individual parameter of the second model to be learned d Dimension parameters, m =1,2,…,K, K represents the total number of common model parameter individuals, Indicates m The second spiral learning information corresponding to the common model parameter individuals d dimension parameter, cos represents the cosine function, represents pi, represents a random number between [-1,1], b represents the spiral morphology constant, and b is set to 1.2; || represents the absolute value sign.

[0120] According to the adaptive inertia weight and the second model parameter individual to be learned, the second current update speed is obtained as:

[0121]

[0122] in, Indicates t During the training m The common model parameter individual d The speed corresponding to the dimension parameter, Indicates t +1 training session m The common model parameter individual d The speed corresponding to the dimension parameter, that is, the second current update speed d Dimension parameters, represents the third learning coefficient, Indicates m The Euclidean distance between the common model parameter individuals and the second model parameter individuals to be learned;

[0123] According to the second current update speed and the second spiral learning information, the common model parameter individuals are interval searched, and the common model parameter individuals after the interval search are obtained as follows:

[0124]

[0125] in, Indicates m The first individual of the common model parameter after the interval search d Dimension parameters.

[0126] The embodiment of the present invention adopts an adaptive collaborative search method to perform interval search on common model parameter individuals, which can effectively realize the search of common model parameter individuals in local areas, improve the search capability of more unsearched areas, and thus improve the possibility of finding the global optimal solution.

[0127] In the embodiment of the present invention, for the elite model parameter individuals after the local search and the common model parameter individuals after the interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after the global search, including:

[0128] According to the upper limit individuals and lower limit individuals corresponding to the model parameter individuals, the midpoint information of the feasible domain is determined as follows:

[0129]

[0130] in, The first d Dimension parameters, Represents the d-th dimension parameter of the upper limit individual, that is, the upper limit of the d-th dimension parameter; Represents the d-th dimension parameter of the lower limit individual, that is, the lower limit of the d-th dimension parameter.

[0131] A perturbation factor is randomly generated between (0, 1), and the midpoint information of the feasible domain is perturbed by the perturbation factor. The midpoint information of the feasible domain after perturbation is obtained as follows: ;in, represents the disturbance factor.

[0132] For any target model parameter individual, the perturbation factor is used to perturb the elite model parameter individual or the common model parameter individual, and the target model parameter individual after perturbation is obtained as follows: ; Wherein, the target model parameter individual represents the elite model parameter individual after local search or the common model parameter individual after interval search; Indicates s The target model parameter individual d Dimension parameters, s= 1,2,…,(M-1).

[0133] According to the midpoint information of the feasible domain, the midpoint information of the feasible domain after disturbance, and the target model parameter individual after disturbance, a global search is performed on the target model parameter individual to obtain the target model parameter individual after the global search, that is, the elite model parameter individual or the ordinary model parameter individual after the global search is obtained as follows:

[0134]

[0135] in, represents the first individual of the target model parameter after global search. d Dimension parameters.

[0136] The embodiment of the present invention uses a reverse learning method to perform a global search on elite model parameter individuals and common model parameter individuals, which can effectively achieve a global search and help the algorithm jump out of the local optimum, thereby achieving a better training effect. During the global search process, a simulated annealing algorithm can be used for control, thereby improving the algorithm's search accuracy in the later stage.

[0137] Optionally, after each search, each model parameter individual may be processed for out-of-bounds to ensure parameter validity.

[0138] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0142] A person of ordinary skill in the art can understand that all or part of the steps in realizing the above-mentioned facts and methods can be completed by instructing the relevant hardware through a program, and the program involved or the program described can be stored in a computer-readable storage medium. When the program is executed, it includes the following steps: At this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a disk, an optical disk, etc.

[0143] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A data security control method based on biometrics, characterized in that: include: Obtain the biometric information of the user to be authorized and the data authority management key entered by the administrator, verify the data authority management key, and obtain the authority verification result; When the permission verification result is verification passed, an authorization category label corresponding to the biometric information of the user to be authorized is generated; An identity information recognition model is constructed using an artificial intelligence algorithm, and the identity information recognition model is trained using the biometric information corresponding to the user to be authorized and the corresponding authorization category label to obtain a trained identity information recognition model; Deploy the trained identity information recognition model on the data platform, and simultaneously obtain the data management permissions corresponding to each user to be authorized entered by the administrator; When a user accesses the data platform, the biometric information corresponding to the current access user is collected, and the deployed identity information recognition model is used to recognize the biometric information corresponding to the current access user to obtain the identity recognition result; When the identity recognition result is that the identity authentication is passed, the current access user is allowed to access the data platform, and the access process is controlled according to the data management authority; An artificial intelligence algorithm is used to construct an identity information recognition model, and the biometric information corresponding to the user to be authorized and the corresponding authorization category label are used to train the identity information recognition model to obtain the trained identity information recognition model, including: An identity information recognition model is constructed by using an artificial intelligence algorithm, and model parameters of the identity information recognition model are initialized to determine a plurality of model parameter individuals; wherein each model parameter individual includes all parameters to be trained of the identity information recognition model; For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the biometric information corresponding to the user to be authorized is used as the input of the identity information recognition model, and the authorization category label corresponding to the biometric information is used as the expected output of the identity information recognition model to determine the fitness corresponding to the model parameter individual; According to the fitness corresponding to each model parameter individual, the optimal model parameter individual, the elite model parameter individual and the common model parameter individual are determined among all the model parameter individuals; For elite model parameter individuals, according to the optimal model parameter individuals, the elite model parameter individuals are searched by using a spiral adaptive local optimal search method to obtain elite model parameter individuals after local search; For common model parameter individuals, an adaptive collaborative search method is used to perform interval search on the common model parameter individuals to obtain common model parameter individuals after interval search; For the elite model parameter individuals after local search and the common model parameter individuals after interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after global search; Determine whether the training end conditions are met. If so, re-obtain the optimal model parameter individual based on the elite model parameter individual and the ordinary model parameter individual after the global search, and use the re-obtained optimal model parameter individual as the final parameter of the identity information recognition model to obtain the identity information recognition model after training. Otherwise, return to the step of obtaining fitness.

2. The biometric-based data security control method according to claim 1, characterized in that: When a new user is added to be authorized, the transfer learning algorithm is used to re-update and deploy the identity information recognition model.

3. The biometric-based data security control method according to claim 1, characterized in that: Obtain the biometric information of the user to be authorized and the data authority management key entered by the administrator, verify the data authority management key, and obtain the authority verification result, including: Obtain a facial image of the user to be authorized and obtain the biometric information corresponding to the user to be authorized; Obtaining a data rights management key input by an administrator; wherein the data rights management key is a digital key or a key stored in a physical key device; When the data permission management key input by the administrator is the same as the permission management key pre-stored in the data platform, the permission verification result is determined to be verification passed, otherwise the permission verification result is determined to be verification failed.

4. The biometric-based data security control method according to claim 3, characterized in that: When the authority verification result is a verification failure, the administrator is denied from entering the information of the authorized user.

5. The data security control method based on biometrics according to claim 1, characterized in that: For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the biometric information corresponding to the user to be authorized is used as the input of the identity information recognition model, and the authorization category label corresponding to the biometric information is used as the expected output of the identity information recognition model, and the fitness corresponding to the model parameter individual is determined, including: For any model parameter individual, after applying the model parameters corresponding to the model parameter individual to the identity information recognition model, the identity information recognition model after the parameters are applied is obtained; Using the authorization category label corresponding to the biometric information as the input of the identity information recognition model after applying the parameters, and obtaining the actual output of the identity information recognition model; Taking the authorization category label corresponding to the biometric information as the expected output of the identity information recognition model, and obtaining the cross entropy loss function of the identity information recognition model according to the actual output and the expected output of the identity information recognition model; After adding the cross entropy loss function of the identity information recognition model to the preset constant term, the intermediate numerical term is obtained, and the reciprocal of the intermediate numerical term is taken to obtain the fitness corresponding to the individual model parameter.

6. The biometric-based data security control method according to claim 5, characterized in that: According to the fitness corresponding to each model parameter individual, the optimal model parameter individual, the elite model parameter individual and the common model parameter individual are determined among all the model parameter individuals, including: According to the fitness corresponding to each model parameter individual, the model parameter individual with the largest fitness is determined as the optimal model parameter individual; For other model parameter individuals except the optimal model parameter individual, the first N model parameter individuals are taken as elite model parameter individuals in descending order of fitness, and the remaining model parameter individuals are taken as ordinary model parameter individuals; where N is less than M / 3, M represents the total number of all model parameter individuals, and both M and N are positive integers.

7. The biometric-based data security control method according to claim 6, characterized in that: For the elite model parameter individuals, according to the optimal model parameter individuals, the elite model parameter individuals are searched by using the spiral adaptive local optimal search method to obtain the elite model parameter individuals after the local search, including: Based on the current training times and the maximum training times, the adaptive inertia weight is obtained by combining the sine and cosine functions; For any elite model parameter individual, randomly match another elite model parameter individual to obtain the first model parameter individual to be learned corresponding to the elite model parameter individual; According to the first model parameter individual to be learned corresponding to the elite model parameter individual, a spiral function is used to obtain the first spiral learning information corresponding to the elite model parameter individual; Determine the historical optimal state corresponding to each elite model parameter individual, and obtain the first current update speed by combining the adaptive inertia weight, the historical optimal state and the optimal model parameter individual; Based on the first current update speed and the first spiral learning information, the elite model parameter individuals are searched to obtain the elite model parameter individuals after the local search.

8. The biometric-based data security control method according to claim 7, characterized in that: For common model parameter individuals, an adaptive collaborative search method is used to perform interval search on the common model parameter individuals, and the common model parameter individuals after the interval search are obtained, including: For any common model parameter individual, a model parameter individual is randomly matched to the common model parameter individual to obtain a second model parameter individual to be learned corresponding to the common model parameter individual; According to the second model parameter individuals to be learned corresponding to the common model parameter individuals, a spiral function is used to obtain the second spiral learning information corresponding to the common model parameter individuals; According to the adaptive inertia weight and the second model parameter individual to be learned, the second current update speed is obtained, and according to the second current update speed and the second spiral learning information, the common model parameter individual is interval searched to obtain the common model parameter individual after the interval search.

9. The biometric-based data security control method according to claim 8, characterized in that: For the elite model parameter individuals after local search and the common model parameter individuals after interval search, a reverse learning method is used to perform a global search on the elite model parameter individuals and the common model parameter individuals to obtain the elite model parameter individuals and the common model parameter individuals after global search, including: According to the upper limit individuals and lower limit individuals corresponding to the model parameter individuals, determine the midpoint information of the feasible domain; A perturbation factor is randomly generated between (0, 1), and the midpoint information of the feasible domain is perturbed by the perturbation factor to obtain the midpoint information of the feasible domain after perturbation; For any target model parameter individual, the elite model parameter individual or the common model parameter individual is perturbed by a perturbation factor to obtain the target model parameter individual after perturbation; wherein the target model parameter individual represents the elite model parameter individual after local search or the common model parameter individual after interval search; According to the midpoint information of the feasible domain, the midpoint information of the feasible domain after disturbance and the target model parameter individual after disturbance, a global search is performed on the target model parameter individual to obtain the target model parameter individual after the global search, that is, the elite model parameter individual or the ordinary model parameter individual after the global search is obtained.

Images