Virtual Card Library Access Control System with Multi-factor Authentication

Abstract

The present invention relates to the field of access control technology, and in particular to a virtual card library access control system with multi-factor authentication, comprising the following steps: a user identity authentication module, which verifies the user identity through a multi-factor authentication mechanism; the multi-factor authentication mechanism automatically selects the optimal authentication combination in different authentication scenarios through authentication context association technology; a multi-level artifact detection module: an artifact recognition model is constructed through a historical data set, and an anti-artifact model is generated. The anti-artifact model is used as part of a training process to generate virtual artifact data. The virtual artifact data generated by the anti-artifact model is used to expand a training set to improve the recognition capability of the artifact recognition model. The present invention greatly improves the detection capability of the artifact recognition model for unknown artifacts, and enhances the overall defense capability of the system.

Description

Technical Field

[0001] The present invention relates to the technical field of access control, in particular to a virtual card library access control system with multi-factor authentication. Background Art

[0002] With the rapid development of information technology, more and more sensitive information and key resources are stored in virtual card vaults (VCVs), which are used to protect users' digital identities, payment information and other important data. Therefore, their security has become a top priority. However, existing virtual card vault access control technologies face multiple security challenges, especially in terms of resisting artifact attacks and ensuring user experience. Existing multi-factor authentication (MFA) systems improve security by combining multiple authentication methods (such as password authentication, biometric authentication, token authentication, etc.), but there are still problems when dealing with complex authentication scenarios and dealing with artifact attacks.

[0003] Artifact data refers to data that attackers use to attempt to deceive the authentication system through forged biometrics (such as fake fingerprints, simulated facial images) or forged behavioral data (such as simulated mouse tracks, keyboard input patterns). This type of data usually imitates the operation behavior of legitimate users with a very high degree of simulation, aiming to bypass the security defenses of multi-factor authentication systems. Existing artifact detection technologies often seem to be unable to cope with such complex artifact data, especially when the artifact data involves multimodal characteristics. Single-modal detection methods are difficult to effectively identify such attacks.

[0004] Current artifact detection technology usually focuses on the analysis of single-modal data, such as artifact identification only for biometric data or behavioral data. This method often has difficulty coping with the combined attack of multi-modal artifacts when faced with complex attack scenarios, and is easily deceived by forged multi-modal data, leading to security vulnerabilities.

[0005] Existing artifact detection models usually rely on limited artifact samples for training, resulting in insufficient model recognition capabilities for unknown artifacts. In addition, traditional artifact data generation methods often only consider a single modality and lack the ability to simulate complex scenarios and multimodal data, which makes artifact detection models perform poorly in practical applications.

[0006] Existing multi-factor authentication systems lack the ability to comprehensively analyze and adjust authentication scenarios in real time, and are unable to dynamically adjust authentication strategies based on changes in factors such as access time, geographic location, network environment, and device type. This not only reduces the security of the system, but also affects the user experience. Especially in high-risk scenarios, existing systems may fail to strengthen authentication requirements in a timely manner, leading to potential security threats. Summary of the invention

[0007] Based on the above objectives, the present invention provides a virtual card library access control system with multi-factor authentication.

[0008] The virtual card library access control system with multi-factor authentication includes the following steps:

[0009] A user identity authentication module verifies the user identity through a multi-factor authentication mechanism, including password authentication, biometric authentication (fingerprint, facial recognition), token authentication, geographic location authentication, and device authentication; the multi-factor authentication mechanism automatically selects the optimal authentication combination in different authentication scenarios through authentication context association technology to improve the ability to resist artifact attacks;

[0010] Multi-level artifact detection module: The artifact recognition model is constructed through a large-scale historical data set, and an anti-artifact model is generated. The artifact recognition model identifies forged biometric or behavioral data through the convolutional neural network (CNN). The anti-artifact model generates virtual artifact data through the generative adversarial network (GAN). The anti-artifact model is used to generate virtual artifact data as part of the training process. The virtual artifact data generated by the anti-artifact model is used to expand the training set to improve the recognition ability of the artifact recognition model. During user authentication, the trained artifact recognition model is used to compare the received multi-factor authentication data with known normal behaviors and artifact behaviors to analyze whether there are any abnormalities. If the artifact recognition model detects suspected artifact behaviors, the following measures are taken according to the detection results:

[0011] Trigger multi-level authentication: Require users to provide additional authentication factors to confirm their identity;

[0012] Limit or deny access: If the detection results show obvious artifacts, directly reject the authentication request or lock the account.

[0013] Optionally, the user identity authentication module further includes authentication context association technology, which automatically selects the optimal authentication combination in different authentication contexts, specifically including:

[0014] Contextual data collection: real-time collection of user access contextual data, including access time, geographic location, network environment, device type, and current security threat level;

[0015] Contextual relevance analysis: Analyze various contextual factors in the current authentication context through the pre-built contextual relevance model, and determine the relevance and security risks between the contextual factors;

[0016] Optimal authentication combination selection: Based on the results of situational association analysis, the authentication combination suitable for the current situation is automatically selected. For low-risk situations, loose authentication factors are selected to improve user experience; for high-risk situations, strict multi-factor authentication combinations are selected.

[0017] Optionally, the contextual factors include access time , expressed as the hour of the day or the weight of a specific time period, geographic location , expressed as the coordinates of the user's current location or the distance to a common location, network environment , expressed as the security level of the network, such as IP address credibility or network latency, device type , which is represented by the matching degree of the hardware features of the device or the device fingerprint;

[0018] The situational correlation model analyzes the correlation between various situational factors and calculates a comprehensive safety risk value , including:

[0019] Situational Factor Quantification: Each situational factor Through a normalization function Quantify and convert the original data into a risk value between [0,1], where 0 represents the safest and 1 represents the highest risk;

[0020] Weighting: Each Situational Factor Different contributions to the overall risk need to be given different weights , weights are optimized through historical data; If there are situational factors, the weights satisfy the following conditions: ;

[0021] The comprehensive risk value calculation is calculated by the weighted sum of each scenario factor: ; This formula means that the quantitative values ​​of all situational factors are weighted and summed to obtain an overall situational risk value R. The larger the value of R, the higher the risk of the current authentication situation, and the system will require a more stringent authentication combination.

[0022] Optionally, the optimal authentication combination is selected based on the calculated comprehensive risk value , select appropriate authentication combinations and set multiple risk thresholds ,according to Choose different certification combinations according to the size;

[0023] By judging You can select different numbers and combinations of authentication factors based on the risk value. The higher the risk value, the more authentication factors are required.

[0024] Optionally, the construction of the artifact recognition model specifically includes:

[0025] Collect a variety of real users’ biometric data (fingerprints, facial images, irises) and behavioral data (keyboard input patterns), as well as known forged data, to build a large-scale training dataset containing both normal and artifact behaviors;

[0026] Artifact recognition model training: Use convolutional neural network (CNN) to train the training data set, extract key features from the data, and learn the differences between normal biometric and behavioral data and forged data through multi-layer convolution and pooling operations to generate an artifact recognition model for real-time detection of artifact behavior.

[0027] Optionally, the training process of the convolutional neural network CNN includes optimization of a convolutional layer, a pooling layer, a fully connected layer, and a first loss function;

[0028] The convolution layer extracts key feature patterns by locally weighting and summing the input data, and the feature patterns help identify normal and forged biometric features, and the input data is biometric data;

[0029] The pooling layer reduces the data dimension and retains key features by selecting the maximum value (or average value) in the local area;

[0030] By minimizing the first loss function, the weight and bias of the convolution kernel are adjusted to improve the recognition accuracy of artifact data.

[0031] Optionally, in the generation of the anti-artifact model, training is performed through a generative adversarial network GAN, wherein the GAN consists of a generator and a discriminator, the generator generates forged biometric and behavioral data, and the discriminator evaluates whether the generated data is authentic; during the training process, the generator continuously improves the fidelity of the generated artifact data, and the discriminator improves its ability to discriminate artifact data through adversarial training with the generator;

[0032] It also includes virtual artifact data generation: using the trained anti-artifact model to generate a large amount of realistic virtual artifact data, including a variety of forged biometric features (such as counterfeit fingerprints, synthetic facial images) and abnormal behavior patterns (such as simulated keyboard input patterns). The virtual artifact data is used to expand the training set;

[0033] Training set expansion and artifact recognition model optimization: The virtual artifact data generated by the anti-artifact model is added to the original training data set. The expanded training set contains more artifact samples, allowing the artifact recognition model to learn a wide range of artifact patterns and improve the ability to detect unknown artifacts. The expanded data set is retrained to optimize the accuracy and robustness of the artifact recognition model.

[0034] Optionally, the generator not only generates realistic artifact data, but also considers the multimodal characteristics of multi-factor authentication data, including the joint generation of biometric and behavioral data. Therefore, the generator is designed to simultaneously generate multimodal artifact data:

[0035] ,in, is the generation of fake biometric data (such as fingerprints, facial images), is the generated fake behavior data (such as keyboard input mode), is a random noise vector, is a vector of situational conditions that control the characteristics of the generated data, including specifying the type of biometrics or behavior patterns to be forged;

[0036] The generator passes the situational condition The type and characteristics of the generated data are controlled to generate multimodal artifact data associated with specific situations. This design can make the artifact data more complex and simulate more realistic attack scenarios.

[0037] Optionally, the discriminator processes multimodal data simultaneously and determines whether the combination of multimodal data is artifact data. A joint discriminator is designed to make a joint judgment by combining multimodal inputs, which is expressed as:

[0038] ,in, is the input biometric data, and is the weight matrix of the discriminator corresponding to the biometric data and behavioral data, is the bias term;

[0039] The discriminator receives biometric data and behavioral data simultaneously, and evaluates whether it is an artifact by jointly judging the correlation between the biometric data and the behavioral data. This design takes into account the correlation between multimodal data and enhances the detection capability of complex artifacts.

[0040] Optionally, the GAN also includes a second loss function. In order to better adapt to artifact detection of multimodal data, the second loss function is adopted as a joint loss function to optimize the discrimination results of biometric data and behavioral data respectively.

[0041] Beneficial effects of the present invention:

[0042] The present invention significantly improves the accuracy of artifact detection by introducing multimodal data processing and analysis technology. It extracts and analyzes biometric data (such as fingerprints and facial images) through convolutional neural networks (CNNs), and combines behavioral data (keyboard input mode) with joint judgment to achieve effective recognition of complex artifact behaviors. Through this joint detection of multimodal data, the system can accurately identify various types of artifacts in different scenarios, especially those that are difficult to detect through single modality detection, which significantly improves the security of virtual card library access control.

[0043] In the present invention, the anti-artifact model utilizes the adaptive deformation of the generative adversarial network (GAN) to control the generation of artifact data with different characteristics and multi-modality through the situational condition vector c, including the joint generation of biometrics and behavioral data. Compared with the traditional single-modality generation, the generated artifact data is more diverse and realistic in multi-modal characteristics, and can simulate more complex attack scenarios. By adding these multi-modal artifact data to the training set of the artifact recognition model, the diversity of the training data is expanded, so that the artifact recognition model's detection capability for unknown artifacts is greatly improved, and the overall defense capability of the system is enhanced.

[0044] The present invention adopts authentication context association technology, and dynamically adjusts the combination strategy of multi-factor authentication through real-time analysis and weight calculation of contextual factors such as access time, geographic location, network environment, and device type. This technology can automatically select the optimal authentication combination according to the risk level of the current situation, and dynamically adjust the authentication strategy according to situation changes during the authentication process. The flexibility enables the system to optimize the user experience while ensuring high security and avoid the inconvenience caused by excessive authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0046] Figure 1 A schematic diagram of a system module according to an embodiment of the present invention;

[0047] Figure 2 Schematic diagram of a generative adversarial network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0048] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. At the same time, it is explained here that in order to make the embodiments more detailed, the following embodiments are the best and preferred embodiments, and those skilled in the art may also adopt other alternatives to implement some known technologies; and the accompanying drawings are only for more specific description of the embodiments, and are not intended to specifically limit the present invention.

[0049] It should be noted that the references to "one embodiment", "embodiment", "exemplary embodiments", "some embodiments" and the like in the specification indicate that the embodiments described may include specific features, structures or characteristics, but not every embodiment may include the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in conjunction with an embodiment, it should be within the knowledge of a person skilled in the art to implement such feature, structure or characteristic in conjunction with other embodiments (whether or not explicitly described).

[0050] In general, a term can be understood, at least in part, from its use in context. For example, depending, at least in part, on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending, at least in part, on the context, allow for the presence of other factors that are not necessarily explicitly described.

[0051] like Figure 1-Figure 2 As shown, the multi-factor authentication virtual card library access control system includes the following steps:

[0052] User authentication module verifies user identity through a multi-factor authentication mechanism, including password authentication, biometric authentication (fingerprint, facial recognition), token authentication, geographic location authentication, and device authentication. The multi-factor authentication mechanism automatically selects the optimal authentication combination in different authentication scenarios through authentication context association technology to improve the ability to resist artifact attacks.

[0053] Password authentication: Identity verification is performed through a static password entered by the user. The system dynamically adjusts the weight of password verification based on the complexity and historical usage of the password, and combines other authentication factors to enhance the verification effect.

[0054] Biometric authentication: including fingerprint recognition and facial recognition, obtaining the user's biometric information through multimodal biometric data collection equipment.

[0055] Token authentication: A one-time password or key is generated by a hardware token or software token, and the user needs to enter the key within a specified time to complete the authentication.

[0056] Geographic location authentication: The user's current location is obtained through the GPS module or IP address information of the user's device and matched with the preset geographic location whitelist; when the user's location does not match his or her usual location or the preset safe area, the system will increase the weight of other authentication factors or trigger additional verification steps.

[0057] Device authentication: Based on device fingerprint technology, the hardware characteristics of the user's access device (such as CPU serial number, MAC address, device configuration, etc.) are identified; device authentication determines the legitimacy of the device by comparing the user's historical access device information with the current device information. If the device information changes, the system will increase the verification strength of other authentication factors.

[0058] Multi-level artifact detection module: The artifact recognition model is constructed through a large-scale historical data set, and an anti-artifact model is generated. The artifact recognition model identifies forged biometric or behavioral data through the convolutional neural network (CNN). The anti-artifact model generates virtual artifact data through the generative adversarial network (GAN). The anti-artifact model is used to generate virtual artifact data as part of the training process. The virtual artifact data generated by the anti-artifact model is used to expand the training set to improve the recognition ability of the artifact recognition model. During user authentication, the trained artifact recognition model is used to compare the received multi-factor authentication data with known normal behaviors and artifact behaviors to analyze whether there are any abnormalities. If the artifact recognition model detects suspected artifact behaviors, the following measures are taken according to the detection results:

[0059] Trigger multi-level authentication: requiring the user to provide an additional authentication factor (such as an additional biometric or a one-time password) to confirm the user's identity;

[0060] Limit or deny access: If the detection results show obvious artifacts, directly reject the authentication request or lock the account;

[0061] Multi-level verification also includes cross-checking using a multi-level verification mechanism during the user authentication process. By simultaneously verifying multimodal biometric and behavioral data, data cross-matching is performed between different verification levels to ensure consistency and authenticity in the authentication process and prevent artifacts from bypassing a single verification point.

[0062] The user identity authentication module also includes authentication context association technology, which automatically selects the optimal authentication combination in different authentication contexts, including:

[0063] Contextual data collection: real-time collection of user access contextual data, including access time, geographic location, network environment, device type, and current security threat level;

[0064] Contextual relevance analysis: Analyze various contextual factors in the current authentication context through the pre-built contextual relevance model, and determine the relevance and security risks between the contextual factors;

[0065] Optimal authentication combination selection: Based on the results of contextual correlation analysis, the authentication combination suitable for the current context is automatically selected. For low-risk contexts, loose authentication factors (such as password-only authentication) are selected to improve user experience; for high-risk contexts, strict multi-factor authentication combinations (such as adding biometric authentication, token authentication, or requiring multiple verifications) are selected.

[0066] During the authentication process, situational changes are monitored in real time, and the authentication combination is dynamically adjusted based on new situational data. If an increase in risk level is detected during the authentication process, additional authentication steps are immediately triggered to ensure safety. Conversely, if the situational risk is reduced, the authentication steps are simplified to improve efficiency.

[0067] Situational factors include time of visit , expressed as the hour of the day or the weight of a specific time period, geographic location , expressed as the coordinates of the user's current location or the distance to a common location, network environment , expressed as the security level of the network, such as IP address credibility or network latency, device type , which is represented by the matching degree of the hardware features of the device or the device fingerprint;

[0068] The situational correlation model analyzes the correlation between various situational factors and calculates a comprehensive safety risk value , including:

[0069] Situational Factor Quantification: Each situational factor Through a normalization function Quantify and convert the original data into a risk value between [0,1], where 0 represents the safest and 1 represents the highest risk. The standardization function ;

[0070] Weighting: Each Situational Factor Different contributions to the overall risk need to be given different weights , weights are optimized through historical data; If there are situational factors, the weights satisfy the following conditions: ;

[0071] The comprehensive risk value calculation is calculated by the weighted sum of each scenario factor: ; This formula means that the quantitative values ​​of all situational factors are weighted and summed to obtain an overall situational risk value , The larger the value, the higher the risk of the current authentication situation, and the system will require a stricter authentication combination.

[0072] The optimal certification combination is selected based on the calculated comprehensive risk value , select appropriate authentication combinations and set multiple risk thresholds ,according to Choose from different certification combinations for your size:

[0073]

[0074] in, Represents password authentication, stands for biometric authentication, stands for token authentication, Stands for Geographic Location Authentication, Stands for equipment authentication;

[0075] By judging You can select different numbers and combinations of authentication factors based on the risk value. The higher the risk value, the more authentication factors are required.

[0076] During the authentication process, the system collects new situational data in real time and dynamically adjusts the risk value according to changes in the situation. and certification combination :

[0077] ;

[0078] ;

[0079] in, is the risk value adjusted based on the new data. It is the increase or decrease in risk value caused by the change in situation. It is an authentication combination selected based on the new risk value. During the authentication process, the system will dynamically adjust the authentication strategy according to the changes in the situation. If the situation becomes more dangerous, the system will increase the authentication requirements, otherwise it will simplify the process.

[0080] The construction of the artifact recognition model specifically includes:

[0081] Collect a variety of real users’ biometric data (fingerprints, facial images, irises) and behavioral data (keyboard input patterns), as well as known forged data, to build a large-scale training dataset containing both normal and artifact behaviors;

[0082] Artifact recognition model training: Use convolutional neural network (CNN) to train the training data set, extract key features from the data, and learn the differences between normal biometric and behavioral data and forged data through multi-layer convolution and pooling operations to generate an artifact recognition model for real-time detection of artifact behavior.

[0083] The training process of the convolutional neural network CNN includes the optimization of the convolution layer, the pooling layer, the fully connected layer, and the first loss function;

[0084] The convolution layer extracts key feature patterns by locally weighting and summing the input data. The feature patterns help identify normal and forged biometrics. The input data is biometric data. In the convolution layer, each convolution operation is expressed as:

[0085] ,in, It is The convolution kernels are located at The output value at is the input feature map at position The value at It is The weight of the convolution kernel, It is The bias term of the convolution kernel, and is the height and width of the convolution kernel;

[0086] The pooling layer reduces the data dimension and retains key features by selecting the maximum value (or average value) in the local area, thereby improving the computational efficiency of the model. The pooling layer is used for downsampling and uses maximum pooling. The expression of maximum pooling is: ;

[0087] By minimizing the first loss function, adjusting the weight and bias of the convolution kernel, the recognition accuracy of artifact data is improved; the first loss function uses the cross entropy loss function to measure the difference between the model prediction result and the true label, which is expressed as: ,in, is the loss value, is the number of categories (normal data and artifact data), is the true label (1 for the correct category, 0 for the wrong category), is the predicted probability of the model.

[0088] In the generation of the anti-artifact model, training is performed through a generative adversarial network (GAN). GAN consists of a generator and a discriminator. The generator generates fake biometric and behavioral data, and the discriminator evaluates whether the generated data is real. During the training process, the generator continuously improves the fidelity of the generated artifact data, and the discriminator improves its ability to distinguish artifact data through adversarial training with the generator.

[0089] It also includes virtual artifact data generation: using the trained anti-artifact model to generate a large amount of realistic virtual artifact data, including a variety of forged biometric features (such as counterfeit fingerprints, synthetic facial images) and abnormal behavior patterns (such as simulated keyboard input patterns). The virtual artifact data is used to expand the training set;

[0090] Training set expansion and artifact recognition model optimization: The virtual artifact data generated by the anti-artifact model is added to the original training data set. The expanded training set contains more artifact samples, allowing the artifact recognition model to learn a wide range of artifact patterns and improve the ability to detect unknown artifacts. The expanded data set is retrained to optimize the accuracy and robustness of the artifact recognition model.

[0091] The generator not only generates realistic artifact data, but also considers the multimodal characteristics of multi-factor authentication data, including the joint generation of biometric and behavioral data. Therefore, the generator is designed to generate multimodal artifact data simultaneously:

[0092] ,in, is the generation of fake biometric data (such as fingerprints, facial images), is the generated fake behavior data (such as keyboard input mode), is a random noise vector, is a vector of situational conditions that control the characteristics of the generated data, including specifying the type of biometrics or behavior patterns to be forged;

[0093] Generators through situational conditions The type and characteristics of the generated data are controlled to generate multimodal artifact data associated with specific situations. This design can make the artifact data more complex and simulate more realistic attack scenarios.

[0094] The discriminator processes multimodal data at the same time and determines whether the combination of multimodal data is artifact data. A joint discriminator is designed to make a joint judgment by combining multimodal inputs, which is expressed as:

[0095] ,in, is the input biometric data, and is the weight matrix of the discriminator corresponding to the biometric data and behavioral data, is the bias term;

[0096] The discriminator receives biometric data and behavioral data at the same time, and evaluates whether it is an artifact by jointly judging the correlation between the biometric data and the behavioral data. This design takes into account the correlation between multimodal data and enhances the detection capability of complex artifacts.

[0097] GAN also includes a second loss function. In order to better adapt to the artifact detection of multimodal data, the second loss function is adopted as a joint loss function, which optimizes the discrimination results of biometric data and behavioral data respectively, expressed as:

[0098]

[0099] The deformed loss function takes into account the authenticity of both biometric data and behavioral data. By jointly optimizing the generator and the discriminator, the generated artifact data is closer to the real data in multiple modalities, thereby enhancing the artifact detection capability of the overall system.

[0100] Finally, the anti-artifact model generates multimodal artifact data Added to the training set of the artifact recognition model, the artifact recognition model with expanded training set can recognize more complex and diverse artifacts:

[0101] ,in, is the output of the artifact recognition model on the biometric data, It is the output of the artifact recognition model on the behavior data. By using the extended multimodal artifact data, the artifact recognition model can better cope with complex attack scenarios and improve the detection ability of different types of artifacts.

[0102] The present invention covers any substitution, modification, equivalent method and scheme made on the essence and scope of the present invention. In order to make the public have a thorough understanding of the present invention, specific details are described in detail in the following preferred embodiments of the present invention, but those skilled in the art can fully understand the present invention without the description of these details. In addition, in order to avoid unnecessary confusion about the essence of the present invention, well-known methods, processes, procedures, components and circuits are not described in detail.

[0103] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A multi-factor authentication virtual card library access control system, characterized in that: Includes the following modules: A user identity authentication module verifies the user identity through a multi-factor authentication mechanism, including password authentication, biometric authentication, token authentication, geographic location authentication, and device authentication; the multi-factor authentication mechanism automatically selects the optimal authentication combination in different authentication scenarios through authentication context association technology; Multi-level artifact detection module: constructs an artifact recognition model through historical data sets and generates an anti-artifact model. The artifact recognition model identifies forged biometric or behavioral data through the convolutional neural network (CNN). The anti-artifact model generates virtual artifact data through the generative adversarial network (GAN). The anti-artifact model is used to generate virtual artifact data as part of the training process. The virtual artifact data generated by the anti-artifact model is used to expand the training set to improve the recognition ability of the artifact recognition model. During user authentication, the trained artifact recognition model is used to compare the received multi-factor authentication data with known normal behaviors and artifact behaviors to analyze whether there are any abnormalities. If the artifact recognition model detects suspected artifact behaviors, the following measures are taken according to the detection results: Trigger multi-level authentication: Require users to provide additional authentication factors to confirm their identity; Limit or deny access: If the detection results show obvious artifacts, directly reject the authentication request or lock the account; In the generation of the anti-artifact model, training is performed through a generative adversarial network (GAN), which consists of a generator and a discriminator. The generator generates forged biometric and behavioral data, and the discriminator evaluates whether the generated data is authentic. During the training process, the generator continuously improves the fidelity of the generated artifact data, and the discriminator is trained through adversarial training with the generator. It also includes virtual artifact data generation: using the trained anti-artifact model to generate realistic virtual artifact data, the data includes a variety of forged biometric features and abnormal behavior patterns, and the virtual artifact data is used to expand the training set; Training set expansion and artifact recognition model optimization: Add virtual artifact data generated by the anti-artifact model to the original training data set. The expanded training set contains artifact samples, allowing the artifact recognition model to learn a wide range of artifact patterns. Retrain the expanded data set. The generator not only generates realistic artifact data, but also considers the multimodal characteristics of multi-factor authentication data, including the joint generation of biometric and behavioral data. The generator is designed to generate multimodal artifact data simultaneously: ,in, is the generated fake biometric data, is the generated fake behavior data, is a random noise vector, is a vector of situational conditions that control the characteristics of the generated data, including specifying the type of biometrics or behavior patterns to be forged; The generator is constructed by using the context condition vector Control the type and characteristics of the generated data so that it generates multimodal artifact data associated with specific situations; The discriminator processes multimodal data simultaneously and determines whether the combination of multimodal data is artifact data. A joint discriminator is designed to make a joint judgment by combining multimodal inputs, which is expressed as: ,in, is the input biometric data, and is the weight matrix of the discriminator corresponding to the biometric data and behavioral data, is the bias term; The discriminator receives the biometric data and the behavioral data at the same time, and evaluates whether it is an artifact by jointly judging the correlation between the biometric data and the behavioral data; The GAN also includes a second loss function, which is a joint loss function, and is optimized for the discrimination results of biometric data and behavioral data, respectively, and is expressed as: 。 2. The multi-factor authentication virtual card library access control system according to claim 1, characterized in that: The user identity authentication module also includes authentication context association technology, which automatically selects the optimal authentication combination in different authentication contexts, specifically including: Contextual data collection: real-time collection of user access contextual data, including access time, geographic location, network environment, device type, and current security threat level; Contextual relevance analysis: Analyze various contextual factors in the current authentication context through the pre-built contextual relevance model, and determine the relevance and security risks between the contextual factors; Optimal authentication combination selection: Based on the results of situational association analysis, the authentication combination suitable for the current situation is automatically selected. For low-risk situations, loose authentication factors are selected to improve user experience; for high-risk situations, strict multi-factor authentication combinations are selected.

3. The multi-factor authentication virtual card library access control system according to claim 2, characterized in that: The contextual factors include the time of visit , Geographical location , Network environment And the device type ; The situational correlation model analyzes the correlation between various situational factors and calculates a comprehensive safety risk value , including: Situational Factor Quantification: Each situational factor Through a normalization function Quantify and convert the original data into a risk value between [0,1], where 0 represents the safest and 1 represents the highest risk; Weighting: Each Situational Factor The contribution to the overall risk is different, so different weights are required , weights are optimized through historical data; The comprehensive risk value is calculated by the weighted sum of each scenario factor. 。 4. The multi-factor authentication virtual card library access control system according to claim 3, characterized in that: The optimal authentication combination is selected based on the calculated comprehensive risk value , select authentication combinations and set multiple risk thresholds ,according to Choose different certification combinations according to the size; By judging You can select different numbers and combinations of authentication factors based on the risk value. The higher the risk value, the more authentication factors are required.

5. The multi-factor authentication virtual card library access control system according to claim 1, characterized in that: The construction of the artifact recognition model specifically includes: Collect biometric and behavioral data of multiple real users, as well as known forged data, to build a training dataset containing normal and artifact behaviors; Artifact recognition model training: Use convolutional neural network (CNN) to train the training data set, extract key features from the data, and learn the differences between normal biometric and behavioral data and forged data through multi-layer convolution and pooling operations to generate an artifact recognition model for real-time detection of artifact behavior.

6. The multi-factor authentication virtual card library access control system according to claim 5, characterized in that: The training process of the convolutional neural network CNN includes optimization of a convolutional layer, a pooling layer, a fully connected layer and a first loss function; The convolution layer extracts characteristic patterns by locally weighting and summing the input data, and the characteristic patterns help identify normal and forged biometric features, and the input data is biometric data; The pooling layer reduces the data dimension and retains key features by selecting the maximum value in the local area; By minimizing the first loss function, the weight and bias of the convolution kernel are adjusted to improve the recognition accuracy of artifact data.

Images