Biological characteristic intelligent management system and method based on artificial intelligence

Abstract

The invention relates to the technical field of information management, and discloses a biological characteristic intelligent management system and method based on artificial intelligence, and the method comprises the following steps: responding to a business request initiated by a user, and collecting at least two kinds of biological characteristic information of the user; inputting the biological characteristic information into a pre-trained multi-modal fusion model to generate a composite biological characteristic code; wherein the multi-modal fusion model adopts an attention mechanism to dynamically weight and fuse contribution degrees of different biological characteristics; and based on the context information of the service request, the hardware fingerprint of the user terminal and the current network behavior mode. According to the method, multi-modal biological feature fusion and dynamic risk assessment are combined, the accuracy and reliability of an identity authentication system are remarkably improved, different biological features are adaptively weighted by using an attention mechanism, the distinction degree and the anti-counterfeiting capability of a composite feature code are effectively enhanced, and the security of the identity authentication system is improved. Therefore, the user identity can be discriminated more accurately in a complex application scene.

Description

Technical Field

[0001] This invention relates to the field of information management technology, and in particular to an intelligent management system and method based on artificial intelligence and biometrics. Background Technology

[0002] Existing identity authentication systems mostly rely on single biometric features or static password verification. When faced with complex and ever-changing business scenarios and increasingly sophisticated forgery attacks, their security and reliability face severe challenges. At the same time, traditional systems generally lack the ability to dynamically perceive user behavior context and environmental risks, and cannot adaptively adjust authentication strategies according to real-time risk levels. This makes it difficult to balance security strength and user experience, resulting in either security vulnerabilities due to insufficient verification or disruption to normal business processes due to excessive verification.

[0003] An existing patent discloses a data security control method based on biometrics (publication number CN119337355A). The technical solution disclosed in this existing patent mainly adopts single biometric verification. However, single biometric verification is susceptible to forgery attacks, lacks the ability to dynamically perceive the real-time risk environment, cannot adaptively adjust the authentication strength according to the risk level, and suffers from insufficient long-term adaptability due to the static nature of the system. Summary of the Invention

[0004] This invention provides an artificial intelligence-based intelligent management system and method for biometrics to address existing technical problems, solving the problem of the inability to adaptively adjust authentication strategies according to real-time risk levels.

[0005] To address the aforementioned technical problems, according to one aspect of the present invention, more specifically, an artificial intelligence-based intelligent management method for biometrics includes the following steps:

[0006] S1. In response to a user-initiated business request, collect at least two types of biometric information from the user; input the biometric information into a pre-trained multimodal fusion model to generate a composite biometric code; wherein, the multimodal fusion model uses an attention mechanism to dynamically weight and fuse the contributions of different biometric features;

[0007] S2. Based on the context information of the service request, the hardware fingerprint of the user terminal, and the current network behavior pattern, extract dynamic risk factors; the dynamic risk factors include at least: the geolocation anomaly index, the terminal device change identifier anomaly index, and the deviation of the keystroke rhythm.

[0008] S3. Based on historical business requests in the database and their corresponding dynamic risk factors, as well as the actual negative consequences caused by each historical business request in subsequent business processes, assess and quantify the degree of historical risk; using the quantified degree of historical risk and the corresponding dynamic risk factors, construct a risk assessment model through regression analysis to predict the risk of new business requests.

[0009] S4. Input the dynamic risk factors of the current business request into the risk assessment model, calculate the real-time risk coefficient of the current business request, and classify the real-time risk level of the business request according to the real-time risk coefficient.

[0010] S5. Based on the real-time risk level, dynamically select the corresponding authentication method from the preset verification strategy library; the verification strategy library includes at least the following levels:

[0011] Low-risk business requests are verified once based on the composite biometric code.

[0012] Medium-risk level business requests are used to superimpose secondary verification of behavioral features based on temporal convolutional networks on top of biometric verification.

[0013] High-risk business requests trigger multi-factor authentication, which includes at least composite biometric code, physical key device verification, and one-time dynamic password verification.

[0014] S6. Once identity verification is successful, the verification event is uniquely bound to the business request, generating an authentication record containing a timestamp, business type, verification method, and risk level, and storing it in a blockchain or tamper-proof database for auditing and traceability.

[0015] Furthermore, in step S1, the multimodal fusion model employs an attention mechanism to dynamically weight and fuse the contributions of different biometric features, specifically including:

[0016] For each type of biometric information input, feature extraction is performed to obtain the corresponding feature vector;

[0017] Each feature vector is input into an attention network, which outputs an attention weight for each feature vector. This weight represents the importance of the biometric feature in the current business scenario.

[0018] Based on the attention weights, the feature vectors are weighted and summed to generate the final composite biometric code.

[0019] Furthermore, in step S3, a risk assessment model consisting of two dynamic risk factors is constructed, which are represented as dynamic risk factor A and dynamic risk factor B, respectively. The specific steps are as follows:

[0020] S301. Filter historical business requests in the sample set that have the same variable as dynamic risk factor B; obtain dynamic risk factor A and corresponding risk level in the filtered historical business requests, and use this to construct feature model A;

[0021] S302. Again, filter the historical business requests in the sample set that have the same variable as dynamic risk factor A; obtain the dynamic risk factor B and the corresponding risk level in the filtered historical business requests, and use this to construct feature model B;

[0022] S303. Obtain the corresponding dynamic risk factor A and dynamic risk factor B from the historical business requests in the sample set according to the order of risk level, and determine the relationship between feature model A and feature model B based on dynamic risk factor A and dynamic risk factor B, thereby constructing a risk assessment model.

[0023] Furthermore, the risk assessment model is a predictive model built by analyzing the inherent correlation between these factors and risk levels, based on multiple dynamic risk factors extracted from historical business requests and their corresponding risk levels assessed by a third party. This model is used to calculate the risk coefficient of the current business request in real time and classify the risk level accordingly, thereby providing a basis for decision-making for dynamically selecting the authentication method.

[0024] Furthermore, in step S5, the secondary verification of behavioral features based on the temporal convolutional network specifically involves:

[0025] Synchronously collect keystroke timing data and mouse movement trajectory data of users during the verification process to form a multimodal behavior sequence;

[0026] The multimodal behavior sequence is input into a pre-trained temporal convolutional network to extract its temporal correlation features;

[0027] The temporal correlation features are matched with the user's historical behavior template to calculate the matching degree;

[0028] If the matching degree is higher than the preset threshold, the secondary verification will pass.

[0029] Furthermore, the AI-based intelligent biometric management method also collects successful and failed case data during the verification process, as well as corresponding contextual information.

[0030] The collected case data is used to perform incremental learning or periodic retraining on the multimodal fusion model and the risk assessment model.

[0031] An AI-based intelligent management system for biometrics includes: a biometric acquisition module, a multimodal fusion processing module, a dynamic risk factor extraction module, a risk assessment model construction module, a real-time risk assessment module, a verification strategy execution module, and an authentication record and storage module.

[0032] Furthermore, this also includes:

[0033] The model update module is used to collect successful and failed case data and context information during the verification process, and to perform incremental learning or periodic retraining on the multimodal fusion model and the risk assessment model.

[0034] This invention provides an artificial intelligence-based intelligent management system and method for biometrics. Compared with existing technologies, the advantages of this method are as follows:

[0035] 1. This invention significantly improves the accuracy and reliability of the identity authentication system by combining multimodal biometric fusion with dynamic risk assessment. It utilizes an attention mechanism to adaptively weight different biometric features, effectively enhancing the distinguishability and anti-counterfeiting capability of composite feature codes, thereby achieving more accurate identification of user identities in complex application scenarios.

[0036] 2. This invention constructs a risk assessment model based on real business feedback, which integrates multiple dynamic risk factors into a unified risk coefficient for decision-making. This overcomes the problem of high misjudgment rate of single indicators, significantly reduces the false alarm rate of the system, and achieves an intelligent balance between security strength and ease of operation through dynamic verification strategies.

[0037] 3. This invention introduces a behavioral feature-based secondary verification mechanism based on temporal convolutional networks. Without increasing the user's additional operational burden, it constructs a non-intrusive second line of defense by capturing subtle behavioral biometric features, effectively preventing identity impersonation and imitation attacks, and significantly improving the security level in medium-risk scenarios.

[0038] 4. This invention establishes a continuous learning model optimization mechanism. By collecting actual case data during the verification process, the core algorithm model is iteratively updated, enabling the system to have self-evolution capabilities and continuously adapt to new attack methods and changes in user behavior, ensuring that the system maintains optimal protection status in the long term. Attached Figure Description

[0039] Figure 1 This is a flowchart of the present invention;

[0040] Figure 2 This is a graph showing the relationship between the terminal device change identifier anomaly index D and the risk level A in this invention;

[0041] Figure 3This is a graph showing the relationship between the geographical location anomaly index G and the risk level B in this invention;

[0042] Figure 4 This is a graph showing the relationship between the terminal device change identifier anomaly index D and the geographical location anomaly index G in this invention. Detailed Implementation

[0043] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0044] Example 1:

[0045] like Figure 1 As shown, according to one aspect of the present invention, an artificial intelligence-based intelligent management method for biometrics is provided, comprising: in response to a user-initiated business request, collecting at least two types of biometric information from the user; inputting the biometric information into a pre-trained multimodal fusion model to generate a composite biometric code; wherein, the multimodal fusion model employs an attention mechanism to dynamically weight and fuse the contributions of different biometrics; specifically, the step of the multimodal fusion model employing an attention mechanism to dynamically weight and fuse the contributions of different biometrics includes:

[0046] For each type of biometric information input, feature extraction is performed to obtain the corresponding feature vector;

[0047] Each feature vector is input into an attention network, which outputs an attention weight for each feature vector. This weight represents the importance of the biometric feature in the current business scenario.

[0048] Based on the attention weights, the feature vectors are weighted and summed to generate the final composite biometric code.

[0049] Among them, dynamic risk factors are extracted based on the context information of the business request, the hardware fingerprint of the user terminal, and the current network behavior pattern. In this step, the dynamic risk factors include at least: the geolocation anomaly index, the terminal device change identifier anomaly index, and the deviation of the keystroke rhythm.

[0050] The geographical location anomaly index G is:

[0051] ;

[0052] In the formula, Indicates the current login location; Indicates the user's frequently used geographical location (e.g., the center of the N most recent login locations); This indicates the calculation of the geographical distance between two locations; This indicates the maximum acceptable distance set by the system; This indicates the number of times a user logs in from frequently used historical locations, used to smooth out outliers.

[0053] A geolocation anomaly index (G) of 0 indicates login from a frequently used location with no anomalies. A G value between 0 and 0.1 indicates activity within a reasonable geographical range, such as different areas within the same city or distances caused by slight GPS drift.

[0054] The terminal device change identifier anomaly index D is:

[0055] ;

[0056] In the formula, This indicates the current device's hardware fingerprint (such as device ID, C address, browser fingerprint, etc.). Represents the user's historical device fingerprint set; This represents the fingerprint similarity calculation function (such as Jaccard similarity or cosine similarity).

[0057] A terminal device change identifier anomaly index D of 0 indicates that a fully trusted historical device is being used. A terminal device change identifier anomaly index D of 0 to 0.1 indicates that the device fingerprint has undergone slight changes (such as browser upgrades or the installation of new plugins), but the core features are highly similar.

[0058] The deviation K of the keystroke rhythm is:

[0059] ;

[0060] In the formula, This represents the time interval between pressing the i-th key; This represents the average press interval of the i-th key in the user's history; The standard deviation of the historical key press interval is represented by n; the length of the keystroke sequence is represented by n.

[0061] A keystroke rhythm deviation K of 0 indicates that the keystroke rhythm perfectly matches the user's template. A keystroke rhythm deviation K of 0 to 0.2 indicates slight fluctuations, which are within the user's normal daily or state variation range.

[0062] Example 2:

[0063] like Figures 1-4 As shown, at least two dynamic risk factors are extracted from historical business requests in the database, and the historical risk level is quantified based on the actual negative consequences caused by each historical business request in subsequent business processes, in order to construct a risk assessment model; the dynamic risk factors of new business requests are input into the risk assessment model to calculate the real-time risk coefficient of the current business request, and the real-time risk level of the business request is divided according to the real-time risk coefficient.

[0064] Specifically, the quantification of historical risk levels is as follows:

[0065] (1) For historical business requests of the same type in the database, statistical analysis is conducted based on the actual negative impacts (such as financial losses, security incidents, operational disputes, etc.) generated in subsequent use or processing, and the degree of negative impact is quantified into a risk assessment value.

[0066] (2) Sort the historical business requests that participated in the assessment according to the quantified risk assessment value;

[0067] (3) Collect historical business requests with different risk levels and form a sample set.

[0068] For example, collect 1,000 historical business requests of the same type (sample set) from the database (such as expense reimbursement requests from enterprises), and calculate a quantitative risk value for each request based on the degree of negative impact such as financial anomalies and audit issues actually caused by these 1,000 historical business requests in subsequent processing.

[0069] The 1,000 historical business requests were sorted according to this quantified risk value (the quantified risk values ​​are different for each historical business request, that is, the degree of negative impact caused by each historical business request is quantified into a unique score).

[0070] If the quantified risk value of a certain business request is higher than that of the other 500 historical business requests, then the risk level of that business request can be quantified as 50.0%.

[0071] Furthermore, in this step, a risk assessment model consisting of two dynamic risk factors is constructed, which are represented as dynamic risk factor A and dynamic risk factor B, respectively. The specific steps are as follows:

[0072] (1) Filter historical business requests with the same variable B in any of the sample sets; obtain the dynamic risk factor A and its corresponding quantitative risk value in the filtered historical business requests, and use this to construct feature model A;

[0073] For example, data for 100 historical reimbursement requests is collected from 1000 historical requests. The value of the geographic location anomaly index G (dynamic risk factor B) in these 100 historical reimbursement requests is controlled as a variable (e.g., the variable of geographic location anomaly index G is controlled to 0.3). However, the terminal device change identifier anomaly index D (dynamic risk factor A) is not controlled. In addition, the quantitative risk value corresponding to these 100 historical reimbursement requests is also collected and recorded as the risk level A.

[0074] Then, the real-time risk coefficient output by the risk assessment model To establish a feature model A (the red dots in the figure represent the distribution of the 100 collected samples) regarding the relationship between the terminal device change identifier anomaly index D (dynamic risk factor A) and the terminal device change identifier anomaly index D, we have:

[0075] (Formula 1);

[0076] In Formula 1 above, , Used for control A constant that approximates the risk level A. And by... Figure 2 The data in the middle can be determined , When, in Formula 1 It approaches the level of risk A (where, express Figure 2 The curve in the image, with red dots representing... Figure 2 Risk level A of each historical reimbursement request.

[0077] (2) Filter the historical business requests with the same variable, dynamic risk factor A, in the sample set again; obtain dynamic risk factor B and its corresponding quantitative risk value in the filtered historical business requests, and use them to construct feature model B;

[0078] For example, data for 100 historical reimbursement requests is collected from 1000 historical requests. The value of the terminal device change identifier anomaly index D (dynamic risk factor A) in these 100 historical reimbursement requests is controlled as a variable (e.g., the variable of terminal device change identifier anomaly index D is controlled to 0.2). However, the geographical location anomaly index G (dynamic risk factor B) is not controlled. In addition, the quantitative risk value corresponding to these 100 historical reimbursement requests is also collected and recorded as the risk level B.

[0079] Then, the real-time risk coefficient output by the risk assessment model To establish a feature model B (the red dots in the figure represent the distribution of the 100 collected samples) regarding the relationship between the geographic location anomaly index G (dynamic risk factor B) and the index, we have:

[0080] (Formula 2);

[0081] In formula 2 above, , Used for control A constant that approximates the risk level B. And this can be determined from the data in the graph. , When, in Formula 2 It approaches the level of risk B (where, express Figure 3 The curve in the image, with blue dots representing... Figure 3 Risk level B of each historical reimbursement request.

[0082] (3) Obtain the corresponding dynamic risk factor A and dynamic risk factor B for historical business requests in the sample set according to the order of their quantitative risk values, and determine the relationship between feature model A and feature model B based on dynamic risk factor A and dynamic risk factor B, thereby constructing a risk assessment model.

[0083] For example, data for 100 historical reimbursement requests is collected from 1,000 historical reimbursement requests. In these 100 historical reimbursement requests, the terminal device change identifier anomaly index D (dynamic risk factor A) and the geographical location anomaly index G (dynamic risk factor B) are not controlled variables. In addition, the quantitative risk value corresponding to these 100 historical reimbursement requests is also collected and recorded as the risk level E.

[0084] To determine the relationship between Formula 1 and Formula 2 above, we need to examine the relationship between risk level E and the terminal equipment change identifier anomaly index D (dynamic risk factor A) and geographical location anomaly index G (dynamic risk factor B). Then we have:

[0085] ;

[0086] In the above, Used to control formulas A constant that approximates the risk level E. And by... Figure 4 The data in the middle can be determined When, formula It approaches the level of risk E.

[0087] Therefore, the mathematical expression for the risk assessment model obtained based on historical expense reimbursement requests is:

[0088] ;

[0089] In the above formula, C represents the risk coefficient output by the risk assessment model for the new business application; D represents the terminal equipment change identifier anomaly index of the new business application; and G represents the geographical location anomaly index of the new business application.

[0090] The risk coefficient for this newly entered similar business application is output based on the risk assessment model obtained from historical expense reimbursement requests. For example:

[0091] Table 1. Partial Implementation Data and Business Application Risks Assessed by Third-Party Assessment

[0092]

[0093] As shown in Table 1 above, a real-time risk coefficient C = 63.6% can be used as the standard for classifying high-risk business applications. For example, if the calculated real-time risk coefficient C > 63.6% for a subsequent new application of the same type, it can be considered a high-risk application, requiring multi-factor authentication, including at least composite biometric code, physical key device verification, and one-time dynamic password verification.

[0094] By fitting multiple indicators (geographic location anomaly index, terminal device change identifier anomaly index, and keystroke rhythm deviation) into a comprehensive risk coefficient, this essentially upgrades a simple rule-based system for "individual combat" to a multi-dimensional collaborative intelligent prediction system. By quantifying the complex relationships between indicators, it can more intelligently distinguish between real threats and normal anomalies, thereby significantly reducing false alarms and improving user experience without sacrificing security, while also enabling the system to continuously learn and optimize. This is a key advantage and innovation of this invention compared to existing technologies.

[0095] Example 3:

[0096] like Figure 1 As shown, based on the real-time risk level, the corresponding authentication method is dynamically selected from a pre-set verification strategy library. The verification strategy library includes at least the following levels: low-risk business requests are verified once based on a composite biometric code; medium-risk business requests are used to superimpose a secondary verification based on behavioral features from a temporal convolutional network on top of biometric verification; high-risk business requests trigger multi-factor authentication, which includes at least composite biometric code, physical key device verification, and one-time dynamic password verification. In this step, the secondary verification based on behavioral features from a temporal convolutional network specifically involves:

[0097] (1) Synchronously collect the keystroke timing data and mouse movement trajectory data of the user during the verification process to form a multimodal behavior sequence;

[0098] (2) Input the multimodal behavior sequence into a pre-trained temporal convolutional network and extract its temporal correlation features;

[0099] (3) Match the temporal correlation features with the user's historical behavior template and calculate the matching degree;

[0100] (4) If the matching degree is higher than the preset threshold, the secondary verification is passed.

[0101] Furthermore, once identity verification is successful, the verification event is uniquely bound to the business request, generating an authentication record containing a timestamp, business type, verification method, and risk level, which is then stored in a blockchain or tamper-proof database for auditing and traceability.

[0102] By employing a pre-built, dynamically linked verification strategy library that is tied to real-time risk levels, and combining this with secondary verification based on behavioral features from Temporal Convolutional Networks (TCNs), adaptive adjustment of authentication strength is achieved. The core principle lies in the fact that TCNs can effectively capture long-range temporal dependencies in behavioral sequences such as keystrokes and mouse movements, thereby extracting fine-grained behavioral biometric features that are difficult to mimic. The advantage of this design is that, based on biometric verification, it provides a non-invasive, highly discriminative second line of defense for medium-risk scenarios, achieving an optimal balance between security and user experience, and preventing the abuse of high-risk verification strategies.

[0103] Example 4:

[0104] like Figure 1 As shown, an AI-based biometric intelligent management method also collects successful and failed case data during the verification process, as well as corresponding contextual information; and uses the collected case data to perform incremental learning or periodic retraining of the multimodal fusion model and risk assessment model.

[0105] By introducing a closed-loop model optimization mechanism, the system systematically collects successful and failed cases and their contextual information during the verification process. This real-world feedback data is then used for incremental learning or periodic retraining of the multimodal fusion model and risk assessment model. The principle behind this is the construction of a data-driven closed loop from practical feedback to model optimization, enabling the system to continuously adapt to new attack patterns and changes in user behavior. A key advantage of this mechanism is that it endows the system with the ability to self-evolve and continuously optimize, effectively improving the model's accuracy and generalization ability, thereby ensuring high reliability and low maintenance costs during long-term system operation.

[0106] Example 5:

[0107] like Figure 1 As shown, an artificial intelligence-based biometric intelligent management system includes: a biometric acquisition module, a multimodal fusion processing module, a dynamic risk factor extraction module, a risk assessment model construction module, a real-time risk assessment module, a verification strategy execution module, an authentication record and storage module, and a model update module.

[0108] The biometric data acquisition module is used to collect at least two types of biometric information from users in response to business requests initiated by users.

[0109] The multimodal fusion processing module is used to input biometric information into a pre-trained multimodal fusion model to generate a composite biometric code. The multimodal fusion model uses an attention mechanism to dynamically weight and fuse the contributions of different biometric features.

[0110] The dynamic risk factor extraction module is used to extract dynamic risk factors based on the context information of the business request, the hardware fingerprint of the user terminal, and the current network behavior pattern.

[0111] The risk assessment model building module is used to extract at least two dynamic risk factors based on historical business requests in the database and combine them with the risk level assessed by a third party to build a risk assessment model.

[0112] The real-time risk assessment module is used to input the dynamic risk factors of the current business request into the risk assessment model, calculate the real-time risk coefficient, and classify the real-time risk level.

[0113] The verification strategy execution module is used to dynamically select the corresponding authentication method from the pre-set verification strategy library based on the real-time risk level and execute the verification process.

[0114] The authentication record and storage module is used to uniquely bind the authentication event to the business request after the authentication is successful, generate an authentication record containing a timestamp, business type, authentication method and risk level, and store it in a blockchain or tamper-proof database.

[0115] The model update module is used to collect data on successful and failed cases and contextual information during the validation process, and to perform incremental learning or periodic retraining on the multimodal fusion model and risk assessment model.

[0116] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An artificial intelligence-based biometric intelligent management method, characterized in that, The method comprises the following steps: S1, in response to a user-initiated service request, collecting at least two kinds of biological characteristic information of the user; inputting the biological characteristic information into a pre-trained multi-modal fusion model to generate a composite biological characteristic code; wherein the multi-modal fusion model dynamically weights and fuses the contribution degrees of different biological characteristics using an attention mechanism; S2, based on the context information of the service request, the hardware fingerprint of the user terminal and the current network behavior mode, extracting dynamic risk factors; the dynamic risk factors at least include: geographical location abnormality index, terminal device change identification abnormality index and keystroke rhythm deviation degree; S3, based on the historical service requests and the corresponding dynamic risk factors in the database, and the negative consequences actually caused by each historical service request in the subsequent service process, evaluating and quantifying the historical risk degree; using the quantified historical risk degree and the corresponding dynamic risk factor, a risk assessment model is constructed through regression analysis for predicting the risk of new service requests; S4, inputting the dynamic risk factors of the current service request into the risk assessment model to calculate the real-time risk coefficient of the current service request, and dividing the service request into real-time risk levels according to the real-time risk coefficient; S5, according to the real-time risk level, dynamically selecting the corresponding identity verification mode from the pre-set verification strategy library; the verification strategy library at least includes the following levels: The service request of low risk level is verified based on the composite biological characteristic code for single verification; The service request of medium risk level is used for secondary verification of behavior characteristics based on time convolution network on the basis of biological characteristic verification; The service request of high risk level triggers multi-factor authentication, which at least includes composite biological characteristic code, physical key device verification and one-time dynamic password verification; S6, when the identity verification is passed, the current verification event is uniquely bound with the service request to generate an authentication record containing timestamp, service type, verification mode and risk level, and stored in the blockchain or tamper-proof database for auditing and tracing. 2.The AI-based biometric intelligent management method of claim 1, wherein: In step S1, the multi-modal fusion model dynamically weights and fuses the contribution degrees of different biological characteristics using an attention mechanism, which specifically includes: Each biological characteristic information is inputted for feature extraction to obtain the corresponding feature vector; Each feature vector is inputted into an attention network, which outputs the attention weight of each feature vector, which represents the importance of the biological characteristic in the current service scenario; According to the attention weight, the feature vectors are weighted and summed to generate the final composite biological characteristic code. 3.The AI-based biometric intelligent management method of claim 1, wherein: In step S3, a risk assessment model is constructed based on two dynamic risk factors, which are represented as dynamic risk factor A and dynamic risk factor B, and the specific steps are as follows: S301, filtering the historical service requests with the same variable of dynamic risk factor B in the sample set; obtaining the dynamic risk factor A and the corresponding risk degree in the filtered historical service request, and constructing feature model A based on the same. S302, screening the same historical business request of the dynamic risk factor A variable in the sample set again; obtaining the dynamic risk factor B and the corresponding risk degree in the historical business request after screening, and constructing a feature model B based on the same; S303, obtaining the corresponding dynamic risk factor A and dynamic risk factor B of the historical business request in the sample set according to the size order of the risk degree, and determining the relationship between the feature model A and the feature model B according to the dynamic risk factor A and the dynamic risk factor B, and then constructing a risk assessment model. 4.The method of claim 1, wherein: The risk assessment model is a prediction model constructed based on the multiple dynamic risk factors extracted from the historical business request and the corresponding risk degree evaluated by the third party, by analyzing the internal correlation between the factors and the risk degree, for real-time calculation of the risk coefficient of the current business request and division of the risk level according to the same. 5.The method of claim 1, wherein: In step S5, the behavior feature secondary verification based on the time convolution network is specifically: Synchronize the collection of keyboard typing time sequence data and mouse moving trajectory data of the user in the verification process to form a multi-modal behavior sequence; Input the multi-modal behavior sequence into the pre-trained time convolution network to extract its time sequence correlation feature; Match the time sequence correlation feature with the user historical behavior template to calculate the matching degree; If the matching degree is higher than the preset threshold, the secondary verification is passed. 6.The method of claim 1, wherein: The biological feature intelligent management method based on artificial intelligence also collects case data of success and failure in the verification process, and corresponding context information; The collected case data is used for incremental learning or periodic retraining of the multi-modal fusion model and the risk assessment model.

7. An artificial intelligence-based biometric intelligent management system, characterized by, The biological feature intelligent management system based on artificial intelligence is applied to the biological feature intelligent management method based on artificial intelligence in any one of claims 1-6, and comprises a biological feature collection module, a multi-modal fusion processing module, a dynamic risk factor extraction module, a risk assessment model construction module, a real-time risk assessment module, a verification strategy execution module, and an authentication record and storage module. 8.The AI-based biometric intelligent management system according to claim 7, characterized in that, Further comprising: A model updating module is configured to collect case data of success and failure in the verification process and context information, and perform incremental learning or periodic retraining on the multi-modal fusion model and the risk assessment model.

Images