Intelligent security authentication system based on multi-modal large model

Abstract

The invention discloses an intelligent security authentication system based on a multi-modal large model. Comprising a multi-modal data acquisition module, a data preprocessing and synchronization module, a multi-modal feature extraction and fusion module, a large model dynamic judgment engine, a self-adaptive security authentication strategy module, a privacy protection module, a security storage and transmission module and a real-time feedback learning and model fine tuning module. The multi-modal data acquisition module is used for acquiring multi-modal data of a user; the data preprocessing and synchronizing module is used for denoising and normalizing the acquired multi-modal data and aligning the data through a synchronizing signal; the multi-modal feature extraction and fusion module is used for extracting features from different modal data by using a deep learning algorithm and generating unified multi-modal feature representation through a fusion network; the large model dynamic judgment engine is used for processing historical authentication data according to the fused feature data; the method has the characteristics of high efficiency and safety.

Description

Technical Field

[0001] This invention relates to the field of information security authentication technology, specifically to an intelligent security authentication system based on a multimodal large model. Background Technology

[0002] In the digital age, information security has become a critical issue. Traditional security authentication methods, such as single passwords or biometric authentication, can no longer meet the current dual demands for security and convenience. These methods struggle to balance security and convenience and often lack privacy protection measures when processing multimodal data. Furthermore, traditional systems have limited capabilities in detecting abnormal behavior, relying heavily on human intervention, lacking objectivity, and being heavily dependent on specific scenarios. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent security authentication system based on a multimodal large model to solve the problems mentioned in the background art.

[0004] To address the aforementioned technical problems, this invention provides the following technical solution: an intelligent security authentication system based on a multimodal large model, comprising a multimodal data acquisition module, a data preprocessing and synchronization module, a multimodal feature extraction and fusion module, a large model dynamic judgment engine, an adaptive security authentication strategy module, a privacy protection module, a secure storage and transmission module, and a real-time feedback learning and model fine-tuning module; wherein,

[0005] The multimodal data acquisition module is used to collect the user's multimodal data;

[0006] The data preprocessing and synchronization module is used to denoise and normalize the collected multimodal data, and to align the data through a synchronization signal.

[0007] The multimodal feature extraction and fusion module is used to extract features from different modal data using deep learning algorithms, and generate a unified multimodal feature representation through a fusion network;

[0008] The large model dynamic judgment engine is used to process historical authentication data based on the fused feature data;

[0009] The adaptive security authentication strategy module is used to dynamically adjust the authentication strictness based on the results of historical authentication data processing;

[0010] The privacy protection module is used to protect user privacy;

[0011] The secure storage and transmission module is used to protect authentication data through encryption technology during storage and transmission;

[0012] The real-time feedback learning and model fine-tuning module is used to update and optimize the authentication model in real time based on user feedback and behavioral data.

[0013] According to the above technical solution, the intelligent security authentication system includes the following steps in the real-time security authentication process:

[0014] Step S1: Collect the user's voice, facial expressions, eye movements, gestures, and tactile feedback data through the multimodal data acquisition module;

[0015] Step S2: The collected multimodal data is denoised and normalized by the data preprocessing and synchronization module, and the data is time-aligned by the synchronization signal.

[0016] Step S3: Extract features from each modality of data through the multimodal feature extraction and fusion module, and fuse them to generate a unified multimodal feature representation;

[0017] Step S4: Process the fused feature data through the large model dynamic judgment engine, and dynamically adjust the authentication strategy by combining historical authentication data and external environment information;

[0018] Step S5: Anonymize, encrypt, store, and transmit user data through the privacy protection module, clean up temporary storage data according to the principle of data minimization, and provide user privacy protection options to ensure that user data is safe and reliable throughout the authentication process;

[0019] Step S6: Through the real-time feedback learning and model fine-tuning module, the model is fine-tuned based on user feedback and behavioral data to optimize the authentication strategy.

[0020] According to the above technical solution, the specific method for extracting features from each modality data and fusing them to generate a unified multimodal feature representation in step S3 is as follows:

[0021] Pre-trained CNN and RNN neural networks are used to extract features from data of different modalities. CNN is suitable for processing image and video data, while RNN is suitable for speech and sequence data. Features are extracted for each modality M. i ;

[0022] Set the input data to X i Then the feature extraction calculation expression for the i-th mode is F. i =Net i (X i ), where Net i Let X represent the feature extraction network for the i-th modality. i It is the raw input data, including images, voice signals, or sensor data, F i It is the extracted feature vector;

[0023] Obtain all extracted feature vectors F i Then, through the weighted fusion formula Perform weighted fusion, where n is the number of modes and α i F represents the weights of the i-th modality, which are obtained through adaptive learning during model training. i F is the feature vector extracted from the i-th modality. a This is the weighted fusion result value of all modal features;

[0024] For the fused feature vector F a The final unified representation is obtained through processing using a Transformer network.

[0025] Based on the above technical solution, the weight α in the weighted fusion formula... i The backpropagation algorithm is used to optimize the training process, ensuring that the contribution of each modality to the final authentication result is the optimal value.

[0026] According to the above technical solution, adaptive threshold judgment is applied to the multimodal feature representation after feature fusion.

[0027] According to the above technical solution, step S4 specifically includes:

[0028] Step S41: The weighted fused multimodal feature vector F output from step S3 a The current historical authentication data is used as input, including: past user behavior characteristics, historical authentication results, and security threat level L. w ;

[0029] Step S42: Place F a The input is fed into the credibility calculation model, and combined with historical authentication data, an unnormalized credibility value S is generated. k ;

[0030] Step S43: Apply the Sigmoid activation function to the confidence value S k After normalization, its calculation expression is: Where S k P is the original confidence value calculated by the system. k The final authentication credibility is determined by a value ranging from 0 to 1, with values ​​closer to 1 indicating high user credibility.

[0031] Step S44: Based on external sensor data and historical authentication failure data, dynamically generate the current environment's security threat level L. w Among them, the security threat level L w The calculation expression is: L w=β·L h +γ·L x In the formula L h Threat levels generated based on the external environment, L x The threat level is generated based on the degree of abnormality in user behavior. β and γ are both weight parameters, dynamically generated according to the environmental risk assessment model. β reflects the influence weight of external environmental threats, and γ reflects the influence weight of abnormal user behavior, ensuring that the threat level calculated in different scenarios is targeted and used to balance the impact of environmental threats and behavioral threats.

[0032] Step S45: Place P k and L w The input is fed into the dynamic policy generation model, and processed through formula S. c =S j +λ·(1-P k )+μ·L w Calculate the authentication strictness S c In the formula S j The system's default authentication policy is defined by λ and μ, which are adjustment coefficients that control the impact of decreased trust level and increased threat level on policy strictness, respectively. c Indicates the rigor of the final certification;

[0033] Step S46: When S c If the value exceeds the set threshold, multi-factor authentication will be enabled; otherwise, Dangdang S c If the value is below the set threshold, the authentication process will be simplified, retaining only the basic verification steps.

[0034] According to the above technical solution, step S5 specifically includes:

[0035] The privacy protection module anonymizes users' multimodal data, removes sensitive information, and ensures that the data cannot be directly associated with specific users.

[0036] Encryption algorithms are used to protect data during storage and transmission, ensuring data confidentiality and integrity;

[0037] After data processing is complete, temporary storage data is cleaned up according to the principle of data minimization, and only necessary historical data is retained for model optimization.

[0038] Provide users with privacy protection options so that they can understand and adjust the privacy protection policies during the authentication process.

[0039] According to the above technical solution, step S6 specifically includes:

[0040] Step S61: Collect user feedback data, including user satisfaction with the authentication process, feedback on the accuracy of the authentication results, and user behavior data;

[0041] Step S62: Preprocess the collected feedback data and behavioral data;

[0042] Step S63: Analyze user feedback and behavioral data using machine learning algorithms;

[0043] Step S64: Based on the analysis results, fine-tune the authentication model;

[0044] Step S65: Deploy the fine-tuned model to the system and continuously monitor the system's performance;

[0045] Step S66: Establish a mechanism for regular iterative updates to continuously optimize the authentication model based on new user feedback and behavioral data.

[0046] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: This invention effectively resolves the contradiction between security and convenience in traditional authentication systems through multimodal behavioral feature extraction, adaptive authentication strategies, privacy-preserving data processing, and abnormal behavior detection, while simultaneously protecting user privacy. Furthermore, the above technologies collectively constitute an efficient, secure, and user-friendly intelligent security authentication system with broad application prospects. Attached Figure Description

[0047] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0048] In the attached diagram:

[0049] Figure 1 This is a schematic diagram of the system module composition of the present invention;

[0050] Figure 2 This is a flowchart of the system operation method of the present invention. Detailed Implementation

[0051] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] Please see Figure 1This invention provides a technical solution: an intelligent security authentication system based on a multimodal large model, comprising a multimodal data acquisition module, a data preprocessing and synchronization module, a multimodal feature extraction and fusion module, a large model dynamic judgment engine, an adaptive security authentication strategy module, a privacy protection module, a secure storage and transmission module, and a real-time feedback learning and model fine-tuning module; wherein,

[0053] The multimodal data acquisition module is used to collect the user's multimodal data, including the user's voice, facial expressions, eye movements, gestures, tactile feedback, etc.

[0054] The data preprocessing and synchronization module is used to denoise and normalize the collected multimodal data, and to align the data through a synchronization signal to ensure the consistency of each modality of data, providing an accurate data foundation for subsequent feature extraction and fusion.

[0055] The multimodal feature extraction and fusion module is used to extract features from different modal data using deep learning algorithms and generate a unified multimodal feature representation through a fusion network;

[0056] The large model dynamic judgment engine is used to process historical authentication data based on the fused feature data;

[0057] The adaptive security authentication strategy module is used to dynamically adjust the authentication strictness based on the results of historical authentication data processing;

[0058] The privacy protection module is used to protect user privacy;

[0059] The secure storage and transmission module is designed to protect authenticated data using encryption technology during storage and transmission.

[0060] The real-time feedback learning and model fine-tuning module is used to update and optimize the authentication model in real time based on user feedback and behavioral data.

[0061] Please see Figure 2 The intelligent security authentication system includes the following steps in the real-time security authentication process:

[0062] Step S1: Collect user's voice, facial expressions, eye movements, gestures, and tactile feedback data through the multimodal data acquisition module; for example, the system collects facial expression and eye movement data through a high-definition camera, voice data through a microphone, gesture data through a tactile sensor, and location and motion data through GPS and IMU sensors;

[0063] Step S2: The collected multimodal data is denoised and normalized by the data preprocessing and synchronization module, and the data is time-aligned by the synchronization signal.

[0064] Step S3: Extract features from each modality of data through the multimodal feature extraction and fusion module, and fuse them to generate a unified multimodal feature representation;

[0065] Step S4: Process the fused feature data through the large model dynamic judgment engine, and dynamically adjust the authentication strategy by combining historical authentication data and external environment information;

[0066] Step S5: Anonymize, encrypt, store, and transmit user data through the privacy protection module, clean up temporary storage data according to the principle of data minimization, and provide user privacy protection options to ensure that user data is safe and reliable throughout the authentication process;

[0067] Step S6: Through the real-time feedback learning and model fine-tuning module, the model is fine-tuned based on user feedback and behavioral data to optimize the authentication strategy.

[0068] In step S3, the specific method for extracting features from each modality and fusing them to generate a unified multimodal feature representation is as follows:

[0069] Pre-trained CNN and RNN neural networks are used to extract features from data of different modalities. CNN is suitable for processing image and video data, while RNN is suitable for speech and sequence data. Features are extracted for each modality M. i ;

[0070] Set the input data to X i Then the feature extraction calculation expression for the i-th mode is F. i =Net i (X i ), where Net i Let X represent the feature extraction network for the i-th modality. i It is the raw input data, including images, voice signals, or sensor data, F i It is the extracted feature vector;

[0071] Obtain all extracted feature vectors F i Then, through the weighted fusion formula Perform weighted fusion, where n is the number of modes and α i F represents the weights of the i-th modality, which are obtained through adaptive learning during model training. i F is the feature vector extracted from the i-th modality. a This is the weighted fusion result value of all modal features;

[0072] For the fused feature vector F aThe Transformer network processes the data to obtain the final unified representation, which will be used for subsequent identity authentication decisions and authentication policy adjustments.

[0073] The weight α in the weighted fusion formula i The backpropagation algorithm is used to optimize the system during training, ensuring that the contribution of each modality to the final authentication result is optimal, thereby improving the accuracy and robustness of the multimodal authentication system.

[0074] An adaptive threshold judgment is applied to the multimodal feature representation after feature fusion, so as to adjust the authentication strategy according to the actual authentication scenario and further improve the authentication accuracy and system response speed.

[0075] Step S4 specifically includes:

[0076] Step S41: The weighted fused multimodal feature vector F output from step S3 a The current historical authentication data is used as input, including: past user behavior characteristics, historical authentication results, and security threat level L. w ;

[0077] Step S42: Place F a The input is fed into the credibility calculation model, and combined with historical authentication data, an unnormalized credibility value S is generated. k ;

[0078] Step S43: Apply the Sigmoid activation function to the confidence value S k After normalization, its calculation expression is: Where S k P is the original confidence value calculated by the system. k The final authentication credibility is determined by a value ranging from 0 to 1, with values ​​closer to 1 indicating high user credibility.

[0079] Step S44: Based on external sensor data such as device status, network environment, time and location, and combined with historical authentication failure data, dynamically generate the current security threat level L. w Among them, the security threat level L w The calculation expression is: L w =β·L h +γ·L x In the formula L h Threat levels generated based on the external environment, L xThe threat level, β, and γ are weighted parameters generated based on user behavior such as input patterns and the degree of abnormality in device operation. They are dynamically generated according to the environmental risk assessment model. β reflects the influence weight of external environmental threats, and γ reflects the influence weight of abnormal user behavior. This ensures that the threat level calculated in different scenarios is targeted and is used to balance the influence of environmental threats and behavioral threats.

[0080] Step S45: Place P k and L w The input is fed into the dynamic policy generation model, and processed through formula S. c =S j +λ·(1-P k )+μ·L w Calculate the authentication strictness S c In the formula S j The system's default authentication strategy, such as multimodal authentication or single-factor authentication, is used. λ and μ are adjustment coefficients that control the impact of decreased credibility and increased threat level on the policy's strictness, respectively. c This indicates the rigor of the final certification, such as adding or removing certification steps;

[0081] Step S46: When S c If the value exceeds the set threshold, multi-factor authentication will be enabled; otherwise, Dangdang S c If the threshold is lower than the set threshold, the authentication process is simplified, and only the basic verification steps are retained; multi-factor authentication includes adding biometric detection and strengthening user input verification.

[0082] Step S5 specifically includes:

[0083] The privacy protection module anonymizes users' multimodal data, removes sensitive information, and ensures that the data cannot be directly associated with specific users.

[0084] Encryption algorithms are used to protect data during storage and transmission, ensuring data confidentiality and integrity;

[0085] After data processing is complete, temporary storage data is cleaned up according to the principle of data minimization, and only necessary historical data is retained for model optimization.

[0086] Provide users with privacy protection options so that they can understand and adjust the privacy protection policies during the authentication process.

[0087] Step S6 specifically includes:

[0088] Step S61: Collect user feedback data, including user satisfaction with the authentication process, feedback on the accuracy of the authentication results, and user behavior data;

[0089] Step S62: Preprocess the collected feedback data and behavioral data;

[0090] Step S63: Analyze user feedback and behavioral data using machine learning algorithms;

[0091] Step S64: Based on the analysis results, fine-tune the authentication model, including adjusting the parameters of the feature extraction network, optimizing the fusion strategy, and adjusting the weights of the credibility calculation model, in order to improve the accuracy and user-friendliness of authentication.

[0092] Step S65: Deploy the fine-tuned model into the system and continuously monitor the system's performance, including indicators such as authentication accuracy and user satisfaction, in order to identify and resolve problems in a timely manner.

[0093] Step S66: Establish a mechanism for regular iterative updates, continuously optimize the authentication model based on new user feedback and behavioral data, and ensure that the system's security and user experience are constantly improved;

[0094] This system collects multimodal data from users, including voice, facial expressions, and gestures, and uses deep learning algorithms for feature extraction and fusion to generate a unified feature representation, significantly improving the accuracy and robustness of authentication. Based on dynamic assessment of user trustworthiness and environmental threat levels, combined with real-time adjusted authentication strategies, the system can provide more stringent multi-factor authentication in high-risk scenarios while simplifying the authentication process in low-risk scenarios, balancing security and convenience and optimizing user experience. Furthermore, a robust privacy protection module is designed, employing anonymization, encrypted storage and transmission, and dynamic masking techniques to comprehensively ensure user data security and prevent the leakage of sensitive information. Combined with access control and auditing functions, this further enhances user trust in the system. Leveraging real-time feedback learning and model fine-tuning technology, the system can dynamically optimize authentication strategies, ensuring consistently efficient and reliable operation across diverse application scenarios.

[0095] This application has broad application prospects, as exemplified by...

[0096] In intelligent vehicles, this system can provide personalized security authentication services based on driver behavior and environmental changes. For example, when the driver is driving on a familiar route, the system may reduce authentication steps, while increasing authentication steps when the driver deviates from the usual route. Intelligent vehicle unlocking: In intelligent vehicles, this system can provide personalized security authentication services based on driver behavior and environmental changes. For example, when the driver approaches the vehicle, the system automatically unlocks the doors by analyzing the driver's facial expressions, gestures, and location information. If the system detects abnormal behavior or security threats, it will require additional authentication steps, such as entering a PIN code or using biometric technology for secondary authentication.

[0097] Before application: On a cold winter day, Zhang San, wearing a scarf and hat, tried to unlock his car using a facial recognition system. Because his facial features were partially obscured, the system could not accurately recognize him, forcing Zhang San to adjust his scarf and hat multiple times and even expose himself to the cold air before he could unlock the car.

[0098] After application: By adopting the large-model dynamic judgment engine, even if Zhang San is wearing a scarf and hat, the system can accurately determine his identity and quickly unlock the vehicle by fusing multimodal data such as facial features, body movements, and surrounding environmental information. This fully utilizes the system's ability to learn and adapt to different occlusion conditions, improving the flexibility and accuracy of recognition.

[0099] In financial services, this system can provide a convenient user authentication experience while ensuring transaction security. For example, the system may request additional authentication information when it detects an unusual login attempt.

[0100] Before application: Li Si needed to make a large mobile banking transfer, but due to the lack of intelligent judgment in the system, she had to go through a series of complicated authentication steps, including entering a password, answering security questions and performing biometric identification. The whole process was time-consuming and cumbersome.

[0101] After application: After the adaptive security authentication strategy module was implemented, when Li Si attempted a large transfer, the system intelligently assessed the risk level of the operation based on her historical behavior patterns, current device information, and usage environment. Since the system determined the risk to be low, the authentication process was simplified, requiring only a quick fingerprint verification for Li Si to complete the transfer. The system learned Li Si's typical behavior and was able to simplify the operation steps for trusted users while ensuring security.

[0102] In smart home systems, intelligent security authentication based on user behavior and habits can be provided. For example, the system may trigger a security alarm when it detects unusual patterns of home activity.

[0103] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0104] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent security authentication system based on a multimodal large model, characterized in that: It includes a multimodal data acquisition module, a data preprocessing and synchronization module, a multimodal feature extraction and fusion module, a large-model dynamic judgment engine, an adaptive security authentication strategy module, a privacy protection module, a secure storage and transmission module, and a real-time feedback learning and model fine-tuning module; among which, The multimodal data acquisition module is used to collect the user's multimodal data; The data preprocessing and synchronization module is used to denoise and normalize the collected multimodal data, and to align the data through a synchronization signal. The multimodal feature extraction and fusion module is used to extract features from different modal data using deep learning algorithms, and generate a unified multimodal feature representation through a fusion network; The large model dynamic judgment engine is used to process historical authentication data based on the fused feature data; The adaptive security authentication strategy module is used to dynamically adjust the authentication strictness based on the results of historical authentication data processing; The privacy protection module is used to protect user privacy; The secure storage and transmission module is used to protect authentication data through encryption technology during storage and transmission; The real-time feedback learning and model fine-tuning module is used to update and optimize the authentication model in real time based on user feedback and behavioral data.

2. The intelligent security authentication system based on a multimodal large model according to claim 1, characterized in that: The intelligent security authentication system includes the following steps in the real-time security authentication process: Step S1: Collect the user's voice, facial expressions, eye movements, gestures, and tactile feedback data through the multimodal data acquisition module; Step S2: The collected multimodal data is denoised and normalized by the data preprocessing and synchronization module, and the data is time-aligned by the synchronization signal. Step S3: Extract features from each modality of data through the multimodal feature extraction and fusion module, and fuse them to generate a unified multimodal feature representation; Step S4: Process the fused feature data through the large model dynamic judgment engine, and dynamically adjust the authentication strategy by combining historical authentication data and external environment information; Step S5: Anonymize, encrypt, store, and transmit user data through the privacy protection module, clean up temporary storage data according to the principle of data minimization, and provide user privacy protection options to ensure that user data is safe and reliable throughout the authentication process; Step S6: Through the real-time feedback learning and model fine-tuning module, the model is fine-tuned based on user feedback and behavioral data to optimize the authentication strategy.

3. The intelligent security authentication system based on a multimodal large model according to claim 2, characterized in that: In step S3, the specific method for extracting features from each modality and fusing them to generate a unified multimodal feature representation is as follows: Pre-trained CNN and RNN neural networks are used to extract features from data of different modalities. CNN is suitable for processing image and video data, while RNN is suitable for speech and sequence data. Features are extracted for each modality M. i ; Set the input data to X i Then the feature extraction calculation expression for the i-th mode is F. i =Net i (X i ), where Net i Let X represent the feature extraction network for the i-th modality. i It is the raw input data, including images, voice signals, or sensor data, F i It is the extracted feature vector; Obtain all extracted feature vectors F i Then, through the weighted fusion formula Perform weighted fusion, where n is the number of modes and α i F represents the weights of the i-th modality, which are obtained through adaptive learning during model training. i F is the feature vector extracted from the i-th modality. a This is the weighted fusion result value of all modal features; For the fused feature vector F a The final unified representation is obtained through processing using a Transformer network.

4. The intelligent security authentication system based on a multimodal large model according to claim 3, characterized in that: The weight α in the weighted fusion formula i The backpropagation algorithm is used to optimize the training process, ensuring that the contribution of each modality to the final authentication result is the optimal value.

5. The intelligent security authentication system based on a multimodal large model according to claim 3, characterized in that: An adaptive threshold judgment is applied to the multimodal feature representation after feature fusion.

6. The intelligent security authentication system based on a multimodal large model according to claim 2, characterized in that: Step S4 specifically includes: Step S41: The weighted fused multimodal feature vector F output from step S3 a The current historical authentication data is used as input, including: past user behavior characteristics, historical authentication results, and security threat level L. w ; Step S42: Place F a The input is fed into the credibility calculation model, and combined with historical authentication data, an unnormalized credibility value S is generated. k ; Step S43: Apply the Sigmoid activation function to the confidence value S k After normalization, its calculation expression is: Where S k P is the original confidence value calculated by the system. k The final authentication credibility is determined by a value ranging from 0 to 1, with values ​​closer to 1 indicating high user credibility. Step S44: Based on external sensor data and historical authentication failure data, dynamically generate the current environment's security threat level L. w Among them, the security threat level L w The calculation expression is: L w =β·L h +γ·L x In the formula L h Threat levels generated based on the external environment, L x The threat level is generated based on the degree of abnormality in user behavior. β and γ are both weight parameters, dynamically generated according to the environmental risk assessment model. β reflects the influence weight of external environmental threats, and γ reflects the influence weight of abnormal user behavior, ensuring that the threat level calculated in different scenarios is targeted and used to balance the impact of environmental threats and behavioral threats. Step S45: Place P k and L w The input is fed into the dynamic policy generation model, and processed through formula S. c =S j +λ·(1-P k )+μ·L w Calculate the authentication strictness S c In the formula S j This is the system's default authentication policy. λ and μ are adjustment coefficients, controlling the impact of decreased trust level and increased threat level on policy strictness, respectively. c Indicates the rigor of the final certification; Step S46: When S c If the value exceeds the set threshold, multi-factor authentication will be enabled; otherwise, Dangdang S c If the value is below the set threshold, the authentication process will be simplified, retaining only the basic verification steps.

7. The intelligent security authentication system based on a multimodal large model according to claim 2, characterized in that: Step S5 specifically includes: The privacy protection module anonymizes users' multimodal data and removes sensitive information. Encryption algorithms are used to protect data during storage and transmission; After data processing is complete, clean up temporary storage data according to the principle of data minimization; Offer users privacy protection options and adjust privacy protection policies during the authentication process.

8. The intelligent security authentication system based on a multimodal large model according to claim 2, characterized in that: Step S6 specifically includes: Step S61: Collect user feedback data, including user satisfaction with the authentication process, feedback on the accuracy of the authentication results, and user behavior data; Step S62: Preprocess the collected feedback data and behavioral data; Step S63: Analyze user feedback and behavioral data using machine learning algorithms; Step S64: Based on the analysis results, fine-tune the authentication model; Step S65: Deploy the fine-tuned model to the system and continuously monitor the system's performance; Step S66: Establish a mechanism for regular iterative updates to continuously optimize the authentication model based on new user feedback and behavioral data.

Images