Login method and system based on 5G message voiceprint verification code

Abstract

The invention provides a login method and system based on a 5G message voiceprint verification code, and the method comprises the steps: generating a 5G message, receiving a message receiving confirmation feedback of a user, obtaining a composite audio stream through combining with environment sound collected based on a local voiceprint collection device, carrying out the processing through a blind source separation algorithm, obtaining a preliminary composite voiceprint sample, and carrying out the verification of the voiceprint verification code. A target composite voiceprint sample is obtained after optimization; performing encryption processing based on a preset encryption protocol, obtaining a decrypted target composite voiceprint sample in combination with a quantum key distribution mechanism and a cloud authentication platform, and calculating a preliminarily adjusted comprehensive score by using a convolutional neural network, a recurrent neural network and a generative adversarial network combination model; performing processing by utilizing a behavior analysis algorithm to obtain a final adjusted comprehensive score, and performing access permission confirmation to obtain an access confirmation result; according to the invention, the security and accuracy of user identity verification are enhanced, efficient user experience is provided, and the security and reliability of the user login process are ensured.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of communication technology, and in particular, to a login method and system based on a 5G message voiceprint verification code. Background Art

[0002] With the popularization of 5G networks and the widespread application of IoT devices, the login method based on 5G messages has gradually become an important means of identity authentication. Especially in the fields of finance, medical care, and corporate office with high security requirements, the traditional login method of username and password can no longer meet the growing security needs. Users need a more secure, convenient and efficient login method to ensure the security of access to sensitive information and systems.

[0003] At present, common identity authentication schemes mainly include static passwords and SMS verification codes. The existing static password method is vulnerable to attacks and lacks security. The existing SMS verification code method uses a one-time verification code to send a one-time verification code to the user's mobile phone, and the user enters the verification code to complete the login. This method has the risk of SMS being intercepted and relies on the speed and reliability of traditional SMS channels. Summary of the invention

[0004] The embodiment of the present invention provides a login method and system based on a 5G message voiceprint verification code, which is used to solve the problems in the prior art that it is vulnerable to attacks, has insufficient security, has the risk of being intercepted, and relies on traditional SMS channels.

[0005] In a first aspect, an embodiment of the present invention provides a login method based on a 5G message voiceprint verification code, comprising: Generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; Based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, a composite audio stream is obtained, and the composite audio stream is separated and processed by a blind source separation algorithm to extract the original ambient sound and the one-time random sequence code, and the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, and the preliminary composite voiceprint sample is optimized to obtain a target composite voiceprint sample; The target composite voiceprint sample is encrypted based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is transmitted to a cloud authentication platform in combination with a quantum key distribution mechanism, and the decrypted target composite voiceprint sample is obtained after being received and decrypted by the cloud authentication platform, and the decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features; Based on historical data of user behavior patterns, a behavior analysis algorithm is used to adjust the score weight of the preliminary adjusted comprehensive score to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result.

[0006] Optionally, the target composite voiceprint sample is encrypted based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is transmitted to a cloud authentication platform in combination with a quantum key distribution mechanism, and the decrypted target composite voiceprint sample is obtained after being received and decrypted by the cloud authentication platform, and the decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features, including: Encrypting the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and using homomorphic encryption technology in the encryption process to perform a specific type of calculation on the encrypted target composite voiceprint sample, the specific type including additive homomorphic, multiplicative homomorphic and fully homomorphic encryption; The encrypted target composite voiceprint sample is transmitted from the user terminal device to the cloud authentication platform using the quantum key distribution mechanism and the post-quantum cryptography algorithm. During the transmission process, the quantum key distribution generation mechanism is used to add encryption keys to each sample in the encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is processed in combination with the post-quantum cryptography algorithm. Based on the cloud authentication platform, the received encrypted target composite voiceprint sample is decrypted, and the decryption process is verified by zero-knowledge proof technology. After the verification process is completed, the decrypted target composite voiceprint sample is obtained; The decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework. During the analysis process, time-frequency domain features are extracted by a convolutional neural network, time series characteristics are extracted by a recurrent neural network, and a synthetic voiceprint sample is generated by a generative adversarial network to obtain an enhanced multimodal voiceprint feature set; Combining the multimodal voiceprint feature set and the user's biological features, constructing a user identity model, performing multimodal data analysis based on the user identity model to obtain multimodal user features, and using a graph neural network to analyze user behavior based on the multimodal user features and historical data of user behavior patterns to obtain user behavior pattern features, wherein the biological features include facial features and fingerprint features; In combination with the user behavior pattern characteristics and the multimodal user characteristics, an initial comprehensive score is calculated, the user identity model is optimized to obtain a comprehensive user identity model, and based on the comprehensive user identity model, the score weight of the initial comprehensive score is dynamically adjusted to obtain a preliminary adjusted comprehensive score.

[0007] Optionally, combining the user behavior pattern characteristics and the multimodal user characteristics, calculating an initial comprehensive score, optimizing the user identity model to obtain a comprehensive user identity model, and dynamically adjusting the score weight of the initial comprehensive score based on the comprehensive user identity model to obtain a preliminary adjusted comprehensive score, including: The user behavior pattern features and the multimodal user features are integrated to obtain a user feature set, based on the user feature set, a preset scoring algorithm is used to preliminarily evaluate the user identity to obtain a user identity evaluation result, and an initial comprehensive score is calculated based on the user identity evaluation result; Based on historical verification results and login requests, an adaptive scoring mechanism is used to dynamically adjust key parameters in a preset scoring algorithm in the process of calculating an initial comprehensive score to obtain an adjusted preliminary score. Based on the adjusted preliminary score, a federated learning framework and a reinforcement learning algorithm are applied to update and optimize the user identity model to obtain a comprehensive user identity model. Based on the comprehensive user identity model, the existing scoring weight adjustment strategy is optimized in combination with Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy. The initial comprehensive score is analyzed by using a variational autoencoder and anomaly detection algorithm, and the abnormal behavior is modeled to evaluate whether the login request is abnormal, thereby obtaining an anomaly detection result; Based on the comprehensive user identity model, the optimized score weight adjustment strategy and the anomaly detection result, the initial comprehensive score is dynamically adjusted to obtain a preliminary adjusted comprehensive score.

[0008] Optionally, based on historical verification results and login request conditions, an adaptive scoring mechanism is used to dynamically adjust key parameters in a preset scoring algorithm in the process of calculating an initial comprehensive score to obtain an adjusted preliminary score. Based on the adjusted preliminary score, a federated learning framework and a reinforcement learning algorithm are applied to update and optimize the user identity model to obtain a comprehensive user identity model. Based on the comprehensive user identity model, the existing scoring weight adjustment strategy is optimized in combination with Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy, including: Based on historical identity authentication records and login request characteristics used to reflect login request situations, an adaptive scoring mechanism is used to adjust key parameters in the preset scoring algorithm in real time to obtain an adjusted preliminary score, wherein the adjusted preliminary score is used to reflect the actual behavior pattern of the user; Optimizing the user behavior pattern features by using a federated learning framework to obtain optimized user behavior pattern features, and performing distributed training and updating on the user identity model based on the optimized user behavior pattern features to obtain a user identity model to be optimized; Based on the user identity model to be optimized, monitor and record the user's behavior, generate user behavior feedback data, select the optimal model parameter adjustment path by applying the reward mechanism in the reinforcement learning algorithm based on the user behavior feedback data, optimize the user identity model to be optimized based on the optimal model parameter adjustment path, and obtain a comprehensive user identity model, wherein the user behavior feedback data includes the user's login behavior, operation frequency, and usage habits; Based on the comprehensive user identity model, the best scoring weight configuration is selected using Bayesian optimization technology, and the existing scoring weight adjustment strategy is optimized based on the best scoring weight configuration to generate an optimized scoring weight adjustment strategy.

[0009] Optionally, based on the comprehensive user identity model, using Bayesian optimization technology, selecting the best scoring weight configuration, optimizing the existing scoring weight adjustment strategy based on the best scoring weight configuration, and generating an optimized scoring weight adjustment strategy, including: A probability model is constructed using Bayesian optimization technology, and based on the probability model, the scoring performance results under different scoring weight configurations are predicted. Based on the scoring performance results, an optimal solution is searched, and the optimal solution represents the best scoring weight configuration. In the prediction process, a meta-learning framework is used to adjust the hyperparameters of the probability model according to the scoring performance results and historical optimization experience to accelerate convergence to the optimal solution. In the process of searching for the optimal solution, multi-task learning technology is applied to share the underlying feature representation so that the Bayesian optimization technology can consider multiple related tasks while searching for the optimal solution, including anomaly detection and behavior prediction; Based on the optimal scoring weight configuration, adversarial training is applied to introduce adversarial samples, simulate attack situations, obtain simulation results, and based on the simulation results, optimize the existing scoring weight adjustment strategy to generate an optimized scoring weight adjustment strategy.

[0010] Optionally, the composite audio stream is separated and processed by using a blind source separation algorithm to extract the original ambient sound and the one-time random sequence code, the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, and the preliminary composite voiceprint sample is optimized to obtain a target composite voiceprint sample, including: The composite audio stream is separated by using a blind source separation algorithm to extract the original ambient sound and the one-time random sequence code, and during the separation process, a separation matrix is ​​initialized and iteratively optimized to improve the restoration degree of the original ambient sound and the one-time random sequence code; Synchronously aligning the original ambient sound and the one-time random sequence code to obtain the aligned original ambient sound and the one-time random sequence code, and performing superposition synthesis processing on the aligned original ambient sound and the one-time random sequence code to generate a preliminary composite voiceprint sample, wherein the preliminary composite voiceprint sample contains the real sound information of the user environment and the one-time random sequence code; Based on the preliminary composite voiceprint sample, a noise suppression algorithm is used to perform denoising processing to obtain a denoised preliminary composite voiceprint sample. Based on the denoised preliminary composite voiceprint sample, a feature enhancement technique is used to highlight key features to obtain a target composite voiceprint sample to be adjusted, and the audio quality parameters of the target composite voiceprint sample to be adjusted are adjusted to generate a target composite voiceprint sample, wherein the audio quality parameters include sampling rate and quantization bit number. (Key features include frequency domain features, time domain features and other related features, such as zero crossing rate, energy and zero crossing rate, etc.) Optionally, based on historical data of user behavior patterns, a behavior analysis algorithm is used to adjust the score weight of the preliminary adjusted comprehensive score to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result, including: Based on historical data of user behavior patterns, a risk level assessment is performed on the behavior characteristics of the login request to obtain a risk assessment result, target characteristics are extracted based on the risk assessment result, a user behavior model is constructed based on the target characteristics, and based on the user behavior model, a behavior analysis result is generated and abnormal situations are identified, wherein the target characteristics include login time distribution, geographic location changes, and device fingerprints, and the abnormal situations include abnormal login time and geographic location mutations; Based on the behavior analysis results and the abnormal situation, adjusting the score weight of the preliminary adjusted comprehensive score using a behavior analysis algorithm to obtain a final adjusted comprehensive score; An access permission threshold is set according to an existing security policy, and access permission is confirmed based on the access permission threshold. If the final adjusted comprehensive score is higher than the access permission threshold, access is confirmed to be allowed; if the final adjusted comprehensive score is lower than the access permission threshold, access is confirmed to be not allowed, thereby obtaining a confirmed access result.

[0011] In a second aspect, an embodiment of the present invention provides a login system based on a 5G message voiceprint verification code, comprising: A generation module, used to generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; An extraction module is used to obtain a composite audio stream based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, separate the composite audio stream using a blind source separation algorithm, extract the original ambient sound and the one-time random sequence code, combine the original ambient sound and the one-time random sequence code to obtain a preliminary composite voiceprint sample, optimize the preliminary composite voiceprint sample, and obtain a target composite voiceprint sample; An analysis module is used to encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, transmit the encrypted target composite voiceprint sample to a cloud authentication platform in combination with a quantum key distribution mechanism, and obtain a decrypted target composite voiceprint sample after the cloud authentication platform receives and decrypts the sample. The decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features; The adjustment module is used to adjust the score weight of the preliminary adjusted comprehensive score based on the historical data of the user behavior pattern by using the behavior analysis algorithm to obtain the final adjusted comprehensive score, and to confirm the access permission based on the final adjusted comprehensive score to obtain the confirmed access result so that the user can log in according to the confirmed access result.

[0012] In a third aspect, an embodiment of the present invention provides a computing device, comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a login method based on a 5G message voiceprint verification code as described in any one of the first aspects.

[0013] In a fourth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a login method based on a 5G message voiceprint verification code as described in any one of the first aspects.

[0014] In an embodiment of the present invention, a 5G message including a one-time random sequence code is generated, the 5G message is sent to a user terminal device, and a message reception confirmation feedback is received from the user terminal device; based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receipt of the 5G message, a composite audio stream is obtained in combination with the message reception confirmation feedback, the composite audio stream is separated and processed using a blind source separation algorithm, the original ambient sound and the one-time random sequence code are extracted, the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, the preliminary composite voiceprint sample is optimized, and a target composite voiceprint sample is obtained; the target composite voiceprint sample is encrypted based on a preset encryption protocol to obtain an encrypted The encrypted target composite voiceprint sample is transmitted to the cloud authentication platform in combination with the quantum key distribution mechanism. The decrypted target composite voiceprint sample is received and decrypted by the cloud authentication platform to obtain the decrypted target composite voiceprint sample. The decrypted target composite voiceprint sample is analyzed by using the convolutional neural network, recurrent neural network and generative adversarial network combined model in the deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features. Based on the historical data of user behavior patterns, the scoring weight of the preliminary adjusted comprehensive score is adjusted by using the behavior analysis algorithm to obtain a final adjusted comprehensive score. Access rights are confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result. The technical solution provided by the present invention significantly enhances the security of user identity authentication and prevents replay attacks and data leakage by introducing a one-time random sequence code, a quantum key distribution mechanism and a preset encryption protocol; the blind source separation algorithm and the deep learning framework are used to improve the restoration degree and recognition accuracy of the audio signal and reduce the possibility of misjudgment; combined with the adaptive scoring mechanism and the optimization of the preliminary composite voiceprint samples, the system can flexibly respond to different user behaviors and environmental changes, and provide a personalized user experience; all voiceprint data are collected on the local device, and federated learning and distributed training are used to protect user privacy; with the help of 5G message transmission and automated processing procedures, efficient response speed and operational efficiency are ensured; through anomaly detection, behavior analysis and adversarial training, the anti-attack capability is improved, and potential security threats can be discovered and prevented in a timely manner; in summary, the solution not only improves security, accuracy and dynamic adaptability, but also protects user privacy, provides an efficient user experience, and ensures the security and reliability of the user login process.Among them, through the adaptive scoring mechanism, federated learning framework and reinforcement learning algorithm, the system can adjust the scoring parameters in real time, optimize the characteristics of user behavior patterns, and intelligently update the user identity model, effectively improving the security of identity authentication; combining multiple technologies and algorithms, the system can more accurately reflect the user's actual behavior patterns, reduce the possibility of misjudgment, and improve the reliability of identity authentication; all adjustment and optimization processes are carried out dynamically, and the system can flexibly adjust according to the user's real-time behavior and environmental changes, providing a personalized user experience; the federated learning framework ensures that user data is not shared and protects user privacy; through distributed training and intelligent optimization, the system can operate efficiently in different environments, improving response speed and efficiency; the introduction of reinforcement learning and anomaly detection algorithms enhances the system's ability to combat abnormal behavior and promptly detect and prevent potential security threats.

[0015] These and other aspects of the present invention will become more apparent from the following description of the embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 A flowchart of a login method based on a 5G message voiceprint verification code provided by an embodiment of the present invention; Figure 2 A structural diagram of a login system based on a 5G message voiceprint verification code provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a computing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0019] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0021] Figure 1 A flowchart of a login method based on a 5G message voiceprint verification code is provided for an embodiment of the present invention, such as Figure 1 As shown, the method includes: With the rapid development of mobile communication technology, 5G network, with its high speed, low latency and large capacity, provides a solid foundation for various innovative applications. However, traditional login authentication methods are powerless in the face of increasingly complex network security threats. In order to improve the security and convenience of user login, based on this, the present invention provides a login method based on 5G message voiceprint verification code, such as Figure 1 ,include: Step 101: Generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; In this step, the one-time random sequence code refers to a unique and unpredictable code used to verify the user's identity and ensure the security of communication. 5G messages refer to messages transmitted based on the fifth generation of mobile communication technology, which have the characteristics of high speed and low latency. Message reception confirmation feedback refers to the confirmation information returned to the sender by the user terminal device after receiving the 5G message.

[0022] This step generates a unique one-time random sequence code and embeds it into a 5G message. This message will be sent to the user terminal device to ensure the uniqueness and security of each verification process; the message containing the one-time random sequence code is sent to the user terminal device through the 5G network. After the user terminal device receives the message, it immediately receives a message reception confirmation feedback to confirm that the message has been successfully received.

[0023] Step 102: Based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, a composite audio stream is obtained, and the composite audio stream is separated and processed by a blind source separation algorithm to extract the original ambient sound and the one-time random sequence code, and the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, and the preliminary composite voiceprint sample is optimized to obtain a target composite voiceprint sample; In this step, ambient sound refers to the sound environment around the user terminal device, including background noise and other non-target sounds. Composite audio stream refers to an audio data stream composed of ambient sound and a one-time random sequence code. The preliminary composite voiceprint sample refers to a preliminary audio sample obtained by combining the original ambient sound and the one-time random sequence code. The composite audio stream before separation contains a lot of background noise and other interference, while the preliminary composite voiceprint sample after recombination has been denoised and optimized, and the signal quality is significantly improved; the separated random sequence code and the original ambient sound are clearer, and the preliminary composite voiceprint sample after recombination can better retain and highlight these features, which is convenient for subsequent processing.

[0024] In this step, when the user terminal device confirms the receipt of the 5G message, the local voiceprint collection device collects the surrounding ambient sound. These ambient sounds, together with the message reception confirmation feedback, form a composite audio stream; the composite audio stream is processed using a blind source separation algorithm to separate the original ambient sound and the one-time random sequence code. This step ensures the purity of the audio data and provides an accurate basis for subsequent analysis; the separated original ambient sound and the one-time random sequence code are combined to form a preliminary composite voiceprint sample. Then, a specific optimization algorithm is used to improve the quality and accuracy of the sample to obtain the target composite voiceprint sample.

[0025] Step 103: Encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, transmit the encrypted target composite voiceprint sample to the cloud authentication platform in combination with the quantum key distribution mechanism, receive and decrypt the target composite voiceprint sample on the cloud authentication platform to obtain a decrypted target composite voiceprint sample, analyze the decrypted target composite voiceprint sample using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and calculate a preliminary adjusted comprehensive score based on the time-frequency domain features; In this step, the cloud authentication platform refers to the server or system located in the cloud, which is responsible for processing and verifying the user's identity information.

[0026] This step is based on the pre-; then the quantum key distribution mechanism is used to securely transmit the encrypted sample to the cloud authentication platform; the cloud authentication platform receives and decrypts the target composite voiceprint sample, and then uses the convolutional neural network, recurrent neural network and generative adversarial network combined model in the deep learning framework to analyze the decrypted sample, extract its time-frequency domain features, and calculate the preliminary adjusted comprehensive score.

[0027] Step 104: Based on the historical data of the user's behavior pattern, the score weight of the preliminary adjusted comprehensive score is adjusted using a behavior analysis algorithm to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result; This step uses the behavior analysis algorithm to adjust the score weight of the preliminary adjusted comprehensive score based on the historical data of the user's behavior pattern to reflect the user's current behavior pattern and habits, thereby obtaining the final adjusted comprehensive score; based on the final adjusted comprehensive score, determine whether to grant the user access rights. If the score reaches the set threshold, the user's identity is confirmed and login is allowed; otherwise, the access request is denied.

[0028] The embodiments of the present invention can achieve the following beneficial effects through the above steps: The one-time random sequence code and quantum key distribution mechanism ensure the security of communication and data transmission, effectively preventing the risk of man-in-the-middle attacks and data leakage; The use of blind source separation algorithms and deep learning frameworks improves the accuracy and reliability of voiceprint recognition and reduces the possibility of misjudgment; Combined with historical data of user behavior patterns, the system can dynamically adjust the scoring weights, making the authentication process more personalized and intelligent, and improving the user experience; All operations are performed between the local device and the cloud, ensuring that user data is not obtained by third parties and protecting user privacy; The high speed and low latency characteristics of the 5G network enable the entire authentication process to be completed quickly, improving the response speed and efficiency of the system.

[0029] In order to ensure the security and reliability of user identity authentication, based on this, the present invention provides a specific embodiment, in which step 103, based on a preset encryption protocol, the target composite voiceprint sample is encrypted to obtain an encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is transmitted to a cloud authentication platform in combination with a quantum key distribution mechanism, and the decrypted target composite voiceprint sample is obtained after being received and decrypted by the cloud authentication platform, and the decrypted target composite voiceprint sample is analyzed by using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features, specifically comprising the following steps: Step 301: Encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample. In the encryption process, homomorphic encryption technology is used to perform a specific type of calculation on the encrypted target composite voiceprint sample. The specific type includes additive homomorphic encryption, multiplicative homomorphic encryption, and fully homomorphic encryption. This step encrypts the target composite voiceprint sample according to the preset encryption protocol. Homomorphic encryption technology is used in this process, so that the encrypted sample can be calculated by additive homomorphic, multiplicative homomorphic or fully homomorphic encryption without decryption. This ensures the security of data during transmission and processing while retaining computing power.

[0030] Step 302: Transmit the encrypted target composite voiceprint sample from the user terminal device to the cloud authentication platform using the quantum key distribution mechanism and the post-quantum cryptography algorithm. During the transmission process, the quantum key distribution generation mechanism is used to add encryption keys to each sample in the encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is processed in combination with the post-quantum cryptography algorithm. This step uses the quantum key distribution mechanism and post-quantum cryptography algorithm to securely transmit the encrypted target composite voiceprint sample from the user terminal device to the cloud authentication platform; during the transmission process, the quantum key distribution generation mechanism is used to add encryption keys to each sample, and combined with the post-quantum cryptography algorithm to further enhance security; quantum key distribution ensures that the key transmission cannot be eavesdropped, while the post-quantum cryptography algorithm provides protection against quantum computer attacks.

[0031] Step 303: Based on the cloud authentication platform, the received encrypted target composite voiceprint sample is decrypted, and the decryption process is verified by zero-knowledge proof technology. After the verification process is completed, the decrypted target composite voiceprint sample is obtained; This step is based on the cloud authentication platform to decrypt the received encrypted target composite voiceprint sample, and verify the correctness and integrity of the decryption process through zero-knowledge proof technology; zero-knowledge proof technology ensures the transparency and security of the decryption process and prevents any potential information leakage; after the verification is completed, the decrypted target composite voiceprint sample is obtained.

[0032] Step 304: Analyze the decrypted target composite voiceprint sample using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework. During the analysis process, extract time-frequency domain features through a convolutional neural network, extract time series characteristics through a recurrent neural network, and generate a synthetic voiceprint sample through a generative adversarial network to obtain an enhanced multimodal voiceprint feature set. This step uses the combined model of convolutional neural network (CNN), recurrent neural network (RNN) and generative adversarial network (GAN) in the deep learning framework to analyze the decrypted target composite voiceprint sample. Specifically, the convolutional neural network is used to extract time-frequency domain features and capture the time and frequency characteristics of the audio signal; the recurrent neural network is used to extract time series characteristics and analyze the time dependency of the audio signal; the generative adversarial network is used to generate synthetic voiceprint samples and enhance the quality and diversity of the multimodal voiceprint feature set. These features together constitute the enhanced multimodal voiceprint feature set.

[0033] Step 305: construct a user identity model by combining the multimodal voiceprint feature set and the user's biological features, perform multimodal data analysis based on the user identity model to obtain multimodal user features, and analyze user behavior using a graph neural network based on the multimodal user features and historical data of user behavior patterns to obtain user behavior pattern features, wherein the biological features include facial features and fingerprint features; This step combines the multimodal voiceprint feature set and the user's biometric features, such as facial features and fingerprint features, to build a user identity model; this user identity model integrates multiple features through multimodal data analysis to improve the accuracy and reliability of recognition; based on the user identity model, the user behavior is analyzed using a graph neural network to extract user behavior pattern features. These features reflect the user's habits and behavior patterns, enhancing the personalization and intelligence of identity authentication.

[0034] Step 306: Calculate an initial comprehensive score by combining the user behavior pattern characteristics and the multimodal user characteristics, optimize the user identity model, obtain a comprehensive user identity model, and dynamically adjust the score weight of the initial comprehensive score based on the comprehensive user identity model to obtain a preliminary adjusted comprehensive score; This step combines user behavior pattern characteristics and multimodal user characteristics to calculate the initial comprehensive score. Then optimize the user identity model and dynamically adjust the score weight of the initial comprehensive score to obtain the preliminary adjusted comprehensive score. This process ensures the accuracy and fairness of the score and improves the adaptability of the system and user experience.

[0035] The embodiments of the present invention can achieve the following beneficial effects through the above steps: Homomorphic encryption technology and quantum key distribution mechanism ensure the security of data during transmission and processing, effectively preventing the risk of man-in-the-middle attacks and data leakage; Using multiple neural network models in the deep learning framework improves the accuracy and reliability of voiceprint recognition and reduces the possibility of misjudgment; Combining historical data and multimodal features of user behavior patterns, the system can dynamically adjust the scoring weights, making the authentication process more personalized and intelligent, and improving the user experience; All operations are performed between the local device and the cloud, ensuring that user data is not obtained by third parties and protecting user privacy; The high speed and low latency of the 5G network enable the entire authentication process to be completed quickly, improving the system's response speed and efficiency; The introduction of post-quantum cryptography algorithms ensures the security of the system in the face of future quantum computer attacks.

[0036] The present invention provides a specific embodiment, wherein step 306 combines the user behavior pattern characteristics and the multimodal user characteristics to calculate an initial comprehensive score, optimizes the user identity model, obtains a comprehensive user identity model, and dynamically adjusts the score weight of the initial comprehensive score based on the comprehensive user identity model to obtain a preliminary adjusted comprehensive score, specifically comprising the following steps: Step 311: fusing the user behavior pattern features and the multimodal user features to obtain a user feature set, and based on the user feature set, preliminarily evaluating the user identity using a preset scoring algorithm to obtain a user identity evaluation result, and calculating an initial comprehensive score based on the user identity evaluation result; In this step, the preset scoring algorithm refers to a predefined algorithm for evaluating the user's identity, usually based on multiple features and historical data. The initial comprehensive score refers to a preliminary score calculated based on the user's feature set, which is used to measure the credibility of the user's identity authentication.

[0037] This step combines the user behavior pattern features and multimodal user features to form a comprehensive user feature set. This step ensures the diversity and integrity of the features and provides a solid foundation for subsequent evaluation. Based on the user feature set, the user identity is preliminarily evaluated using a preset scoring algorithm to obtain the user identity evaluation result. Based on the result, the initial comprehensive score is calculated to measure the initial credibility of the user identity verification.

[0038] Step 312: Based on historical verification results and login requests, the adaptive scoring mechanism is used to dynamically adjust key parameters in the preset scoring algorithm in the process of calculating the initial comprehensive score to obtain an adjusted preliminary score. Based on the adjusted preliminary score, the federated learning framework and the reinforcement learning algorithm are applied to update and optimize the user identity model to obtain a comprehensive user identity model. Based on the comprehensive user identity model, the existing scoring weight adjustment strategy is optimized in combination with the Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy. In this step, historical verification results refer to the user's past identity verification records and their results, which are used for reference and adjustment of scoring strategies. Login request status refers to the specific situation of the current user initiating a login request, including time, location, device and other information.

[0039] This step combines historical verification results and current login requests, and uses an adaptive scoring mechanism to dynamically adjust key parameters in the preset scoring algorithm. This process takes into account the user's real-time behavior and historical data to ensure that the scoring is more accurate and personalized, thereby obtaining an adjusted preliminary score; based on the adjusted preliminary score, the federated learning framework and reinforcement learning algorithm are applied to update and optimize the user identity model. The federated learning framework allows multiple clients to train collaboratively, while the reinforcement learning algorithm helps the system continuously improve its decision-making strategy, and finally obtains a comprehensive user identity model; based on the comprehensive user identity model, combined with Bayesian optimization technology, the existing scoring weight adjustment strategy is optimized. Bayesian optimization technology is used to find the optimal parameter setting, making the scoring weight adjustment more scientific and reasonable, and obtaining the optimized scoring weight adjustment strategy.

[0040] Step 313: Analyze the initial comprehensive score using a variational autoencoder and anomaly detection algorithm, and evaluate whether the login request is abnormal by modeling the abnormal behavior to obtain an anomaly detection result; This step uses a variational autoencoder and anomaly detection algorithm to analyze the initial comprehensive score, model abnormal behavior, evaluate whether there are any abnormalities in the login request, and obtain anomaly detection results. This step enhances the security and reliability of the system and promptly detects and prevents potential security threats.

[0041] Step 314: Based on the comprehensive user identity model, the optimized score weight adjustment strategy and the anomaly detection result, dynamically adjust the initial comprehensive score to obtain a preliminary adjusted comprehensive score; This step dynamically adjusts the initial comprehensive score based on the comprehensive user identity model, the optimized score weight adjustment strategy and the anomaly detection results to obtain a preliminary adjusted comprehensive score. This process ensures the accuracy and fairness of the score and improves the system's adaptability and user experience.

[0042] The embodiments of the present invention can achieve the following beneficial effects through the above steps: By integrating multimodal user features and user behavior pattern features and combining them with anomaly detection algorithms, the security of user identity authentication is effectively improved, reducing the risk of misjudgment and attack; The use of preset scoring algorithms, adaptive scoring mechanisms and deep learning models improves the accuracy and reliability of scoring and reduces the possibility of misjudgment; Combining historical verification results and real-time login requests, the system can dynamically adjust the scoring weights, making the authentication process more personalized and intelligent, and improving the user experience; All operations are performed between the local device and the cloud, ensuring that user data is not obtained by third parties and protecting user privacy; The application of the federated learning framework and reinforcement learning algorithm enables the system to efficiently train and optimize models in a distributed environment, improving response speed and efficiency; The introduction of variational autoencoders and anomaly detection algorithms enhances the system's ability to combat abnormal behaviors and promptly detect and prevent potential security threats.

[0043] With the increasing complexity of network security threats, traditional user identity authentication methods face multiple difficulties, including insufficient security, weak privacy protection, and weak anti-attack capabilities. In response to these challenges, the present invention provides a comprehensive technical solution by introducing an adaptive scoring mechanism, a federated learning framework, a reinforcement learning algorithm, and Bayesian optimization technology. Specifically, the scoring parameters are adjusted in real time using historical identity authentication records and login request characteristics, which improves the accuracy and dynamic adaptability of the scoring; the federated learning framework is used for distributed training and model update, which protects user privacy and enhances the generalization ability of the model; the reinforcement learning algorithm is used to select the optimal model parameter adjustment path and continuously optimize system performance; the Bayesian optimization technology is used to select the best scoring weight configuration to ensure the scientific rationality of the scoring adjustment. In addition, with the help of 5G message transmission and automated processing procedures, the response speed and efficiency of the system are improved, and the anti-attack capability and robustness of the system are enhanced through anomaly detection and adversarial training. In summary, the design of the present invention not only solves the key problems in the prior art, but also achieves higher security, accuracy and flexibility, ensures the safety and reliability of the user login process, and meets the needs of modern technology. The present invention provides a specific embodiment, in which step 312 dynamically adjusts key parameters in a preset scoring algorithm in the process of calculating an initial comprehensive score based on historical verification results and login request conditions by using an adaptive scoring mechanism to obtain an adjusted preliminary score, and based on the adjusted preliminary score, applies a federated learning framework and a reinforcement learning algorithm to update and optimize the user identity model to obtain a comprehensive user identity model, and based on the comprehensive user identity model, optimizes an existing scoring weight adjustment strategy in combination with Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy, specifically comprising the following steps: Step 321: Based on the historical identity authentication records and the login request characteristics used to reflect the login request situation, the key parameters in the preset scoring algorithm are adjusted in real time using the adaptive scoring mechanism to obtain an adjusted preliminary score, wherein the adjusted preliminary score is used to reflect the actual behavior pattern of the user; In this step, historical authentication records refer to the user's past authentication records and their results, which are used for reference and adjustment of scoring strategies. Login request characteristics refer to the specific circumstances of the current user's login request, including time, location, device and other information.

[0044] This step uses an adaptive scoring mechanism to adjust key parameters in the preset scoring algorithm in real time based on historical authentication records and login request characteristics. This step ensures that the scoring algorithm can more accurately reflect the user's actual behavior patterns and reduce misjudgments.

[0045] Step 322: optimizing the user behavior pattern features by using a federated learning framework to obtain optimized user behavior pattern features, and performing distributed training and updating on the user identity model based on the optimized user behavior pattern features to obtain a user identity model to be optimized; This step uses the federated learning framework to optimize the user behavior pattern features and obtain the optimized user behavior pattern features. This process not only protects user privacy, but also improves the quality and generalization ability of the features; based on the optimized user behavior pattern features, the user identity model is distributedly trained and updated to obtain the user identity model to be optimized. Through the federated learning framework, multiple clients can collaborate in training without having to share sensitive data.

[0046] Step 323: Based on the user identity model to be optimized, monitor and record the user's behavior, generate user behavior feedback data, select the optimal model parameter adjustment path by applying the reward mechanism in the reinforcement learning algorithm based on the user behavior feedback data, optimize the user identity model to be optimized based on the optimal model parameter adjustment path, and obtain a comprehensive user identity model, wherein the user behavior feedback data includes the user's login behavior, operation frequency, and usage habits; In this step, the optimal model parameter adjustment path refers to a path that can optimize the performance of the user identity model from multiple possible parameter adjustment paths selected through the reward mechanism in the reinforcement learning algorithm. This path ensures that the system can continuously optimize in different environments and provide secure and reliable user identity authentication services.

[0047] This step is based on the user identity model to be optimized, monitors and records user behavior, and generates user behavior feedback data, which provides an important basis for subsequent optimization; based on the user behavior feedback data, the reward mechanism in the reinforcement learning algorithm is used to select the optimal model parameter adjustment path, and the user identity model is optimized according to this path to obtain a comprehensive user identity model to help the model continuously improve its decision-making strategy.

[0048] More specifically, the present invention provides a calculation formula for a comprehensive user identity model, which is used to optimize the user identity model to be optimized in combination with the optimal model parameter adjustment path to maximize the cumulative reward and ensure that the comprehensive user identity model performs best over a long time series. The specific calculation formula is as follows: ; in, represents a comprehensive user identity model; Indicates the user identity model to be optimized; Indicates that among all possible user identity models to be optimized Find the user identity model to be optimized that maximizes the cumulative reward ; Represents the total number of time steps in the reinforcement learning process; Represents the time step The reward function is used to measure the user identity model to be optimized and optimal model parameter adjustment path Instant rewards for: is the reward discount factor of The power is expressed in time steps The reward discount on the said; represents the optimal model parameter adjustment path; This comprehensive user identity model formula ensures that the final optimized user identity model The performance at each time point is optimal, maximizing the overall performance of the system and user experience. Specifically, the formula is optimized based on the characteristics of user behavior patterns, user behavior feedback data, and the optimal model parameter adjustment path, which ensures that the model can better adapt to the user's actual behavior patterns and improve the security and convenience of the login process; by continuously optimizing the model parameters, the system can more accurately identify the user's identity, reduce the risk of misjudgment and attack, and thus improve the overall security performance; by maximizing the cumulative rewards, the model is encouraged to continuously improve itself and gradually approach the optimal solution, which not only improves the accuracy of the model, but also enhances the system's adaptive ability, enabling it to maintain efficient and stable performance in different environments.

[0049] Among them, the calculation formula of the optimal model parameter adjustment path is as follows: ; in, represents the optimal model parameter adjustment path, which represents the best path from the current state to the target state; represents a reinforcement learning algorithm; Indicates user behavior and user behavior records; Represents user behavior feedback data, which indicates feedback information generated by user behavior; Represents the reward discount factor, which is used to evaluate the importance of future rewards; Represents the balance factor, which is used to find a suitable balance between the known path and the new path obtained by adjustment to achieve better optimization effect; Indicates that among all possible paths Find a path that maximizes the cumulative reward ; Represents the total number of time steps in the reinforcement learning process; Indicates The reward function of the step is used to measure the and balance factor Instant rewards for: The optimal model parameter adjustment path formula uses a reinforcement learning algorithm to select the optimal model parameter adjustment path to maximize the cumulative reward. This path is based on user behavior and user behavior feedback data, and by adjusting the balance factor and reward discount factor, a global optimal path is found, so that the model can flexibly adjust strategies at different stages to improve adaptability and efficiency; by adjusting the balance factor to explore the balance between new paths and using known paths, the decision-making process is smoother, which is conducive to extensively exploring potential optimal paths in the early stages of optimization, avoiding falling into local optimal solutions too early, and gradually reducing the optimal path in the later stages. To accelerate the convergence to the global optimal solution.

[0050] The calculation formula of the user identity model to be optimized is as follows: ; in, Indicates the user identity model to be optimized; Represents all possible user identity models Find the user identity model that minimizes the loss function ; Represents the loss function, measuring the user identity model User behavior pattern characteristics after optimization The performance below; Represents the user identity model; Represents the regularization parameter in federated learning, which is used to balance the loss function The weight between and the regularization term; Represents the weight coefficient of the regularization term in the user identity model, controlling the regularization strength; represents the initial model or reference model, usually the global average model; The user identity model formula to be optimized optimizes the user identity model by minimizing the loss function. , ensuring that the user behavior pattern characteristics after optimization The model performs best under the new features, improving the accuracy and reliability of the model. By minimizing the loss function, the model can be effectively adjusted under new features to ensure its optimal performance; the regularization term The variation range of model parameters is limited, the instability of the model caused by large fluctuations in parameters is avoided, and the reliability of the system is improved.

[0051] Among them, the calculation formula of the optimized user behavior pattern characteristics is as follows: ; in, Indicates the characteristics of user behavior patterns after optimization; represents the federated learning framework; Indicates the user behavior pattern characteristics and the user's original behavior data; Represents the learning rate of federated learning, which is used to control the speed of model update; Represents the regularization parameter, which is used to balance the loss function The weight between and the regularization term; Indicates all possible user behavior pattern characteristics Find the one that minimizes the loss function ; Represents the loss function, which measures the current user behavior pattern characteristics Initial user behavior pattern characteristics The difference between Represents the weight coefficient of the regularization term in federated learning and controls the regularization strength; Represents the initial feature or reference feature, usually the global average feature.

[0052] The optimized user behavior pattern feature formula optimizes the user behavior pattern feature by minimizing the loss function. , ensuring its optimal performance in the federated learning process and improving the generalization and stability of the model. Through federated learning, user data is retained locally, and only updated model parameters are uploaded to the cloud, ensuring the security of user privacy; regularization term It effectively prevents the model from overfitting to a specific data set during training and improves the generalization and robustness of the model in different scenarios.

[0053] Step 324: Based on the comprehensive user identity model, using Bayesian optimization technology, select the best scoring weight configuration, optimize the existing scoring weight adjustment strategy based on the best scoring weight configuration, and generate an optimized scoring weight adjustment strategy; This step uses Bayesian optimization technology to select the best scoring weight configuration based on the comprehensive user identity model. This process ensures that the scoring weight adjustment is more scientific and reasonable, and that the scoring strategy can dynamically adapt to different user behaviors and environmental changes; based on the best scoring weight configuration, the existing scoring weight adjustment strategy is optimized to generate an optimized scoring weight adjustment strategy, which can dynamically adjust the scoring weight to ensure the timeliness and effectiveness of the scoring.

[0054] The embodiments of the present invention can achieve the following beneficial effects through the above steps: Through the adaptive scoring mechanism, federated learning framework and reinforcement learning algorithm, the system can adjust the scoring parameters in real time, optimize the characteristics of user behavior patterns, and intelligently update the user identity model, effectively improving the security of identity authentication; By combining multiple technologies and algorithms, the system can more accurately reflect the user's actual behavior patterns, reduce the possibility of misjudgment, and improve the reliability of identity authentication; All adjustments and optimization processes are carried out dynamically. The system can flexibly adjust according to the user's real-time behavior and environmental changes, providing a personalized user experience. The federated learning framework ensures that user data is not shared and protects user privacy; Through distributed training and intelligent optimization, the system can operate efficiently in different environments, improving response speed and efficiency; The introduction of reinforcement learning and anomaly detection algorithms enhances the system's ability to combat abnormal behaviors and promptly detect and prevent potential security threats.

[0055] The present invention provides a specific embodiment, wherein step 324, based on the comprehensive user identity model, uses Bayesian optimization technology to select the best scoring weight configuration, optimizes the existing scoring weight adjustment strategy based on the best scoring weight configuration, and generates an optimized scoring weight adjustment strategy, specifically comprising the following steps: Step 331: construct a probability model using Bayesian optimization technology, predict the scoring performance results under different scoring weight configurations based on the probability model, search for an optimal solution based on the scoring performance results, the optimal solution represents the best scoring weight configuration, and use a meta-learning framework in the prediction process to adjust the hyperparameters of the probability model according to the scoring performance results and historical optimization experience to accelerate convergence to the optimal solution; In this step, the scoring performance results refer to the scoring performance indicators predicted based on different scoring weight configurations, which are used to evaluate the effectiveness of the configurations. Historical optimization experience refers to the data and knowledge accumulated during the previous optimization process, including but not limited to past parameter configurations and their performance results, optimization path records, convergence speed and stability, and task relevance analysis.

[0056] In this step, Bayesian optimization technology is used to build a probability model, and the scoring performance results under different scoring weight configurations are predicted based on the model. This step provides a basis for the subsequent search for the optimal solution. The probability model is used to predict the scoring performance under different configurations to ensure that the search process has a clear goal and direction; based on the scoring performance results, Bayesian optimization technology is used to search for the optimal solution, that is, to find the best scoring weight configuration. In this process, the meta-learning framework is used to adjust the hyperparameters of the probability model according to the scoring performance results and historical optimization experience to accelerate convergence to the optimal solution and improve optimization efficiency.

[0057] Step 332: In the process of searching for the optimal solution, multi-task learning technology is applied to share the underlying feature representation so that the Bayesian optimization technology considers multiple related tasks while searching for the optimal solution, wherein the related tasks include anomaly detection and behavior prediction; In this step, multi-task learning technology is applied in the process of searching for the optimal solution. By sharing the underlying feature representation, Bayesian optimization technology can consider multiple related tasks while searching for the optimal solution, such as anomaly detection and behavior prediction, which helps to improve the overall performance and generalization ability of the system. Through multi-task learning, the adaptability and flexibility of the system are improved, enabling it to efficiently switch between different tasks, thereby enhancing the robustness and reliability of the system.

[0058] Step 333: Based on the optimal scoring weight configuration, adversarial training is applied to introduce adversarial samples, simulate attack situations, obtain simulation results, and based on the simulation results, optimize the existing scoring weight adjustment strategy to generate an optimized scoring weight adjustment strategy; This step is based on the optimal scoring weight configuration, and uses adversarial training to enhance the system's anti-attack capabilities, ensuring that it can maintain high performance and accuracy in the face of potential threats; by introducing adversarial samples, simulating attack situations, and obtaining simulation results. By analyzing these results, the existing scoring weight adjustment strategy is optimized and the optimized scoring weight adjustment strategy is generated.

[0059] The embodiments of the present invention can achieve the following beneficial effects through the above steps: Through Bayesian optimization technology and meta-learning framework, the system can quickly converge to the optimal solution, reducing optimization time and computing resource consumption; Multi-task learning technology enables the system to handle multiple related tasks simultaneously, improving overall performance and generalization capabilities, and enhancing the robustness and adaptability of the system; With the introduction of adversarial training, the system can simulate attack situations, detect and fix potential security vulnerabilities in advance, and improve the security and anti-attack capabilities of the system; Based on probability model predictions and historical optimization experience, the system can make more scientific and reasonable decisions to ensure that the scoring weight adjustment strategy is always in the optimal state; All adjustment and optimization processes are carried out dynamically. The system can flexibly adjust according to the user's real-time behavior and environmental changes, providing a personalized user experience.

[0060] The present invention provides a specific embodiment, step 102, using a blind source separation algorithm to separate the composite audio stream, extracting the original ambient sound and the one-time random sequence code, combining the original ambient sound and the one-time random sequence code to obtain a preliminary composite voiceprint sample, optimizing the preliminary composite voiceprint sample, and obtaining a target composite voiceprint sample, specifically includes the following steps: Step 201: using a blind source separation algorithm to separate the composite audio stream, extracting the original ambient sound and the one-time random sequence code, and during the separation process, initializing a separation matrix and iteratively optimizing the separation matrix to improve the restoration degree of the original ambient sound and the one-time random sequence code; In this step, initializing the separation matrix means setting the initial value for the separation matrix in the blind source separation process as the starting point for iterative optimization. Iteratively optimizing the separation matrix means gradually improving the separation effect by repeatedly adjusting the parameters of the separation matrix so that the separated signal is closer to the original signal. The degree of restoration refers to the degree of similarity between the separated signal and the original signal, reflecting the effect of the separation process.

[0061] This step uses a blind source separation algorithm to separate the composite audio stream and extract the original ambient sound and the one-time random sequence code. In this process, by initializing the separation matrix and iteratively optimizing the matrix, the degree of restoration of the original ambient sound and the one-time random sequence code is gradually improved, ensuring that the separated signal is as close to the original signal as possible, thereby improving the accuracy of subsequent processing.

[0062] Step 202: synchronously aligning the original ambient sound and the one-time random sequence code to obtain the aligned original ambient sound and the one-time random sequence code, superimposing and synthesizing the aligned original ambient sound and the one-time random sequence code to generate a preliminary composite voiceprint sample, wherein the preliminary composite voiceprint sample contains the real sound information of the user environment and the one-time random sequence code; In this step, synchronization alignment refers to the precise time alignment of audio signals from different sources to ensure that they can be correctly superimposed when synthesized.

[0063] In this step, the extracted original ambient sound and the one-time random sequence code are synchronously aligned to obtain the aligned original ambient sound and the one-time random sequence code. This step ensures the consistency of the synthesized audio signal in time and content, and avoids distortion caused by time difference; the aligned original ambient sound and the one-time random sequence code are superimposed and synthesized to generate a preliminary composite voiceprint sample containing all necessary information. This sample contains the real sound information of the user environment and the one-time random sequence code.

[0064] Step 203: Based on the preliminary composite voiceprint sample, a noise suppression algorithm is used to perform denoising processing to obtain a denoised preliminary composite voiceprint sample; based on the denoised preliminary composite voiceprint sample, a feature enhancement technology is used to highlight key features to obtain a target composite voiceprint sample to be adjusted; and the audio quality parameters of the target composite voiceprint sample to be adjusted are adjusted to generate a target composite voiceprint sample, where the audio quality parameters include a sampling rate and a quantization bit number; In this step, key features include frequency domain features, time domain features and other related features, such as zero crossing rate, energy and zero crossing rate.

[0065] This step is based on the preliminary composite voiceprint sample, and uses the noise suppression algorithm to perform denoising to obtain the preliminary composite voiceprint sample after denoising; this step aims to reduce background noise and other interference in the audio to ensure the clarity and reliability of the audio signal; based on the preliminary composite voiceprint sample after denoising, the feature enhancement technology is used to highlight the key features to obtain the target composite voiceprint sample to be adjusted. These key features may include frequency, amplitude, etc. By enhancing the key features, the recognition and analysis efficiency of the audio signal are improved; the audio quality parameters of the target composite voiceprint sample to be adjusted, such as the sampling rate and the number of quantization bits, are adjusted to generate the final target composite voiceprint sample. This step ensures the high quality and high fidelity of the final audio sample.

[0066] The embodiments of the present invention can achieve the following beneficial effects through the above steps: Through the blind source separation algorithm and iterative optimization of the separation matrix, the restoration degree of the original ambient sound and the one-time random sequence code is significantly improved, ensuring the accuracy and reliability of the separation process; The synchronization alignment step ensures the consistency of the audio signal in time and content, avoids distortion caused by time difference, and improves the quality of the synthesized audio; The noise suppression algorithm effectively reduces background noise and other interference in the audio, making the audio signal clearer and enhancing the system reliability and user experience; Feature enhancement technology highlights key features in audio signals, improves the recognition and analysis efficiency of audio signals, and helps the subsequent identity verification process; By adjusting the audio quality parameters, the high quality and high fidelity of the final audio samples are ensured, improving the overall performance of the system and user experience.

[0067] The present invention provides a specific embodiment, wherein step 104, based on historical data of user behavior patterns, uses a behavior analysis algorithm to adjust the score weight of the preliminary adjusted comprehensive score to obtain a final adjusted comprehensive score, and performs access permission confirmation based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result, specifically comprising the following steps: Step 401: Based on historical data of user behavior patterns, a risk level assessment is performed on the behavior characteristics of the login request to obtain a risk assessment result, target characteristics are extracted based on the risk assessment result, a user behavior model is constructed based on the target characteristics, and a behavior analysis result is generated based on the user behavior model and abnormal situations are identified, wherein the target characteristics include login time distribution, geographic location changes, and device fingerprints, and the abnormal situations include abnormal login time and geographic location mutations; In this step, the behavior characteristics of the login request refer to the specific behavior characteristics of the current login request, such as login time, geographic location, and device fingerprint used. The user behavior model refers to a mathematical model built based on the target characteristics, which is used to describe the normal behavior pattern of the user. Abnormal situations refer to situations that deviate from the normal behavior pattern of the user, such as abnormal login time and geographic location mutation.

[0068] This step evaluates the risk level of the behavior characteristics of the current login request based on the historical data of the user behavior pattern and obtains the risk assessment result. This process determines the risk level by comparing the current behavior characteristics with the normal patterns in the historical data to ensure that potential risk behaviors can be accurately identified; based on the risk assessment results, extract key target features, such as login time distribution, geographic location changes, and device fingerprints. These features are the basis for building a user behavior model and can fully describe the user's normal behavior pattern; based on the extracted target features, build a user behavior model. By learning the user's historical behavior pattern, the model can predict the user's future behavior and identify any abnormalities, providing a scientific basis for behavior analysis; use the user behavior model to generate behavior analysis results and identify abnormalities, such as abnormal login time and geographic location mutations. These abnormalities may indicate potential security threats, which can help operators discover potential security issues in advance and enhance the security of the system.

[0069] Step 402: Based on the behavior analysis result and the abnormal situation, the score weight of the preliminary adjusted comprehensive score is adjusted using a behavior analysis algorithm to obtain a final adjusted comprehensive score; This step uses the behavioral analysis algorithm to adjust the scoring weights of the preliminary adjusted comprehensive score based on the behavioral analysis results and identified anomalies to obtain the final adjusted comprehensive score. This step ensures that the score more accurately reflects the user's actual behavior patterns, making the score more scientific and reasonable, and improving the accuracy of identity authentication.

[0070] Step 403: setting an access permission threshold according to the existing security policy, and confirming the access permission based on the access permission threshold. If the final adjusted comprehensive score is higher than the access permission threshold, confirming that the access is allowed; if the final adjusted comprehensive score is lower than the access permission threshold, confirming that the access is not allowed, and obtaining a confirmed access result; This step sets the access permission threshold according to the existing security policy. The threshold is used to decide whether to allow the user to access the system. Generally, the threshold setting will consider the balance between the system's security requirements and user experience to ensure that the access permission setting meets security requirements while not affecting the normal user experience. Based on the access permission threshold, access permission is confirmed. If the final adjusted comprehensive score is higher than the access permission threshold, access is confirmed to be allowed; if it is lower than the access permission threshold, access is confirmed to be not allowed. The access confirmation result is obtained to ensure that only verified legitimate users can access the system, thereby improving the security and reliability of the system.

[0071] The embodiments of the present invention can achieve the following beneficial effects through the above steps: Through risk level assessment and behavior analysis, the system can promptly identify potential security threats, such as abnormal login time and geographic location mutations, thus enhancing the security of the system; Combining historical data of user behavior patterns with current behavior characteristics, the system can more accurately assess the user's identity, reducing the possibility of misjudgment; The dynamic adjustment of scoring weights enables the system to flexibly respond to different user behaviors and environmental changes, providing a personalized user experience; All operations are performed between the local device and the cloud, ensuring that user data is not obtained by third parties and protecting user privacy; Through automated behavior analysis and score adjustment, the system can complete identity authentication in a short time, improving response speed and efficiency; Based on user behavior models and behavior analysis algorithms, the system can make more scientific and reasonable decisions to ensure that the score weight adjustment is always in the optimal state.

[0072] Figure 2 A structural diagram of a login system based on a 5G message voiceprint verification code is provided for an embodiment of the present invention, such as Figure 2 As shown, the system includes: A generating module 21 is used to generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; An extraction module 22 is used to obtain a composite audio stream based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, separate the composite audio stream using a blind source separation algorithm, extract the original ambient sound and the one-time random sequence code, combine the original ambient sound and the one-time random sequence code to obtain a preliminary composite voiceprint sample, optimize the preliminary composite voiceprint sample, and obtain a target composite voiceprint sample; The analysis module 23 is used to encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, transmit the encrypted target composite voiceprint sample to the cloud authentication platform in combination with the quantum key distribution mechanism, and obtain the decrypted target composite voiceprint sample after being received and decrypted by the cloud authentication platform. The decrypted target composite voiceprint sample is analyzed by using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features; The adjustment module 24 is used to adjust the score weight of the preliminary adjusted comprehensive score based on the historical data of the user behavior pattern by using the behavior analysis algorithm to obtain the final adjusted comprehensive score, and to confirm the access permission based on the final adjusted comprehensive score to obtain the confirmed access result so that the user can log in according to the confirmed access result.

[0073] Figure 2 The login system based on 5G message voiceprint verification code can be executed Figure 1 The implementation principle and technical effect of the login method based on 5G message voiceprint verification code described in the illustrated embodiment will not be repeated. For a login system based on 5G message voiceprint verification code in the above embodiment, the specific way in which each module and unit performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0074] In one possible design, Figure 2 A login system based on a 5G message voiceprint verification code in the illustrated embodiment can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0075] The processing component 32 is used to: generate a 5G message containing a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receipt of the 5G message, combined with the message reception confirmation feedback, obtain a composite audio stream, use a blind source separation algorithm to separate the composite audio stream, extract the original ambient sound and the one-time random sequence code, combine the original ambient sound and the one-time random sequence code to obtain a preliminary composite voiceprint sample, optimize the preliminary composite voiceprint sample, and obtain a target composite voiceprint sample; encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted The encrypted target composite voiceprint sample is transmitted to the cloud authentication platform in combination with the quantum key distribution mechanism. The decrypted target composite voiceprint sample is received and decrypted by the cloud authentication platform to obtain the decrypted target composite voiceprint sample. The decrypted target composite voiceprint sample is analyzed by using the convolutional neural network, recurrent neural network and generative adversarial network combined model in the deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features. Based on the historical data of user behavior patterns, the scoring weight of the preliminary adjusted comprehensive score is adjusted by using the behavior analysis algorithm to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result.

[0076] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above method.

[0077] The storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0078] Of course, the computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0079] The input / output interface provides an interface between the processing component and the peripheral interface module, which may be an output device, an input device, etc.

[0080] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0081] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0082] The embodiment of the present invention further provides a computer storage medium storing a computer program, which can achieve the above-mentioned Figure 1 A login method based on a 5G message voiceprint verification code in the illustrated embodiment.

[0083] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0084] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0085] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0086] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A login method based on 5G message voiceprint verification code, characterized in that: include: Generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; Based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, a composite audio stream is obtained, the composite audio stream is separated and processed by a blind source separation algorithm, the original ambient sound and the one-time random sequence code are extracted, the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, the preliminary composite voiceprint sample is optimized, and a target composite voiceprint sample is obtained; The target composite voiceprint sample is encrypted based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is transmitted to a cloud authentication platform in combination with a quantum key distribution mechanism, and the decrypted target composite voiceprint sample is obtained after being received and decrypted by the cloud authentication platform, and the decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features; Based on historical data of user behavior patterns, a behavior analysis algorithm is used to adjust the score weight of the preliminary adjusted comprehensive score to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result.

2. The method according to claim 1, characterized in that The target composite voiceprint sample is encrypted based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is transmitted to a cloud authentication platform in combination with a quantum key distribution mechanism. The decrypted target composite voiceprint sample is obtained after being received and decrypted by the cloud authentication platform. The decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features, including: Encrypting the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, and using homomorphic encryption technology in the encryption process to perform a specific type of calculation on the encrypted target composite voiceprint sample, the specific type including additive homomorphic, multiplicative homomorphic and fully homomorphic encryption; The encrypted target composite voiceprint sample is transmitted from the user terminal device to the cloud authentication platform using the quantum key distribution mechanism and the post-quantum cryptography algorithm. During the transmission process, the quantum key distribution generation mechanism is used to add encryption keys to each sample in the encrypted target composite voiceprint sample, and the encrypted target composite voiceprint sample is processed in combination with the post-quantum cryptography algorithm. Based on the cloud authentication platform, the received encrypted target composite voiceprint sample is decrypted, and the decryption process is verified by zero-knowledge proof technology. After the verification process is completed, the decrypted target composite voiceprint sample is obtained; The decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework. During the analysis process, time-frequency domain features are extracted by a convolutional neural network, time series characteristics are extracted by a recurrent neural network, and a synthetic voiceprint sample is generated by a generative adversarial network to obtain an enhanced multimodal voiceprint feature set; Combining the multimodal voiceprint feature set and the user's biological features, constructing a user identity model, performing multimodal data analysis based on the user identity model to obtain multimodal user features, and using a graph neural network to analyze user behavior based on the multimodal user features and historical data of user behavior patterns to obtain user behavior pattern features, wherein the biological features include facial features and fingerprint features; In combination with the user behavior pattern characteristics and the multimodal user characteristics, an initial comprehensive score is calculated, the user identity model is optimized to obtain a comprehensive user identity model, and based on the comprehensive user identity model, the score weight of the initial comprehensive score is dynamically adjusted to obtain a preliminary adjusted comprehensive score.

3. The method according to claim 2, characterized in that In combination with the user behavior pattern characteristics and the multimodal user characteristics, an initial comprehensive score is calculated, the user identity model is optimized to obtain a comprehensive user identity model, and based on the comprehensive user identity model, the score weight of the initial comprehensive score is dynamically adjusted to obtain a preliminary adjusted comprehensive score, including: The user behavior pattern features and the multimodal user features are integrated to obtain a user feature set, based on the user feature set, a preset scoring algorithm is used to preliminarily evaluate the user identity to obtain a user identity evaluation result, and an initial comprehensive score is calculated based on the user identity evaluation result; Based on historical verification results and login requests, an adaptive scoring mechanism is used to dynamically adjust key parameters in a preset scoring algorithm in the process of calculating an initial comprehensive score to obtain an adjusted preliminary score. Based on the adjusted preliminary score, a federated learning framework and a reinforcement learning algorithm are applied to update and optimize the user identity model to obtain a comprehensive user identity model. Based on the comprehensive user identity model, the existing scoring weight adjustment strategy is optimized in combination with Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy. The initial comprehensive score is analyzed by using a variational autoencoder and anomaly detection algorithm, and the abnormal behavior is modeled to evaluate whether the login request is abnormal, thereby obtaining an anomaly detection result; Based on the comprehensive user identity model, the optimized score weight adjustment strategy and the anomaly detection result, the initial comprehensive score is dynamically adjusted to obtain a preliminary adjusted comprehensive score.

4. The method according to claim 3, characterized in that Based on historical verification results and login requests, an adaptive scoring mechanism is used to dynamically adjust key parameters in a preset scoring algorithm in the process of calculating an initial comprehensive score to obtain an adjusted preliminary score. Based on the adjusted preliminary score, a federated learning framework and a reinforcement learning algorithm are applied to update and optimize the user identity model to obtain a comprehensive user identity model. Based on the comprehensive user identity model, the existing scoring weight adjustment strategy is optimized in combination with Bayesian optimization technology to obtain an optimized scoring weight adjustment strategy, including: Based on historical identity authentication records and login request characteristics used to reflect login request situations, an adaptive scoring mechanism is used to adjust key parameters in the preset scoring algorithm in real time to obtain an adjusted preliminary score, wherein the adjusted preliminary score is used to reflect the actual behavior pattern of the user; Optimizing the user behavior pattern features by using a federated learning framework to obtain optimized user behavior pattern features, and performing distributed training and updating on the user identity model based on the optimized user behavior pattern features to obtain a user identity model to be optimized; Based on the user identity model to be optimized, monitor and record the user's behavior, generate user behavior feedback data, select the optimal model parameter adjustment path by applying the reward mechanism in the reinforcement learning algorithm based on the user behavior feedback data, optimize the user identity model to be optimized based on the optimal model parameter adjustment path, and obtain a comprehensive user identity model, wherein the user behavior feedback data includes the user's login behavior, operation frequency, and usage habits; Based on the comprehensive user identity model, the best scoring weight configuration is selected using Bayesian optimization technology, and the existing scoring weight adjustment strategy is optimized based on the best scoring weight configuration to generate an optimized scoring weight adjustment strategy.

5. The method according to claim 4, characterized in that Based on the comprehensive user identity model, the best scoring weight configuration is selected by using Bayesian optimization technology, and the existing scoring weight adjustment strategy is optimized based on the best scoring weight configuration to generate an optimized scoring weight adjustment strategy, including: A probability model is constructed using Bayesian optimization technology, and based on the probability model, the scoring performance results under different scoring weight configurations are predicted. Based on the scoring performance results, an optimal solution is searched, and the optimal solution represents the best scoring weight configuration. In the prediction process, a meta-learning framework is used to adjust the hyperparameters of the probability model according to the scoring performance results and historical optimization experience to accelerate convergence to the optimal solution. In the process of searching for the optimal solution, multi-task learning technology is applied to share the underlying feature representation so that the Bayesian optimization technology can consider multiple related tasks while searching for the optimal solution, including anomaly detection and behavior prediction; Based on the optimal scoring weight configuration, adversarial training is applied to introduce adversarial samples, simulate attack situations, obtain simulation results, and based on the simulation results, optimize the existing scoring weight adjustment strategy to generate an optimized scoring weight adjustment strategy.

6. The method according to claim 1, characterized in that The composite audio stream is separated and processed by using a blind source separation algorithm, the original ambient sound and the one-time random sequence code are extracted, the original ambient sound and the one-time random sequence code are combined to obtain a preliminary composite voiceprint sample, and the preliminary composite voiceprint sample is optimized to obtain a target composite voiceprint sample, including: The composite audio stream is separated by using a blind source separation algorithm to extract the original ambient sound and the one-time random sequence code, and during the separation process, a separation matrix is ​​initialized and iteratively optimized to improve the restoration degree of the original ambient sound and the one-time random sequence code; Synchronously aligning the original ambient sound and the one-time random sequence code to obtain the aligned original ambient sound and the one-time random sequence code, and performing superposition synthesis processing on the aligned original ambient sound and the one-time random sequence code to generate a preliminary composite voiceprint sample, wherein the preliminary composite voiceprint sample contains the real sound information of the user environment and the one-time random sequence code; Based on the preliminary composite voiceprint sample, denoising is performed using a noise suppression algorithm to obtain a denoised preliminary composite voiceprint sample; based on the denoised preliminary composite voiceprint sample, feature enhancement technology is used to highlight key features to obtain a target composite voiceprint sample to be adjusted; and audio quality parameters of the target composite voiceprint sample to be adjusted are adjusted to generate a target composite voiceprint sample, wherein the audio quality parameters include sampling rate and number of quantization bits.

7. The method according to claim 3, characterized in that Based on historical data of user behavior patterns, the score weight of the preliminary adjusted comprehensive score is adjusted using a behavior analysis algorithm to obtain a final adjusted comprehensive score, and access permission is confirmed based on the final adjusted comprehensive score to obtain a confirmed access result, so that the user can log in according to the confirmed access result, including: Based on historical data of user behavior patterns, a risk level assessment is performed on the behavior characteristics of the login request to obtain a risk assessment result, target characteristics are extracted based on the risk assessment result, a user behavior model is constructed based on the target characteristics, and based on the user behavior model, a behavior analysis result is generated and abnormal situations are identified, wherein the target characteristics include login time distribution, geographic location changes, and device fingerprints, and the abnormal situations include abnormal login time and geographic location mutations; Based on the behavior analysis results and the abnormal situation, adjusting the score weight of the preliminary adjusted comprehensive score using a behavior analysis algorithm to obtain a final adjusted comprehensive score; An access permission threshold is set according to an existing security policy, and access permission is confirmed based on the access permission threshold. If the final adjusted comprehensive score is higher than the access permission threshold, access is confirmed to be allowed; if the final adjusted comprehensive score is lower than the access permission threshold, access is confirmed to be not allowed, thereby obtaining a confirmed access result.

8. A login system based on 5G message voiceprint verification code, characterized in that: include: A generation module, used to generate a 5G message including a one-time random sequence code, send the 5G message to a user terminal device, and receive a message reception confirmation feedback from the user terminal device; An extraction module is used to obtain a composite audio stream based on the ambient sound of the surrounding environment collected by the local voiceprint collection device when the user terminal device confirms receiving the 5G message, combined with the message reception confirmation feedback, separate the composite audio stream using a blind source separation algorithm, extract the original ambient sound and the one-time random sequence code, combine the original ambient sound and the one-time random sequence code to obtain a preliminary composite voiceprint sample, optimize the preliminary composite voiceprint sample, and obtain a target composite voiceprint sample; An analysis module is used to encrypt the target composite voiceprint sample based on a preset encryption protocol to obtain an encrypted target composite voiceprint sample, transmit the encrypted target composite voiceprint sample to a cloud authentication platform in combination with a quantum key distribution mechanism, and obtain a decrypted target composite voiceprint sample after the cloud authentication platform receives and decrypts the sample. The decrypted target composite voiceprint sample is analyzed using a convolutional neural network, a recurrent neural network, and a generative adversarial network combined model in a deep learning framework to extract time-frequency domain features, and a preliminary adjusted comprehensive score is calculated based on the time-frequency domain features; The adjustment module is used to adjust the score weight of the preliminary adjusted comprehensive score based on the historical data of the user behavior pattern by using the behavior analysis algorithm to obtain the final adjusted comprehensive score, and to confirm the access permission based on the final adjusted comprehensive score to obtain the confirmed access result so that the user can log in according to the confirmed access result.

9. A computing device, characterized in that It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a login method based on a 5G message voiceprint verification code as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a computer, a login method based on a 5G message voiceprint verification code as described in any one of claims 1 to 7 is implemented.

Images