Multi-factor security authentication method and device based on AI adaptive recognition and blockchain

Abstract

The embodiment of the present application provides a multi-factor security authentication method and device based on AI adaptive recognition and blockchain. By collecting user biometrics, behavioral data and device information, convolutional neural networks and long short-term memory networks are used to extract static and dynamic features respectively. A feature fusion network is innovatively constructed, and an attention mechanism is introduced to dynamically weight and fuse different features, and feature correction is performed based on the user behavior baseline model. Blockchain networks and smart contracts are deployed at the same time to achieve trusted storage of authentication data and risk level assessment. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

Description

Technical Field

[0001] The present application relates to the field of data processing, and specifically to a multi-factor security authentication method and device based on AI adaptive recognition and blockchain. Background Art

[0002] Traditional identity authentication technologies rely on a single authentication method, such as passwords or fingerprints, which poses a great security risk. With the continuous escalation of cyber attack methods, single-dimensional authentication mechanisms have become difficult to cope with complex security threats. Although existing multi-factor authentication schemes have added verification links, they often adopt a simple superposition method, lack dynamic evaluation of the importance of different authentication factors, and fail to fully utilize user behavior characteristics for risk identification.

[0003] Especially in terms of trusted storage of authentication data, traditional centralized architectures are easy targets for attack, and once data leaks occur, serious consequences will occur. At the same time, existing authentication systems generally lack adaptive capabilities and cannot dynamically adjust security policies based on user behavior patterns and environmental changes, resulting in the authentication process being either too strict, affecting user experience, or too loose, bringing security risks.

[0004] In addition, with the continuous evolution of identity fraud and attack methods, how to ensure the credibility and non-tamperability of the authentication process and achieve secure storage and effective traceability of authentication data has become an important challenge facing the current identity authentication field. Solving these problems is of great significance to building a safer and smarter identity authentication system. Summary of the invention

[0005] In response to the problems in the prior art, the present application provides a multi-factor security authentication method and device based on AI adaptive recognition and blockchain, which can break through the limitations of traditional single authentication, achieve an effective balance between security and user experience, and provide advanced technical solutions for the field of identity authentication.

[0006] In order to solve at least one of the above problems, the present application provides the following technical solutions:

[0007] In the first aspect, the present application provides a multi-factor security authentication method based on AI adaptive recognition and blockchain, including:

[0008] Collecting user biometric data, user behavior data and device information, performing liveness detection and obtaining user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, obtaining hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0009] Constructing a feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, inputting the first feature vector, the second feature vector and the device information into the feature fusion network, sorting the importance and assigning weights to the input features based on the attention mechanism layer, weightedly fusing features with different weights to obtain a fused feature vector, constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation;

[0010] A blockchain network is constructed, and permission chain nodes and smart contracts are deployed in the blockchain network. The corrected fused feature vector is input into the smart contract for identity authentication, and the authentication confidence is calculated to obtain the risk level. When the risk level is within a preset range, the authentication data is encrypted by a hardware security module, and the encrypted data is written into the blockchain network to complete the authentication process.

[0011] Furthermore, the step of inputting the biometric data into a pre-trained convolutional neural network to extract static features to obtain a first feature vector includes:

[0012] Preprocessing the biometric data, normalizing the biometric data to obtain standardized image data with pixel values ​​between 0 and 1, using an image enhancement algorithm based on a multi-scale convolution kernel to perform spatial feature enhancement on the standardized image data, and performing feature dimension reduction on the enhanced image data through a maximum pooling operation to obtain preprocessed image data;

[0013] The preprocessed image data is input into a convolutional neural network comprising multiple convolutional layers and a fully connected layer, a 3×3 convolution kernel and a ReLU activation function are set in the convolutional layer to extract local features of the image, and the extracted local features are mapped to a 128-dimensional feature space through the fully connected layer to obtain the first feature vector.

[0014] Furthermore, the step of inputting the user behavior data into a pre-trained long short-term memory network to perform dynamic feature extraction to obtain a second feature vector includes:

[0015] The user behavior data is sampled and segmented according to the time series, the user's continuous operation information is divided into time windows of fixed length, the behavior data in each time window is numerically encoded and standardized, and a behavior feature sequence including the user operation sequence, operation time interval and operation duration is generated;

[0016] The behavioral feature sequence is input into a long short-term memory network with a forget gate and an input gate, the historical state information is filtered through the forget gate, the importance of the current input information is controlled through the input gate, the state information is nonlinearly transformed using the tanh activation function, and the transformed features are mapped to a 64-dimensional feature space to obtain the second feature vector.

[0017] Furthermore, the constructing of the feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, and inputting the first feature vector, the second feature vector and the device information into the feature fusion network, includes:

[0018] Constructing a feature fusion network with a three-branch network structure, each branch includes two fully connected layers for feature dimension reduction and mapping, inputting the first feature vector, the second feature vector and the encoded device information into corresponding network branches respectively, and normalizing the features through a batch normalization layer to obtain initial fusion features;

[0019] A multi-head attention mechanism layer is constructed in the feature fusion network, and a query matrix, a key matrix and a value matrix are calculated for the initial fusion feature. The correlation between different features is calculated by scaling the dot product attention to obtain the attention weight, and the attention weight is weightedly summed with the corresponding feature to obtain the fusion feature.

[0020] Furthermore, the importance of input features is sorted and weighted based on the attention mechanism layer, and features with different weights are weightedly fused to obtain a fused feature vector, including:

[0021] Normalizing the attention weights to obtain importance scores, sorting the input features in descending order according to the importance scores, converting the sorting results into probability distributions using a soft maximum function, and assigning initial fusion weights to different features based on the probability distributions;

[0022] The initial fusion weight is multiplied by the corresponding feature, all weighted features are fused through a weighted average pooling operation, and the fusion result is mapped to a feature space of uniform dimension using a nonlinear transformation to obtain the fused feature vector.

[0023] Furthermore, the user behavior baseline model is constructed based on the historical authentication data, the degree of deviation between the current authentication behavior and the baseline model is calculated, and the fused feature vector is corrected according to the degree of deviation, including:

[0024] Extract the user's authentication time distribution, operation behavior sequence and geographic location information from the historical authentication data, use the Gaussian mixture model to perform probability modeling on these features, obtain the probability density function of the user behavior through maximum likelihood estimation, and use the probability density function as the user behavior baseline model;

[0025] The current authentication behavior is input into the baseline model to calculate the behavior probability value, the deviation coefficient is calculated based on the difference between the behavior probability value and the preset threshold, and an adaptive weight adjustment algorithm is used to dynamically correct the components of each dimension of the fused feature vector according to the deviation coefficient to obtain the corrected fused feature vector.

[0026] Furthermore, the corrected fusion feature vector is input into the smart contract for identity authentication, the authentication confidence is calculated to obtain the risk level, and when the risk level is within a preset range, the authentication data is encrypted by the hardware security module, and the encrypted data is written into the blockchain network to complete the authentication process, including:

[0027] Constructing a multi-layer neural network authentication model in the smart contract, inputting the corrected fusion feature vector into the authentication model to calculate identity similarity, calculating authentication confidence based on the similarity score, and mapping the authentication confidence to a predefined risk level interval to obtain the current authentication risk level;

[0028] The authentication data is encrypted using the elliptic curve encryption algorithm through the encryption chip built into the hardware security module, a transaction block containing the encrypted data is generated according to the consensus mechanism of the blockchain network, the transaction block is broadcast to the permission chain node in the blockchain network and waits for confirmation to complete the authentication process.

[0029] In the second aspect, the present application provides a multi-factor security authentication device based on AI adaptive recognition and blockchain, including:

[0030] A feature extraction module, used to collect user biometric data, user behavior data and device information, perform liveness detection and obtain user facial image data when collecting the biometric data, record user continuous operation information when collecting the user behavior data, obtain hardware identification and geographic location information when collecting the device information, input the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and input the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0031] A feature fusion module, used to construct a feature fusion network, set a fully connected layer and an attention mechanism layer in the feature fusion network, input the first feature vector, the second feature vector and the device information into the feature fusion network, sort the importance and assign weights to the input features based on the attention mechanism layer, perform weighted fusion on features with different weights to obtain a fused feature vector, build a user behavior baseline model based on historical authentication data, calculate the degree of deviation between the current authentication behavior and the baseline model, and correct the fused feature vector according to the degree of deviation;

[0032] The security encryption module is used to build a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0033] In a third aspect, the present application provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain are implemented.

[0034] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain.

[0035] In a fifth aspect, the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain.

[0036] It can be seen from the above technical solutions that the present application provides a multi-factor security authentication method and device based on AI adaptive recognition and blockchain. By collecting user biometrics, behavioral data and device information, convolutional neural networks and long short-term memory networks are used to extract static and dynamic features respectively. A feature fusion network is innovatively constructed, and an attention mechanism is introduced to dynamically weight and fuse different features, and feature correction is performed based on the user behavior baseline model. Blockchain networks and smart contracts are deployed at the same time to achieve trusted storage and risk level assessment of authentication data. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is one of the flow charts of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0039] Figure 2 This is the second flow chart of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0040] Figure 3 This is the third flow chart of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0041] Figure 4 This is a fourth flow chart of a multi-factor security authentication method based on AI adaptive recognition and blockchain in an embodiment of the present application;

[0042] Figure 5 This is the fifth flow chart of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0043] Figure 6 This is the sixth flow chart of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0044] Figure 7 This is the seventh flow chart of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the embodiment of the present application;

[0045] Figure 8 This is a structural diagram of a multi-factor security authentication device based on AI adaptive recognition and blockchain in an embodiment of the present application;

[0046] Fig. 9 It is a schematic diagram of the structure of an electronic device in an embodiment of the present application.

[0047] Reference numerals:

[0048] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver program storage unit 9144, antenna 9111, speaker 9131, microphone 9132. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0050] The acquisition, storage, use, and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0051] Taking into account the problems existing in the prior art, this application provides a multi-factor security authentication method and device based on AI adaptive recognition and blockchain. By collecting user biometrics, behavioral data and device information, convolutional neural networks and long short-term memory networks are used to extract static and dynamic features respectively. A feature fusion network is innovatively constructed, and an attention mechanism is introduced to dynamically weight and fuse different features, and feature correction is performed based on the user behavior baseline model. Blockchain networks and smart contracts are deployed at the same time to achieve trusted storage of authentication data and risk level assessment. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0052] In order to break through the limitations of traditional single authentication, achieve an effective balance between security and user experience, and provide advanced technical solutions for the field of identity authentication, this application provides an embodiment of a multi-factor security authentication method based on AI adaptive recognition and blockchain, see Figure 1 The multi-factor security authentication method based on AI adaptive recognition and blockchain specifically includes the following contents:

[0053] Step S101: collecting user biometric data, user behavior data and device information, performing liveness detection and acquiring user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, acquiring hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0054] Optionally, this embodiment designs a multi-dimensional user feature collection and extraction solution. In the biometric data collection stage, the front camera collects the user's facial image and combines it with the infrared depth sensor to obtain three-dimensional spatial information. Liveness detection adopts a multimodal fusion strategy, which effectively prevents deception methods such as photo and video playback by detecting natural movements such as blinking and opening the mouth, combined with facial texture analysis and depth information.

[0055] In the facial image processing process, this embodiment first performs face detection and key point positioning to determine the precise position and contour features of the facial area. Then, illumination compensation and posture normalization are performed to ensure the stability of subsequent feature extraction. The pre-trained convolutional neural network adopts an improved ResNet structure, which effectively solves the problem of deep network training difficulties through residual connections and improves the accuracy of feature extraction.

[0056] The collection of user behavior data covers multiple dimensions of user-device interaction. This embodiment records the user's touch operation characteristics, including fine-grained information such as sliding trajectory, pressing force and duration. At the same time, it captures the user's application usage habits, such as the order of opening commonly used applications, usage time distribution and other time series characteristics. After these behavior data are preprocessed in time series, they are input into the long short-term memory network for feature learning.

[0057] The device information collection adopts a multi-layer verification mechanism. This embodiment obtains the inherent identification of the device such as the hardware serial number and MAC address, and collects the network environment characteristics of the device. The geographic location information uses a combination of GPS, base station positioning and WiFi positioning to improve the accuracy of location determination. In indoor scenarios, Bluetooth beacons can also be combined to achieve more accurate spatial positioning.

[0058] In the static feature extraction process, this embodiment introduces a spatial pyramid pooling module in the convolutional neural network, which can process input images of different scales. An attention module is set in the middle layer of the network to focus on the key regional features of the face. The feature vector finally obtained has good discriminability and stability.

[0059] Dynamic feature extraction uses the gating mechanism of the long short-term memory network to effectively capture the long-term dependencies in the user's behavior pattern. This embodiment adds a time attention layer to the network structure, which can adaptively determine the importance of behavioral features at different time points. At the same time, a forgetting mechanism is introduced to timely update the changes in the user's behavior pattern.

[0060] A typical application scenario of this embodiment is mobile payment security authentication. When a user makes a large payment operation, the system will collect facial features, payment operation behavior and device environment information at the same time. Through comprehensive analysis of multi-dimensional features, it is possible to effectively identify whether it is the normal transaction behavior of the user himself. Even if the facial features are similar, if the behavior pattern or device environment is abnormal, risks can be discovered in time.

[0061] In the enterprise access control scenario, this embodiment can be deeply integrated with the access control system. In addition to traditional facial recognition, continuous authentication can be performed based on the employee's operating habits and the characteristics of the device they hold. This multi-factor dynamic authentication method not only improves security, but also provides a smooth experience for employees.

[0062] Through the collaborative analysis of multi-dimensional features, this embodiment significantly improves the security and reliability of identity authentication. Especially in high-risk scenarios such as remote office and mobile payment, it can effectively prevent various identity theft and fraud. At the same time, the real-time analysis of dynamic features also provides strong support for early warning of abnormal behavior.

[0063] Step S102: constructing a feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, inputting the first feature vector, the second feature vector and the device information into the feature fusion network, sorting the importance and assigning weights to the input features based on the attention mechanism layer, weightedly fusion of features with different weights to obtain a fused feature vector, constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation;

[0064] Optionally, this embodiment designs an adaptive feature fusion architecture to achieve dynamic weight adjustment and feature fusion according to the importance differences of various features in different scenarios. The feature fusion network adopts a three-branch structure to process biometric features, behavioral features and device information respectively. Each branch contains two fully connected layers for feature dimensionality reduction and nonlinear mapping.

[0065] In the feature preprocessing stage, this embodiment normalizes the first feature vector (biometric feature) and the second feature vector (behavioral feature) to ensure that the numerical ranges of features from different sources are consistent. After independent encoding, the device information is converted into a fixed-dimensional vector representation, which contains key information such as device identification and geographic location.

[0066] The attention mechanism layer uses a multi-head self-attention structure to calculate the correlation between different features. This embodiment first converts the input features into a query vector, a key vector, and a value vector, and calculates the attention score through a scaled dot product operation. In a financial transaction scenario, when a user performs operations at an unconventional time or place, the attention mechanism will automatically increase the weight of behavioral features and device information, and enhance the ability to identify abnormal behaviors.

[0067] The feature importance ranking adopts a soft attention mechanism, and the attention score is converted into a probability distribution through a softmax function. This embodiment sets the basic weight according to the application scenario, such as increasing the basic weight of biometric features in high-security scenarios and appropriately increasing the weight of device features in mobile scenarios. The final fusion weight is determined by the basic weight and the attention weight.

[0068] The user behavior baseline model is built based on historical authentication data and uses probability density estimation to characterize the user's normal behavior pattern. This embodiment uses kernel density estimation technology to establish a multidimensional probability distribution model for user operation time, geographic location, and behavior sequence. In the corporate office scenario, the baseline model can accurately describe the regular work mode of employees, including working hours, office location, and operating habits.

[0069] The deviation degree calculation adopts an adaptive threshold strategy. This embodiment calculates the standardized deviation score by comparing the probability difference between the current behavior characteristics and the baseline model. In the remote office scenario, the system will consider the change of the employee's temporary office location and appropriately adjust the tolerance of the deviation calculation.

[0070] The feature vector correction adopts a dynamic weighting mechanism. When a large behavioral deviation is detected, this embodiment will reduce the weight contribution of the corresponding feature while increasing the importance of features in other dimensions. For example, when an abnormal user location is detected, the system will increase the weight ratio of the biometric feature to ensure the reliability of identity authentication.

[0071] In practical applications, this embodiment can effectively handle the situation of missing features. When some features cannot be obtained, the fusion network will automatically adjust the weights of other available features to maintain the stability of the authentication system. For example, when the quality of facial features is reduced due to insufficient light, the system will rely more on behavioral features and device information for identity authentication.

[0072] The dynamic adaptive nature of feature fusion enables this embodiment to adapt to different application scenarios and security requirements. In daily mobile payment scenarios, the system can balance user experience and security; in high-risk transaction scenarios, it will automatically raise the authentication standard and require stricter feature matching to effectively prevent potential identity fraud risks.

[0073] This embodiment significantly improves the accuracy and reliability of identity authentication through intelligent fusion and dynamic correction of multi-dimensional features, while maintaining a good user experience. Especially in complex enterprise security environments, it can effectively respond to various abnormal situations and attack threats.

[0074] Step S103: construct a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0075] Optionally, this embodiment builds an identity authentication network based on the alliance chain architecture, adopts a distributed node deployment method to ensure the security and traceability of authentication data. The access mechanism of the permission chain node adopts the PKI certificate system, and realizes identity authorization and access control through certificate issuance and revocation list management.

[0076] Node deployment adopts a layered architecture, including consensus nodes, verification nodes, and ordinary nodes. This embodiment deploys an improved PBFT consensus algorithm on the consensus node and implements transaction confirmation through a voting mechanism. The verification node is responsible for executing smart contracts and data verification, and the ordinary node is used for data synchronization and query. In enterprise application scenarios, each department can participate in the authentication process as an independent permission chain node.

[0077] The design of the smart contract adopts a modular structure, including a feature verification module, a risk assessment module, and a data storage module. This embodiment implements a multi-dimensional feature matching algorithm in the feature verification module, and performs identity authentication through a configurable similarity threshold. For example, when processing the fused feature vector, feature decomposition is first performed to calculate the matching degree of biometric features, behavioral features, and device features respectively.

[0078] The authentication confidence is calculated using a multi-layer neural network model to map the fused feature vector to the risk scoring space. This embodiment adds a time decay factor to the model so that the impact of historical authentication records decreases over time. In financial transaction scenarios, the system will dynamically adjust the confidence threshold based on the transaction amount and adopt stricter verification standards for large transactions.

[0079] The risk level assessment adopts a grading strategy to divide the risk into three levels: low, medium, and high. This embodiment designs an adaptive risk threshold adjustment mechanism to dynamically adjust the threshold range according to the security requirements of different scenarios. In the remote office scenario, when an abnormal login location is detected, the system will increase the risk level and require additional verification steps.

[0080] The hardware security module uses the Trusted Execution Environment (TEE) technology to achieve secure storage and processing of authentication data. This embodiment uses a hardware encryption chip to generate a unique device key, combined with a hierarchical key management mechanism to ensure the security of the key. The authentication data is encrypted in the TEE environment, using the AES-256 algorithm for symmetric encryption, and the key is protected by an asymmetric encryption algorithm.

[0081] The blockchain data is written using a sharding storage strategy to improve the scalability and performance of the system. This embodiment designs data sharding rules to store authentication data in shards according to time and user dimensions. Each data block contains information such as authentication timestamp, encrypted feature data, and risk level assessment results. In the enterprise identity management scenario, cross-departmental authentication data sharing and auditing can be achieved.

[0082] The transaction processing of the authentication process adopts a two-phase commit protocol to ensure data consistency. This embodiment implements a transaction rollback mechanism in the smart contract, which can promptly cancel related operations when an exception occurs during the authentication process. For example, when a potential attack is detected, the system will trigger a transaction rollback and record the exception information on the blockchain.

[0083] Through the application of blockchain technology, this embodiment realizes the immutability and traceability of authentication data. In the data sharing scenario, each participant can access data in a controllable manner based on smart contracts. Especially in the cross-institutional authentication scenario, the blockchain network provides a trusted data exchange environment.

[0084] This embodiment adopts an asynchronous processing mechanism in high-concurrency scenarios, buffering authentication requests through message queues to ensure system stability. For urgent authentication needs, a fast-track mechanism is designed to prioritize high-priority authentication requests. In large-scale enterprise applications, it can effectively support a hybrid deployment mode of centralized authentication and distributed authentication.

[0085] This implementation plan builds a secure and reliable identity authentication infrastructure through the combination of blockchain technology and hardware security modules, which can effectively prevent security threats such as identity fraud and data tampering, while supporting flexible business expansion and cross-organizational collaboration.

[0086] From the above description, it can be seen that the multi-factor security authentication method based on AI adaptive recognition and blockchain provided in the embodiment of the present application can extract static and dynamic features by collecting user biometrics, behavioral data and device information, respectively using convolutional neural networks and long short-term memory networks. Innovatively construct a feature fusion network, introduce an attention mechanism to dynamically weight and fuse different features, and perform feature correction based on the user behavior baseline model. At the same time, blockchain networks and smart contracts are deployed to achieve trusted storage and risk level assessment of authentication data. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0087] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 2 , and can also include the following:

[0088] Step S201: preprocessing the biometric data, normalizing the biometric data to obtain standardized image data with pixel values ​​between 0 and 1, using an image enhancement algorithm based on a multi-scale convolution kernel to perform spatial feature enhancement on the standardized image data, and performing feature dimension reduction on the enhanced image data through a maximum pooling operation to obtain preprocessed image data;

[0089] Step S202: Input the preprocessed image data into a convolutional neural network comprising multiple convolutional layers and a fully connected layer, set a 3×3 convolution kernel and a ReLU activation function in the convolutional layer to extract local features of the image, and map the extracted local features to a 128-dimensional feature space through the fully connected layer to obtain the first feature vector.

[0090] Optionally, this embodiment designs a complete set of image preprocessing and feature extraction processes for biometric data processing. In the preprocessing stage, the pixel values ​​of the input biometric data are first normalized, and the Min-Max normalization method is used to map the pixel values ​​to the [0,1] interval to reduce the difference in data distribution and improve the stability of subsequent processing.

[0091] Image enhancement uses a multi-scale convolution kernel structure, including convolution kernels of three different scales: 3×3, 5×5, and 7×7. This embodiment extracts spatial features of different scales through parallel convolution branches. Smaller convolution kernels focus on capturing local detail features, while larger convolution kernels focus on larger structural information. In face recognition scenarios, this multi-scale feature extraction method can retain both facial texture details and contour structure information.

[0092] In order to deal with the difference in image quality under different lighting conditions, this embodiment introduces adaptive histogram equalization processing before multi-scale convolution. By processing the image in blocks and limiting the amplitude of contrast enhancement, the image quality in low-light environments is effectively improved, while avoiding the problem of noise amplification caused by over-enhancement.

[0093] The maximum pooling operation uses a 2×2 sliding window with a step size of 2 to achieve dimensionality reduction processing of the feature map. This embodiment adds a batch normalization layer before the pooling layer to accelerate the network training process by standardizing the feature distribution. In the remote office scenario, this feature dimensionality reduction method can effectively extract significant features in the camera-captured images and reduce computational complexity.

[0094] The convolutional neural network adopts an improved VGG structure, which includes 5 convolution blocks and 3 fully connected layers. This embodiment uses multiple layers of 3×3 convolution kernels in each convolution block for feature extraction, which has fewer parameters and better feature extraction capabilities than single-layer large-size convolution kernels. The use of ReLU activation function solves the gradient vanishing problem in deep network training.

[0095] In the feature extraction process, this embodiment designs a residual connection structure to directly transfer shallow features to the deep network. This skip connection mechanism can retain more original image information and help extract richer identity features. In the mobile payment scenario, the user's facial features can be accurately extracted even in complex backgrounds.

[0096] The fully connected layer adopts a dimensionality reduction strategy to compress high-dimensional features into a 128-dimensional feature space through multi-layer nonlinear mapping. In this embodiment, a dropout layer is added before the last fully connected layer to randomly discard a certain proportion of neurons to enhance the generalization ability of the model. The final feature vector has strong discriminability and can effectively distinguish the identity characteristics of different users.

[0097] During the model training phase, this embodiment adopts an online difficult example mining strategy to dynamically adjust the weights of training samples. For easily confused facial features, the sampling frequency is increased during the training process to improve the model's ability to distinguish similar faces. In the corporate attendance scenario, even employees with similar appearances can be accurately identified.

[0098] Through this carefully designed feature processing flow, this embodiment achieves efficient and stable biometric feature extraction. In particular, in actual application environments, it can cope with complex situations such as changes in lighting and posture, and provide reliable feature support for subsequent identity authentication. While ensuring the accuracy of authentication, this solution also achieves low computational complexity and is suitable for deployment on mobile devices.

[0099] This embodiment significantly improves the robustness and accuracy of biometric recognition by combining multi-scale feature extraction and deep learning models. In various practical application scenarios, it can quickly and accurately extract the identity features of users, laying the foundation for safe and reliable identity authentication.

[0100] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 3 , and can also include the following:

[0101] Step S301: sampling and segmenting the user behavior data according to the time series, dividing the user's continuous operation information into time windows of fixed length, numerically encoding and standardizing the behavior data in each time window, and generating a behavior feature sequence including the user operation sequence, operation time interval and operation duration;

[0102] Step S302: Input the behavior feature sequence into a long short-term memory network with a forget gate and an input gate, filter the historical state information through the forget gate, control the importance of the current input information through the input gate, use the tanh activation function to perform a nonlinear transformation on the state information, and map the transformed features to a 64-dimensional feature space to obtain the second feature vector.

[0103] Optionally, this embodiment implements a complete user behavior analysis solution, which accurately describes the behavior pattern by modeling the temporal characteristics of user operation behavior. First, the user behavior data is divided into time windows, the window length is set to 30 seconds, and the overlap rate of adjacent windows is 50%, which not only ensures the continuity of the behavior sequence, but also captures local behavior characteristics.

[0104] In the digital coding stage, this embodiment uses one-hot encoding to convert discrete operation types into vector representations. Common user operations such as click, slide, input, etc. are assigned unique codes. The operation time interval is processed by logarithmic transformation, mapping the millisecond time difference to a more appropriate numerical range, effectively dealing with the problem of uneven time interval distribution.

[0105] The extraction of duration features takes into account the characteristics of different operation types. For click operations, this embodiment records the time difference from pressing to releasing; for sliding operations, the duration of the entire sliding trajectory is counted; for input operations, the completion time of a single input is calculated. This fine-grained time feature characterization provides an important basis for subsequent behavioral pattern recognition.

[0106] The normalization of the behavior feature sequence adopts a sliding window normalization method, and the features in each time window are independently normalized. This embodiment retains the relative order and time relationship of user operations during the processing, and eliminates the numerical differences caused by different user operation habits.

[0107] The design of the long short-term memory network (LSTM) adopts a two-layer structure, each layer contains 128 hidden units. This embodiment introduces an adaptive threshold mechanism in the forget gate to dynamically adjust the forget ratio according to the importance of the historical state. For example, when a significant change in the user's operation mode is detected, the threshold of the forget gate is increased to quickly update the internal state.

[0108] The design of the input gate uses an attention mechanism to assign different weights to input information at different times. This embodiment adaptively adjusts the importance of input information by calculating the correlation between the current input and the historical pattern. In the identity authentication scenario, this mechanism can highlight the key operation sequence with identity recognition value.

[0109] During the state update process, the tanh activation function is used for nonlinear transformation to compress the state value to the interval [-1,1]. This embodiment adds residual connection to the state calculation to alleviate the gradient vanishing problem in long sequence processing. This design shows good feature extraction ability when processing longer operation sequences.

[0110] The feature mapping stage uses a multi-layer perceptron structure to compress the output features of LSTM into a 64-dimensional space through two fully connected layers. This embodiment retains the key features of temporal behavior during the feature compression process, and the generated feature vector has a strong user identity discrimination capability.

[0111] In practical applications, this embodiment can effectively identify the user's operating habits and behavior patterns. For example, in the mobile payment scenario, by analyzing the user's sliding trajectory, key rhythm and other characteristics, it can accurately identify whether it is the user's own operation. In the enterprise office system, combined with the user's access path and operation timing characteristics, abnormal account usage behavior can be discovered in a timely manner.

[0112] This solution uses deep learning technology to accurately model user behavior, ensuring authentication accuracy while having strong real-time and adaptability. Especially in complex application environments, it can quickly adapt to the dynamic changes in user behavior patterns and provide reliable identity feature support.

[0113] Through this behavior analysis solution, this embodiment realizes continuous verification of user identity. Combined with biometric recognition, a multi-dimensional identity authentication system is constructed, which significantly improves the security and reliability of identity authentication.

[0114] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 4 , and can also include the following:

[0115] Step S401: construct a feature fusion network with a three-branch network structure, each branch includes two fully connected layers for feature dimension reduction and mapping, the first feature vector, the second feature vector and the encoded device information are respectively input into the corresponding network branches, and the features are normalized by a batch normalization layer to obtain initial fusion features;

[0116] Step S402: construct a multi-head attention mechanism layer in the feature fusion network, calculate the query matrix, key matrix and value matrix for the initial fusion feature, calculate the correlation between different features by scaling the dot product attention to obtain the attention weight, and perform weighted summation of the attention weight and the corresponding feature to obtain the fusion feature.

[0117] Optionally, this embodiment designs an innovative multimodal feature fusion solution to achieve effective fusion of biometric features, behavioral features, and device features through a three-branch network structure. In the feature input stage, the first branch processes a 128-dimensional biometric feature vector, the second branch processes a 64-dimensional behavioral feature vector, and the third branch processes the feature vector after device information encoding.

[0118] The fully connected layer of each branch adopts a progressive dimensionality reduction structure. The first layer reduces the input features to half of the original dimension, and the second layer further compresses the features into a 32-dimensional unified space. This embodiment adds a ReLU activation function between the fully connected layers to enhance the nonlinear expression ability of the network. The introduction of the batch normalization layer effectively reduces the distribution differences between different feature branches and improves the stability of feature fusion.

[0119] In order to process device information, this embodiment designs a special encoding scheme. For discrete information such as device model and operating system version, an embedding layer is used to convert them into dense vector representation. For continuous values ​​such as device performance indicators and network status, they are mapped to a suitable numerical range through linear transformation. This encoding method retains the semantic information of device features.

[0120] In the feature fusion stage, this embodiment adopts an 8-head attention mechanism, and each attention head independently learns the association pattern between different features. By dividing the 32-dimensional feature vector into 8 4-dimensional subspaces, each attention head calculates the feature correlation in the corresponding subspace, realizing multi-angle feature correlation analysis.

[0121] The query matrix, key matrix and value matrix are constructed by linear transformation, and the transformation matrix is ​​obtained by back propagation optimization. In this embodiment, a temperature parameter is introduced when calculating the attention weight, and the problem of too sharp attention distribution is avoided by adjusting the smoothness of the softmax function.

[0122] In practical applications, this embodiment can effectively handle identity authentication requirements in different scenarios. For example, in a remote office environment, when a user logs in with a new device, the system will rely more on biometrics and behavioral features for identity authentication. On the devices that users use daily, the three features will be more evenly weighted.

[0123] The multi-head design of the attention mechanism enables the system to capture feature associations at different levels. In the mobile payment scenario, some attention heads may focus on the consistency of biometric and behavioral features, while other attention heads are responsible for verifying the match between device features and user history.

[0124] This embodiment obtains the final fusion feature by weighted summation, and the weight value reflects the importance of different features in the current authentication scenario. This adaptive feature fusion method can dynamically adjust the contribution ratio of different features according to specific application scenarios and security requirements.

[0125] In the enterprise security office scenario, the feature fusion solution of this embodiment shows excellent adaptability. When abnormal login behavior is detected, the system automatically increases the weight of the biometric feature to improve the reliability of identity authentication. In a normal office environment, the weights of the three features will tend to be balanced, providing a smoother user experience.

[0126] Through this multimodal feature fusion solution, this embodiment achieves more comprehensive and reliable identity authentication. Especially in complex application environments, it can effectively resist various identity forgery attacks while maintaining good user friendliness. The adaptive characteristics of this solution enable it to flexibly respond to security requirements in different scenarios, providing strong technical support for the identity authentication system.

[0127] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 5 , and can also include the following:

[0128] Step S501: normalizing the attention weights to obtain importance scores, sorting the input features in descending order according to the importance scores, converting the sorting results into probability distributions using a soft maximum function, and assigning initial fusion weights to different features based on the probability distributions;

[0129] Step S502: multiply the initial fusion weight by the corresponding feature, fuse all weighted features through weighted average pooling operation, and use nonlinear transformation to map the fusion result to a feature space of uniform dimension to obtain the fused feature vector.

[0130] Optionally, this embodiment designs an adaptive feature fusion weight allocation scheme, which realizes accurate quantification of feature importance by deeply analyzing and processing the output of the attention mechanism. First, the attention weights are normalized by L2 norm, and the weight values ​​are scaled to the [0,1] interval to ensure the comparability of different feature weights.

[0131] In the feature importance evaluation stage, this embodiment adopts an improved sorting algorithm that takes into account the interrelationship between features. By calculating the sum of the attention scores of each feature and other features, the global importance index of the feature is obtained. This method is particularly suitable for remote office scenarios and can accurately identify the most recognizable identity features.

[0132] The sorting result is converted into probability distribution through softmax function, and the temperature parameter is set to 0.5, which not only maintains the discrimination of feature importance but also avoids excessive weight distribution. This embodiment introduces a smoothing factor in probability calculation to reduce the impact of noise on weight distribution.

[0133] The allocation of initial fusion weights adopts a dynamic adjustment strategy. This embodiment sets weight adjustment thresholds according to the security requirements of different application scenarios. For example, when performing large-amount payment operations, the weight of biometric features will be appropriately increased; while in daily office environments, behavioral features will be given more weight.

[0134] The element-level product operation is used in the feature weighting process, which retains the fine-grained information of the original feature. This embodiment standardizes the features before the product operation to avoid the weight imbalance problem caused by the difference in the value range. In the enterprise access control scenario, this fine weight adjustment mechanism can accurately reflect the credibility of different features.

[0135] The weighted average pooling operation uses an adaptive window size, and the window size is dynamically adjusted according to the feature dimension. This embodiment optimizes the parameter setting of the pooling operation and improves the stability of feature fusion by introducing feedback information of the attention mechanism.

[0136] The nonlinear transformation uses a two-layer neural network structure. The first layer uses the ReLU activation function to enhance the expressiveness of the features, and the second layer uses the tanh function to map the features to a uniform numerical range. This embodiment retains the residual connection during the transformation process, effectively preventing information loss.

[0137] In the mobile office scenario, the fusion solution of this embodiment shows excellent environmental adaptability. When the user switches between different devices, the fusion weight will automatically adjust according to the device credibility. For newly connected devices, the system will rely more on biometrics and behavioral features for identity authentication.

[0138] In the final stage of feature fusion, this embodiment normalizes the fused features through a batch normalization layer to ensure the distribution stability of the output features. This processing method makes the subsequent identity matching process more reliable, especially when processing large-scale user data.

[0139] Through this sophisticated feature fusion mechanism, this embodiment realizes the intelligent integration of multimodal features. In practical applications, this solution can dynamically adjust the feature fusion strategy according to different usage scenarios and security requirements, which not only ensures the accuracy of identity authentication, but also provides a good user experience.

[0140] The adaptive nature of this fusion solution makes it particularly suitable for use in complex and ever-changing office environments. Through an intelligent weight distribution mechanism, it can effectively respond to various abnormal situations and provide reliable technical support for enterprise-level identity authentication systems.

[0141] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 6 , and can also include the following:

[0142] Step S601: extract the user's authentication time distribution, operation behavior sequence and geographic location information from the historical authentication data, use the Gaussian mixture model to perform probability modeling on these features, obtain the probability density function of the user behavior through maximum likelihood estimation, and use the probability density function as the user behavior baseline model;

[0143] Step S602: Input the current authentication behavior into the baseline model to calculate the behavior probability value, calculate the deviation coefficient based on the difference between the behavior probability value and the preset threshold, and use an adaptive weight adjustment algorithm to dynamically correct the components of each dimension of the fused feature vector according to the deviation coefficient to obtain a corrected fused feature vector.

[0144] Optionally, this embodiment builds an intelligent user behavior baseline model, which achieves accurate modeling of user authentication behavior through in-depth analysis of historical authentication data. In the feature extraction stage, the authentication time distribution is analyzed using a 24-hour time series, and the user's activity time is divided into multiple time periods to capture the user's work and rest patterns.

[0145] The processing of the operation behavior sequence uses a sequence matching algorithm to extract the user's operation habit characteristics. This embodiment focuses on the temporal relationship and duration of the operation. For example, in an enterprise office system, the order and residence time of users accessing various functional modules often have strong personalized characteristics.

[0146] The modeling of geographic location information takes into account multiple dimensions, including latitude and longitude coordinates, IP address segments, and network environment characteristics. This embodiment identifies the user's common activity areas through cluster analysis and establishes a multi-level description model of location information. In a remote office scenario, this location model can effectively identify abnormal login locations.

[0147] The Gaussian mixture model is constructed using the expectation maximization algorithm. The model contains multiple Gaussian components, each of which corresponds to a typical mode of user behavior. This embodiment automatically determines the optimal number of components through the Bayesian information criterion, thereby avoiding the problem of model overfitting.

[0148] In the process of estimating the probability density function, the kernel density estimation method is used to smooth the sparse area. This embodiment introduces an adaptive bandwidth selection mechanism to dynamically adjust the smoothing parameters according to the data density, thereby improving the model's ability to express atypical behaviors.

[0149] The behavior probability value is calculated using a log-likelihood function to weight the probability values ​​of multiple feature dimensions. In this embodiment, the correlation between features is considered in the probability calculation, and the weight contribution of different features is adjusted through the covariance matrix.

[0150] The calculation of the deviation coefficient adopts an adaptive threshold mechanism, and the threshold is dynamically adjusted according to the fluctuation range of the user's historical behavior. In the enterprise security office scenario, this embodiment will adopt a stricter threshold standard for sensitive operations to improve security.

[0151] The weight adjustment algorithm adopts the gradient descent method to adjust the fused feature vector in real time according to the deviation coefficient. In this embodiment, an attenuation factor is designed so that the adjustment amplitude varies nonlinearly with the deviation degree, thus avoiding the problem of over-correction.

[0152] In a mobile office environment, the baseline model of this embodiment shows excellent environmental adaptability. For example, when it is detected that a user accesses at an unusual time or location, the model will automatically increase the verification weight of biometrics and behavioral features to improve the reliability of identity authentication.

[0153] During the correction process, a soft threshold strategy is used to gently adjust deviations that are beyond the normal range but not serious. This embodiment achieves a balance between security and user experience through this mechanism, avoiding overly frequent identity authentication requests.

[0154] Through this behavioral baseline model and dynamic correction mechanism, this embodiment achieves continuous verification of user identity. Especially in complex corporate office environments, it can accurately identify abnormal access behaviors while maintaining a low false alarm rate. The adaptive characteristics of this solution enable it to continuously optimize as user behavior patterns change, providing reliable security for enterprise-level identity authentication systems.

[0155] In one embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain of the present application, see Figure 7 , and can also include the following:

[0156] Step S701: construct a multi-layer neural network authentication model in the smart contract, input the corrected fusion feature vector into the authentication model to calculate identity similarity, calculate authentication confidence based on the similarity score, and map the authentication confidence to a predefined risk level interval to obtain the current authentication risk level;

[0157] Step S702: The authentication data is encrypted using an elliptic curve encryption algorithm through the encryption chip built into the hardware security module, a transaction block containing the encrypted data is generated according to the consensus mechanism of the blockchain network, the transaction block is broadcast to the permission chain node in the blockchain network, and waits for confirmation that the authentication process is complete.

[0158] Optionally, this embodiment designs a multi-layer neural network authentication architecture based on smart contracts, and realizes high-precision identity authentication through deep learning technology. The authentication model adopts a five-layer fully connected network structure. The first three layers use the ReLU activation function to enhance the nonlinear expression ability of the features, and the last two layers use the sigmoid function to map the output to the [0,1] interval.

[0159] To adapt to enterprise-level identity authentication scenarios, this embodiment adds attention mechanism and residual connection to the network structure. The attention mechanism can automatically identify key features in different scenarios, while the residual connection effectively prevents the gradient vanishing problem of deep networks. In scenarios with high security requirements such as financial transactions, this structural design significantly improves the accuracy of authentication.

[0160] The identity similarity calculation uses the cosine distance metric, and the basic similarity score is obtained by calculating the similarity between the current fused feature vector and the user's historical feature template. This embodiment introduces a dynamic threshold mechanism to adaptively adjust the similarity judgment standard according to different business scenarios and security level requirements.

[0161] The calculation of authentication confidence comprehensively considers multiple factors, including similarity score, user historical authentication performance and current environmental risk assessment. This embodiment designs an adaptive weight fusion algorithm to dynamically adjust the importance of each factor so that the authentication result is more in line with actual application requirements.

[0162] The risk level classification adopts a multi-level quantitative scheme, mapping the authentication confidence to three basic risk levels of low, medium and high, and subdividing each basic level. This embodiment formulates corresponding access control policies according to different risk levels. For example, high risk levels may require additional biometric verification.

[0163] In terms of data security protection, this embodiment adopts a new elliptic curve encryption algorithm and selects the P-256 curve parameter set to ensure encryption strength while maintaining high computing efficiency. The encryption chip of the hardware security module provides a trusted execution environment, effectively preventing key leakage and data tampering.

[0164] In the process of constructing blockchain transactions, this embodiment packages key elements such as authentication data, timestamp, device information, etc. into a transaction body and uses the user's private key for digital signature. In an enterprise-level blockchain network, this approach ensures the immutability and traceability of authentication records.

[0165] The consensus mechanism of the permission chain node adopts an improved practical Byzantine fault-tolerant algorithm, and ensures the validity of transactions through a voting mechanism. This embodiment introduces a reputation evaluation mechanism in the consensus process, increases the voting weight of trusted nodes, and speeds up transaction confirmation.

[0166] In practical applications, the authentication scheme of this embodiment demonstrates excellent security and scalability. For example, in a cross-border payment scenario, multi-party collaborative authentication can be achieved through a blockchain network, and each participant can independently verify the reliability of the authentication result.

[0167] The integrity verification of the authentication process uses a multi-signature mechanism, requiring multiple authorized nodes within the enterprise to jointly confirm important transactions. This embodiment effectively prevents single point failures and internal collusion risks through this mechanism, and is particularly suitable for enterprise application scenarios that require strict access control.

[0168] By combining deep learning technology with blockchain security mechanisms, this embodiment builds a reliable and efficient identity authentication framework. This solution can not only accurately identify the user's identity, but also ensure the transparency and immutability of the authentication process through blockchain technology, providing all-round security protection for enterprise-level applications.

[0169] In order to break through the limitations of traditional single authentication, achieve an effective balance between security and user experience, and provide advanced technical solutions for the field of identity authentication, this application provides an embodiment of a multi-factor security authentication device based on AI adaptive recognition and blockchain for implementing all or part of the multi-factor security authentication method based on AI adaptive recognition and blockchain, see Figure 8 The multi-factor security authentication device based on AI adaptive recognition and blockchain specifically includes the following contents:

[0170] The feature extraction module 10 is used to collect user biometric data, user behavior data and device information, perform liveness detection and obtain user facial image data when collecting the biometric data, record user continuous operation information when collecting the user behavior data, obtain hardware identification and geographic location information when collecting the device information, input the biometric data into a pre-trained convolutional neural network to perform static feature extraction to obtain a first feature vector, and input the user behavior data into a pre-trained long short-term memory network to perform dynamic feature extraction to obtain a second feature vector;

[0171] A feature fusion module 20 is used to construct a feature fusion network, in which a fully connected layer and an attention mechanism layer are set, the first feature vector, the second feature vector and the device information are input into the feature fusion network, the input features are ranked and weighted based on the attention mechanism layer, features with different weights are weightedly fused to obtain a fused feature vector, a user behavior baseline model is constructed based on historical authentication data, the degree of deviation between the current authentication behavior and the baseline model is calculated, and the fused feature vector is corrected according to the degree of deviation;

[0172] The security encryption module 30 is used to build a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0173] From the above description, it can be seen that the multi-factor security authentication device based on AI adaptive recognition and blockchain provided in the embodiment of the present application can extract static and dynamic features by collecting user biometrics, behavioral data and device information, respectively using convolutional neural networks and long short-term memory networks. Innovatively construct a feature fusion network, introduce an attention mechanism to dynamically weight and fuse different features, and perform feature correction based on the user behavior baseline model. Deploy blockchain networks and smart contracts at the same time to achieve trusted storage of authentication data and risk level assessment. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0174] From the hardware level, in order to break through the limitations of traditional single authentication, achieve an effective balance between security and user experience, and provide advanced technical solutions for the field of identity authentication, this application provides an embodiment of an electronic device for implementing all or part of the multi-factor security authentication method based on AI adaptive recognition and blockchain, and the electronic device specifically includes the following contents:

[0175] Processor, memory, communication interface and bus; wherein the processor, memory and communication interface communicate with each other through the bus; the communication interface is used to realize information transmission between the multi-factor security authentication device based on AI adaptive recognition and blockchain and related devices such as core business systems, user terminals and related databases; the logic controller can be a desktop computer, a tablet computer and a mobile terminal, etc., but the present embodiment is not limited thereto. In the present embodiment, the logic controller can be implemented with reference to the embodiment of the multi-factor security authentication method based on AI adaptive recognition and blockchain, and the embodiment of the multi-factor security authentication device based on AI adaptive recognition and blockchain, and the contents thereof are incorporated herein, and the repeated parts are not repeated.

[0176] It is understandable that the user terminal may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, etc. Among them, the smart wearable device may include smart glasses, a smart watch, a smart bracelet, etc.

[0177] In practical applications, part of the multi-factor security authentication method based on AI adaptive recognition and blockchain can be executed on the electronic device side as described above, or all operations can be completed in the client device. The specific selection can be based on the processing capability of the client device and the limitations of the user's usage scenario. This application does not limit this. If all operations are completed in the client device, the client device may also include a processor.

[0178] The client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission with the server. The server may include a server on the task scheduling center side, and other implementation scenarios may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, or a server cluster consisting of multiple servers, or a server structure of a distributed device.

[0179] Fig. 9 FIG. 9 is a schematic block diagram of the system structure of the electronic device 9600 according to an embodiment of the present application. Fig. 9 As shown, the electronic device 9600 may include a central processor 9100 and a memory 9140; the memory 9140 is coupled to the central processor 9100. It is worth noting that Fig. 9 is exemplary; other types of structures may also be used to supplement or replace this structure to implement telecommunication functions or other functions.

[0180] In one embodiment, the multi-factor security authentication method based on AI adaptive recognition and blockchain can be integrated into the central processor 9100. The central processor 9100 can be configured to perform the following control:

[0181] Step S101: collecting user biometric data, user behavior data and device information, performing liveness detection and acquiring user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, acquiring hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0182] Step S102: constructing a feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, inputting the first feature vector, the second feature vector and the device information into the feature fusion network, sorting the importance and assigning weights to the input features based on the attention mechanism layer, weightedly fusion of features with different weights to obtain a fused feature vector, constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation;

[0183] Step S103: construct a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0184] From the above description, it can be seen that the electronic device provided in the embodiment of the present application, by collecting user biometrics, behavioral data and device information, uses convolutional neural networks and long short-term memory networks to extract static and dynamic features respectively. Innovatively construct a feature fusion network, introduce an attention mechanism to dynamically weight and fuse different features, and perform feature correction based on the user behavior baseline model. At the same time, the blockchain network and smart contracts are deployed to achieve trusted storage of authentication data and risk level assessment. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0185] In another embodiment, the multi-factor security authentication device based on AI adaptive recognition and blockchain can be configured separately from the central processor 9100. For example, the multi-factor security authentication device based on AI adaptive recognition and blockchain can be configured as a chip connected to the central processor 9100, and the function of the multi-factor security authentication method based on AI adaptive recognition and blockchain can be implemented through the control of the central processor.

[0186] like Fig. 9 As shown, the electronic device 9600 may also include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily have to include Fig. 9 In addition, the electronic device 9600 may also include Fig. 9 For components not shown, reference may be made to the prior art.

[0187] like Fig. 9As shown, the central processing unit 9100 is sometimes also referred to as a controller or an operation control, and may include a microprocessor or other processor device and / or logic device. The central processing unit 9100 receives input and controls the operation of various components of the electronic device 9600.

[0188] The memory 9140 may be, for example, one or more of a cache, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory or other suitable devices. The above-mentioned information related to the failure may be stored, and a program for executing the relevant information may also be stored. The CPU 9100 may execute the program stored in the memory 9140 to implement information storage or processing, etc.

[0189] The input unit 9120 provides input to the central processing unit 9100. The input unit 9120 is, for example, a key or a touch input device. The power supply 9170 is used to provide power to the electronic device 9600. The display 9160 is used to display display objects such as images and texts. The display may be, for example, an LCD display, but is not limited thereto.

[0190] The memory 9140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, etc. It may also be a memory that saves information even when the power is off, can be selectively erased, and is provided with more data, examples of which are sometimes referred to as EPROMs, etc. The memory 9140 may also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142, which is used to store application programs and function programs or processes for executing the operation of the electronic device 9600 through the central processor 9100.

[0191] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for communication functions of the electronic device and / or for executing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0192] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processor 9100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

[0193] Based on different communication technologies, multiple communication modules 9110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module and / or a wireless LAN module. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide an audio output via the speaker 9131 and receive an audio input from the microphone 9132, thereby realizing a common telecommunication function. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. In addition, the audio processor 9130 is also coupled to the central processor 9100, so that recording can be performed on the local machine through the microphone 9132, and the sound stored on the local machine can be played through the speaker 9131.

[0194] The embodiments of the present application also provide a computer-readable storage medium capable of implementing all the steps in the multi-factor security authentication method based on AI adaptive recognition and blockchain in the above embodiments, where the execution subject is a server or a client. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, all the steps in the multi-factor security authentication method based on AI adaptive recognition and blockchain in the above embodiments are implemented. For example, when the processor executes the computer program, the following steps are implemented:

[0195] Step S101: collecting user biometric data, user behavior data and device information, performing liveness detection and acquiring user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, acquiring hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0196] Step S102: constructing a feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, inputting the first feature vector, the second feature vector and the device information into the feature fusion network, sorting the importance and assigning weights to the input features based on the attention mechanism layer, weightedly fusion of features with different weights to obtain a fused feature vector, constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation;

[0197] Step S103: construct a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0198] From the above description, it can be seen that the computer-readable storage medium provided in the embodiment of the present application, by collecting user biometrics, behavioral data and device information, uses convolutional neural networks and long short-term memory networks to extract static and dynamic features respectively. Innovatively construct a feature fusion network, introduce an attention mechanism to dynamically weight and fuse different features, and perform feature correction based on the user behavior baseline model. At the same time, blockchain networks and smart contracts are deployed to achieve trusted storage and risk level assessment of authentication data. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0199] The embodiments of the present application also provide a computer program product capable of implementing all the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain in the above embodiments, where the execution subject is a server or a client. When the computer program / instruction is executed by a processor, the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain are implemented. For example, the computer program / instruction implements the following steps:

[0200] Step S101: collecting user biometric data, user behavior data and device information, performing liveness detection and acquiring user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, acquiring hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector;

[0201] Step S102: constructing a feature fusion network, setting a fully connected layer and an attention mechanism layer in the feature fusion network, inputting the first feature vector, the second feature vector and the device information into the feature fusion network, sorting the importance and assigning weights to the input features based on the attention mechanism layer, weightedly fusion of features with different weights to obtain a fused feature vector, constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation;

[0202] Step S103: construct a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

[0203] From the above description, it can be seen that the computer program product provided in the embodiment of the present application, by collecting user biometrics, behavioral data and device information, uses convolutional neural networks and long short-term memory networks to extract static and dynamic features respectively. Innovatively construct a feature fusion network, introduce an attention mechanism to dynamically weight and fuse different features, and perform feature correction based on the user behavior baseline model. At the same time, blockchain networks and smart contracts are deployed to achieve trusted storage of authentication data and risk level assessment. This method breaks through the limitations of traditional single authentication, achieves an effective balance between security and user experience, and provides an advanced technical solution for the field of identity authentication.

[0204] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0205] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0206] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1A function specified in one or more boxes.

[0207] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0208] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A multi-factor security authentication method based on AI adaptive recognition and blockchain, characterized in that: The method comprises: Collecting user biometric data, user behavior data and device information, performing liveness detection and obtaining user facial image data when collecting the biometric data, recording user continuous operation information when collecting the user behavior data, obtaining hardware identification and geographic location information when collecting the device information, inputting the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and inputting the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector; A feature fusion network with a three-branch network structure is constructed, each branch contains two fully connected layers for feature dimension reduction and mapping, the first feature vector, the second feature vector and the encoded device information are respectively input into the corresponding network branches, and the features are standardized by the batch normalization layer to obtain the initial fusion feature; a multi-head attention mechanism layer is constructed in the feature fusion network, the query matrix, the key matrix and the value matrix are calculated for the initial fusion feature, the correlation between different features is calculated by scaling the dot product attention to obtain the attention weight, and the attention weight is weighted and summed with the corresponding feature to obtain the fusion feature; the attention weight is normalized to obtain the importance score, the input features are sorted in descending order according to the importance score, the sorting result is converted into a probability distribution by using the soft maximum function, and the initial fusion weight is assigned to different features based on the probability distribution; the initial fusion weight is multiplied by the corresponding feature, all weighted features are fused by weighted average pooling operation, and the fusion result is mapped to the feature space of uniform dimension by nonlinear transformation to obtain the fusion feature vector, a user behavior baseline model is constructed based on historical authentication data, the deviation degree between the current authentication behavior and the baseline model is calculated, and the fusion feature vector is corrected according to the deviation degree; A blockchain network is constructed, and permission chain nodes and smart contracts are deployed in the blockchain network. The corrected fused feature vector is input into the smart contract for identity authentication, and the authentication confidence is calculated to obtain the risk level. When the risk level is within a preset range, the authentication data is encrypted by a hardware security module, and the encrypted data is written into the blockchain network to complete the authentication process.

2. The multi-factor security authentication method based on AI adaptive recognition and blockchain according to claim 1 is characterized in that: The step of inputting the biometric data into a pre-trained convolutional neural network to extract static features to obtain a first feature vector includes: Preprocessing the biometric data, normalizing the biometric data to obtain standardized image data with pixel values ​​between 0 and 1, using an image enhancement algorithm based on a multi-scale convolution kernel to perform spatial feature enhancement on the standardized image data, and performing feature dimension reduction on the enhanced image data through a maximum pooling operation to obtain preprocessed image data; The preprocessed image data is input into a convolutional neural network comprising multiple convolutional layers and a fully connected layer, a 3×3 convolution kernel and a ReLU activation function are set in the convolutional layer to extract local features of the image, and the extracted local features are mapped to a 128-dimensional feature space through the fully connected layer to obtain the first feature vector.

3. The multi-factor security authentication method based on AI adaptive recognition and blockchain according to claim 1 is characterized in that: The step of inputting the user behavior data into a pre-trained long short-term memory network to perform dynamic feature extraction to obtain a second feature vector includes: The user behavior data is sampled and segmented according to the time series, the user's continuous operation information is divided into time windows of fixed length, the behavior data in each time window is numerically encoded and standardized, and a behavior feature sequence including the user operation sequence, operation time interval and operation duration is generated; The behavioral feature sequence is input into a long short-term memory network with a forget gate and an input gate, the historical state information is filtered through the forget gate, the importance of the current input information is controlled through the input gate, the state information is nonlinearly transformed using the tanh activation function, and the transformed features are mapped to a 64-dimensional feature space to obtain the second feature vector.

4. The multi-factor security authentication method based on AI adaptive recognition and blockchain according to claim 1 is characterized in that: The step of constructing a user behavior baseline model based on historical authentication data, calculating the degree of deviation between the current authentication behavior and the baseline model, and correcting the fused feature vector according to the degree of deviation includes: Extract the user's authentication time distribution, operation behavior sequence and geographic location information from the historical authentication data, use the Gaussian mixture model to perform probability modeling on these features, obtain the probability density function of the user behavior through maximum likelihood estimation, and use the probability density function as the user behavior baseline model; The current authentication behavior is input into the baseline model to calculate the behavior probability value, the deviation coefficient is calculated based on the difference between the behavior probability value and the preset threshold, and an adaptive weight adjustment algorithm is used to dynamically correct the components of each dimension of the fused feature vector according to the deviation coefficient to obtain the corrected fused feature vector.

5. The multi-factor security authentication method based on AI adaptive recognition and blockchain according to claim 1 is characterized in that: The corrected fusion feature vector is input into the smart contract for identity authentication, the authentication confidence is calculated to obtain the risk level, and when the risk level is within a preset range, the authentication data is encrypted by the hardware security module, and the encrypted data is written into the blockchain network to complete the authentication process, including: Constructing a multi-layer neural network authentication model in the smart contract, inputting the corrected fusion feature vector into the authentication model to calculate identity similarity, calculating authentication confidence based on the similarity score, and mapping the authentication confidence to a predefined risk level interval to obtain the current authentication risk level; The authentication data is encrypted using the elliptic curve encryption algorithm through the encryption chip built into the hardware security module, a transaction block containing the encrypted data is generated according to the consensus mechanism of the blockchain network, the transaction block is broadcast to the permission chain node in the blockchain network and waits for confirmation to complete the authentication process.

6. A multi-factor security authentication device based on AI adaptive recognition and blockchain, characterized in that: The device comprises: A feature extraction module, used to collect user biometric data, user behavior data and device information, perform liveness detection and obtain user facial image data when collecting the biometric data, record user continuous operation information when collecting the user behavior data, obtain hardware identification and geographic location information when collecting the device information, input the biometric data into a pre-trained convolutional neural network for static feature extraction to obtain a first feature vector, and input the user behavior data into a pre-trained long short-term memory network for dynamic feature extraction to obtain a second feature vector; A feature fusion module is used to construct a feature fusion network with a three-branch network structure. Each branch contains two fully connected layers for feature dimension reduction and mapping. The first feature vector, the second feature vector and the encoded device information are respectively input into the corresponding network branches, and the features are standardized through the batch normalization layer to obtain the initial fusion features; a multi-head attention mechanism layer is constructed in the feature fusion network, and the query matrix, key matrix and value matrix are calculated for the initial fusion features. The correlation between different features is calculated by scaling the dot product attention to obtain the attention weight, and the attention weight is weighted summed with the corresponding feature to obtain the fusion feature; the attention The weights are normalized to obtain importance scores, the input features are sorted in descending order according to the importance scores, the sorting results are converted into probability distributions using a soft maximum function, and initial fusion weights are assigned to different features based on the probability distribution; the initial fusion weights are multiplied by the corresponding features, all weighted features are fused through a weighted average pooling operation, the fusion results are mapped to a feature space of uniform dimension using a nonlinear transformation to obtain the fused feature vector, a user behavior baseline model is constructed based on historical authentication data, the degree of deviation between the current authentication behavior and the baseline model is calculated, and the fused feature vector is corrected according to the degree of deviation; The security encryption module is used to build a blockchain network, deploy permission chain nodes and smart contracts in the blockchain network, input the corrected fusion feature vector into the smart contract for identity authentication, calculate the authentication confidence to obtain the risk level, and when the risk level is within a preset range, encrypt the authentication data through the hardware security module, and write the encrypted data into the blockchain network to complete the authentication process.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain are implemented as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the multi-factor security authentication method based on AI adaptive recognition and blockchain are implemented as described in any one of claims 1 to 5.

Images