Multi-factor identity authenticated payment transaction encryption verification system

Abstract

The application discloses a payment transaction encryption verification system of multi-factor identity authentication, relates to the technical field of payment security authentication, and comprises a registration management, real-time data acquisition, risk decision, multi-factor authentication, token generation and transaction verification module. When the payment transaction is initiated, the system dynamically calculates an environmental risk index by analyzing network environment characteristics in real time, and selects a required authentication factor combination for verification in real time according to the index. The system uses an encryption synthesis algorithm to fuse and calculate a plurality of generated independent authentication results and transaction plaintext information, generates an encrypted payment token, and sends the token to a payment gateway to complete decryption and comparison verification. The authentication strength is intelligently and dynamically adjusted according to the environmental risk, the identity authentication evidence is deeply bound with the transaction information by using cryptography, and the security and reliability of the transaction process are improved.

Description

Technical Field

[0001] This invention belongs to the field of payment security authentication technology, specifically a payment transaction encryption verification system with multi-factor authentication. Background Technology

[0002] Current payment transaction security systems generally employ multi-factor authentication technology. During user registration, a fixed combination of authentication factors, including biometrics and physical tokens, is preset, and the same static verification process is executed for each transaction. The authentication process is independent of the payment token generation stage; successful authentication only generates an internal system authorization status, and subsequent transactions are processed using a separate encryption module to generate payment credentials. This is the mainstream technical implementation method currently used.

[0003] Such static schemes have flaws. Their authentication strategies lack real-time awareness and response capabilities to environmental risks, and cannot dynamically adjust authentication requirements according to the security level of the transaction scenario, making it difficult to balance security protection and operational efficiency. Furthermore, identity authentication and final payment authorization are disconnected at both the logical and cryptographic levels. The authentication result serves only as an intermediate state in the process and does not directly participate in forming the cryptographic credentials of the final payment token, allowing attackers to exploit gaps in the connections between system modules to carry out bypass attacks. Summary of the Invention

[0004] This invention aims to solve one of the technical problems existing in the prior art;

[0005] To address this, the present invention proposes a multi-factor authentication payment transaction encryption verification system, comprising: a registration management module, which collects and stores a multi-dimensional biometric baseline data set of the user, a set of physical token device feature codes held by the user, and a set of behavioral habit baseline data set formed by the user's operations in a preset secure environment during the user registration phase; a real-time data acquisition module, which acquires in real-time network environment feature data of the current transaction request, a set of real-time biometric data accompanying the current transaction request, and a set of real-time user operation behavior data when a payment transaction is initiated; a risk decision module, which compares the network environment feature data with a historical trusted environment database, calculates an environmental risk index, and dynamically selects the required authentication factor combination based on the environmental risk index; a multi-factor authentication module, which calls the corresponding verification algorithm to verify the real-time biometric data set and the real-time user operation behavior data set according to the dynamically selected authentication factor combination, generating two independent authentication results; a token generation module, which uses a preset encryption synthesis algorithm to fuse the two independent authentication results with the plaintext transaction information of the current transaction request to generate the final encrypted payment token; and a transaction verification module, which sends the encrypted payment token to the payment gateway for decryption and comparison to complete the transaction verification.

[0006] Further, the calculation of the environmental risk index based on the comparison between the network environment feature data and the historical trusted environment database includes: extracting network access point identifiers, geographic location information, current timestamps, and device hardware fingerprints from the network environment feature data; calculating the matching degree between the network access point identifiers, geographic location information, current timestamps, and device hardware fingerprints and the historical network access point identifier sequences, historical geographic location trajectories, historical frequently used transaction time periods, and bound device hardware fingerprints corresponding to the user stored in the historical trusted environment database; obtaining a preliminary environmental trust score through weighted fusion based on the matching degree calculation results; and correcting the preliminary environmental trust score by combining the amount deviation of the current transaction request and the unfamiliarity of the payee to obtain the environmental risk index, wherein the environmental risk index is inversely proportional to the preliminary environmental trust score.

[0007] Further, the dynamic selection of the required authentication factor combination based on the environmental risk index includes: selecting the authentication factor combination from the multi-dimensional biometric baseline data set, the physical token device feature code set, and the behavioral habit baseline data set; presetting multiple environmental risk index thresholds, each environmental risk index threshold corresponding to a predefined authentication strength level; comparing the calculated environmental risk index with the multiple environmental risk index thresholds to determine the current authentication strength level at the exchange; each authentication strength level is associated with a preset authentication factor type and quantity requirement, wherein the higher the authentication strength level, the more authentication factor types are required and the more complex the verification steps; based on the determined authentication strength level and its associated authentication factor type and quantity requirements, selecting corresponding type and quantity of baseline data items from the multi-dimensional biometric baseline data set, the physical token device feature code set, and the behavioral habit baseline data set to constitute the dynamically selected authentication factor combination.

[0008] Further, the verification of the real-time biometric data set by calling the corresponding verification algorithm includes: when the dynamically selected authentication factor combination contains biometric factors, extracting the corresponding real-time biometric sample from the real-time biometric data set; comparing the real-time biometric sample with the corresponding benchmark template in the multi-dimensional biometric benchmark data set; the feature comparison process adopts a similarity measurement method based on deep feature embedding to calculate the distance between the real-time biometric sample and the corresponding benchmark template in the feature space; if the distance is less than a preset biometric matching threshold, an independent authentication result of successful biometric authentication is generated.

[0009] Furthermore, the verification algorithm for the real-time user operation behavior data set includes: when the dynamically selected authentication factor combination includes behavioral feature factors, extracting operation timing data from the real-time user operation behavior data set during the payment transaction input process; the operation timing data includes touch pressure change sequence, input speed curve, and device grip posture fine-tuning angle sequence; inputting the operation timing data into a pre-trained behavior habit verification model, which is trained on the behavior habit benchmark data set; the behavior habit verification model outputs a behavior feature matching confidence score, and if the behavior feature matching confidence score exceeds a preset behavior verification threshold, an independent authentication result of successful behavior feature authentication is generated.

[0010] Furthermore, the step of fusing the two independent authentication results with the plaintext transaction information of the current transaction request using a preset encryption synthesis algorithm includes: converting each successful independent authentication result into a unique authentication result code; concatenating the authentication result code, the environmental risk index, and a dynamically generated random number to form an authentication information digest; encrypting the authentication information digest for the first time using a session key derived from the user's private key to obtain primary encrypted data; performing an XOR and cyclic shift operation on the primary encrypted data and the plaintext transaction information to generate intermediate synthesized data; and encrypting the intermediate synthesized data using the public key of the payment gateway to generate the final encrypted payment token.

[0011] Further, the step of sending the encrypted payment token to the payment gateway for decryption and comparison includes: the payment gateway using its private key to decrypt the received encrypted payment token to obtain the intermediate synthesized data; the payment gateway reverse-parses the plaintext transaction information and the primary encrypted data from the intermediate synthesized data; the payment gateway using the user's pre-stored public key to decrypt the primary encrypted data to restore the authentication information digest; the payment gateway extracts the authentication result code and the environmental risk index from the authentication information digest, and verifies whether the selection of the authentication factor combination conforms to the strategy based on the environmental risk index; the payment gateway performs a consistency check between the parsed plaintext transaction information and the original transaction request information, and confirms that the authentication result code indicates that all required authentication factors have been successfully verified; when the consistency check passes and the authentication verification passes, the payment gateway approves the payment transaction.

[0012] Furthermore, the construction of the behavioral habit verification model is achieved through the following steps: collecting multi-dimensional behavioral time-series data formed by users in historical secure payment transactions as training samples, with each training sample labeled as a positive sample of the user's own operation; the multi-dimensional behavioral time-series data includes a pressure sequence of touch pressure changing over time, a speed curve of input speed changing over time, and a sequence of fine-tuning angles of the device's grip posture in three-dimensional space; preprocessing the training samples, including data cleaning to remove abnormal noise points and time series alignment operations to unify data length and sampling frequency; extracting deep spatiotemporal features characterizing the user's stable operating habits from the preprocessed time-series data, the deep spatiotemporal features being... A stacked convolutional neural network and long short-term memory network hybrid model is automatically learned; the extracted deep spatiotemporal features are input into a multilayer perceptron classifier for training, and the output of the classifier is the confidence score of the user's operation behavior belonging to the user; the overall model composed of the convolutional neural network, long short-term memory network and multilayer perceptron classifier is jointly trained end-to-end using the backpropagation algorithm and labeled positive samples until the model converges, forming the initial behavior habit verification model; during the system operation phase, new positive samples obtained using a continuous update mechanism are used to fine-tune the parameters of the behavior habit verification model through online learning or periodic batch retraining to achieve incremental optimization of the model.

[0013] Furthermore, the feature comparison process employs a similarity metric based on deep feature embedding to calculate the distance between the real-time biometric sample and the corresponding benchmark template in the feature space. This includes: using a pre-trained deep neural network feature extractor to perform forward propagation calculations on the real-time biometric sample and the corresponding benchmark template in the multi-dimensional biometric benchmark dataset to extract the first feature vector of the real-time biometric sample and the second feature vector of the benchmark template; calculating the metric distance between the feature vector of the real-time biometric sample and the feature vector of the benchmark template, wherein the metric distance is Euclidean distance or cosine distance; and comparing the calculated metric distance with the preset biometric matching threshold to determine whether the biometrics match.

[0014] Furthermore, it also includes continuous updates to the behavioral habit benchmark data set: after each payment transaction is successfully verified and no fraud report is received, the current real-time user operation behavior dataset is used as a new positive sample; the new positive sample is denoised and standardized to extract stable behavioral features; the extracted stable behavioral features are then incorporated into the corresponding user historical behavior feature model in the behavioral habit benchmark data set with certain weights, thereby realizing the progressive update of the behavioral habit benchmark data set.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows.

[0016] By collecting and analyzing network environment characteristic data of transaction initiation in real time and comparing it with a historical trusted environment database, a quantified environmental risk index is calculated. This index directly drives the system to dynamically select and combine the currently required biometrics, behavioral habits, or physical token verification items from the available authentication factor pool. This makes authentication strength no longer fixed but flexible enough to adapt to different risk scenarios. In low-risk environments, unnecessary authentication interference is reduced, improving transaction smoothness; in high-risk environments, more authentication factors are automatically added to build a more robust security barrier. The authentication process has transformed from rigid execution to context-aware intelligent decision-making, achieving a dynamic balance between security protection level and operational convenience without user intervention.

[0017] Multiple independent authentication results output by the multi-factor authentication module are used as key input parameters and fed into the encrypted synthesis algorithm along with the plaintext information of the transaction for fusion calculation. The generated encrypted payment token is no longer simply ciphertext of transaction data, but a composite that cryptographically binds the evidence of successful identity authentication with transaction details. The payment gateway's decoding and verification process of this token, while verifying the transaction content, also necessarily verifies the validity of the multi-factor authentication status on which the token was generated. Any attempt to forge or replay the transaction token must simultaneously compromise both the front-end multi-factor authentication system and the back-end encrypted synthesis algorithm. This deep coupling mechanism eliminates the possibility of logical gaps between the authentication process and the final payment authorization process, enhancing the integrity and non-repudiation of the entire transaction chain. Attached Figure Description

[0018] Figure 1 This is a timing diagram of the multi-factor authentication payment transaction encryption verification system described in this invention; Figure 2 A flowchart for the process of dynamically selecting a combination of authentication factors; Figure 3 Heatmap of encryption algorithm performance at each stage in a multi-factor authentication payment transaction encryption verification system; Figure 4 A bar chart showing the success rate of various authentication factors in a multi-factor authentication system; Figure 5 Performance trend chart for batch updates of behavioral habit benchmark data. Detailed Implementation

[0019] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] See Figure 1 The multi-factor authentication payment transaction encryption verification system of this invention is implemented as follows: During the user registration phase, the registration management module collects and stores the user's multi-dimensional biometric baseline data set, physical token device feature code set, and behavioral habit baseline data set. When a payment transaction is initiated, the real-time data acquisition module simultaneously acquires the current network environment feature data, the real-time biometric data set, and the real-time user operation behavior data set. The risk decision module compares the network environment feature data with a historical trusted environment database, calculates an environmental risk index, and dynamically selects the authentication factor combination required for the current transaction based on this index. The multi-factor authentication module calls the corresponding verification algorithm to verify the real-time biometric data set and the real-time user operation behavior data set according to the dynamically selected authentication factor combination, generating two independent authentication results. The token generation module uses a preset encryption synthesis algorithm to fuse the two independent authentication results with the plaintext transaction information of the current transaction request to generate the final encrypted payment token. The transaction verification module sends the encrypted payment token to the payment gateway for decryption and comparison, thereby completing the entire transaction verification process.

[0021] In one embodiment of the present invention, see [reference] Figure 2 In its implementation, the system acquires network environment characteristic data through a real-time data acquisition module when a payment transaction is initiated. This data includes network access point identifiers (NATs), geographic location information, current timestamps, and device hardware fingerprints. The system also accesses a historical trusted environment database, which stores the user's corresponding historical NAT sequence, historical geographic location trajectory, historical frequently used transaction time periods, and bound device hardware fingerprints. The system calculates the NAT matching degree by matching the real-time NAT identifier with the historical NAT sequence. The matching degree is calculated using the following formula: .in, Represents the degree of matching of network access point identifiers. This represents the total number of historical network access point identifier sequences. For real-time network access point identifiers, The first in the historical network access point identifier sequence One element, For indicator functions, when The value is 1 if the condition is met, and 0 otherwise.

[0022] The system calculates the matching degree between real-time geolocation information and historical geolocation trajectories. The matching degree is calculated using the following formula: .in, For the degree of matching of geographic location information, Provides real-time geolocation coordinates. A collection of historical geographical location trajectories. The Haversine distance between the two points. This is the preset distance threshold.

[0023] The system calculates the matching degree of the current timestamp by comparing it with historically frequently used transaction time periods. The matching degree is calculated using the following formula: .in, For the current timestamp matching degree, For real-time timestamps, This refers to historically common trading periods. This is the preset time buffer window.

[0024] The system calculates the matching degree between the real-time device hardware fingerprint and the bound device hardware fingerprint. The matching degree is calculated using the following formula: .in, For device hardware fingerprint matching accuracy, For real-time device hardware fingerprint string, For the bound device hardware fingerprint string, This function is used to calculate the length of the longest common subsequence of two strings. This is a function for calculating the length of a string.

[0025] In some embodiments, the system performs a weighted fusion of network access point identifier matching, geolocation information matching, current timestamp matching, and device hardware fingerprint matching. The network access point identifier matching has a weight coefficient α, the geolocation information matching has a weight coefficient β, the current timestamp matching has a weight coefficient γ, and the device hardware fingerprint matching has a weight coefficient δ. After weighted fusion, a preliminary environmental credibility score is obtained. The system then corrects the preliminary environmental credibility score by considering the amount deviation of the current transaction request and the unfamiliarity of the payee. The amount deviation is calculated based on the ratio of the current transaction amount to the user's historical average transaction amount. The unfamiliarity of the payee is determined by whether the payee's account exists in the user's historical transaction payee list. The corrected score yields an environmental risk index, which is inversely proportional to the preliminary environmental credibility score. The formula for calculating the environmental risk index is as follows: .in, Represents the environmental risk index. This represents a preliminary environmental credibility score. This represents a correction factor determined by both the amount deviation and the unfamiliarity of the payee.

[0026] In practical implementation, the system presets multiple environmental risk index thresholds, including threshold values. Threshold and threshold and satisfy The system will calculate the real-time environmental risk index R and compare it with the threshold. Threshold Threshold A comparison is made to determine the current level of certification strength at the exchange. When it corresponds to the first certification strength level, when When it corresponds to the second certification strength level, when When it corresponds to the third certification strength level, when The system corresponds to the fourth authentication strength level. Each authentication strength level is associated with a preset authentication factor type and quantity requirement. The higher the authentication strength level, the more authentication factor types are required and the more complex the verification steps are. The authentication factor combination is selected from the multi-dimensional biometric baseline data set, the physical token device feature code set, and the behavioral habit baseline data set. Based on the determined authentication strength level and its associated authentication factor type and quantity requirements, the system selects the corresponding type and quantity of baseline data items from the multi-dimensional biometric baseline data set, the physical token device feature code set, and the behavioral habit baseline data set to form a dynamically selected authentication factor combination.

[0027] It is understandable that the network access point identifier matching degree is calculated by checking whether the real-time network access point identifier exists in the historical network access point identifier sequence; the geolocation information matching degree is calculated by calculating the physical distance between the real-time geolocation point and the historical geolocation trajectory point; the current timestamp matching degree is calculated by determining whether the current timestamp falls within the historical commonly used transaction time period; and the device hardware fingerprint matching degree is calculated by comparing the consistency between the real-time device hardware fingerprint string and the bound device hardware fingerprint string. Optionally, the first authentication strength level is associated with one authentication factor type, the second authentication strength level is associated with two authentication factor types, the third authentication strength level is associated with three authentication factor types, and the fourth authentication strength level is associated with three authentication factor types with additional verification steps.

[0028] In one embodiment of the present invention, when the dynamically selected combination of authentication factors includes biometric factors, the multi-factor authentication module extracts the corresponding real-time biometric samples from the real-time biometric dataset. The multi-factor authentication module then performs feature comparison between the real-time biometric samples and the corresponding benchmark templates in the multi-dimensional biometric benchmark dataset. The feature comparison process employs a similarity measurement method based on deep feature embedding. The multi-factor authentication module uses a pre-trained deep neural network feature extractor to perform forward propagation calculations on the real-time biometric samples to extract a first feature vector. The deep neural network feature extractor also performs forward propagation calculations on the benchmark templates to extract a second feature vector. The calculation uses deep neural network feature extraction technology. The multi-factor authentication module calculates the metric distance between the first and second feature vectors in the feature space. If the metric distance is less than a preset biometric matching threshold, the multi-factor authentication module generates an independent authentication result indicating successful biometric authentication.

[0029] In some embodiments, the distance measurement is performed using cosine distance, and the formula for calculating cosine distance is as follows: .in, Represents the cosine distance between a real-time biometric sample and a baseline template. This represents the first feature vector extracted from the real-time biometric sample. This represents the second feature vector extracted from the baseline template, and the symbol "·" represents the vector dot product operation. The magnitude of the first eigenvector is represented by the length of the vector. This represents the magnitude of the second feature vector. When the dynamically selected combination of authentication factors includes behavioral feature factors, the multi-factor authentication module extracts the operation timing data of the current payment transaction input process from the real-time user operation behavior data set. The operation timing data includes the touch pressure change sequence, input speed curve, and device grip posture fine-tuning angle sequence. The multi-factor authentication module inputs the operation timing data into the pre-trained behavior habit verification model. The behavior habit verification model outputs a behavior feature matching confidence score. If the behavior feature matching confidence score exceeds the preset behavior verification threshold, the multi-factor authentication module generates an independent authentication result indicating successful behavior feature authentication.

[0030] In its implementation, the behavioral habit verification model is constructed through the following steps: The system collects multi-dimensional behavioral time-series data generated by users in historical secure payment transactions as training samples. Each training sample is labeled as a positive sample of the user's own operation. The multi-dimensional behavioral time-series data includes a pressure sequence of touch pressure changing over time, a speed curve of input speed changing over time, and a sequence of fine-tuning angles of the device's grip posture in three-dimensional space. The system preprocesses the training samples, including data cleaning to remove abnormal noise points and time series alignment to unify data length and sampling frequency. The system extracts deep spatiotemporal features representing the user's stable operating habits from the preprocessed time-series data. The deep spatiotemporal features are automatically learned through a hybrid model of a stacked convolutional neural network and a long short-term memory network. The system inputs the extracted deep spatiotemporal features into a multilayer perceptron classifier for training. The output of the multilayer perceptron classifier is the confidence score that the user's operation belongs to the user. The system uses the backpropagation algorithm and labeled positive samples to perform end-to-end joint training on the overall model composed of the convolutional neural network, the long short-term memory network, and the multilayer perceptron classifier until the model converges to form the initial behavioral habit verification model.

[0031] It is understandable that the distance measurement during feature comparison can also adopt the Euclidean distance method, which is the square root of the sum of the squares of the differences between the first and second feature vectors in each dimension. During system operation, the behavior habit verification model can utilize newly acquired samples obtained through a continuous update mechanism to fine-tune its parameters through online learning or periodic batch retraining, achieving incremental optimization of the model. Specifically, during system operation, the continuous update mechanism for the behavior habit verification model refers to the following: after each successful payment transaction verification without fraud reports, the system uses the real-time user operation behavior data as a new positive sample. After denoising and standardization, stable behavioral features are extracted and fused into the user's historical behavior feature model according to preset weights, achieving incremental optimization of the model parameters. This mechanism dynamically adjusts the model parameters through online learning or periodic batch retraining, enabling the model to continuously adapt to changes in user operation habits. Optionally, the pre-training process of the biometric deep neural network feature extractor uses large-scale public biometric datasets and user-private biometric data for supervised training. The acquisition of touch pressure change sequences relies on the pressure-sensitive sensor on the mobile device screen, while the acquisition of device grip posture fine-tuning angle sequences relies on the gyroscope and accelerometer sensors built into the mobile device.

[0032] In one embodiment of the present invention, in a specific implementation, the token generation module receives two independent authentication results generated by the multi-factor authentication module and an environmental risk index calculated by the risk decision module. The token generation module converts each successful independent authentication result into a unique authentication result code. The token generation module concatenates all authentication result codes, the environmental risk index, and a dynamically generated random number according to a predefined order to form an authentication information digest. In some embodiments, the token generation module uses a session key derived from the user's private key to perform an initial encryption of the authentication information digest. The session key is generated based on the user's master private key and the current transaction session identifier through a key derivation function. The initial encryption of the authentication information digest uses a symmetric encryption algorithm, resulting in primary encrypted data. The token generation module performs a fusion calculation with the plaintext transaction information of the current transaction request. The fusion calculation includes an XOR operation and a circular shift operation. The operation process can be described as follows: .in, This represents the i-th data block of the intermediate composite data. The i-th data block representing plaintext transaction information. Represents the i-th data block of the initial encrypted data, symbol Represents the bitwise XOR operation, symbol This represents a circular left shift operation. This represents the number of shifts derived from the initial encrypted data. The intermediate synthesized data is generated through the above operations.

[0033] In implementation, the token generation module obtains the public key of the payment gateway, which is pre-stored in the system. The token generation module uses this public key to encrypt the intermediate synthesized data using an asymmetric encryption algorithm, generating the final encrypted payment token. The authentication result encoding can be generated by creating a fixed-length hash value based on the type of authentication factor and the verification pass status. Dynamic random numbers are generated by a cryptographically secure random number generator to ensure unpredictability. It can be understood that the purpose of the XOR operation and the cyclic shift operation is to obfuscate the initial encrypted data with the plaintext transaction information, ensuring that the intermediate synthesized data simultaneously contains both authentication and transaction information. The number of bits in the cyclic shift... The token can be obtained by modulo operation on the corresponding byte value of the initial encrypted data, thus ensuring the dynamic nature of the shift operation. The public key encryption operation of the payment gateway ensures that only the payment gateway with the corresponding private key can decrypt the encrypted payment token.

[0034] Optionally, the symmetric encryption algorithm can be the Advanced Encryption Standard (AES), and the asymmetric encryption algorithm can be either RSA or Elliptic Curve Cryptography (ECC). The order of concatenating the authentication message digest can be: authentication result encoding first, environmental risk index in the middle, and dynamic random number last.

[0035] See Figure 3 This is a heatmap showing the performance of encryption algorithms at each stage of a multi-factor authentication payment transaction encryption verification system. The AES algorithm performs best in the "initial encryption" stage, scoring near-perfect (approximately 98 points). This aligns with AES's characteristics as a symmetric encryption algorithm: high encryption efficiency and low computational overhead, making it ideal for scenarios involving rapid encryption of large amounts of data. The RSA algorithm performs weakly overall, especially in the "initial encryption" stage, where it scores the lowest (approximately 75 points). This is because RSA, as an asymmetric encryption algorithm, has high computational complexity and is slow, making it unsuitable for encrypting large volumes of data. For lightweight encryption scenarios such as environmental risk assessment, AES or ECC can be chosen based on requirements, avoiding the use of RSA.

[0036] In one embodiment of the present invention, in a specific implementation, the transaction verification module sends the encrypted payment token to the payment gateway. The payment gateway uses its private key to decrypt the received encrypted payment token. The private key of the payment gateway is paired with the public key previously used for encryption. This decryption uses a decryption algorithm corresponding to the asymmetric encryption algorithm. After decryption, intermediate synthetic data is obtained. The payment gateway reverse-parses the plaintext transaction information and the primary encrypted data from the intermediate synthetic data. The reverse parsing process is the inverse operation of the fusion calculation process in the token generation module.

[0037] In some embodiments, the payment gateway uses the user's pre-stored public key to decrypt the initial encrypted data to recover the authentication message digest. The user's pre-stored public key is stored by the system in the secure storage area of ​​the payment gateway during user registration or initial device binding. The decryption of the initial encrypted data uses the symmetric decryption algorithm corresponding to the initial encryption. The session key used needs to be regenerated based on the user's master public key and the session identifier parsed from the authentication message digest using the same key derivation function. The payment gateway extracts the authentication result code and environmental risk index from the recovered authentication message digest. The payment gateway verifies whether the selection of authentication factor combinations conforms to the preset strategy based on the environmental risk index. The verification process involves comparing the environmental risk index with a predefined threshold and checking whether the actual type and number of authentication factors used match the strategy requirements. The matching relationship is shown in Table 1. Table 1: Correspondence between Environmental Risk Index and Certification Requirement Factors

[0038] The payment gateway performs a consistency check between the parsed plaintext transaction information and the original transaction request information. This consistency check includes comparing key fields such as transaction amount, payee account, and transaction timestamp. The information matching degree M can be calculated as follows: .in, Represents the degree of information matching. This represents the total number of key fields involved in the comparison. This represents the value of the j-th key field parsed from the encrypted payment token. This represents the value of the j-th key field in the original transaction request information stored by the payment gateway. This is an indicator function; it takes the value 1 when the condition within the parentheses is true, and 0 otherwise. The payment gateway also confirms that the authentication result code extracted from the authentication message digest indicates that all required authentication factors have been successfully verified. It can be understood that decrypting the payment gateway's private key is the only way to recover the intermediate synthesized data, ensuring the integrity and authenticity of the encrypted payment token. Reverse parsing the plaintext transaction information and the initial encrypted data from the intermediate synthesized data requires precisely reversing the XOR and cyclic shift operations; the direction and number of bits in the cyclic shift must be consistent with the encryption end.

[0039] Optionally, the symmetric decryption algorithm is the decryption mode of the Advanced Encryption Standard (AES) algorithm. The indicator function ensures that the comparison result for each field is a Boolean value, and the information matching degree M is equal to the ratio of the number of completely matched fields to the total number of fields. When the consistency check and authentication verification pass, the payment gateway approves the payment transaction. In specific implementations, if the environmental risk index verification finds that the authentication factor combination does not meet the requirements, or the information matching degree M is lower than a preset threshold, or the authentication result encoding indicates that the required authentication factor verification has failed, the payment gateway will reject the payment transaction and generate a security alert log. The payment gateway's storage of the original transaction request information is time-sensitive, only temporarily saved during the transaction verification session.

[0040] See Figure 4 This is a bar chart showing the success rates of various authentication factors in a multi-factor authentication system. Physical tokens have the highest success rate, approaching 1.0. This is because the verification logic of physical tokens (such as hardware USB keys and dynamic password cards) is based on preset encryption algorithms and synchronization mechanisms, making them minimally affected by environmental and user behavior interference and exhibiting the strongest stability. In high-security scenarios, physical tokens and biometric factors should be used in combination, leveraging their high success rates and low correlation to enhance overall authentication strength. To balance user experience and security, behavioral factors can be used as supplementary verification, adding a layer of security without increasing the user's workload. When the success rate of environmental factor verification decreases, the system should automatically increase authentication strength to address potential risks.

[0041] In one embodiment of the present invention, in a specific implementation, after each payment transaction is successfully verified and no fraud report is received, the system triggers a continuous update process for the behavioral habit benchmark data set. The system combines the real-time user operation behavior dataset after the successful verification of this payment transaction without fraud reports into a new positive sample. The system performs denoising processing on the new positive sample to eliminate random operation noise, and performs standardization processing on the new positive sample to unify the data scale and format. In some embodiments, the system extracts stable behavioral features from the new positive sample after denoising and standardization processing. Stable behavioral features refer to operation pattern features that repeatedly appear in multiple successful transactions and have a variance lower than a preset threshold. The extraction process can use a sliding time window and feature aggregation method. The system integrates the extracted stable behavioral features with certain weights into the corresponding user historical behavior feature model in the behavioral habit benchmark data set. The integration process uses an incremental update algorithm, and the update formula of the user historical behavior feature model is expressed as: .in, This represents the updated user history behavior feature model. This represents the user's historical behavior characteristics model before the update. This represents stable behavioral features extracted from new positive samples. Represents the preset weighting coefficients and The above calculations enable the gradual updating of the behavioral habit baseline data set.

[0042] It is understandable that denoising can employ filter algorithms to remove outliers in the real-time user behavior data set caused by device jitter or accidental touches. Standardization can transform data of different dimensions or ranges in the real-time user behavior data set, such as touch pressure values ​​and angle values, into a unified numerical range. The value of the weighting coefficient λ determines the relative proportion of historical and latest behavioral habits in the updated model. A smaller λ value makes the model update more conservative, while a larger λ value allows the model to adapt more quickly to gradual changes in operating habits. In specific implementations, the user historical behavior feature model can be a statistical model storing the mean and covariance matrix of feature vectors, or it can be an embedded representation model based on neural networks. The gradual update of the behavioral habit baseline data set occurs in the background asynchronous process after the payment transaction is successfully verified and securely confirmed, without blocking the main process of this transaction. Optionally, the weighting coefficient λ can be dynamically adjusted according to the environmental risk index of this transaction. A relatively larger λ value can be used when the environmental risk index is low, and a relatively smaller λ value can be used when the environmental risk index is high. Before new positive samples are integrated into the behavioral habit benchmark data set, they are first stored in a temporary buffer queue. After a predetermined number of successful transaction verifications, the system performs batch feature extraction and fusion on multiple new samples in the buffer queue, and then performs an update operation to improve computational efficiency and stability.

[0043] See Figure 5 This is a performance trend chart for batch updates of behavioral habit benchmark data, showing the relationship between the number of batch update samples and the model normalization performance index in a progressive update process. From a cost-benefit perspective, choosing 20-25 samples as the batch update unit is optimal. At this point, the accuracy is close to saturation, while the update stability is also at a high level, avoiding the waste of computational resources caused by excessively increasing the number of samples. In a practical system, a dynamic threshold can be set: when the number of new samples in the buffer queue reaches 20, a batch update is triggered to achieve a balance between model performance and computational efficiency. For high-risk transaction scenarios (such as large payments), the number of batch samples can be appropriately increased (e.g., 25-30) to pursue higher model stability. For low-risk scenarios, the number of batch samples can be reduced to 15-20 to reduce computational overhead and accelerate model iteration.

[0044] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A payment transaction encryption verification system with multi-factor authentication, characterized in that, include: The registration management module collects and stores a set of multi-dimensional biometric baseline data of users, a set of physical token device feature codes held by users, and a set of behavioral habit baseline data formed by users operating in a preset secure environment during the user registration phase. The real-time data acquisition module acquires network environment characteristic data, real-time biometric data set, and real-time user operation behavior data set in real time when a payment transaction is initiated. The risk decision-making module compares the network environment characteristic data with the historical trusted environment database, calculates the environmental risk index, and dynamically selects the required combination of authentication factors based on the environmental risk index. The multi-factor authentication module, based on the dynamically selected combination of authentication factors, calls the corresponding verification algorithm to verify the real-time biometric data set and the real-time user operation behavior data set, generating two independent authentication results; The token generation module uses a preset encryption synthesis algorithm to fuse the two independent authentication results with the plaintext transaction information of the current transaction request to generate the final encrypted payment token. The transaction verification module sends the encrypted payment token to the payment gateway for decryption and comparison to complete the transaction verification.

2. The payment transaction encryption verification system with multi-factor authentication as described in claim 1, characterized in that, The calculation of the environmental risk index based on the comparison between the network environment characteristic data and the historical trusted environment database includes: Extract the network access point identifier, geolocation information, current timestamp, and device hardware fingerprint from the network environment feature data; The matching degree of the network access point identifier, geolocation information, current timestamp and device hardware fingerprint is calculated by comparing them with the historical network access point identifier sequence, historical geolocation trajectory, historical common transaction time period and bound device hardware fingerprint of the user stored in the historical trusted environment database. Based on the matching degree calculation results, a preliminary environmental credibility score is obtained through weighted fusion. The preliminary environmental credibility score is revised by combining the deviation of the current transaction request amount and the unfamiliarity of the recipient to obtain the environmental risk index, wherein the environmental risk index is inversely proportional to the preliminary environmental credibility score.

3. The payment transaction encryption verification system with multi-factor authentication as described in claim 2, characterized in that, The dynamic selection of the required certification factor combination based on the environmental risk index includes: The authentication factor combination is selected from the multi-dimensional biometric baseline data set, the physical token device feature code set, and the behavioral habit baseline data set; Multiple environmental risk index thresholds are preset, and each environmental risk index threshold corresponds to a predefined certification strength level; The calculated environmental risk index is compared with the multiple environmental risk index thresholds to determine the current authentication strength level of the exchange. Each certification strength level is associated with a preset certification factor type and quantity requirement. The higher the certification strength level, the more certification factor types are required and the more complex the verification steps are. Based on the determined authentication strength level and its associated authentication factor type and quantity requirements, the corresponding type and quantity of benchmark data items are selected from the multi-dimensional biometric benchmark data set, physical token device feature code set, and behavioral habit benchmark data set to form the dynamically selected authentication factor combination.

4. The payment transaction encryption verification system with multi-factor authentication as described in claim 3, characterized in that, Verifying the real-time biometric dataset by invoking the corresponding verification algorithm includes: When the dynamically selected authentication factor combination includes biometric factors, the corresponding real-time biometric samples are extracted from the real-time biometric data set. The real-time biometric sample is compared with the corresponding benchmark template in the multi-dimensional biometric benchmark dataset; The feature comparison process uses a similarity measurement method based on deep feature embedding to calculate the distance between the real-time biometric sample and the corresponding benchmark template in the feature space; If the distance is less than the preset biometric matching threshold, an independent authentication result of successful biometric authentication is generated.

5. The payment transaction encryption verification system for multi-factor authentication as described in claim 3, characterized in that, The process of calling the corresponding verification algorithm to verify the real-time user behavior data set includes: When the dynamically selected authentication factor combination includes behavioral feature factors, the operation timing data of the current payment transaction input process is extracted from the real-time user operation behavior data set. The operation timing data includes a touch pressure change sequence, an input speed curve, and a device grip posture fine-tuning angle sequence. The operation time series data is input into a pre-trained behavior habit verification model, which is trained based on the behavior habit benchmark data set. The behavior habit verification model outputs a behavior feature matching confidence score. If the behavior feature matching confidence score exceeds a preset behavior verification threshold, an independent authentication result is generated indicating successful behavior feature authentication.

6. The payment transaction encryption verification system for multi-factor authentication as described in claim 4 or 5, characterized in that, The step of fusing the two independent authentication results with the plaintext transaction information of the current transaction request using a preset encryption synthesis algorithm includes: Each successful independent authentication result is converted into a unique authentication result code; The authentication result code, the environmental risk index, and a dynamically generated random number are concatenated to form an authentication information digest; The authentication message digest is initially encrypted using a session key derived from the user's private key to obtain primary encrypted data; The primary encrypted data and the plaintext transaction information are XORed and cyclically shifted to generate intermediate composite data. The intermediate synthesized data is encrypted using the public key of the payment gateway to generate the final encrypted payment token.

7. The payment transaction encryption verification system for multi-factor authentication as described in claim 6, characterized in that, The step of sending the encrypted payment token to the payment gateway for decryption and comparison includes: The payment gateway uses its private key to decrypt the received encrypted payment token to obtain the intermediate composite data; The payment gateway reverse-parses the plaintext transaction information and the primary encrypted data from the intermediate synthesized data; The payment gateway uses the user's pre-stored public key to decrypt the primary encrypted data and restore the authentication information digest; The payment gateway extracts the authentication result code and the environmental risk index from the authentication information digest, and verifies whether the selection of the authentication factor combination conforms to the strategy based on the environmental risk index; The payment gateway verifies the consistency between the parsed plaintext transaction information and the original transaction request information, and confirms that the authentication result encoding indicates that all required authentication factors have been successfully verified. When the consistency check and authentication verification pass, the payment gateway approves the payment transaction.

8. The payment transaction encryption verification system for multi-factor authentication according to claim 5, characterized in that, The behavioral habit verification model is constructed through the following steps: Collect multi-dimensional behavioral time-series data of users in historical secure payment transactions as training samples, and label each training sample as a positive sample of the user's own operation; The multi-dimensional behavioral timing data includes a pressure sequence of touch pressure changing over time, a speed curve of input speed changing over time, and a sequence of fine-tuning angles of the device's grip posture in three-dimensional space. The training samples are preprocessed, including data cleaning to remove abnormal noise points and time series alignment to unify data length and sampling frequency. Deep spatiotemporal features characterizing stable user operating habits are extracted from preprocessed time-series data. These deep spatiotemporal features are automatically learned through a hybrid model of stacked convolutional neural networks and long short-term memory networks. The extracted deep spatiotemporal features are input into a multilayer perceptron classifier for training. The output of the classifier is a confidence score that the user's operation behavior belongs to the user himself. The overall model consisting of the convolutional neural network, long short-term memory network, and multilayer perceptron classifier is jointly trained end-to-end using the backpropagation algorithm and labeled positive samples until the model converges, forming the initial behavioral habit verification model. During the system operation phase, new positive samples obtained through a continuous update mechanism are used to fine-tune the parameters of the behavioral habit verification model through online learning or periodic batch retraining, so as to achieve incremental optimization of the model.

9. The payment transaction encryption verification system for multi-factor authentication as described in claim 4, characterized in that, The feature comparison process employs a similarity metric based on deep feature embedding to calculate the distance between the real-time biometric sample and the corresponding benchmark template in the feature space, including: A pre-trained deep neural network feature extractor is used to perform forward propagation calculations on the real-time biometric samples and the corresponding benchmark templates in the multi-dimensional biometric benchmark dataset, respectively, to extract the first feature vector of the real-time biometric samples and the second feature vector of the benchmark templates. Calculate the metric distance between the feature vector of the real-time biometric sample and the feature vector of the baseline template, wherein the metric distance is Euclidean distance or cosine distance; The calculated metric distance is compared with the preset biometric matching threshold to determine whether the biometrics match.

10. The payment transaction encryption verification system for multi-factor authentication as described in claim 1, characterized in that, This also includes continuous updates to the behavioral habit benchmark dataset: After each payment transaction is successfully verified and no fraud report is received, the dataset of real-time user behavior is collected as a new positive sample. The new positive samples are denoised and standardized to extract stable behavioral features; The extracted stable behavioral features are incorporated into the corresponding user historical behavioral feature model in the behavioral habit benchmark data set with certain weights, thereby realizing the gradual update of the behavioral habit benchmark data set.

Images