Cross-border digital certificate authentication method supporting multi-modal semantic alignment and cross-chain mutual recognition

Abstract

This invention discloses a cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain interoperability, belonging to the field of information security and digital identity authentication technology. It includes: constructing a multimodal unified identity identification model; achieving cross-modal semantic alignment of X.509 certificates, DID identifiers, and biometric data through a semantic autoencoder nonlinear mapping function; achieving lossless multi-protocol conversion using an adaptive protocol conversion function and multi-dimensional signature mapping; realizing AI-powered end-to-end intelligent authentication and privacy enhancement through CRNN+CTC document recognition, homomorphic encryption nonlinear aggregation, and secret sharing; and constructing a deep neural network cross-chain data format conversion, multi-signature, and lightweight PKI verification system to achieve interoperability of digital certificates between blockchain platforms. This invention effectively solves key technical challenges such as semantic fusion, multi-protocol interoperability, privacy protection, and cross-chain interoperability of cross-border heterogeneous identity data, and can be widely applied in scenarios such as cross-border e-commerce, cross-border financial services, and supply chain management.

Description

Technical Field

[0001] This invention relates to the field of information security and digital identity authentication technology, specifically to a cross-border digital certificate authentication method that supports multimodal semantic alignment and cross-chain mutual recognition. Background Technology

[0002] With the rapid development of the global digital economy, the demand for cross-border digital identity authentication is becoming increasingly urgent in business scenarios such as cross-border e-commerce, cross-border financial services, and cross-border supply chain management. Economic and trade exchanges between China and ASEAN countries are becoming increasingly close; however, due to significant differences in digital certificate authentication systems, technical standards, and laws and regulations among countries, the mutual recognition of cross-border digital certificates faces numerous technical challenges.

[0003] Currently, existing digital certificate authentication technologies have the following main shortcomings: First, multi-source, heterogeneous identity data is difficult to integrate and interoperate with. Existing authentication systems differ significantly in data structure, encoding rules, and security verification methods. Different modalities of identity data, such as X.509 certificates, decentralized identity identifiers (DIDs), and biometric data, each employ independent data models and semantic representations. Traditional processing methods often use simple linear mapping or data concatenation, failing to effectively preserve the key semantic information of each modality, leading to a significant decrease in the accuracy of subsequent data fusion and identity reasoning. For example, X.509 certificates use the ASN.1 encoding format and a strict hierarchical trust system, DIDs are based on W3C standards and the JSON-LD data model, while biometric data refers to standards such as ISO / IEC 19794. The lack of a unified semantic alignment mechanism among these three makes seamless cross-platform and cross-domain data exchange and interoperability difficult.

[0004] Second, the coexistence of multiple protocols and signature algorithms leads to poor interoperability. In the China-ASEAN cross-border scenario, different countries and industries use varying standards for authentication protocols (such as SAML, OAuth, OpenID Connect, SM2, etc.) and signature algorithms (RSA, ECDSA, SM2, etc.). Current technologies often employ fixed mapping rules for converting protocol fields and signature formats, making it difficult to adapt to the dynamic changes in different systems. Inaccurate conversion can compromise the integrity of the original data and the legitimacy of the signature, posing serious security risks. Furthermore, the lack of a unified trust assessment mechanism makes it difficult to dynamically select the optimal trust anchor from multiple candidate trust points during cross-domain authentication, failing to effectively address the security risks of multi-source data flows.

[0005] Third, the automation and privacy protection levels of cross-border identity authentication are insufficient. In existing identity authentication processes, the recognition and verification of document images often rely on manual review or single technical means, which is inefficient and prone to errors. Simultaneously, cross-border data transmission involves the cross-domain flow of sensitive user information, and existing technologies have significant shortcomings in privacy protection. Traditional encryption schemes either cannot support data operations in encrypted states or are too inefficient to meet actual business needs, posing a risk of user privacy leakage during cross-border authentication.

[0006] Fourth, there is a lack of effective technical support for the mutual recognition of digital certificates between blockchain platforms. Currently, digital identity authentication in China and ASEAN countries is fragmented across different blockchain platforms, with differences in data formats, consensus mechanisms, and smart contracts. Existing cross-chain technologies either rely on centralized trust intermediaries or have excessively high communication complexity (e.g., n organizations require n...). 2 The communication complexity makes it difficult to achieve efficient, secure, and decentralized cross-chain certificate mutual recognition and joint auditing.

[0007] Fifth, the limited terminal equipment capabilities in the ASEAN region place an excessive burden on traditional PKI systems. Some ASEAN countries have relatively limited network resources and terminal equipment capabilities. Traditional RSA-based Public Key Infrastructure (PKI) systems, with their long key lengths and high computational demands, are difficult to operate efficiently in resource-constrained environments. Existing lightweight solutions often sacrifice security, making it difficult to achieve a good balance between security and efficiency.

[0008] In summary, existing technologies lack a systematic solution that can simultaneously address a series of issues such as unified modeling of multimodal identity data, adaptive conversion of multiple protocols, intelligent authentication and privacy protection throughout the entire process, cross-chain mutual recognition, and lightweight deployment. Summary of the Invention

[0009] This invention addresses the aforementioned shortcomings of existing technologies by providing a cross-border digital certificate authentication method that supports multimodal semantic alignment and cross-chain interoperability. It constructs a multimodal unified identity model to achieve high-precision alignment and fusion of multi-source heterogeneous identity data, such as X.509 certificates, DID identifiers, and biometric data, within a unified semantic space. It enables adaptive lossless conversion between multiple protocol and multi-signature algorithms and intelligent selection of dynamic trust anchors. It achieves AI-driven end-to-end intelligent authentication and enhanced privacy protection. It enables cross-chain interoperability and lightweight, efficient verification of digital certificates between blockchain platforms. Finally, it supports trusted interoperability and business linkage of digital certificates in cross-border multi-scenario applications.

[0010] To achieve the above-mentioned objectives, the present invention provides the following technical solution: A cross-border digital certificate authentication method that supports multimodal semantic alignment and cross-chain interoperability includes the following steps: Step S1: Construct a multimodal unified identity identification model, and perform nonlinear semantic alignment and fusion on identity data from different sources to obtain identity data to be authenticated. The identity data includes at least two or more of the following: X.509 certificate data, decentralized identity identification (DID) data, and biometric identification data. The X.509 certificate data is parsed using ASN.1 parsing rules combined with the JWT standard, through a mapping function. Extract issuing authority information, entity information, validity period information, and public key information; Calculate the second-order interaction term ,in The SHA-256 hash function is used. This represents the covariance between the issuing authority information and the subject information. For covariance weighting parameters; For the biometric identification data, a unified template was constructed according to the ISO / IEC 19794 standard. Multimodal images of fingerprints, irises, and faces were collected and features were extracted. Kernel principal component analysis was used for nonlinear dimensionality reduction mapping. ,in For kernel function The constructed characteristic matrix, This is the transformation matrix obtained through eigenvalue decomposition; Using a semantic autoencoder to learn a nonlinear mapping function This involves mapping identity data from different sources to a unified latent space for semantic alignment. For learnable matrices, The bias vector is used to construct a unified identity knowledge graph using the aligned latent space vectors. To achieve cross-platform interoperability of identity data; Step S2: Based on the adaptive protocol conversion and dynamic trust anchor mechanism, realize lossless conversion between multiple protocol and signature algorithms and intelligent selection of trust anchors; build a protocol template library covering SAML, OAuth, OpenID Connect and local authentication standards of various countries. For the source protocol data to be converted Adaptive protocol conversion function is used Perform field-level format conversion and adaptively solve parameters using online machine learning. , , This ensures that the converted data retains complete semantic information; For the signature to be converted Employing a multidimensional signature mapping function Signature format conversion is performed, and the matrix is ​​adjusted using online machine learning. sum vector This ensures that the original signature can be verified as a valid signature by the target system after conversion; Each authentication event is represented as a node in a graph structure, and a corresponding feature vector is constructed. Feature updates are performed using graph convolutional networks: ; in, Given a weighted adjacency matrix containing self-loops, Representing the Layer feature representation, The learned weight matrix, The bias matrix, For activation functions; A hybrid time-series prediction model combining LSTM and ARIMA is constructed based on historical authentication data, and a trust scoring function is defined. ,in This provides real-time scores for various trust indicators. For activation function, As the indicator weights, based on the aforementioned trust scoring function Dynamically select the optimal trust anchor from multiple candidate trust anchors; Step S3: Perform AI-driven end-to-end intelligent authentication and privacy enhancement processing: The user-submitted ID document image is preprocessed, and OCR recognition is performed using a CRNN architecture. This CRNN architecture combines convolutional neural networks and recurrent neural networks, and uses the CTC loss function for sequence output correction. ,in Image features extracted by the convolutional layer The set of parameters obtained during training; Calculate forgery sensitive indicators The authenticity of documents is determined by multi-scale convolution and frequency domain analysis. If the value exceeds the preset threshold, it will be judged as a forged document and the authentication process will be terminated. A residual graph convolutional network is used to construct a behavioral graph of authentication events for real-time feature updates. ; It also incorporates long short-term memory networks for risk time series prediction. ,in For the noise term, correction is performed using the minimum mean square error. Homomorphic encryption algorithm is used to encrypt plaintext data Convert to ciphertext Based on the nonlinear weighted aggregation operation rules Data processing is performed in encrypted form, where For aggregation parameters; Data is broken down into multiple parts using secret sharing technology. Each computing node independently stores data fragments and jointly computes to recover the data; Step S4: Based on cross-chain mutual recognition and a lightweight PKI system, achieve mutual recognition and efficient verification of digital certificates between blockchain platforms. Addressing the differences in authentication data formats between different blockchain platforms, a deep neural network model is used for cross-chain data format conversion. ,in For source chain authentication data, For deep neural network transformation functions, For network parameters, As compensation; Adopting a multi-signature scheme To achieve the fusion of signatures from multiple participating nodes, where For the first The signature of each node, For the corresponding weighting coefficients, Modulus; Cross-chain data is verified and synchronized using PBFT or Raft consensus algorithms, and each on-chain operation is recorded in a chain log with a random salt value. Among them The salt value is random. Elliptic curve cryptography algorithm The alternative to the RSA algorithm, in which For fixed generators on elliptic curves, For the private key; construct a distributed key management center and use secret sharding technology to decompose the private key into multiple parts. Each node only stores a fragment of the key; Lightweight verification formula is used in the certificate verification process. Complete certificate verification, among which The hash value is calculated using a lightweight hash function. , and functions Preset parameters for the system. This is the fault tolerance threshold; Step S5: Perform cross-border digital identity verification for multiple application scenarios: Digital certificates, after being processed by a unified identity identification model, protocol conversion, smart authentication, and cross-chain mutual recognition, will be used for identity authentication and business linkage in cross-border service scenarios.

[0011] Preferably, in step S1, a unified identity knowledge graph is constructed. Specifically, it includes: After the parsed X.509 certificate data, DID data, and biometric data are processed by the middleware platform, a heterogeneous data fusion function is used. Achieve seamless conversion from different formats to a unified JSON-LD structure, where This includes various types of parsed identity data; By combining graph embedding algorithms to update the RDF graph structure, a matrix reflecting the semantic relationships between the data is formed. Based on the unified identity knowledge graph Perform cross-platform identity reasoning and security verification.

[0012] Preferably, step S1 further includes: During cross-border data transmission, a timestamp based on timestamp synchronization is appended to each data item. Generate a signature by combining elliptic curve digital signature mechanism ,in For hash functions, For private key; Using Merkle trees to verify the overall integrity of multiple data segments: Let the data segment be... Calculate the hash value for each fragment. The parent node is constructed by combining adjacent hash values ​​and hashing them again. until a unique root hash is constructed. The system employs a distributed microservice architecture to deploy various functional modules, with communication between modules achieved through lightweight message queues. Role-based access control (RBAC) and TLS protocols are used to protect data transmission.

[0013] Preferably, the graph convolutional network based on the authentication event graph in step S2 is also used for real-time detection of abnormal patterns: The feature vector of each authentication event The input graph convolutional network performs multi-layer feature updates. The feature representation output by the last layer is compared with a preset anomaly detection threshold. When abnormal behavior is detected, a security warning is triggered and the trust anchor point is switched.

[0014] Preferably, the use of homomorphic encryption and secret sharing technology to ensure data privacy in step S3 specifically includes: A hybrid encryption mode of AES and RSA is used during data transmission. Furthermore, IPSec and TLS tunneling technologies are used to enhance the security of cross-border transmissions; Embedding hybrid digital signatures at every stage of the authentication process Combined with timestamps Ensure that operation logs are non-repudiable and traceable.

[0015] Preferably, the deep neural network model for cross-chain data format conversion in step S4 The training process includes: Certificate Authority (CA) certification data from multiple blockchain platforms were collected as training samples, using source chain data. For input and target chain data Using labels, network parameters are optimized by minimizing the transformation error loss function. This makes the converted data Maintain the integrity and encryption properties of the original data.

[0016] Preferably, in step S4, a timeout retry strategy is introduced into the smart contract, and the random delay compensation formula is as follows: ,in For average delay, The standard deviation of the delay. The random numbers are standard normally distributed to ensure data transmission is completed in high-latency or packet-loss environments.

[0017] Preferably, the cross-border service scenarios in step S5 include at least one of the following: cross-border service with one certificate, cross-border linkage of digital identity, and trusted mutual recognition of supply chain finance. In the aforementioned cross-border service using a single certificate, dynamic semantic linking functions are utilized. Combine identity information with the current spatiotemporal context Combined, after successful authentication, a state transition function is used. Automatically triggers the linkage and approval of customs declaration, logistics and fund settlement systems; In the aforementioned cross-border digital identity linkage, the Diffie-Hellman protocol is used to generate a shared key: Real-time data synchronization and secure transmission are achieved; an improved signature format is adopted in the aforementioned trusted mutual recognition of supply chain finance. Sign the contract and invoice, among which The time perturbation function is used; a risk warning model is employed. Enable real-time risk monitoring.

[0018] Compared with the prior art, the present invention has the following advantages: 1. This invention innovatively proposes a nonlinear mapping function based on a semantic autoencoder. With second-order interaction terms The combined approach involves designing learnable matrices A, vectors b, and covariance weights. It achieves high-precision alignment of identity data from different sources, such as X.509 certificates, DID identifiers, and biometric data, in a unified latent space. This not only preserves the inherent characteristics of each system but also eliminates structural differences, significantly improving the accuracy and robustness of cross-platform fusion and identity reasoning. 2. This invention constructs a protocol template library covering SAML, OAuth, OIDC, and local standards of various countries, and innovatively designs a system based on... Field-level adaptive conversion function and multidimensional signature mapping By continuously optimizing transformation parameters through online machine learning, end-to-end lossless protocol transformation and signature compatibility were achieved for the first time. Simultaneously, a trust scoring function was designed by combining graph convolutional networks with LSTM-ARIMA hybrid temporal prediction. This enables intelligent selection of dynamic trust anchors and real-time anomaly warnings, significantly improving the security protection capabilities of cross-border authentication. 3. This invention is the first to deeply integrate CRNN+CTC text recognition with multi-scale convolution and frequency domain analysis to create forgery-sensitive indicators throughout the entire authentication process. An end-to-end automated OCR parsing and forgery screening pipeline was constructed, and a residual GCN+LSTM model was used to predict the risk of authentication behavior timing. For privacy protection, homomorphic encryption and nonlinear aggregation operations were employed. Secretly shared with MPC The organic combination of these technologies enables data computation and multi-party collaborative computation in encrypted form, ensuring that sensitive user information remains encrypted throughout the transmission and processing process.

[0019] 4. This invention proposes a cross-chain data format conversion based on deep neural networks. With multiple joint signatures Combining PBFT / Raft consensus and timeout compensation strategies This achievement marks the first time that seamless interoperability and joint auditing between blockchain platforms have been realized. Furthermore, considering the resource-constrained network and terminal conditions in the ASEAN region, ECC has been introduced to replace RSA, and a compact certificate structure and distributed key sharding update mechanism have been designed, enabling low-power, high-speed certificate management and verification. 5. Through multi-scenario application verification such as cross-border services with a single certificate, cross-border linkage of digital identity, and trusted mutual recognition of supply chain finance, this invention has realized business linkage and automatic approval under a unified data standard, providing secure and efficient digital identity authentication technology support for China-ASEAN cross-border trade, financial services, and personnel exchanges. Attached Figure Description

[0020] Figure 1This is a schematic diagram of the overall process of the cross-border digital certificate authentication method that supports multimodal semantic alignment and cross-chain mutual recognition according to the present invention. Figure 2 This is a flowchart illustrating the construction of the multimodal unified identity identification model and nonlinear semantic alignment in step S1 of the present invention. Figure 3 This is a flowchart illustrating the adaptive protocol conversion and dynamic trust anchor selection process in step S2 of the present invention. Figure 4 This is a flowchart illustrating the AI-driven intelligent authentication and privacy enhancement process in step S3 of the present invention. Figure 5 This is a flowchart illustrating the cross-chain mutual recognition and lightweight PKI process in step S4 of the present invention. Figure 6 This is a schematic diagram of the module structure of the cross-border digital certificate authentication system that supports multimodal semantic alignment and cross-chain mutual recognition according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] Example 1: This embodiment provides a cross-border digital certificate authentication method that supports multimodal semantic alignment and cross-chain mutual recognition.

[0023] I. Construction and Semantic Alignment of a Multimodal Unified Identity Model: This embodiment first obtains the identity data to be authenticated. In cross-border electronic authentication scenarios, users may submit various types of data, including X.509 digital certificates, decentralized identity (DID) documents, and biometric data (such as fingerprints, facial images, iris images, etc.).

[0024] For X.509 certificate data, this embodiment uses ASN.1 parsing rules combined with the JWT standard for parsing. An X.509 certificate is a binary-encoded ASN.1 data structure. An ASN.1 decoder decodes the binary data X into readable certificate fields, and then a mapping function is used... Extract the issuing authority information, entity information, validity period information, and public key information.

[0025] Based on this, this embodiment calculates the second-order interaction term. .in, Using the SHA-256 hash function, This represents the covariance between the issuing authority information and the subject information. Specifically, the issuing authority information I and the subject information S are encoded into vector form, and the covariance matrix between them is calculated to capture the intrinsic relationship between the issuing authority and the subject. Covariance weighting parameters. The optimal value is determined through cross-validation, typically ranging from 0.1 to 0.5. This second-order interaction term effectively enhances the certificate parsing results' ability to resist noise and format differences.

[0026] For biometric identification data, this embodiment constructs a unified template based on the ISO / IEC 19794 standard. After acquiring multimodal images such as fingerprints, irises, and faces, image processing techniques (including histogram equalization, Gaussian filtering, and edge enhancement) are used to extract features. For high-dimensional feature data, kernel principal component analysis (KPCA) is used for nonlinear dimensionality reduction mapping. Among them, the kernel function Selecting radial basis kernel function ,pass Eigenvalue decomposition yields the transformation matrix U, which maps the original high-dimensional feature data to a low-dimensional manifold space, reducing the data dimensionality while preserving nonlinear structural information.

[0027] After obtaining various parsed identity data, this embodiment uses a semantic autoencoder to learn a nonlinear mapping function. The semantic autoencoder consists of an encoder and a decoder. The encoder encodes identity data from different sources into a unified latent space, and the decoder reconstructs the original data from the latent space representation. By minimizing the reconstruction error and semantic alignment loss, a learnable matrix A and a bias vector b are learned. This nonlinear mapping enables effective alignment of identity data from different sources in the latent space, eliminating differences in data structure between different systems.

[0028] After the parsed data is processed by the middleware platform, heterogeneous data fusion functions are used. Achieve seamless conversion from different formats to a unified JSON-LD structure. JSON-LD is a JSON-based relational data format that can flexibly express semantic relationships between data from different sources. Combined with graph embedding algorithms (such as TransE, DistMult, etc.), update the RDF graph structure to form a matrix reflecting the semantic relationships between the data. A unified identity knowledge graph is constructed. Based on this knowledge graph, cross-platform identity reasoning and security verification can be performed, such as determining whether an identity identifier matches a biometric template through graph queries.

[0029] To ensure data integrity and authenticity during cross-border data transmission, this embodiment employs a dual-digest mechanism. A timestamp based on timestamp synchronization is appended to each data entry. Generate a signature by combining elliptic curve digital signature mechanism Meanwhile, a Merkle tree is used to verify the overall integrity of multiple data segments: Let the data segment be... Calculate the hash value for each fragment. The parent node is constructed by combining adjacent hash values ​​and hashing them again. until a unique root hash is constructed. This structure ensures that any modification to a data segment will result in a change to the root hash, thereby enabling global data integrity verification.

[0030] In terms of system deployment, this embodiment adopts a distributed microservice architecture. Each functional module (certificate parsing module, biometric processing module, semantic alignment module, knowledge graph construction module, etc.) achieves efficient communication through lightweight message queues (such as RabbitMQ and Kafka). Simultaneously, role-based access control (RBAC) and TLS protocol are used to protect data transmission.

[0031] II. Adaptive Protocol Conversion and Dynamic Trust Anchor Selection: This embodiment first constructs a protocol template library covering SAML, OAuth, OpenID Connect, and local authentication standards of various countries. The protocol template library stores information such as field definitions, data types, encoding rules, and message formats for each protocol.

[0032] For the source protocol data to be converted Adaptive protocol conversion function is used Perform field-level format conversion. This sinusoidal conversion function takes parameters... Control the amplitude, Control frequency, By controlling the offset, it can flexibly adapt to the numerical mapping relationships between different protocol fields. Parameters are adaptively solved using online machine learning methods (such as online gradient descent algorithms). , , This ensures that the converted data retains complete semantic information in the target protocol.

[0033] For the signature to be converted Employing a multidimensional signature mapping function Signature format conversion is performed. Signatures generated by different signature algorithms (such as RSA, ECDSA, and SM2) differ in format, length, and encoding. The matrix is ​​adjusted using online machine learning. sum vector This transformation ensures that the original signature can be verified as a valid signature by the target system. Specifically, matrix M performs a linear transformation of the signature vector, and vector c performs offset adjustment. Together, they ensure that the transformed signature meets the verification requirements of the target system.

[0034] Regarding the selection of trust anchors, this embodiment constructs each authentication event as a node in a graph structure and generates a corresponding feature vector. (Including features such as authentication time, source IP, protocol type, and signature verification result), a graph convolutional network is used for feature updating: ; in Given a weighted adjacency matrix containing self-loops, Representing the Layer feature representation, The learned weight matrix, is the bias matrix. Through multi-layer graph convolution, the feature representation of each node incorporates information from its neighboring nodes, thereby enabling the capture of correlation patterns and anomalous behaviors between authentication events.

[0035] A hybrid time-series forecasting model combining LSTM and ARIMA is constructed based on historical authentication data. LSTM (Long Short-Term Memory) excels at capturing long-term dependencies in time series, while ARIMA (Autoregressive Integral Moving Average) excels at capturing linear trends and seasonal patterns. The combination of the two can comprehensively characterize the temporal evolution of trust indicators. A trust rating function is defined. ,in Real-time scores for various trust metrics (including response speed, signature verification reliability, data integrity, etc.). For activation functions (such as ReLU or ELU). These are the indicator weights. The weights of each indicator. and activation function All are dynamically adjusted based on historical data, and continuously optimized by minimizing prediction errors, thereby achieving the real-time selection of the best trust anchor point.

[0036] The feature vector of each authentication event The input graph is processed by a convolutional network to update multiple layers of features. The feature representation output from the last layer is compared with a preset anomaly detection threshold. When abnormal behavior is detected (such as a large number of failed authentication attempts in a short period of time, authentication requests from abnormal geographical locations, etc.), a security alert is triggered and the trust anchor point is automatically switched.

[0037] III. AI-driven end-to-end intelligent authentication and privacy enhancement processing: This embodiment preprocesses user-submitted ID card images, including image enhancement (contrast stretching, histogram equalization), noise filtering (median filtering, Gaussian filtering), and edge detection (Canny operator). OCR recognition is performed using a CRNN architecture, which combines a convolutional neural network (CNN) and a recurrent neural network (RNN). The CNN part (such as ResNet, VGG, etc.) extracts spatial features of the image, while the RNN part (such as bidirectional LSTM) captures sequence features. The CTC loss function is used for sequence output correction. ,in Image features extracted by the convolutional layer This is the set of parameters obtained during training.

[0038] To eliminate the risk of counterfeit materials, this embodiment calculates counterfeit-sensitive indicators. This metric extracts gradient information from images at different scales through multi-scale convolution and combines it with frequency domain analysis (such as Fourier transform) to detect forgery traces in the images. If the threshold is exceeded, the document is identified as counterfeit and the authentication process is terminated. Genuine documents exhibit specific statistical patterns in gradient distribution and frequency domain features, while counterfeit documents often show anomalies in these features.

[0039] A residual graph convolutional network is used to construct a behavioral graph of authentication events for real-time feature updates. ; This residual structure uses the current layer features Adding directly to the updated features effectively alleviates the gradient vanishing problem in deep networks and improves the accuracy of risk prediction. This is combined with a Long Short-Term Memory (LSTM) network for risk time-series prediction. ,in The noise term is corrected using the minimum mean square error.

[0040] In terms of privacy enhancement, this embodiment uses homomorphic encryption algorithms (such as the Paillier encryption scheme or the CKKS scheme) to encrypt plaintext data. Convert to ciphertext Based on the nonlinear weighted aggregation operation rules Data operations are performed in encrypted form. This operation rule introduces a nonlinear term based on standard homomorphic addition. This is suitable for aggregation scenarios that require calculating cross terms (such as feature crosses in certain machine learning models). By appropriately selecting aggregation parameters... It can perform data aggregation operations with nonlinear terms directly in encrypted form without revealing the original information.

[0041] Data is broken down into multiple parts using secret sharing techniques (such as Shamir Secret Share). Each computing node independently stores data fragments and collaborates to recover the data. Even if some nodes are compromised, attackers cannot obtain the complete original data.

[0042] A hybrid encryption mode of AES and RSA is used during data transmission. Specifically, using AES symmetric keys. Encrypt the data m, then use the RSA public key. Encrypting the AES key ensures both encryption efficiency and secure key distribution. Furthermore, IPSec and TLS tunneling technologies are used to enhance cross-border transmission security.

[0043] Embedding hybrid digital signatures at every stage of the authentication process Combined with timestamps Ensure that operation logs are non-repudiable and traceable.

[0044] IV. Cross-chain mutual recognition and lightweight PKI verification To address the differences in authentication data formats across different blockchain platforms, this embodiment employs a deep neural network model for cross-chain data format conversion. Deep neural networks Employing a multi-layer fully connected network or Transformer architecture, the mapping relationship from the source chain data format to the target chain data format is learned through end-to-end training. Compensation term. Used to correct systematic biases that may occur during the conversion process.

[0045] Certificate Authority (CA) certification data from multiple blockchain platforms were collected as training samples, using source chain data. For input and target chain data Using labels, network parameters are optimized by minimizing the transformation error loss function. This makes the converted data Maintain the integrity and encryption properties of the original data.

[0046] Adopting a multi-signature scheme This enables the fusion of signatures from multiple participating nodes. Each participating node signs its own signature on the same piece of data. The signatures are merged into a joint signature through a weighted product. Weighting coefficients It can be dynamically adjusted based on the trust level of each node.

[0047] Cross-chain data is verified and synchronized using either PBFT (Practical Byzantine Fault Tolerance) or Raft consensus algorithms. Each on-chain operation is recorded in a chain log with a random salt value. Random salt value The introduction of this feature effectively prevents replay attacks and dictionary attacks.

[0048] Introducing a timeout retry strategy into smart contracts, the random delay compensation formula is as follows: ,in For average delay, The standard deviation of the delay. The random number is a standard normally distributed number. When a cross-chain transaction fails to be confirmed in a timely manner due to network latency or node failure, the smart contract automatically retryes according to this random delay, avoiding network congestion caused by all nodes retrying at the same time.

[0049] Regarding lightweight PKI, this embodiment employs elliptic curve cryptography. Elliptic curve cryptography (ECC) is an alternative to the RSA algorithm. At the same level of security, ECC has a much shorter key length than RSA (e.g., 256-bit ECC is comparable to 3072-bit RSA in terms of security strength), significantly improving computational speed and storage efficiency.

[0050] Construct a distributed key management center and use secret sharding technology to decompose the private key into multiple parts. Each node stores only a fragment of the key. Even if a node is compromised, the attacker cannot obtain the complete private key. An automatic key update mechanism is also designed to periodically update the key within the certificate's validity period, reducing the security risks associated with using the same key for an extended period.

[0051] Lightweight verification formula is used in the certificate verification process. Certificate verification is completed. This verification formula combines elliptic curve multiplication with a lightweight hash function, avoiding the complex modular exponentiation operations in traditional certificate verification, thus enabling rapid verification even in resource-constrained environments on terminal devices.

[0052] This embodiment uses the digital certificate, after being processed by a unified identity model, protocol conversion, smart authentication, and cross-chain mutual recognition, for the following cross-border service scenarios: "One-Card-for-All" Cross-Border Service Scenario: Utilizing Dynamic Semantic Link Functions Combine identity information with the current spatiotemporal context This involves combining various information. For example, when a company conducts cross-border customs declaration, the system obtains contextual information such as the current time, location (port), and business type, and combines this information with the identity information in the digital certificate to construct a semantic link, ensuring semantic consistency of data during cross-border exchange. After successful authentication, a state transition function is used... It automatically triggers the linkage and approval of customs declaration, logistics and fund settlement systems, and the status of each subsystem is automatically updated. At the same time, it records the operation flow in an unalterable manner for tracking.

[0053] Cross-border digital identity linkage scenario: Generating shared keys using the Diffie-Hellman protocol. Achieve real-time data synchronization and secure transmission. Construct a joint verification vector by combining device fingerprints, dynamic verification codes, and biometric information. Furthermore, linear discriminant analysis is used to establish decision boundaries, enabling multi-dimensional and rapid identity verification.

[0054] Trusted and mutually recognized supply chain finance scenarios: Adopting improved signature formats Sign the contract and invoice. Time perturbation function. This makes each signature more random and unique. During cross-chain data transmission, for each data segment... Calculate checksum It generates precise timestamps to ensure complete recording and non-repudiation. A risk warning model is employed. Enables real-time risk monitoring. Various risk indicators (such as abnormal transaction amounts, abnormal transaction frequencies, changes in counterparty credit ratings, etc.) are combined through non-linear weighted summaries to output a comprehensive risk score. When the score exceeds a threshold, an alert is automatically triggered.

[0055] The overall architecture adopts a microservice architecture, and each functional module (electronic contract generation, signing, data transmission, automatic verification, and risk warning) can be deployed independently using containerization (such as Docker and Kubernetes) and an adaptive load balancing algorithm. Ensure optimal resource scheduling and stable system operation in high-concurrency scenarios.

[0056] Example 2: This embodiment provides a cross-border digital certificate authentication system that supports multimodal semantic alignment and cross-chain mutual recognition.

[0057] The system includes a multimodal unified identity module, used to perform the multimodal unified identity model construction and nonlinear semantic alignment operation described in step S1. This module further includes an X.509 certificate parsing submodule, a DID parsing submodule, a biometric processing submodule, a semantic autoencoder submodule, and a knowledge graph construction submodule.

[0058] The adaptive protocol conversion and trust anchor selection module is used to perform the adaptive protocol conversion and dynamic trust anchor selection operations described in step S2. This module further includes a protocol template library, a protocol conversion submodule, a signature mapping submodule, a graph convolutional network submodule, and a trust score submodule.

[0059] The AI-driven intelligent authentication and privacy enhancement module is used to perform the AI-driven end-to-end intelligent authentication and privacy enhancement processing operations described in step S3. This module further includes an OCR recognition submodule, a forgery detection submodule, a risk prediction submodule, a homomorphic encryption submodule, and a secret sharing submodule.

[0060] The cross-chain mutual recognition and lightweight PKI module is used to perform the cross-chain data format conversion, multi-signature, and lightweight certificate verification operations described in step S4. This module further includes a cross-chain data conversion submodule, a multi-signature submodule, a consensus synchronization submodule, an ECC key management submodule, and a lightweight verification submodule.

[0061] The multi-scenario application verification module is used to perform the cross-border digital identity multi-scenario application verification operation described in step S5. This module further includes a "One-Card-for-All" cross-border service sub-module, a digital identity cross-border linkage sub-module, and a supply chain finance trusted mutual recognition sub-module.

[0062] The modules communicate with each other via RESTful APIs, and the overall deployment adopts a distributed microservice architecture.

[0063] Example 3: This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0064] The computer-readable storage medium can be any medium capable of storing programs, including but not limited to: ROM, RAM, magnetic disk, optical disk, flash memory, hard disk, etc. The processor can be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices.

[0065] Industrial applicability: This invention can be widely applied in the following fields: Cross-border e-commerce: Provide identity verification services for buyers and sellers on the China-ASEAN cross-border e-commerce platform to ensure the authenticity and credibility of the identities of both parties in the transaction.

[0066] Cross-border financial services: Provide digital identity authentication support for financial services such as cross-border payments, cross-border loans, and cross-border insurance, meeting anti-money laundering and customer identification requirements.

[0067] Cross-border supply chain management: providing trusted identity authentication and electronic signature services for cross-border logistics, customs declaration, contract signing and other supply chain links.

[0068] Cross-border personnel exchanges: Provide digital identity mutual recognition services for cross-border personnel exchanges between China and ASEAN countries (such as business travel, study abroad, and employment).

[0069] Government public services: Provide secure and reliable identity authentication infrastructure for government public services such as cross-border government approvals and cross-border data exchange.

[0070] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition, characterized in that: Includes the following steps: Step S1: Construct a multimodal unified identity identification model, and perform nonlinear semantic alignment and fusion on identity data from different sources to obtain identity data to be authenticated. The identity data includes at least two or more of the following: X.509 certificate data, decentralized identity identification (DID) data, and biometric identification data. The X.509 certificate data is parsed using ASN.1 parsing rules combined with the JWT standard, through a mapping function. Extract issuing authority information, entity information, validity period information, and public key information; Calculate the second-order interaction term ,in The SHA-256 hash function is used. This represents the covariance between the issuing authority information and the subject information. For covariance weighting parameters; For the biometric identification data, a unified template was constructed according to the ISO / IEC 19794 standard. Multimodal images of fingerprints, irises, and faces were collected and features were extracted. Kernel principal component analysis was used for nonlinear dimensionality reduction mapping. ,in For kernel function The constructed characteristic matrix, This is the transformation matrix obtained through eigenvalue decomposition; Using a semantic autoencoder to learn a nonlinear mapping function This involves mapping identity data from different sources to a unified latent space for semantic alignment. For learnable matrices, The bias vector is used to construct a unified identity knowledge graph using the aligned latent space vectors. To achieve cross-platform interoperability of identity data; Step S2: Based on the adaptive protocol conversion and dynamic trust anchor mechanism, realize lossless conversion between multiple protocol and signature algorithms and intelligent selection of trust anchors; build a protocol template library covering SAML, OAuth, OpenID Connect and local authentication standards of various countries. For the source protocol data to be converted Adaptive protocol conversion function is used Perform field-level format conversion and adaptively solve parameters using online machine learning. , , This ensures that the converted data retains complete semantic information; For the signature to be converted Employing a multidimensional signature mapping function Signature format conversion is performed, and the matrix is ​​adjusted using online machine learning. sum vector This ensures that the original signature can be verified as a valid signature by the target system after conversion; Each authentication event is represented as a node in a graph structure, and a corresponding feature vector is constructed. Feature updates are performed using graph convolutional networks: ； in, Given a weighted adjacency matrix containing self-loops, Representing the Layer feature representation, The learned weight matrix, The bias matrix, For activation functions; A hybrid time-series prediction model combining LSTM and ARIMA is constructed based on historical authentication data, and a trust scoring function is defined. ,in This provides real-time scores for various trust indicators. For activation function, As the indicator weights, based on the aforementioned trust scoring function Dynamically select the optimal trust anchor from multiple candidate trust anchors; Step S3: Perform AI-driven end-to-end intelligent authentication and privacy enhancement processing: The user-submitted ID document image is preprocessed, and OCR recognition is performed using a CRNN architecture. This CRNN architecture combines convolutional neural networks and recurrent neural networks, and uses the CTC loss function for sequence output correction. ,in Image features extracted by the convolutional layer The set of parameters obtained during training; Calculate forgery sensitive indicators The authenticity of documents is determined by multi-scale convolution and frequency domain analysis. If the value exceeds the preset threshold, it will be judged as a forged document and the authentication process will be terminated. A residual graph convolutional network is used to construct a behavioral graph of authentication events for real-time feature updates. ； It also incorporates long short-term memory networks for risk time series prediction. ,in For the noise term, correction is performed using the minimum mean square error. Homomorphic encryption algorithm is used to encrypt plaintext data Convert to ciphertext Based on the nonlinear weighted aggregation operation rules Data processing is performed in encrypted form, where For aggregation parameters; Data is broken down into multiple parts using secret sharing technology. Each computing node independently stores data fragments and jointly computes to recover the data; Step S4: Based on cross-chain mutual recognition and a lightweight PKI system, achieve mutual recognition and efficient verification of digital certificates between blockchain platforms. Addressing the differences in authentication data formats between different blockchain platforms, a deep neural network model is used for cross-chain data format conversion. ,in For source chain authentication data, For deep neural network transformation functions, For network parameters, As compensation; Adopting a multi-signature scheme To achieve the fusion of signatures from multiple participating nodes, where For the first The signature of each node, For the corresponding weighting coefficients, Modulus; Cross-chain data is verified and synchronized using PBFT or Raft consensus algorithms, and each on-chain operation is recorded in a chain log with a random salt value. Among them The salt value is random. Elliptic curve cryptography algorithm The alternative to the RSA algorithm, in which For fixed generators on elliptic curves, For the private key; construct a distributed key management center and use secret sharding technology to decompose the private key into multiple parts. Each node only stores a fragment of the key; Lightweight verification formula is used in the certificate verification process. Complete certificate verification, among which The hash value is calculated using a lightweight hash function. , and functions Preset parameters for the system. This is the fault tolerance threshold; Step S5: Perform cross-border digital identity verification for multiple application scenarios: Digital certificates, after being processed by a unified identity identification model, protocol conversion, smart authentication, and cross-chain mutual recognition, will be used for identity authentication and business linkage in cross-border service scenarios.

2. The cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition as described in claim 1, characterized in that, In step S1, a unified identity knowledge graph is constructed. Specifically, it includes: After the parsed X.509 certificate data, DID data, and biometric data are processed by the middleware platform, a heterogeneous data fusion function is used. Achieve seamless conversion from different formats to a unified JSON-LD structure, where This includes various types of parsed identity data; By combining graph embedding algorithms to update the RDF graph structure, a matrix reflecting the semantic relationships between the data is formed. Based on the unified identity knowledge graph Perform cross-platform identity reasoning and security verification.

3. The cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition according to claim 1, characterized in that, Step S1 further includes: During cross-border data transmission, a timestamp based on timestamp synchronization is appended to each data item. Generate a signature by combining elliptic curve digital signature mechanism ,in For hash functions, For private key; Using Merkle trees to verify the overall integrity of multiple data segments: Let the data segment be... Calculate the hash value for each fragment. The parent node is constructed by combining adjacent hash values ​​and hashing them again. until a unique root hash is constructed. The system employs a distributed microservice architecture to deploy various functional modules, with communication between modules achieved through lightweight message queues. Role-based access control (RBAC) and TLS protocols are used to protect data transmission.

4. The cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition according to claim 1, characterized in that, In step S2, the graph convolutional network based on the authentication event graph is also used for real-time detection of abnormal patterns: The feature vector of each authentication event The input graph convolutional network performs multi-layer feature updates. The feature representation output by the last layer is compared with a preset anomaly detection threshold. When abnormal behavior is detected, a security warning is triggered and the trust anchor point is switched.

5. The cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition according to claim 1, characterized in that, The use of homomorphic encryption and secret sharing techniques to ensure data privacy in step S3 specifically includes: A hybrid encryption mode of AES and RSA is used during data transmission. Furthermore, IPSec and TLS tunneling technologies are used to enhance the security of cross-border transmissions; Embedding hybrid digital signatures at every stage of the authentication process Combined with timestamps Ensure that operation logs are non-repudiable and traceable.

6. The cross-border digital certificate authentication method supporting multimodal semantic alignment and cross-chain mutual recognition according to claim 1, characterized in that, The deep neural network model for cross-chain data format conversion in step S4 The training process includes: Certificate Authority (CA) certification data from multiple blockchain platforms were collected as training samples, using source chain data. For input and target chain data Using labels, network parameters are optimized by minimizing the transformation error loss function. This makes the converted data Maintain the integrity and encryption properties of the original data.

7. The method according to claim 1, characterized in that, In step S4, a timeout retry strategy is introduced into the smart contract, and the random delay compensation formula is as follows: ,in For average delay, The standard deviation of the delay. The random numbers are standard normally distributed to ensure data transmission is completed in high-latency or packet-loss environments.

8. The method according to claim 1, characterized in that, The cross-border service scenarios in step S5 include at least one of the following: cross-border service with one certificate, cross-border linkage of digital identity, and trusted mutual recognition of supply chain finance. In the aforementioned cross-border service using a single certificate, dynamic semantic linking functions are utilized. Combine identity information with the current spatiotemporal context Combined, after successful authentication, a state transition function is used. Automatically triggers the linkage and approval of customs declaration, logistics and fund settlement systems; In the aforementioned cross-border digital identity linkage, the Diffie-Hellman protocol is used to generate a shared key: Real-time data synchronization and secure transmission are achieved; an improved signature format is adopted in the aforementioned trusted mutual recognition of supply chain finance. Sign the contract and invoice, among which The time perturbation function is used; a risk warning model is employed. Enable real-time risk monitoring.

Images