Financial voucher automatic generation and verification method and system based on intelligent accounting

Abstract

The application provides a multi-source financial data intelligent classification and automatic accounting optimization processing method and system, relates to the technical field of enterprise financial management, and comprises the following steps: text preprocessing of multi-source enterprise financial data, construction of a transaction scene graph for scene clustering, formation of a transaction classification rule set for transaction classification, matching of labels and accounting subject codes for regular transaction data, construction of a transaction traceability graph for analysis of transaction link integrity, generation of an accounting voucher for automatic accounting, improvement of the accuracy and efficiency of financial data processing, reduction of manual intervention cost, and enhancement of the traceability of financial data.

Description

TECHNICAL FIELD

[0001] The present application relates to intelligent financial management technology, and in particular to a financial voucher automatic generation and verification method and system based on intelligent accounting. BACKGROUND

[0002] With the rapid development of information technology, financial management work is gradually realizing digital transformation. Traditional financial voucher processing mainly relies on manual operation of financial personnel, including identifying business document information, preparing accounting entries according to accounting standards, generating financial vouchers and performing audit verification, etc. In recent years, intelligent financial systems have begun to apply artificial intelligence, natural language processing and other technologies to automatically process financial data, and some enterprises have also begun to explore the application of blockchain technology in the financial field to improve the security and credibility of financial data.

[0003] However, the current financial voucher automatic generation and verification technology still has obvious deficiencies. First, the existing technology has low accuracy in extracting information from business documents, especially in processing non-standardized documents, and it is difficult to accurately identify and convert complex and diverse transaction information, which affects the accuracy of subsequent accounting entry generation. Second, the semantic understanding ability is lacking in the process of matching accounting subjects, mainly relying on keyword matching or simple rule judgment, which cannot effectively handle the differences in subject allocation of the same transaction in different business scenarios, and also cannot identify potential abnormal transaction patterns. Third, the effectiveness verification and secure storage mechanism of financial vouchers is not perfect, and the traditional database storage method is easy to be tampered with, lacking a reliable verification mechanism and complete audit tracking ability, which cannot fully guarantee the authenticity and integrity of financial data. SUMMARY

[0004] The embodiments of the present application provide a financial voucher automatic generation and verification method and system based on intelligent accounting, which can solve the problems in the prior art.

[0005] The business document is scanned and recognized to obtain transaction information in the business document, and the transaction information is converted into standardized transaction data containing transaction amount, transaction time, transaction counterparty information, transaction type, commodity information and service information;

[0006] The natural language processing technology is used to extract semantic features from the standardized transaction data, the extracted semantic features are matched with a preset accounting subject feature library, and an initial accounting entry is generated; a transaction relationship network is constructed, transaction abnormalities are identified through graph structure analysis, the initial accounting entry is optimized, and a final accounting entry is generated;

[0007] The final accounting entry generates a financial voucher, analyzes the subject logical relationship, amount matching relationship and voucher integrity of the financial voucher through a deep neural network, analyzes the transaction path involved in the financial voucher through graph calculation, calculates the rule matching degree of the financial voucher based on a preset accounting standard rule library, and judges the validity of the financial voucher;

[0008] The financial voucher passing the validity judgment is block encrypted, the encrypted data is stored in a Merkle tree structure, a smart contract supporting zero-knowledge proof is deployed in a distributed network to verify the authenticity of the financial voucher, and the evidence information of the financial voucher is written into a blockchain network based on a consensus mechanism;

[0009] The financial voucher completing the evidence writing is written into an enterprise financial accounting system, and the check record, evidence writing record and entry record of the financial voucher are written into the blockchain network.

[0010] The semantic feature extraction of the standardized transaction data is performed by using a natural language processing technology, the extracted semantic features are matched with a preset accounting subject feature library, and the steps of generating an initial accounting entry include:

[0011] The standardized transaction data is segmented and tagged with a part of speech, transaction entity information is extracted, and a semantic feature vector is generated;

[0012] A multi-layer subject decision tree is constructed according to accounting standards, a primary accounting subject is set as a root node of the multi-layer subject decision tree, a detailed accounting subject is set as a leaf node of the multi-layer subject decision tree, and an accounting subject attribution relationship is established between the root node and the leaf node;

[0013] Starting from the root node, the semantic feature vector is matched with each layer of accounting subjects, the matching probability of each layer of accounting subjects is calculated, the matching weight of the next layer of accounting subjects is dynamically adjusted based on the matching result of each layer, and an accounting subject matching sequence is generated;

[0014] The semantic similarity between the semantic feature vector and the preset accounting subject feature library is calculated, the business similarity between the transaction rule feature and the preset accounting subject feature library is calculated, the causal relationship strength between the transaction feature and the accounting subject selection is analyzed, the semantic similarity, the business similarity and the causal relationship strength are weighted and fused to obtain a comprehensive similarity;

[0015] The final matched accounting subject is determined according to the accounting subject matching sequence and the comprehensive similarity, the debit and credit direction of the accounting subject is determined based on the causal relationship strength, and an initial accounting entry is generated;

[0016] Calculate a matching confidence of the initial accounting entry, and mark the initial accounting entry as a state to be manually audited when the matching confidence is lower than a preset confidence threshold.

[0017] The step of constructing a transaction relationship network, identifying transaction abnormalities through graph structure analysis, optimizing the initial accounting entry, and generating a final accounting entry includes:

[0018] The transaction relationship network is constructed, the transaction related information is constructed as nodes, and edges are established according to the transaction association relationship;

[0019] The weighted degree distribution feature of each node in the transaction relationship network is calculated, the path importance feature is calculated based on the cumulative edge weight of the transaction path, the network density of the transaction association subgraph is calculated to obtain the global topology feature, and the weighted degree distribution feature, the path importance feature, and the global topology feature are fused to obtain the graph structure feature;

[0020] The link abnormality score of the transaction path, the fluctuation abnormality score of the transaction amount, and the time sequence abnormality score of the transaction frequency are calculated using the graph structure feature, the link abnormality score, the fluctuation abnormality score, and the time sequence abnormality score are weighted and combined, and the abnormality score is adaptively normalized based on the transaction size;

[0021] For transactions with an abnormality score exceeding a first preset threshold, a correction rule is generated using historical normal transaction patterns, and the initial accounting entry is subject to subject re-matching and amount re-allocation based on the correction rule; for transactions with an abnormality score exceeding a second preset threshold, an expert rule library is introduced for deep optimization to generate an optimized accounting entry;

[0022] The basic compliance score of the optimized accounting entry is calculated based on the accounting standard rules, the transaction credibility score is calculated in combination with the abnormality score, and the entry quality evaluation value is generated by fusing the basic compliance score and the transaction credibility score; when the entry quality evaluation value is greater than a preset quality threshold, the optimized accounting entry is determined as a final accounting entry.

[0023] The step of generating a financial voucher for the final accounting entry, analyzing the subject logic relationship, the amount matching relationship, and the voucher integrity of the financial voucher through a deep neural network, analyzing the transaction path involved in the financial voucher through graph calculation, and calculating the rule matching degree of the financial voucher based on a preset accounting standard rule library to determine the validity of the financial voucher includes:

[0024] The subject information and the amount information of the financial voucher are converted into a subject feature vector and an amount feature vector, respectively.

[0025] calculate attention weights for subject nodes in the subject feature vector, calculate correlation strength between subject nodes based on the attention weights to obtain a subject attention vector; fuse context information of historical vouchers with a current voucher state by using a gated recurrent unit to obtain a context vector; and splice the subject attention vector and the context vector to calculate a subject logical relationship score;

[0026] perform an amount matching relationship and integrity check on the financial vouchers to generate an amount matching relationship score and a voucher integrity score;

[0027] construct an initial transaction graph and perform time sequence expansion to obtain a transaction relationship dynamic graph with time sequence connection, iteratively update node features in the graph based on an attention propagation mechanism of transaction scale, transaction time sequence and transaction frequency, identify a transaction path in the transaction relationship dynamic graph by using a bidirectional search strategy and perform sub-path decomposition, calculate a sub-path score in combination with structural importance and time sequence correlation, and obtain a transaction path anomaly score through deep fusion;

[0028] calculate a rule matching degree of the financial vouchers based on a preset accounting standard rule library;

[0029] perform adaptive weighted fusion on the subject logical relationship score, the amount matching relationship score, the voucher integrity score, the transaction path anomaly score and the rule matching degree to generate a comprehensive score of the financial vouchers;

[0030] when the comprehensive score is greater than a preset score threshold, determine that the financial vouchers are valid.

[0031] the steps of constructing an initial transaction graph and performing time sequence expansion to obtain a transaction relationship dynamic graph with time sequence connection, iteratively updating node features in the graph based on an attention propagation mechanism of transaction scale, transaction time sequence and transaction frequency, identifying a transaction path in the transaction relationship dynamic graph by using a bidirectional search strategy and performing sub-path decomposition, calculating a sub-path score in combination with structural importance and time sequence correlation, and obtaining a transaction path anomaly score through deep fusion include:

[0032] construct an initial transaction graph based on the subject feature vector and the amount feature vector, set transaction entities in vouchers as graph nodes, and set fund flow relationships between transaction entities as graph edges, and an initial weight of the graph edges is determined according to transaction amount and transaction frequency;

[0033] perform time sequence expansion on the initial transaction graph, establish time sequence connection for the same transaction entity nodes in different time windows, and calculate dynamic weights of the time sequence connection based on historical transaction data;

[0034] The node features are iteratively updated by attention propagation, for each node, an attention coefficient is calculated based on the transaction size, transaction timing and transaction frequency of its neighbor nodes, the features of the neighbor nodes are weighted and aggregated according to the attention coefficient, and the aggregated features are combined with the node's own features to generate updated node features.

[0035] A bidirectional search strategy is used in the transaction relationship dynamic graph to identify a transaction path, the transaction path is decomposed into transaction sub-paths, and a structural importance score and a timing correlation score are calculated for each sub-path; the structural importance score, the timing correlation score of the sub-path and the feature sequence of the path node are deeply fused to generate a transaction path anomaly score.

[0036] The financial vouchers passing through the validity judgment are block-encrypted, the encrypted data is stored in a Merkle tree structure, a smart contract supporting zero-knowledge proof is deployed in a distributed network to verify the authenticity of the financial vouchers, and the steps of writing the evidence information of the financial vouchers into a blockchain network based on a consensus mechanism include:

[0037] The financial vouchers are structurally divided into multiple data blocks according to the subject dimension, the amount dimension and the time dimension, a block index of the data block is generated, and a mapping relationship between the data block and the block index is established;

[0038] A data block encryption key is generated based on a master key, the block index and a random salt value, the data block and additional authentication data are authentication-encrypted using the data block encryption key to obtain a ciphertext block, and the ciphertext block, an initialization vector and an authentication tag are combined to generate an encrypted data block;

[0039] The encrypted data block and the metadata information are hashed to obtain a leaf node hash value, adjacent leaf node hash values are paired to obtain a parent node hash value, and recursive calculation is performed until a root hash value is generated, and a Merkle proof containing all nodes on the path from the leaf node hash value to the root hash value is generated;

[0040] An arithmetic circuit is constructed based on the encryption verification process, the encrypted data block and the leaf node hash value are used as public inputs, the proof key and the proof private information are used as private inputs, and the arithmetic circuit is substituted to generate a zero-knowledge proof, and the zero-knowledge proof and the root hash value are deployed to a verification smart contract;

[0041] The distributed nodes verify the zero-knowledge proof and the root hash value based on the verification key to obtain a verification result, and the verification results of the nodes are weighted to obtain a comprehensive verification score according to the weights of the nodes;

[0042] The root hash value, the timestamp and the signature information are constructed into a deposit transaction, the deposit transaction is packaged into a block based on a practical Byzantine fault tolerance consensus mechanism and is broadcasted and synchronized in a blockchain network.

[0043] The step of deploying the encrypted data block and the leaf node hash value as public inputs, the proof key and the proof private information as private inputs into the arithmetic circuit to generate a zero-knowledge proof, and the zero-knowledge proof and the root hash value to a verification smart contract comprises:

[0044] An arithmetic circuit is constructed based on an encryption verification process, and the arithmetic circuit comprises a data verification layer, an encryption verification layer and a hash verification layer;

[0045] The encrypted data block is input into the data verification layer and the encryption verification layer to generate a data integrity constraint and a cryptography operation constraint respectively, and the leaf node hash value is input into the hash verification layer to generate a Merkle path constraint;

[0046] The data integrity constraint, the cryptography operation constraint and the Merkle path constraint are respectively generated into sub-proofs based on the proof key and the proof private information, and the sub-proofs are combined to generate an original proof;

[0047] The proof complexity is calculated based on the circuit size and the witness data dimension of the original proof, the proof segmentation size is determined according to the proof complexity and a preset memory limit, and the original proof is divided into a plurality of proof segments according to the proof segmentation size;

[0048] The number of proof items and the correlation degree of the proof items in each proof segment are calculated, the proof segment is restructured based on the number of proof items and the correlation degree of the proof items to obtain an optimized proof segment, and the optimized proof segment is recursively combined according to a tree structure to obtain a combined proof;

[0049] The proof feature of the combined proof is calculated, and the combined proof is directionally compressed based on the proof feature to obtain a compressed proof;

[0050] The static evaluation results of the circuit size and the witness data dimension are obtained, the dynamic evaluation results of the threat type and the historical security events are obtained, and the static evaluation results and the dynamic evaluation results are combined to obtain a security level;

[0051] The security parameter is dynamically adjusted according to the security level, and the security of the compressed proof is enhanced based on the security parameter to obtain a final proof, and the final proof is deployed to a verification smart contract.

[0052] The second aspect of the embodiment of the application provides a financial voucher automatic generation and verification system based on intelligent accounting, comprising:

[0053] The first unit is configured to scan and identify a business document, obtain transaction information in the business document, and convert the transaction information into standardized transaction data containing a transaction amount, a transaction time, transaction counterparty information, a transaction type, commodity information, and service information.

[0054] The second unit is configured to extract semantic features from the standardized transaction data by using a natural language processing technology, perform similarity matching between the extracted semantic features and a preset account subject feature library, generate an initial accounting entry, construct a transaction relationship network, identify transaction abnormalities by graph structure analysis, optimize the initial accounting entry, and generate a final accounting entry.

[0055] The third unit is configured to generate a financial voucher for the final accounting entry, analyze subject logical relationships, amount matching relationships, and voucher integrity of the financial voucher by using a deep neural network, analyze a transaction path involved in the financial voucher by using graph calculation, calculate a rule matching degree of the financial voucher based on a preset accounting standard rule library, and determine validity of the financial voucher.

[0056] The fourth unit is configured to block-encrypt the financial voucher determined as valid, store encrypted data in a Merkle tree structure, deploy a smart contract supporting zero-knowledge proof in a distributed network to verify authenticity of the financial voucher, and write evidence information of the financial voucher into a blockchain network based on a consensus mechanism.

[0057] The fifth unit is configured to write the financial voucher with completed evidence into an enterprise financial accounting system, and write verification records, evidence records, and entry records of the financial voucher into the blockchain network.

[0058] In a third aspect, an electronic device is provided, and the electronic device includes:

[0059] a processor;

[0060] a memory for storing processor-executable instructions;

[0061] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0062] In a fourth aspect, a computer-readable storage medium is provided, and the computer-readable storage medium stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0063] The application provides a financial voucher automatic generation and verification method based on intelligent accounting, which can improve the automation degree and accuracy of financial accounting by intelligently identifying business documents and converting them into standardized transaction data, and automatically generating accurate accounting entries by combining natural language processing and graph structure analysis technology, and effectively reducing the workload and error rate of manual operation.

[0064] The application adopts deep neural network and graph computing technology to perform multi-dimensional verification on financial vouchers, calculates rule matching degree based on a preset accounting standard rule library, and constructs a comprehensive and rigorous voucher validity judgment mechanism, which can timely discover and prevent financial errors and fraud behaviors, and significantly improves the reliability and compliance of financial data.

[0065] The application combines blockchain technology and zero-knowledge proof to perform distributed storage and verification on financial vouchers, realizes the non-tamperability and traceability of voucher data, and protects the security of sensitive financial information, providing efficient, safe and transparent technical support for enterprise financial management, and effectively solving the data security and trust problem in traditional financial systems. BRIEF DESCRIPTION OF DRAWINGS

[0066] Figure 1 The figure is a flowchart of the financial voucher automatic generation and verification method based on intelligent accounting of the embodiment of the application.

[0067] Figure 2 The figure is a flowchart of the financial voucher blockchain storage and zero-knowledge proof verification. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below in combination with the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, not all the embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0069] The technical scheme of the application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0070] Figure 1 The figure is a flowchart of the financial voucher automatic generation and verification method based on intelligent accounting of the embodiment of the application, as shown in the figure, the method comprises: Figure 1

[0071] ​scanning and recognizing a business document, obtaining transaction information in the business document, converting the transaction information into standardized transaction data containing transaction amount, transaction time, transaction counterparty information, transaction type, commodity information, and service information;

[0072] performing semantic feature extraction on the standardized transaction data using natural language processing technology, performing similarity matching of the extracted semantic features with a preset accounting subject feature library, generating an initial accounting entry, constructing a transaction relationship network, identifying transaction abnormalities through graph structure analysis, optimizing the initial accounting entry, and generating a final accounting entry;

[0073] generating a financial voucher for the final accounting entry, analyzing subject logical relationships, amount matching relationships, and voucher integrity of the financial voucher through a deep neural network, analyzing a transaction path involved in the financial voucher through graph calculation, calculating a rule matching degree of the financial voucher based on a preset accounting standard rule library, and judging validity of the financial voucher;

[0074] performing block encryption on the financial voucher that passes the validity judgment, storing the encrypted data in a Merkle tree structure, deploying a smart contract supporting zero-knowledge proof in a distributed network to verify authenticity of the financial voucher, and writing evidence information of the financial voucher into a blockchain network based on a consensus mechanism;

[0075] writing the financial voucher that completes the evidence writing into enterprise financial accounting, and writing verification records, evidence writing records, and entry records of the financial voucher into the blockchain network.

[0076] In an optional implementation, the step of performing semantic feature extraction on the standardized transaction data using natural language processing technology, performing similarity matching of the extracted semantic features with a preset accounting subject feature library, and generating an initial accounting entry includes:

[0077] performing word segmentation and part-of-speech tagging on the standardized transaction data, extracting transaction entity information, and generating a semantic feature vector;

[0078] constructing a multi-layer subject decision tree according to accounting standards, setting a primary accounting subject as a root node of the multi-layer subject decision tree, setting a detailed accounting subject as a leaf node of the multi-layer subject decision tree, and establishing an accounting subject attribution relationship between the root node and the leaf node;

[0079] starting from the root node, matching the semantic feature vector with accounting subjects at each layer, calculating a matching probability of the accounting subjects at each layer, dynamically adjusting a matching weight of the accounting subjects at a next layer based on a matching result at each layer, and generating an accounting subject matching sequence;

[0080] Calculate the semantic similarity of the semantic feature vector and the preset accounting subject feature library, calculate the business similarity of the transaction rule feature and the preset accounting subject feature library, analyze the causal relationship strength of the transaction feature and the accounting subject selection, and fuse the semantic similarity, business similarity and causal relationship strength to obtain a comprehensive similarity;

[0081] Determine the final matched accounting subject based on the accounting subject matching sequence and the comprehensive similarity, determine the debit and credit direction of the accounting subject based on the causal relationship strength, and generate an initial accounting entry;

[0082] Calculate the matching confidence of the initial accounting entry, and mark the initial accounting entry as a manual review state when the matching confidence is lower than a preset confidence threshold.

[0083] For example, when performing basic text processing on standardized transaction data, a conventional lexical analysis method is used to divide the transaction description text into word units. For example, the transaction description "pay office rent 5000 yuan" is divided into "pay", "office", "rent" and "5000 yuan" basic units. Then, the part-of-speech categories of these words are identified using rule matching, such as "pay" as a verb, "office" and "rent" as nouns, and "5000 yuan" as a numerical expression. Based on these basic lexical information, key transaction information such as action type, transaction object and amount value is extracted according to the predefined template. These elements are mapped to numerical representations, such as using simple one-hot encoding or word frequency statistics to represent the characteristics of each element. For example, "rent" is represented as the position index in the preset vocabulary. Finally, these basic features are organized into feature vectors according to the predetermined format, which are used for subsequent subject matching processing.

[0084] When constructing a multi-layer subject decision tree according to accounting standards, the first-level accounting subjects (such as assets, liabilities, owner's equity, cost, expense, and income) are set as the root nodes of the decision tree. The second-level accounting subjects (such as current assets and non-current assets under assets) are set as the second layer nodes. The third-level and below detailed subjects (such as cash, bank deposits, and accounts receivable under current assets) are set as deeper nodes, until the most specific detailed subjects are set as leaf nodes. The subject attribution relationship between nodes is established to ensure that the subject structure conforms to the hierarchical relationship specified by the accounting standards. For example, the "office expenses" node is a child node of the "management expenses" node, and the "management expenses" node is a child node of the "expense" node. The decision tree is stored in a database, containing fields such as subject code, subject name, and parent subject ID, facilitating subject matching and traversal.

[0085] When subject matching starts from the root node, first calculate the matching probability of the semantic feature vector and the first-level accounting subject. For example, for the transaction "pay office rent", calculate its matching probability with the first-level subjects "assets", "expenses", etc. The matching probability of "expenses" is 0.85, and the matching probability of "assets" is 0.15. Based on the matching result of the first-level subject, adjust the matching weight of the second-level subject. Since the matching probability of "expenses" is high, increase the weight of the second-level subjects such as "management expenses" and "sales expenses" in the expense category. In the second-level subject matching, the matching probability of "management expenses" is 0.78, and the matching probability of "sales expenses" is 0.22. Continue to match the lower level, and finally find the leaf node subject "office expenses" under "management expenses", with a matching probability of 0.92. Through this hierarchical matching method, a subject matching sequence is generated: "expenses → management expenses → office expenses", recording the optimal matching path from the root node to the leaf node.

[0086] Calculate the cosine similarity between the semantic feature vector and the features of each subject in the preset accounting subject feature library. The accounting subject feature library contains the name, definition, scope of application, common business description, etc. For example, the cosine similarity between the semantic feature vector of "pay office rent" and the "office expenses" subject feature is 0.88. Secondly, analyze the business rule characteristics of the transaction, such as transaction type, transaction counterparty, amount range, etc., and match with the business rules of the accounting subject to calculate the business similarity. For example, the business similarity between the rent payment transaction and the "office expenses" subject is 0.85. At the same time, based on historical data analysis, the causal relationship strength between transaction characteristics and accounting subject selection is analyzed, for example, the causal relationship strength between "rent" and "office expenses" is 0.90, indicating that the nature of rent transactions into the office expenses subject is very high. Weighted fusion of semantic similarity (0.88), business similarity (0.85) and causal relationship strength (0.90) according to the preset weight (such as 0.3, 0.3, 0.4) to get the comprehensive similarity 0.88.

[0087] When determining the final accounting subject, consider the subject matching sequence and the comprehensive similarity, and select the leaf node subject with the highest comprehensive similarity as the final matching result. In this example, "management expenses - office expenses" is determined as the final matching subject. Based on the causal relationship analysis, determine the debit and credit direction of the subject, such as the analysis that rent payment should be debited to "management expenses - office expenses" and credited to "bank deposits", generating the initial accounting entry: debit: management expenses - office expenses 5000 yuan, credit: bank deposits 5000 yuan.

[0088] The matching confidence of the initial accounting entry is calculated by comprehensively considering factors such as subject matching probability, comprehensive similarity, and historical matching accuracy. For example, the matching confidence of the current accounting entry is calculated as 0.86. When the matching confidence is lower than a preset threshold (such as 0.75), the entry is marked as a state to be manually audited. In this example, since the matching confidence 0.86 is higher than the threshold 0.75, the accounting entry is marked as an automatic pass state and does not need to be manually audited. For cases with lower confidence, such as a transaction of "purchasing office equipment 10000 yuan" generating a debit entry of "management expenses-office expenses", the confidence is only 0.62, and it will be marked as to be manually audited, prompting that it should be recorded as "fixed assets" instead of "office expenses".

[0089] The present application realizes accurate matching of accounting subjects by multi-layer subject decision tree and comprehensive similarity calculation, significantly improves the accuracy of accounting entry generation. The weighted fusion of semantic features, business similarity and causal relationship strength is introduced to solve the problem that traditional methods cannot handle complex business context. At the same time, the confidence evaluation mechanism can intelligently identify situations that need manual intervention, ensuring the reliability of accounting processing while ensuring the efficiency of automated processing, providing more accurate subject matching capability for enterprise financial processing.

[0090] In an optional implementation, the step of constructing a transaction relationship network, identifying transaction abnormalities through graph structure analysis, optimizing the initial accounting entry, and generating a final accounting entry includes:

[0091] The transaction relationship network is constructed, the transaction related information is constructed as nodes, and edges are established according to the transaction association relationship;

[0092] The weighted degree distribution feature of each node in the transaction relationship network is calculated, the path importance feature is calculated based on the cumulative edge weight of the transaction path, and the global topology feature is calculated by calculating the network density of the transaction association subgraph. The weighted degree distribution feature, path importance feature, and global topology feature are fused to obtain the graph structure feature;

[0093] The link abnormality score of the transaction path, the fluctuation abnormality score of the transaction amount, and the time sequence abnormality score of the transaction frequency are calculated using the graph structure feature, the link abnormality score, the fluctuation abnormality score, and the time sequence abnormality score are weighted and combined, and the abnormality score is adaptively normalized based on the transaction size;

[0094] For transactions with an abnormality score exceeding a first preset threshold, a correction rule is generated using historical normal transaction patterns, and the initial accounting entry is subject to subject re-matching and amount re-allocation based on the correction rule. For transactions with an abnormality score exceeding a second preset threshold, an expert rule library is introduced for deep optimization to generate an optimized accounting entry;

[0095] The optimized accounting entry is based on the accounting standard rules to calculate the basic compliance score, combined with the abnormal score to calculate the transaction credibility score, and the basic compliance score and the transaction credibility score are fused to generate the entry quality evaluation value, and when the entry quality evaluation value is greater than the preset quality threshold, the optimized accounting entry is determined as the final accounting entry.

[0096] For example, a transaction relationship network is constructed, for each transaction, the transaction subject (such as a supplier, a customer), the transaction subject, the transaction time and other information are constructed as nodes, and the node attributes include subject type, industry attribute, registered capital and the like; a directed edge is established according to the transaction occurrence relationship, and the edge attributes include transaction amount, transaction frequency, transaction time and the like. For example, when enterprise A pays 1 million yuan to enterprise B, an edge from A to B is established in the network, the edge weight is 1 million yuan, the attribute is "payment of goods", and the time mark is the specific transaction date.

[0097] The in-degree weighted degree represents the total amount of transactions received by the node, and the out-degree weighted degree represents the total amount of transactions initiated by the node. For example, the in-degree weighted degree of a certain supplier node is 5 million yuan, indicating that the total amount of payments received by the supplier; the out-degree weighted degree is 300,000 yuan, indicating that the total amount of payments made by the supplier. By calculating the weighted degree distribution ratio of each node, the transaction amount abnormal node is identified. For example, 95% of the transactions of a certain enterprise are concentrated in two suppliers, and these two suppliers have a large amount of fund flow with another associated entity, so this concentration degree abnormality needs to be marked.

[0098] For each transaction path in the network, the weight of each edge on the path is accumulated, and the path length is normalized. For example, the fund flow path from enterprise A to enterprise C through intermediate enterprise B, if the transaction amount from A to B is 800,000 yuan, the transaction amount from B to C is 780,000 yuan, and the time interval is short (such as within 1 day), the importance of the path is high, and there is fund transfer or associated transaction. The path importance score is calculated as (80+78) / (2x average transaction amount of enterprises) x time decay factor, and the time decay factor is determined according to the transaction time interval, and the shorter the interval, the larger the factor.

[0099] Calculate the global topology feature, and construct a subgraph for each transaction subject as the center of n-hop neighbors, and calculate the ratio of the actual number of edges in the subgraph to the theoretical maximum number of edges. For example, a certain enterprise and its 10 transaction partners form a subgraph, if there are also a large number of transaction relationships between these transaction partners (such as 45 edges, and the theoretical maximum number of edges is 10x9 / 2=45), the network density is 1, indicating that the transaction relationship is highly concentrated, and there is an associated transaction gang.

[0100] The weighted average method is used for the weighted degree distribution feature, the path importance feature, and the global topology feature, and different features are given different weights, such as the weighted degree distribution feature weight 0.4, the path importance feature weight 0.35, and the global topology feature weight 0.25, to generate a comprehensive graph structure feature. For nodes or edges with abnormal feature values, further calculation of abnormal scores is performed.

[0101] The transaction abnormality score is calculated using the graph structure feature. The link abnormality score is calculated based on the deviation of the transaction path pattern from the historical normal pattern. For example, a certain enterprise has never had a transaction with enterprise X in the past, but suddenly a large amount of transaction appears, and enterprise X immediately transfers funds to enterprise Y, and the link abnormality score will be higher. The volatility abnormality score measures the volatility of the transaction amount relative to the historical mean, such as a sudden increase of 500% in the monthly transaction amount, and the volatility abnormality score will be higher. The time series abnormality score is based on the change rule of transaction frequency, such as a large amount of payment that was originally made once every quarter suddenly becoming once every week, and the time series abnormality score will increase.

[0102] The link abnormality score, the volatility abnormality score, and the time series abnormality score are combined by weighting with weights of 0.3, 0.4, and 0.3, respectively, and then self-adaptive normalization is performed based on the transaction size to obtain the final abnormality score. For large transactions (such as more than 5 times the monthly average transaction amount of the enterprise), the normalization factor is adjusted to 0.8, so that large abnormal transactions are more easily detected.

[0103] For transactions with abnormality scores exceeding a first preset threshold (such as 0.75), a correction rule is generated based on the historical normal transaction pattern. For example, if it is found that the payment amount of a certain "management fee" is 10 times the historical mean, and the payment object is a new supplier, the transaction description and amount characteristics are used to determine that the transaction is more consistent with the characteristics of "fixed asset purchase", and the account is re-matched from "management fee" to "fixed asset". The amount redistribution is based on the accounting standards to split or combine unreasonable amounts, such as splitting a 5 million yuan abnormal advertising fee into reasonable advertising fee of 300,000 yuan and service fee of 200,000 yuan. For transactions with abnormality scores exceeding a second preset threshold (such as 0.9), expert rule library is introduced for deep optimization, and a pre-constructed multi-level expert rule library is called, which contains special processing rules for different industries and business scenarios. For example, for cross-border related transactions, the rule library contains transfer pricing evaluation standards, industry profit rate reference values, and tax compliance requirements. When a certain multinational enterprise is detected to pay a service fee of 20 million yuan to an overseas related party, the applicable rule is matched, such as "the service fee of the related party should not exceed 20% of the market price of the same kind", and the reasonable interval is determined to be 15-17 million yuan based on industry benchmark data, and then the excess part is adjusted to "investment" or "profit distribution" account, and corresponding adjustment basis and processing suggestions are generated for reference by financial personnel.

[0104] For the optimized accounting entries, check whether the subject usage conforms to the accounting subject setting specification, whether the debit and credit balance meets the requirements, and whether the attached vouchers are complete, etc., to obtain a compliance score between 0 and 1. Combine the abnormal score to calculate the transaction credibility score, credibility score = (1 - abnormal score) x transaction size adjustment factor, where the transaction size adjustment factor decreases as the transaction amount increases. Combine the basic compliance score and the transaction credibility score to calculate the formula: entry quality evaluation value = 0.6 x basic compliance score + 0.4 x transaction credibility score. When the entry quality evaluation value is greater than the preset quality threshold (such as 0.85), the optimized accounting entries are determined as the final accounting entries.

[0105] The existing financial anomaly detection mainly adopts statistical threshold or simple rule matching method, which lacks in-depth analysis of transaction correlation and is difficult to find complex abnormal patterns, especially the hidden risks such as associated transactions involving multiple parties or fund circulation. The present application proposes a graph structure analysis method with transaction relationship network as the core to transform scattered transaction data into structured network relationship, and reveals potential abnormal patterns through node characteristics, path analysis and topological structure. The innovation lies in the introduction of multi-dimensional fusion of weighted degree distribution characteristics, path importance characteristics and global topological characteristics, realizing comprehensive perception of transaction anomalies. At the same time, through the hierarchical threshold strategy, different degrees of anomalies are distinguished, and targeted optimization measures are adopted to avoid misjudgment caused by simple processing. The accuracy and comprehensiveness of transaction anomaly identification are significantly improved. Compared with traditional methods, complex fund flow patterns and associated transaction networks can be found to prevent financial fraud risks. The multi-dimensional feature fusion mechanism improves the robustness of anomaly detection and reduces the false positive rate. The hierarchical optimization strategy based on abnormal degree realizes the precise adjustment of accounting entries and improves the quality of financial data. The quality evaluation mechanism ensures that the optimization results conform to the accounting standards, enhances the reliability of financial reports, and provides strong support for enterprise financial risk management.

[0106] In an optional implementation, a financial voucher is generated for the final accounting entry. The subject logic relationship, amount matching relationship and voucher integrity of the financial voucher are analyzed by a deep neural network. The transaction path involved in the financial voucher is analyzed by graph calculation. The rule matching degree of the financial voucher is calculated based on a preset accounting standard rule library. The validity of the financial voucher is judged.

[0107] A financial voucher is generated for the final accounting entry. The subject information and amount information of the financial voucher are converted into subject feature vectors and amount feature vectors, respectively.

[0108] An attention weight is calculated for a subject node in the subject feature vector, and a subject attention vector is obtained based on the calculated attention weight and the correlation strength between subject nodes; a context vector is obtained by fusing the context information of the historical voucher and the current voucher state by using a gated recurrent unit; and a subject logical relationship score is calculated by splicing the subject attention vector and the context vector;

[0109] An amount matching relationship and integrity of the financial voucher are checked to generate an amount matching relationship score and a voucher integrity score;

[0110] An initial transaction graph is constructed and time-series expansion is performed to obtain a transaction relationship dynamic graph with time-series connection, and the node features in the graph are iteratively updated based on an attention propagation mechanism of transaction size, transaction time series and transaction frequency; a bidirectional search strategy is used to identify a transaction path in the transaction relationship dynamic graph and to perform sub-path decomposition, a sub-path score is calculated in combination with structural importance and time-series correlation, and a transaction path anomaly score is obtained through deep fusion;

[0111] A rule matching degree of the financial voucher is calculated based on a preset accounting standard rule library;

[0112] The subject logical relationship score, the amount matching relationship score, the voucher integrity score, the transaction path anomaly score and the rule matching degree are adaptively weighted and fused to generate a comprehensive score of the financial voucher;

[0113] When the comprehensive score is greater than a preset score threshold, the financial voucher is determined to be valid.

[0114] For example, a final accounting entry generates a financial voucher, and the subject information in the financial voucher is converted into a subject feature vector, and the amount information is converted into an amount feature vector. The subject information can include the debit subject, the credit subject, the subject level, etc., and these discrete information is mapped to a continuous vector space through an embedding layer. For example, the "accounts receivable" subject can be represented as a vector of [0.32, -0.45, 0.78, 0.21], and the amount information is vectorized according to the amount size, positive and negative characteristics, etc., such as 10000 yuan which can be represented as [0.10, 0.35, -0.42].

[0115] When calculating the attention weight of the subject feature vector, a multi-head self-attention mechanism is used. Each subject node is regarded as a node in the graph, and the correlation strength between the nodes is calculated. The subject feature vector is converted into a query vector, a key vector and a value vector through a query matrix, a key matrix and a value matrix. For the two subjects of "sales income" and "accounts receivable", the calculated attention weight between them is 0.85, indicating high correlation; while the attention weight between "sales income" and "fixed assets" is only 0.12, indicating low correlation. Based on the calculated attention weight, a subject attention vector is obtained.

[0116] The historical voucher information is processed using a gated recurrent unit. For a sales transaction, the context information of similar transactions in the past is fused. For example, the records of sales transactions in the past 3 months are encoded into a state vector, which is fused with the current voucher state to generate a context vector containing the historical transaction patterns. The account attention vector is concatenated with the context vector, and the account logical relationship score is obtained through a fully connected layer. For example, a normal sales transaction voucher scores 0.92, while a voucher that does not conform to the conventional account combination scores 0.35.

[0117] The financial voucher is checked for amount matching relationship. The amount of the debit and credit sides of a normal voucher should be equal, and the total amount difference is calculated and standardized to obtain the amount matching relationship score. For example, a voucher with a balance of 1.0 has a score of 0.95 with a slight difference. The completeness of the voucher is also checked, including the filling of necessary fields, attachment information, etc., to generate a voucher completeness score, such as 0.88 indicating basic completeness but room for improvement.

[0118] The transaction path involved in the voucher is analyzed by constructing a dynamic graph of transaction relationships. Taking inter-enterprise fund transfer as an example, each enterprise is regarded as a node in the graph, and the transaction is regarded as an edge to construct an initial transaction graph. By time series expansion, the transactions at different time points are associated to form a dynamic graph. For transactions that pay similar amounts to the same supplier for three consecutive months, the time series connection relationship is identified. Based on the transaction size, transaction timing and transaction frequency, an attention propagation mechanism is designed, and large, frequent or time series close transactions obtain higher attention weight. The node features are iteratively updated through multiple rounds of message passing.

[0119] A bidirectional search strategy is used in the transaction relationship dynamic graph to identify the transaction path. For example, for the case where A company transfers money to B company, and B company immediately transfers money to C company, a forward search is performed from A, and a reverse search is performed from C, and the intersection is identified at B, identifying the complete path A→B→C. The path is decomposed into two sub-paths A→B and B→C, and the sub-path scores are calculated according to the structural importance and time series association. For example, A→B has an amount of 1 million and is the first transaction, with a score of 0.75; B→C has an amount of 980,000 and occurs in a very short time, with a score of 0.35, indicating high abnormality. The overall transaction path abnormality score is generated by deep fusion of the sub-path scores, such as 0.40, indicating a higher abnormal risk.

[0120] The rule matching degree of the financial voucher is evaluated based on a preset accounting standard rule library. The rule library contains applicable accounting standards and enterprise internal systems, such as the rule that "fixed asset depreciation shall be charged to the corresponding cost expense account". Whether the voucher conforms to these rules is checked, and the rule matching degree is calculated. For example, the matching degree of a voucher that fully conforms to the rules is 0.95, and the matching degree of a voucher that violates a rule is 0.60.

[0121] The subject logic relationship score, the amount matching relationship score, the voucher integrity score, the transaction path abnormality score, and the rule matching degree are adaptively weighted and fused. According to different business scenarios, the weights of the indicators are dynamically adjusted. For example, for a regular business voucher, the weight of the subject logic relationship is 0.3, the weight of the amount matching relationship is 0.25, the weight of the voucher integrity is 0.15, the weight of the transaction path abnormality is 0.1, and the weight of the rule matching degree is 0.2; and for a cross-border transaction voucher, the weight of the transaction path abnormality is increased to 0.25. The comprehensive score is calculated by taking a weighted average value, such as 0.85.

[0122] According to historical data and risk tolerance, a score threshold is set, such as 0.75. Vouchers higher than the threshold are determined to be valid and can enter subsequent processing flows; vouchers lower than the threshold are marked as potential abnormalities and need to be manually reviewed or further analyzed.

[0123] The present application constructs a complete financial voucher validity verification mechanism, analyzes the subject logic relationship, the amount matching, and the voucher integrity through a deep neural network, analyzes the transaction path by combining graph calculation technology, and comprehensively evaluates the compliance of the financial voucher. The adaptive weighted fusion strategy of multi-dimensional scoring dynamically adjusts the importance of each indicator according to different business scenarios, improves the accuracy and adaptability of the verification result, effectively prevents financial fraud risks, and guarantees the authenticity and integrity of enterprise financial data.

[0124] In an optional implementation, an initial transaction graph is constructed and time-series expansion is performed to obtain a transaction relationship dynamic graph with time-series connection, node features in the graph are iteratively updated based on an attention propagation mechanism of transaction size, transaction time sequence, and transaction frequency; a bidirectional search strategy is used to identify a transaction path in the transaction relationship dynamic graph and perform sub-path decomposition, a sub-path score is calculated in combination with structural importance and time sequence correlation, and a transaction path abnormality score is obtained through deep fusion. The steps include:

[0125] An initial transaction graph is constructed based on the subject feature vector and the amount feature vector, transaction entities in the voucher are set as graph nodes, and the fund flow relationship between the transaction entities is set as a graph edge, and the initial weight of the graph edge is determined according to the transaction amount and the transaction frequency;

[0126] performing time series expansion on the initial transaction graph, establishing time series connections between the same transaction entity nodes in different time windows, and calculating dynamic weights of the time series connections based on historical transaction data;

[0127] updating the node features through attention propagation iteration, for each node, calculating an attention coefficient based on the transaction scale, transaction timing, and transaction frequency of its neighbor nodes, weighting and aggregating the features of the neighbor nodes according to the attention coefficient, and combining the aggregated features with the node's own features to generate updated node features;

[0128] adopting a bidirectional search strategy in the dynamic transaction relationship graph to identify a transaction path, decomposing the transaction path into transaction sub-paths, calculating a structural importance score and a timing correlation score for each sub-path, and performing deep fusion calculation on the structural importance score, the timing correlation score of the sub-paths, and the feature sequence of the path nodes to generate a transaction path anomaly score.

[0129] For example, subject feature vectors and amount feature vectors are extracted from the voucher data provided by the enterprise, and then an initial transaction graph is constructed, time series expansion is performed to form a dynamic graph, node features are updated through attention propagation, and finally an abnormal transaction path is identified through bidirectional search.

[0130] In constructing the initial transaction graph, first, the subjects in each transaction voucher are parsed, and transaction entities such as companies, individuals, and bank accounts are set as graph nodes. Each node is assigned initial attributes, including node type (enterprise internal account, external enterprise account, personal account, etc.), initial balance, etc. For the fund transfer relationship recorded in the voucher, an edge of the graph is set, and the direction of the edge represents the direction of the fund flow. The initial weight of the edge is determined according to the transaction amount and the transaction frequency, and the calculation method is to divide the single transaction amount by the average transaction amount in the accounting period to obtain the normalized amount, and then multiply it by the number of transactions between the two parties in the observation period. For example, A company transfers 1 million yuan to B company, if the average transaction amount is 500,000 yuan, and A and B have a total of 3 transactions in the observation period, then the initial weight of the edge is 2x3=6.

[0131] When performing time series expansion on the initial transaction graph, set the time window to one month, and establish time series connections between the same transaction entity nodes in adjacent windows. The dynamic weight of the time series connection is calculated based on historical transaction data, specifically the deviation ratio of the transaction amount in the current time window to the historical average transaction amount. For example, if A company has a total transaction amount of 300,000 yuan to B company in the current month, and the monthly average transaction amount in the past six months is 200,000 yuan, then the dynamic weight of the time series connection is (300-200) / 200=0.5, indicating that the current transaction intensity is 50% higher than the historical average level.

[0132] When updating the node features iteratively using the attention propagation mechanism, three factors are considered for each node: transaction size, transaction timing, and transaction frequency. The transaction size is represented by the ratio of the transaction amount to the overall transaction volume of the enterprise; the transaction timing considers the time interval pattern of the transaction, converting the transaction frequency changes in the past seven days, one month, and three months into timing features; and the transaction frequency is the number of transactions within a unit of time. Based on these three factors, the attention coefficients are calculated, for example, for the neighbor node B of node A, if the transaction amount of A to B accounts for 30% of the total transaction amount of A, and the transaction frequency has significantly increased recently, then B is given a higher attention weight. In specific operations, the normalized values of the three factors are added and subjected to nonlinear transformation to obtain the attention coefficients, and then the coefficients are used to weight and aggregate the features of the neighbor nodes. For example, if node A has three neighbors B, C, and D, with attention coefficients of 0.5, 0.3, and 0.2 respectively, then the updated feature of A is the weighted sum of the features of these three neighbors, and the final feature is generated by connecting the feature of A itself. After three rounds of iteration, the node features can fully integrate the network structure information.

[0133] When identifying abnormal transaction paths in the dynamic transaction relationship graph, a bidirectional search strategy is used. Starting from the suspicious source and destination of funds simultaneously, the depth limit is 4 hops, and when the search paths of the two directions meet, a complete transaction path is formed. The identified transaction path is decomposed into sub-paths with a length of 2. For each sub-path, the structural importance score and the timing correlation score are calculated. The structural importance is evaluated by the centrality and connection strength of the sub-path in the entire graph, and if the nodes connected by the sub-path are in a key position in the graph and the transaction amount is significant, the structural importance is higher. The timing correlation measures the time correlation of the transactions in the sub-path, and if two transactions are very close in time (e.g., less than 24 hours apart), the timing correlation is higher.

[0134] For example, assume that a transaction path is detected: Company A→Company B→Person C→Company D, which is decomposed into three sub-paths A→B, B→C, and C→D. For the sub-path B→C, if B is a core enterprise and has transactions with multiple entities, and C is a newly appeared personal account, then the structural importance score of this sub-path is higher, which is 0.85; if the transfer from B to C is only 2 hours apart from the transfer from C to D, then the timing correlation score is 0.92. By combining these scores with the feature sequence of the nodes in the path, the final abnormal score of the transaction path is calculated by the deep fusion network, which is 0.88, exceeding the set threshold of 0.75, and is judged as a suspicious associated transaction.

[0135] The existing financial transaction analysis technology often relies on fixed rules or simple statistical indicators when dealing with complex transaction paths, making it difficult to discover carefully designed fund transfer paths and abnormal transaction chains closely related in time. The present application integrates time dimension into graph structure analysis by establishing time sequence connections between the same entity nodes in different time windows to capture the time sequence evolution of transaction patterns. The attention propagation mechanism based on transaction size, transaction timing and transaction frequency is introduced to enable dynamic updating of node features and integration of network structure information. At the same time, the bidirectional search and sub-path decomposition strategy are adopted, combined with structural importance and time sequence correlation evaluation, to improve the accuracy of abnormal transaction path identification. The accuracy and efficiency of financial transaction anomaly detection are significantly improved.

[0136] In an alternative embodiment, the financial vouchers passing the validity judgment are block-encrypted, the encrypted data is stored in a Merkle tree structure, a smart contract supporting zero-knowledge proof is deployed in a distributed network to verify the authenticity of the financial vouchers, and the step of writing the evidence information of the financial vouchers into the blockchain network based on a consensus mechanism includes:

[0137] The financial vouchers are structurally divided into multiple data blocks according to the subject dimension, amount dimension and time dimension, a block index of the data blocks is generated, and a mapping relationship between the data blocks and the block index is established;

[0138] A data block encryption key is generated based on a master key, the block index and a random salt value, the data block and additional authentication data are authentication-encrypted using the data block encryption key to obtain a ciphertext block, and the ciphertext block, an initialization vector and an authentication tag are combined to generate an encrypted data block;

[0139] The encrypted data block and metadata information are hashed to obtain leaf node hash values, adjacent leaf node hash values are paired to obtain parent node hash values, and recursive calculation is performed until a root hash value is generated, and a Merkle proof containing all nodes on the path from the leaf node hash value to the root hash value is generated;

[0140] An arithmetic circuit is constructed based on the encryption verification process, the encrypted data block and the leaf node hash value are taken as public inputs, the proof key and the proof private information are taken as private inputs, the arithmetic circuit is substituted to generate a zero-knowledge proof, and the zero-knowledge proof and the root hash value are deployed to a verification smart contract;

[0141] The distributed nodes verify the zero-knowledge proof and the root hash value based on the verification key to obtain a verification result, and the verification results of the nodes are weighted to obtain a comprehensive verification score;

[0142] The root hash value, timestamp and signature information are constructed into a notarization transaction, the notarization transaction is packaged into a block based on a practical Byzantine fault tolerance consensus mechanism and broadcasted and synchronized in a blockchain network.

[0143] Figure 2 A flowchart of a financial voucher blockchain notarization and zero-knowledge proof verification process is shown. The financial vouchers are structured and divided into multiple data blocks according to subject dimension, amount dimension and time dimension. For example, for an original voucher containing multiple accounting subjects, it can be divided into asset subject data block, liability subject data block, income subject data block and expense subject data block; according to the amount dimension, it can be divided into large transaction data block (e.g. amount exceeding 100,000 yuan) and small transaction data block; according to the time dimension, it can be divided into monthly data block, quarterly data block, etc. A unique block index is generated for each data block, such as "asset-2023Q1-large" representing large asset class transaction data in the first quarter of 2023. A mapping table of data block and block index is established in the database, which facilitates subsequent retrieval and verification.

[0144] A 256-bit master key (e.g. "a8b7c6d5e4f3g2h1i0j9k8l7m6n5o4p3") is generated, and a 32-bit random salt value (e.g. "r7d9e2f4") is assigned to each block index. Through a key derivation function, the master key, block index and random salt value are input to generate a 128-bit encryption key specific to each data block. The encryption key is used to encrypt the data block and additional authentication data (such as voucher number, personnel information, etc.) in AEAD (Authenticated Encryption with Associated Data) mode to obtain a ciphertext block. A 16-byte initialization vector (e.g. "9f8e7d6c5b4a3210") and a 16-byte authentication tag are generated, and the ciphertext block is combined with the initialization vector and authentication tag to generate the final encrypted data block.

[0145] The hash value of the leaf node is calculated by hashing the encrypted data block and the metadata information. For example, for the encrypted "assets-2023Q1-large" data block, the hash value is calculated using the SHA-256 hashing algorithm as the leaf node value (e.g., "7f83b1657ff1fc53b92dc18148a1d65dfc2d4b1fa3d677284addd200126d9069"). All leaf nodes are arranged in a predefined order, and adjacent leaf node hash values are paired and concatenated to calculate the parent node hash value. For example, the hash values of leaf nodes A and B are paired to calculate the hash value of parent node P. This process is recursively performed until a single root hash value is generated (e.g., "d82c4eb5261cb9c8aa9855edd67d1bd10482f41529858d925094d173fa662aa7"). A Merkle proof containing all nodes from the leaf nodes to the root hash value on the path is generated for subsequent verification of the integrity and attribution of specific data blocks.

[0146] The encrypted data block and leaf node hash value are used as public inputs, and the proof key (e.g., "e5f6g7h8i9j0k1l2") and proof private information (e.g., master key, random salt value, etc.) are used as private inputs to the arithmetic circuit to generate a zero-knowledge proof. This proof can prove that all the information needed to correctly encrypt the plaintext financial certificate into a ciphertext block is known without exposing these sensitive information. The generated zero-knowledge proof and root hash value are deployed to the verification smart contract on the blockchain network, and the smart contract code contains the verification logic.

[0147] The distributed nodes use the verification key (a 256-bit public key, such as "b9c8a76f5e4d3c2b1a0f9e8d7c6b5a4") to perform cryptographic verification on the zero-knowledge proof. Each node executes the same verification algorithm: first, verify the cryptographic integrity of the zero-knowledge proof, then check whether the root hash value matches the value promised in the zero-knowledge proof. The node weight is determined based on the preset trust level: the weight of the authoritative node (such as regulatory authorities, accounting firms) is 0.5, the weight of the core node within the enterprise is 0.3, and the weight of the ordinary verification node is 0.1. The verification results of each node (1 for pass, 0 for fail) are multiplied by the corresponding weight and summed to obtain the comprehensive verification score. When the score exceeds the preset threshold of 1.5, the overall verification is determined to be passed.

[0148] The 32-byte root hash value, 8-byte UTC timestamp and 64-byte ECDSA digital signature are packaged into a standard format of a deposit transaction. In the PBFT consensus process, the master node first broadcasts a pre-preparation message containing the deposit transaction, and the verification node broadcasts a preparation message after checking the transaction format and signature validity. When the number of preparation messages exceeds 2f+1 (f is the number of allowed malicious nodes), the node broadcasts a confirmation message. Once a node receives 2f+1 confirmation messages, it is considered that the transaction has reached consensus, and it is added to the block candidate set. When the number of candidate transactions reaches the preset block size or the time interval reaches 10 seconds, these transactions are packaged into a block, the previous block hash value is added to form a chain structure, and then broadcast to all nodes through the P2P network. After verification by each node, it is persisted to the local ledger.

[0149] The application establishes a high-security financial certificate deposit mechanism based on blockchain, which guarantees the integrity and privacy of the certificate data through multi-dimensional data block and authentication encryption technology. The application of Merkle tree structure and Merkle proof improves the efficiency of data verification. The introduction of zero-knowledge proof technology enables the verification of certificate authenticity without revealing specific content, solving the privacy protection problem of traditional deposit methods. Distributed verification based on consensus mechanism ensures the credibility of the deposit process, providing an unalterable trust foundation for financial data.

[0150] In an optional implementation, the step of deploying the zero-knowledge proof and the root hash value to the verification smart contract includes:

[0151] An arithmetic circuit is constructed based on the encryption verification process, which includes a data verification layer, an encryption verification layer and a hash verification layer;

[0152] The encryption data block is input into the data verification layer and the encryption verification layer to generate data integrity constraints and cryptographic operation constraints, respectively, and the leaf node hash value is input into the hash verification layer to generate a Merkle path constraint;

[0153] Based on the proof key and the proof private information, the data integrity constraint, the cryptographic operation constraint and the Merkle path constraint are respectively generated into sub-proofs, and the sub-proofs are combined to generate an original proof;

[0154] The proof complexity is calculated based on the circuit size of the original proof and the witness data dimension, the proof segmentation size is determined according to the proof complexity and the preset memory limit, and the original proof is divided into multiple proof segments according to the proof segmentation size;

[0155] The number of proof items and the correlation degree of the proof items in each proof segment are calculated, the proof segments are restructured based on the number of proof items and the correlation degree of the proof items to obtain optimized proof segments, and the optimized proof segments are recursively combined according to a tree structure to obtain a combined proof;

[0156] The proof features of the combined proof are calculated, and the combined proof is directionally compressed based on the proof features to obtain a compressed proof;

[0157] The static evaluation results of the circuit scale and the witness data dimension are obtained, the dynamic evaluation results of the threat type and the historical security events are obtained, and the static evaluation results and the dynamic evaluation results are combined by weighting to obtain a security level;

[0158] The security parameters are dynamically adjusted according to the security level, and the security of the compressed proof is enhanced based on the security parameters to obtain a final proof, and the final proof is deployed to a verification smart contract.

[0159] For example, an arithmetic circuit including a data verification layer, an encryption verification layer, and a hash verification layer is constructed based on an encryption verification process. The design of the circuit takes into account the needs of data integrity verification, cryptography operation verification, and Merkle tree path verification. Taking a blockchain containing 100 user transaction records as an example, each record contains user ID, transaction amount, and timestamp fields, and the total data volume is about 20KB.

[0160] When the encrypted data block is input into the data verification layer, the integrity of the data structure is checked to ensure that each field is correctly formatted and the value is within the valid range. For example, the user ID is an 8-bit string, the transaction amount is a positive integer and does not exceed 1000000 units, and the timestamp conforms to the standard time format. This process generates data integrity constraints, including about 200 logical constraint conditions.

[0161] At the same time, the encrypted data block is also input into the encryption verification layer, which uses the AES-256 encryption algorithm for cryptography verification to generate cryptography operation constraints. In this example, a 16-round iteration structure is used, each round including byte substitution, row shifting, column mixing, and round key addition operations, generating about 3000 constraints related to encryption operations.

[0162] The leaf node hash value is input into the hash verification layer, which uses the SHA-256 hash algorithm to calculate the Merkle tree path and generate Merkle path constraints. For a Merkle tree containing 100 records, the height is 7, and 6 hash operations from the leaf node to the root node need to be verified, generating about 1000 constraints for each hash operation, a total of about 6000 constraint conditions.

[0163] The proof key (containing common circuit parameters) and proof private information (such as the user's private key) are used to generate sub-proofs. For data integrity constraints, a 2KB sub-proof is generated; for cryptographic operation constraints, a 30KB sub-proof is generated; and for Merkel path constraints, a 60KB sub-proof is generated. These three sub-proofs are combined into a single 92KB original proof.

[0164] The complexity of the proof was calculated based on the circuit size of the original proof (approximately 9200 constraints) and the dimensions of the witness data (approximately 3000 variables), yielding a complexity index of 27.6M. Considering the preset memory limit of 64MB, it was determined that the size of each proof segment should not exceed 16KB. Therefore, the 92KB original proof was divided into 6 proof segments.

[0165] For each proof segment, the number of proof items and their correlation are calculated. For example, the first segment contains 32 proof items with an average correlation of 3.5; the second segment contains 28 proof items with an average correlation of 4.2. Based on these metrics, the proof segments are restructured, grouping highly correlated proof items together to obtain six optimized proof segments. These optimized segments are recursively merged according to a tree structure: first, segments 1 and 2 are merged into intermediate proof A; segments 3 and 4 are merged into intermediate proof B; segments 5 and 6 are merged into intermediate proof C; then intermediate proofs A and B are merged into intermediate proof D; finally, intermediate proofs D and C are merged to obtain a merged proof of 45KB in size.

[0166] The proof characteristics of the merged proof are calculated, including the element distribution entropy (approximately 5.6), structural complexity (approximately 12.3), and redundancy (approximately 0.25). Based on these characteristics, a directional compression algorithm is used to compress the merged proof, removing redundant information and optimizing the structure, resulting in a compressed proof of 28KB in size.

[0167] The system retrieved parameters such as the number of constraints (9200 constraints), number of variables (3000 variables), circuit depth (32 layers), and witness vector length (12800 bytes) for the arithmetic circuit. A baseline risk value of 5.0 was set, with the risk value increasing by 0.1 for every 1000 additional constraints in the circuit size, by 0.2 for every 5 additional layers in the circuit depth, and by 0.15 for every 2000 additional bytes of witness data. Based on the calculation results, the static risk index was 6.8. Simultaneously, a dynamic risk assessment was performed. Recent threat intelligence was obtained from the security monitoring center database, identifying major threat types such as replay attacks (occurring 12 times in the past 7 days), man-in-the-middle attacks (occurring 8 times in the past 7 days), and hash collision attacks (occurring 3 times in the past 7 days). The historical security event database was also searched, revealing 23 security vulnerability exploits targeting similar zero-knowledge proofs in the past 30 days, 5 of which were high-risk events. The dynamic risk index is calculated to be 7.5 based on threat type frequency (weight 0.4) and historical security incident severity (weight 0.6). Considering the current network environment and the importance of the verification task, the weight ratio of static assessment to dynamic assessment is determined to be 4:6. The weighted average is calculated as: 6.8 × 0.4 + 7.5 × 0.6 = 7.22. After rounding, the security level is 7.2. According to the security level standards (0-3 low risk, 3-6 medium risk, 6-8 high risk, 8-10 extremely high risk), this verification process is classified as "high risk".

[0168] Based on this security level, security parameters are dynamically adjusted, including increasing the number of random challenges from the default 16 to 24 and the number of verification rounds from 8 to 12. Based on these enhanced security parameters, the compressed proof undergoes security enhancement processing, adding extra verification information and tamper-proof markers, ultimately resulting in a 32KB final proof. This final proof is deployed to the verification smart contract at address 0x7B4A...E2F3 for on-chain verification. In this way, efficient and secure verification of encrypted data is achieved while maintaining data privacy.

[0169] A second aspect of the present invention provides an automatic generation and verification system for financial vouchers based on intelligent accounting, comprising:

[0170] The first unit is used to scan and identify business documents, obtain transaction information from the business documents, and convert the transaction information into standardized transaction data containing transaction amount, transaction time, counterparty information, transaction type, product information, and service information.

[0171] The second unit is used to extract semantic features from the standardized transaction data using natural language processing technology, match the extracted semantic features with a preset accounting subject feature library for similarity, generate initial accounting entries, construct a transaction relationship network, identify transaction anomalies through graph structure analysis, optimize the initial accounting entries, and generate final accounting entries.

[0172] The third unit is used to generate financial vouchers from the final accounting entries, analyze the account logic relationship, amount matching relationship and voucher completeness of the financial vouchers through deep neural network, analyze the transaction path involved in the financial vouchers through graph calculation, calculate the rule matching degree of the financial vouchers based on the preset accounting standard rule base, and determine the validity of the financial vouchers.

[0173] The fourth unit is used to encrypt the financial vouchers that have passed the validity judgment in blocks, store the encrypted data using a Merkle tree structure, deploy smart contracts that support zero-knowledge proofs in a distributed network to verify the authenticity of the financial vouchers, and write the evidence information of the financial vouchers into the blockchain network based on a consensus mechanism.

[0174] The fifth unit is used to write the financial vouchers that have been stored into the enterprise's financial accounting system, and to write the verification records, storage records, and accounting records of the financial vouchers into the blockchain network.

[0175] A third aspect of the present invention provides an electronic device, comprising:

[0176] processor;

[0177] Memory used to store processor-executable instructions;

[0178] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0179] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0180] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for automatic generation and verification of financial vouchers based on intelligent accounting, characterized in that, The method comprises the following steps: scanning and recognizing a business document to obtain transaction information in the business document, and converting the transaction information into standardized transaction data containing transaction amount, transaction time, transaction counterparty information, transaction type, commodity information and service information; extracting semantic features from the standardized transaction data by using natural language processing technology, and performing similarity matching between the extracted semantic features and a preset accounting subject feature library to generate an initial accounting entry, specifically including: performing word segmentation and part-of-speech tagging on the standardized transaction data, extracting transaction entity information, and generating a semantic feature vector; constructing a multi-layer subject decision tree according to accounting standards, setting a first-level accounting subject as a root node of the multi-layer subject decision tree, setting a detailed accounting subject as a leaf node of the multi-layer subject decision tree, and establishing an accounting subject attribution relationship between the root node and the leaf node; starting from the root node, matching the semantic feature vector with accounting subjects at each layer, calculating the matching probability of the accounting subjects at each layer, dynamically adjusting the matching weight of the accounting subjects at the next layer based on the matching result at each layer, and generating an accounting subject matching sequence; calculating the semantic similarity between the semantic feature vector and the preset accounting subject feature library, calculating the business similarity between transaction rule features and the preset accounting subject feature library, analyzing the causal relationship strength between transaction features and accounting subject selection, and weighting and fusing the semantic similarity, business similarity and causal relationship strength to obtain a comprehensive similarity; determining the finally matched accounting subject according to the accounting subject matching sequence and the comprehensive similarity, determining the debit and credit direction of the accounting subject based on the causal relationship strength, and generating the initial accounting entry; calculating the matching confidence of the initial accounting entry, and marking the initial accounting entry as a state to be manually audited when the matching confidence is lower than a preset confidence threshold; constructing a transaction relationship network, identifying transaction abnormalities through graph structure analysis, optimizing the initial accounting entry, and generating a final accounting entry; generating a financial voucher for the final accounting entry, analyzing the subject logical relationship, amount matching relationship and voucher integrity of the financial voucher through a deep neural network, analyzing the transaction path involved in the financial voucher through graph calculation, calculating the rule matching degree of the financial voucher based on a preset accounting standard rule library, and judging the validity of the financial voucher; performing block encryption on the financial voucher that passes the validity judgment, storing the encrypted data in a Merkle tree structure, deploying an intelligent contract supporting zero-knowledge proof in a distributed network to verify the authenticity of the financial voucher, and writing the evidence information of the financial voucher into a blockchain network based on a consensus mechanism; writing the financial voucher that completes the evidence storage into an enterprise financial accounting system, and writing the verification record, evidence storage record and entry record of the financial voucher into the blockchain network.

2. The method of claim 1, wherein, The step of constructing a transaction relationship network, identifying transaction abnormalities through graph structure analysis, optimizing the initial accounting entry, and generating a final accounting entry comprises: constructing a transaction relationship network, constructing transaction-related information as nodes, and establishing edges according to transaction association relationships; The graph structure features are used to calculate link abnormality scores of the transaction path, volatility abnormality scores of the transaction amount, and time sequence abnormality scores of the transaction frequency, the link abnormality scores, the volatility abnormality scores, and the time sequence abnormality scores are combined by weighting, and the abnormality scores are obtained by self-adaptive normalization based on the transaction scale; For transactions with abnormality scores exceeding a first preset threshold, a correction rule is generated based on historical normal transaction patterns, and the initial accounting entries are subjected to subject re-matching and amount re-allocation based on the correction rule; for transactions with abnormality scores exceeding a second preset threshold, expert rule library is introduced for deep optimization to generate optimized accounting entries; The basic compliance scores of the optimized accounting entries are calculated based on the accounting standard rules, the transaction credibility scores are calculated in combination with the abnormality scores, and the entry quality evaluation values are generated by fusing the basic compliance scores and the transaction credibility scores; when the entry quality evaluation values are greater than a preset quality threshold, the optimized accounting entries are determined as final accounting entries. The final accounting entries are used to generate financial vouchers, the subject logical relationship, the amount matching relationship, and the voucher integrity of the financial vouchers are analyzed by a deep neural network, the transaction path involved in the financial vouchers is analyzed by graph calculation, the rule matching degree of the financial vouchers is calculated based on a preset accounting standard rule library, and the effectiveness of the financial vouchers is judged.

3. The method of claim 1, wherein, The subject information and the amount information of the financial vouchers are converted into a subject feature vector and an amount feature vector, respectively; Attention weights of subject nodes in the subject feature vector are calculated, the correlation strength between the subject nodes is calculated based on the attention weights to obtain a subject attention vector, the context information of historical vouchers is fused with the current voucher state by a gated recurrent unit to obtain a context vector, and the subject logical relationship score is calculated by splicing the subject attention vector and the context vector; The amount matching relationship and the integrity of the financial vouchers are checked to generate an amount matching relationship score and a voucher integrity score; An initial transaction graph is constructed and time sequence expansion is performed to obtain a transaction relationship dynamic graph with time sequence connection, the node features in the graph are iteratively updated based on an attention propagation mechanism of transaction scale, transaction time sequence, and transaction frequency, a bidirectional search strategy is used in the transaction relationship dynamic graph to identify transaction paths and perform sub-path decomposition, the sub-path scores are calculated in combination with structural importance and time sequence correlation, and the transaction path abnormality scores are obtained by deep fusion; The subject logical relationship score, the amount matching relationship score, the voucher integrity score, the transaction path abnormality score, and the rule matching degree are adaptively weighted and fused to generate a comprehensive score of the financial vouchers; When the comprehensive score is greater than a preset score threshold, the financial vouchers are determined to be valid. ​ 4. The method of claim 3, wherein, The transaction relationship dynamic graph with time sequence connection is obtained by constructing an initial transaction graph and performing time sequence expansion, and the node features in the graph are iteratively updated based on the attention propagation mechanism of transaction size, transaction time sequence and transaction frequency; The steps of identifying a transaction path in the transaction relationship dynamic graph by using a bidirectional search strategy, decomposing the transaction path into sub-paths, calculating the sub-path scores based on structural importance and time sequence correlation, and obtaining the transaction path anomaly score by deep fusion include: Based on the subject feature vector and the amount feature vector, an initial transaction graph is constructed, the transaction entities in the voucher are set as graph nodes, and the fund flow relationship between the transaction entities is set as graph edges, and the initial weight of the graph edge is determined according to the transaction amount and the transaction frequency; The initial transaction graph is time-expanded, and the same transaction entity nodes in different time windows are time-connected, and the dynamic weight of the time connection is calculated based on historical transaction data; The node features are iteratively updated using the attention propagation, for each node, the attention coefficient is calculated based on the transaction size, transaction time sequence and transaction frequency of its neighbor nodes, the features of the neighbor nodes are weighted and aggregated according to the attention coefficient, and the updated node features are generated by combining the aggregated features with the node features themselves; In the transaction relationship dynamic graph, a bidirectional search strategy is used to identify a transaction path, the transaction path is decomposed into transaction sub-paths, and the structural importance score and the time sequence correlation score of each sub-path are calculated. The structural importance score, the time sequence correlation score of the sub-path and the feature sequence of the path node are deeply fused to generate a transaction path anomaly score.

5. The method of claim 1, wherein, The steps of performing block encryption on the financial vouchers that pass the validity judgment, storing the encrypted data in a Merkle tree structure, deploying a smart contract supporting zero-knowledge proof in a distributed network to verify the authenticity of the financial vouchers, and writing the evidence information of the financial vouchers into the blockchain network based on a consensus mechanism include: The financial vouchers are structurally divided into multiple data blocks according to the subject dimension, the amount dimension and the time dimension, the block index of the data block is generated, and the mapping relationship between the data block and the block index is established; Based on the master key, the block index and the random salt value, a data block encryption key is generated, the data block and additional authentication data are authenticated and encrypted to obtain a ciphertext block using the data block encryption key, and the ciphertext block, the initialization vector and the authentication tag are combined to generate an encrypted data block; The leaf node hash value is calculated by hashing the encrypted data block and the metadata information, the parent node hash value is calculated by pairing adjacent leaf node hash values, and the recursion is performed until the root hash value is generated, and the Merkle proof containing all nodes on the path from the leaf node hash value to the root hash value is generated; Based on the encryption verification process, an arithmetic circuit is constructed, the encrypted data block and the leaf node hash value are taken as public inputs, the proof key and the proof private information are taken as private inputs, and the zero-knowledge proof is generated by substituting the arithmetic circuit, the zero-knowledge proof and the root hash value are deployed to the verification smart contract; The distributed nodes verify the zero-knowledge proof and the root hash value based on a verification key to obtain a verification result, and calculate a comprehensive verification score by weighting the verification results according to the weights of the nodes; The root hash value, timestamp and signature information are constructed into a storage evidence transaction, the storage evidence transaction is packaged into a block based on a practical Byzantine fault tolerance consensus mechanism, and the block is broadcasted and synchronized in a blockchain network.

6. The method of claim 5, wherein, The steps of deploying the zero-knowledge proof and the root hash value to a verification smart contract include: The arithmetic circuit includes a data verification layer, an encryption verification layer and a hash verification layer; The encrypted data block is input into the data verification layer and the encryption verification layer to generate data integrity constraints and cryptographic operation constraints respectively, and the leaf node hash value is input into the hash verification layer to generate a Merkle path constraint; Based on the proof key and the proof private information, the data integrity constraints, the cryptographic operation constraints and the Merkle path constraint are respectively generated into sub-proofs, and the sub-proofs are combined to generate an original proof; The proof complexity is calculated based on the circuit size and the witness data dimension of the original proof, the proof segmentation size is determined according to the proof complexity and a preset memory limit, and the original proof is divided into multiple proof segments according to the proof segmentation size; For each proof segment, the number of proof items and the proof item correlation degree are calculated, the structure of the proof segment is reorganized to obtain an optimized proof segment, and the optimized proof segment is recursively combined according to a tree structure to obtain a combined proof; The proof feature of the combined proof is calculated, and the combined proof is directionally compressed based on the proof feature to obtain a compressed proof; The static evaluation results of the circuit size and the witness data dimension are obtained, the dynamic evaluation results of the threat type and the historical security events are obtained, and the security level is obtained by weighted combination; The security parameter is dynamically adjusted according to the security level, the security of the compressed proof is enhanced based on the security parameter to obtain a final proof, and the final proof is deployed to a verification smart contract.

7. A financial document automatic generation and verification system based on intelligent accounting, for implementing the method of any one of the preceding claims 1-6, characterized in that, Comprise: The first unit is used for scanning and identifying business documents, obtaining transaction information in the business documents, and converting the transaction information into standardized transaction data containing transaction amount, transaction time, transaction counterparty information, transaction type, commodity information and service information; The second unit is used for extracting semantic features from the standardized transaction data by using natural language processing technology, matching the extracted semantic features with a preset accounting subject feature library to generate initial accounting entries, constructing a transaction relationship network, identifying transaction abnormalities by graph structure analysis, optimizing the initial accounting entries, and generating final accounting entries. The third unit is configured to generate a financial voucher for the final accounting entry, analyze subject logical relations, amount matching relations, and voucher integrity of the financial voucher by using a deep neural network, analyze a transaction path involved in the financial voucher by using graph calculation, calculate a rule matching degree of the financial voucher based on a preset accounting standard rule library, and determine validity of the financial voucher. The fourth unit is configured to block-encrypt the financial voucher determined as valid, store encrypted data in a Merkle tree structure, verify authenticity of the financial voucher by using a smart contract supporting zero-knowledge proof in a distributed network, and write evidence information of the financial voucher into a blockchain network based on a consensus mechanism. The fifth unit is configured to write the financial voucher with completed evidence into an enterprise financial accounting system, and write check records, evidence records, and entry records of the financial voucher into the blockchain network.

8. An electronic device, comprising: The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 6. The computer program instructions are executed by the processor to implement the method in any one of claims 1 to 6. ​ ​ 9. A computer-readable storage medium having stored thereon computer program instructions, wherein, ​

Images