Intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion

Abstract

The invention discloses an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion, and relates to the technical field of financial intellectualization and data fusion, and the system comprises an invoice information collection module which is responsible for the recognition and structural processing of multiple types of invoices; the multi-source data fusion module is used for integrating multi-source data and quantifying reliability; the AI compliance detection module is used for evaluating invoice compliance; the abnormity early warning, non-compliance traceability rectification and compliance accounting pushing module is used for realizing early warning, rectification and automatic accounting; according to the method, a progressive technical architecture is constructed, multi-source data are automatically integrated, compliance risks are accurately identified, the judgment accuracy and consistency are improved, and financial risks are prevented; and meanwhile, by utilizing the non-compliance traceability and compliance accounting module, problem nodes are accurately positioned, adaptive rectification suggestions are generated, the processing period is shortened, automatic accounting and block chain evidence storage are realized, and the financial processing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of financial intelligence and data fusion technology, specifically to an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion. Background Technology

[0002] In corporate financial accounting and tax compliance management, invoices serve as core vouchers for recording economic transactions and clarifying rights and obligations. Their compliance directly impacts the authenticity of corporate financial data, the accuracy of tax declarations, and the effectiveness of operational risk control. Currently, businesses handle various types of invoices, including VAT special invoices, VAT general invoices, electronic invoices, and unified invoices for motor vehicle sales. Furthermore, the business data associated with these invoices is fragmented and source-divided—purchase orders and transaction contracts are stored in the company's ERP system, business scope and qualifications are kept on the industrial and commercial administration platform, while invoice information is managed independently in paper or electronic form. With the continuous improvement of the tax supervision system, tax authorities are increasingly stringent in verifying the consistency between invoices and business substance and qualification information. Enterprises urgently need a technical solution that can integrate multi-source data and achieve full-process compliance detection of invoices to meet the needs of efficient compliance management in large-scale business scenarios.

[0003] Currently, enterprises primarily rely on traditional manual review or partial verification by single software systems in the invoice compliance management process, which has significant technical limitations. In the manual review model, finance personnel must compare the information on each invoice with the core content of purchase orders and contracts, consuming substantial manpower and time. Furthermore, differences in personnel experience and data fragmentation can easily lead to missed detections of substantive compliance risks such as transactions exceeding the business scope or lack of corresponding business documentation. Single software systems can only perform format verification or simple logical judgments based on the invoice's own fields, unable to access business data from ERP systems or qualification information from business registration platforms, making it difficult to identify deeper compliance issues such as discrepancies between invoice information and business substance. Moreover, both of these methods lack unified quantitative standards for compliance judgment, resulting in inconsistent judgments across different batches and by different reviewers. This fails to provide stable and reliable compliance management support for enterprises and is ill-suited to the dual demands of expanding business scale and increasing compliance requirements. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion. This system constructs a closed-loop architecture throughout the entire process by collaboratively integrating invoice information collection, multi-source data fusion, AI compliance detection, an anomaly warning module, a non-compliance tracing and rectification module, and a compliance accounting push module. The system automatically integrates multi-source data from invoices, business registration, and ERP systems, and uses a dedicated quantitative formula to achieve accurate identification and tiered warning of compliance risks. At the same time, it accurately locates core non-compliant nodes and generates appropriate rectification suggestions. Compliant invoices can automatically generate standardized accounting vouchers and push them to financial software. Combined with blockchain evidence storage to ensure data credibility, this system effectively solves the pain points of traditional invoice management, such as reliance on manual labor, data fragmentation, and low efficiency, and significantly improves the consistency of compliance judgment and the efficiency of financial processing.

[0005] To solve the above-mentioned technical problems, this invention provides the following technical solution: an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion, the system comprising:

[0006] Invoice information collection module: used to receive various types of invoices, extract structured information and normalize formats, and output basic information related to the credibility of invoice data;

[0007] Multi-source data fusion module: used to connect industrial and commercial big data with enterprise internal ERP data, integrate invoice information, industrial and commercial qualification information and business document information, quantify the reliability of fused data through multi-source data fusion credibility formula, and screen high-quality fused datasets;

[0008] AI Compliance Detection Module: Based on a high-quality fusion dataset, it uses a comprehensive compliance score formula to quantitatively assess the consistency of invoices across the three flows and compliance risks, and determines whether invoices are compliant.

[0009] Anomaly warning module: used to trigger tiered warnings for non-compliant invoices identified by the AI ​​compliance detection module, and record the type of non-compliance and related business information;

[0010] Non-compliance tracing and rectification module: Receives non-compliance judgment results, locates the core business node that caused the invoice non-compliance through the problem tracing correlation formula, and then selects the best rectification suggestions that are suitable for the enterprise scenario from the historical case library through the rectification suggestion adaptation score formula, and pushes them to the corresponding person in charge;

[0011] Compliance accounting push module: Receives compliance assessment results, generates standardized accounting vouchers according to the company's preset financial rules, automatically pushes them to the company's financial software to complete the accounting entry, and links them to the original documents for archiving.

[0012] Furthermore, the invoice information collection module adopts a deep learning-based OCR engine based on the ResNet+CRNN architecture, which supports the recognition of 12 types of invoices, including VAT special invoices, VAT general invoices, electronic invoices, and motor vehicle sales unified invoices. It includes image preprocessing functions, which process blurred and wrinkled invoices through grayscale conversion, noise reduction, and tilt correction, and map the invoice information into standardized fields. The standardized fields include invoice code, invoice number, buyer and seller tax ID, buyer and seller name, name of transaction object, amount, and invoice date, and the fields are normalized in format.

[0013] Furthermore, the multi-source data fusion reliability calculation formula in the multi-source data fusion module is: F=ω1·D f +ω2·D i +ω3·D b Where F represents the final credibility of the integrated data, ranging from 0 to 1, with values ​​closer to 1 indicating higher credibility; ω1 represents the weight of invoice data, ω2 represents the weight of ERP data, and ω3 represents the weight of business registration data, the sum of which is 1. Initial values ​​can be set according to the enterprise's business type and automatically fine-tuned monthly based on manually reviewed data; D f For the reliability of invoice data; D i The credibility of ERP data is derived from the completion rate of the ERP data approval process; D b The credibility of business registration data is derived from the timeliness of its updates.

[0014] Furthermore, the formula for calculating the comprehensive compliance score in the AI ​​compliance detection module is: C = F·(0.6S + 0.4R) -1 The formula is as follows: C is the comprehensive score for invoice compliance, ranging from 0 to 1. Invoices are considered compliant when C is not lower than 0.8; F is the final credibility of the fused data; S is the semantic similarity of the consistency of the three flows, ranging from 0 to 1, which represents the degree of matching of the core information of the purchase order, invoice, and contract; R is the compliance risk coefficient, which is not lower than 1. When there is no compliance risk, R = 1. When there is compliance risk, R increases accordingly according to the risk level.

[0015] Furthermore, the formula for calculating the semantic similarity of the three-stream consistency is: S = 0.4S t +0.3S m +0.3S p Where S is the three-stream consistency semantic similarity; S t The semantic similarity of the subject matter is calculated by the BERT semantic model, yielding the similarity between the subject matter of the invoice and the subject matter of the order and contract; S m The amount matching score is derived from the error rate between the invoice amount and the order amount; S p For consistency between buyer and seller, the buyer and seller must have completely identical tax identification numbers. p=1, S when inconsistent p =0.

[0016] Furthermore, the anomaly warning module has three levels of tiered warnings. Level 1 warnings correspond to high-risk scenarios such as no corresponding purchase order, no corresponding contract, transaction subject exceeding the business scope, and invoices marked as abnormal by the tax system. Level 2 warnings correspond to medium-risk scenarios such as amount deviation greater than 0.1% but not exceeding 1% and purchase order approval not being completed. Level 3 warnings correspond to low-risk scenarios such as missing invoice remarks and mismatched specifications and models of the subject.

[0017] Furthermore, the formula for calculating the correlation degree of problem tracing in the non-compliance tracing and rectification module is as follows: Where T represents the correlation between the problem and the business node, ranging from 0 to 1. A node is considered a core problem node when T is not less than 0.7; C min C represents the minimum threshold for compliance score; C represents the comprehensive compliance score of invoices output by the AI ​​compliance detection module; P represents the process verification coefficient, which characterizes the completeness of the verification process at each business node.

[0018] Furthermore, the formula for calculating the adaptation score of rectification suggestions in the non-compliance tracing and rectification module is: A = T·(0.7S) c +0.3A e ), where A is the fit score of the rectification suggestion, ranging from 0 to 1, and an A score of not less than 0.7 is considered the optimal rectification suggestion; T is the relevance of the problem to its source; S c The historical case similarity is derived from the cosine similarity of the keyword vectors between the current non-compliant scenario and historical cases; A e The enterprise fit is calculated by combining the enterprise's industry attributes and historical rectification habits.

[0019] Furthermore, the standardized accounting vouchers generated by the compliant accounting push module include debit and credit accounts, amounts, number of supporting documents, and information of the preparer. The number of supporting documents is the total number of corresponding invoices, orders, and contracts. The accounting vouchers and corresponding original documents are archived using blockchain evidence storage technology to ensure that the data is tamper-proof and supports one-click retrieval of all original documents by voucher number.

[0020] Compared with existing technologies, this intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion has the following advantages:

[0021] I. This invention constructs a progressive technical architecture consisting of an "invoice information collection module - multi-source data fusion module - AI compliance detection module." Combining multi-source data fusion credibility formulas, compliance comprehensive score formulas, and three-stream consistency semantic similarity calculation methods, it can automatically integrate invoice information, business qualification information, and internal ERP business data. This enables reliable fusion and quantitative analysis of multi-dimensional data, accurately identifying substantive compliance risks such as transactions exceeding the business scope or lacking corresponding business vouchers, avoiding oversights from manual verification. Simultaneously, through standardized extraction and format normalization of invoice information, it ensures stable input data quality for compliance detection, effectively improving the accuracy and consistency of invoice compliance judgments. This provides enterprises with practical technical means to prevent financial risks such as false invoicing, false offsetting, and false accounting, reducing errors caused by manual operation in compliance management and enhancing the enterprise's ability to control financial risks.

[0022] Second, this invention, through the synergistic function of the non-compliance tracing and rectification module and the compliance accounting push module, combined with the problem tracing correlation formula, the rectification suggestion matching score formula, and blockchain evidence storage technology, can accurately locate the core business nodes of invoice non-compliance. It generates specific rectification suggestions tailored to the enterprise's industry attributes, based on historical rectification cases, helping finance personnel quickly close the problem loop and shorten the non-compliance processing cycle. Simultaneously, compliant invoices can automatically generate standardized accounting vouchers and be directly pushed to the enterprise's financial software, eliminating the need for manual entry and significantly reducing repetitive work. Blockchain technology also ensures the authenticity and immutability of accounting vouchers and corresponding original documents, meeting the data traceability needs of financial audits. It achieves intelligent processing throughout the entire process from compliance testing to subsequent accounting and archiving, reducing the manpower and time costs of enterprise financial operations and further improving overall financial processing efficiency.

[0023] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0025] Figure 1 Flowchart of an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion;

[0026] Figure 2 This is a framework diagram of an intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion. Detailed Implementation

[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0028] Example 1: Scenario of Raw Material Purchase Invoice Processing in Manufacturing Enterprises

[0029] This example is applied to a large equipment manufacturing enterprise. The enterprise mainly produces heavy machinery and needs to process invoices for the purchase of various raw materials such as steel, motors, and special parts on a daily basis. It has more than 200 cooperative suppliers and needs to strictly control the matching between invoices and the actual purchase business and supplier qualifications to avoid tax and financial risks such as purchasing beyond the scope of business and recording without corresponding orders.

[0030] The first step involves the invoice information collection module starting up. After receiving the paper VAT invoice from the supplier, the company uses a dedicated scanner from its finance department to transfer the invoice image into the system. The module employs a deep learning-based OCR engine with a ResNet+CRNN architecture to process the invoice image. First, image preprocessing is performed through grayscale conversion, noise reduction, and tilt correction. This processing effectively solves the recognition errors caused by transportation wrinkles and uneven ink distribution on paper invoices, significantly improving the accuracy of extracting key information such as invoice codes and buyer / seller tax identification numbers. Then, the invoice information is automatically extracted and mapped to standardized fields, including invoice code, invoice number, buyer / seller tax identification numbers, buyer / seller names, transaction name, amount, and invoice date. Simultaneously, the fields are formatted for normalization, such as standardizing the invoice date to YYYY-MM-DD format and retaining two decimal places for the amount. Finally, the system outputs standardized and complete basic invoice data, clearing format obstacles for subsequent multi-source data integration. Figure 1 As shown.

[0031] The second step involves the multi-source data fusion module performing data integration. The system automatically connects to the enterprise's internal ERP system and the business administration platform. It retrieves corresponding purchase orders from the ERP system, including information such as the specifications, quantity, agreed amount, and approval status of the purchased goods; and retrieves supplier qualification information such as business scope and operating status from the business administration platform. Subsequently, the reliability of invoice data, ERP data, and business administration data is quantitatively evaluated using a multi-source data fusion reliability formula. The multi-source data fusion reliability calculation formula is: F=ω1·D f +ω2·D i +ω3·D bWhere F represents the final credibility of the merged data; ω1 represents the weight of invoice data, ω2 represents the weight of ERP data, ω3 represents the weight of business registration data; D f For the reliability of invoice data; D i The credibility of ERP data is derived from the completion rate of the ERP data approval process; D b This formula assesses the credibility of business registration data. It effectively eliminates invalid business data from ERP systems that have not completed departmental approvals and expired qualification information from business registration platforms that has not been updated for more than 6 months. Ultimately, it selects high-quality fusion datasets with high data consistency and reliability, providing accurate and reliable data support for subsequent compliance testing and avoiding misjudgments in compliance decisions due to data quality issues.

[0032] The third step involves the AI ​​compliance detection module conducting compliance assessments. Based on a high-quality fused dataset, the module first evaluates the matching degree of core information in purchase orders, invoices, and supporting purchase contracts through three-stream consistency semantic similarity calculation. It focuses on verifying whether the transaction object matches the purchase order, whether the buyer's and seller's names match their tax identification numbers, and whether the invoice amount is within the agreed-upon amount range of the order. This calculation avoids matching omissions caused by differences in the description of the object's name during manual comparison. The three-stream consistency semantic similarity calculation formula is: S = 0.4S t +0.3S m +0.3S p Where S is the three-stream consistency semantic similarity; S t S represents the semantic similarity of the targets; m For monetary matching degree; S p To ensure consistency between the buyer and seller, and combining the credibility results of multi-source data fusion, the overall compliance of the invoice is quantitatively scored using a comprehensive compliance score formula. The final determination of whether the invoice is compliant is made using the formula: C = F * (0.6S + 0.4R). -1 In this scenario, C represents the overall compliance score of the invoice; F represents the final credibility of the fused data; S represents the semantic similarity of the consistency of the three flows; and R represents the compliance risk coefficient. In this scenario, an invoice for special alloy raw materials was deemed non-compliant because the transaction subject exceeded the scope of the supplier's registered business for the sale of ordinary steel. The overall compliance score did not reach the preset threshold, thus effectively preventing potential tax risks in advance.

[0033] The fourth step involves the anomaly warning module triggering a risk alert. Based on the non-compliance assessment results and in accordance with the three-level warning rules, the system determines that the transaction exceeding the business scope constitutes a high-risk scenario that may trigger a tax audit, thus triggering a Level 1 warning. The system simultaneously notifies the purchasing department personnel and the financial audit supervisor via pop-ups in the enterprise ERP system, targeted emails, and WeChat messages. The warning message clearly indicates the invoice ID, non-compliance type, risk level, and associated purchase order number, and automatically records the warning trigger time and recipients. This multi-channel notification method ensures that key personnel receive risk information immediately, preventing non-compliant invoices from reaching the accounting stage. Simultaneously, the clear association information reduces the time required for personnel to query original data, improving risk response efficiency.

[0034] Step 5: The non-compliance tracing and rectification module promotes a closed-loop problem resolution process. After receiving the first-level warning information, the module uses the problem tracing correlation formula to locate the core business node leading to the non-compliance. Specifically, before initiating the procurement of this batch of special alloys, the procurement department did not verify the supplier's business scope through the system and directly signed a procurement contract with the original steel supplier. The problem tracing correlation calculation formula is as follows: Where T represents the correlation between the problem and the business node; C min C represents the minimum threshold for compliance score; C represents the comprehensive compliance score of the invoice output by the AI ​​compliance detection module; and P represents the process verification coefficient. Subsequently, the system's historical case library is accessed, containing rectification plans for procurement issues exceeding the business scope over the past three years. Combined with the company's "Supplier Management System" and the supplier's five-year long-term cooperation history, the optimal rectification suggestion is selected using the rectification suggestion matching score formula. This involves the procurement specialist contacting the supplier to confirm their special alloy sales qualification. If not, the contract is terminated and a qualified supplier is selected. Simultaneously, a "Supplier Qualification Pre-Verification Form" is added to the system archive as a mandatory document for subsequent procurement approvals. The rectification suggestion matching score calculation formula is: A = T * (0.7S) c +0.3A e ), where A is the score for the fit of the rectification suggestions; T is the relevance of the problem to its source; S c For historical case similarity; A e For enterprise adaptation, rectification suggestions are generated into to-do tasks and pushed to the procurement specialist's ERP to-do list, with a rectification deadline of 3 working days. This precise positioning and adaptation suggestions can prevent the procurement department from blindly investigating the entire process, clarify the direction and time limit for rectification, and ensure that problems are quickly closed.

[0035] Step 6: The compliance accounting module completes the accounting and archiving process. After the purchasing specialist replaces the supplier with a compliant one according to the rectification recommendations, a new VAT invoice for special alloys is obtained. The new invoice is verified as compliant by the AI ​​compliance detection module. The module automatically generates standardized accounting vouchers according to the company's preset financial rules, corresponding to the raw material purchase under the "Raw Materials - Special Alloys" accounting subject, and recording the input tax amount under "Taxes Payable - VAT Payable". These vouchers include debit and credit accounts, amounts, number of supporting documents, and the information of the preparer. The number of supporting documents is the total quantity of invoices, new purchase orders, and new contracts. The accounting vouchers are automatically pushed to the company's SAP financial software for accounting, avoiding issues such as incorrect account selection and amount entry during manual data entry. Simultaneously, the system uses blockchain technology to archive the accounting vouchers and corresponding original documents, ensuring data immutability. During subsequent tax audits or internal audits, financial personnel can access the complete document chain with a single click using the voucher number, significantly improving audit efficiency and meeting financial data traceability requirements.

[0036] In summary, this embodiment demonstrates that the intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion achieves end-to-end compliance management in the manufacturing raw material procurement scenario through the collaborative operation of six modules. From the preprocessing and standardized extraction of invoice information by the invoice information collection module, to the reliable data screening by the multi-source data fusion module, and then to the precise interception of non-compliant invoices exceeding the business scope by the AI ​​compliance detection module, each link is closely connected, effectively solving the compliance risk prevention and control problem of multiple suppliers and multiple types of raw material invoices in the manufacturing industry. Subsequent multi-channel notifications for anomaly warnings, precise positioning and adaptation suggestions for non-compliance tracing and rectification, and automated accounting and blockchain storage of compliant invoices not only proactively avoid tax risks but also simplify the problem-solving process and improve financial processing efficiency. This fully demonstrates the system's adaptability to the complex procurement invoice management needs of the manufacturing industry and provides a practical and feasible technical solution for enterprise financial risk prevention and control.

[0037] Example 2: High-frequency, small-amount invoice processing scenario for service industry chain hotel enterprises

[0038] This example is applied to a national chain hotel enterprise with 56 stores in 30 provinces and cities. The enterprise needs to process high-frequency small-amount invoices for office supplies, cleaning supplies, consumables, etc. on a daily basis, with more than 500 invoices per month. The invoice types include electronic invoices and paper invoices. It is necessary to balance processing efficiency with compliance requirements such as noting the store number in the remarks and matching the amount with the purchase application, so as to avoid delays in accounting and the accumulation of low-risk compliance issues.

[0039] The first step involves the efficient processing of invoices by the invoice information collection module. After receiving invoices from suppliers, store purchasing staff upload electronic invoices directly as PDF files to the system, while paper invoices are scanned at the store's front desk. The module employs a deep learning-based OCR engine with a ResNet+CRNN architecture, using a unified recognition logic for both types of invoices. Paper invoices are first processed for grayscale conversion and noise reduction to address image blurring caused by insufficient scanner resolution, improving OCR recognition accuracy. Then, along with electronic invoices, the invoice information is extracted and mapped to standardized fields, including invoice code, invoice number, buyer and seller tax identification numbers, transaction name, amount, invoice date, and remarks. Subsequently, format normalization processing is performed, such as uniformly retaining two decimal places for amounts and mandating a space for the store number in the remarks field. Finally, standardized invoice data is output to the enterprise's cloud database. This unified recognition and standardization process avoids the cumbersome operation of processing different types of invoices separately, while ensuring consistent invoice data format across 56 stores, facilitating centralized financial management at headquarters and subsequent data integration. Figure 2 As shown.

[0040] The second step involves a multi-source data fusion module that integrates disparate data. The system retrieves purchase requisitions from the company's headquarters ERP system for the corresponding stores, including information such as the purchase target, application amount, approver, and approval time. It also retrieves supplier qualification information, such as business status and scope, from the business administration platform. A multi-source data fusion reliability formula is used to assess the reliability of these three types of data. The system focuses on verifying the approval completion status of ERP purchase requisitions, filtering out invalid data from purchases made before approval, and verifying the timeliness of updates to business qualification data to ensure that suppliers are currently operating and to avoid cooperating with suppliers with abnormal operations. Finally, a high-quality fusion dataset is selected, featuring approved purchase requisitions, suppliers operating normally, and complete invoice information. This reduces unnecessary workload in subsequent compliance checks and proactively mitigates the business risks of cooperating with abnormal suppliers.

[0041] The third step involves the AI ​​compliance detection module for rapid compliance determination. Based on a high-quality fused dataset, the module first matches core information from purchase requisitions, invoices, and simplified purchase contracts using three-stream consistency semantic similarity calculation. Given the scattered and small amounts of high-frequency, small-amount invoices, this calculation quickly verifies whether the invoice's subject matter falls within the scope of the purchase requisition and whether the amount exceeds the requisition amount. Then, combined with the credibility results of multi-source data fusion, a comprehensive compliance score formula is used to quantitatively determine the invoice's compliance. In this scenario, most invoices were deemed non-compliant because the remarks field lacked a store number; for example, the remarks on an invoice from the Chaoyang store in Beijing were empty. This quantitative determination method ensures that the 56 stores have consistent standards for judging low-risk issues like missing remarks, avoiding confusion caused by differences in the experience of store finance personnel and improving the uniformity of compliance management within the group.

[0042] The fourth step involves the anomaly warning module triggering tiered alerts. Based on the non-compliance assessment results and the three-tiered warning rules, the system determines that missing remarks are a low-risk scenario that does not affect tax risks and only requires supplementary information; therefore, a level three warning is triggered. The system notifies the corresponding store's finance specialist solely through a WeChat message. The warning message includes the invoice ID, non-compliance type, risk level, and associated purchase requisition number, and prompts the staff to supplement the remarks within 24 hours. This tiered warning design avoids overwhelming headquarters finance personnel with a large amount of low-risk information, while the short-term rectification prompts urge stores to process the issues quickly, preventing delays in accounting due to missing invoice remarks and ensuring timely monthly financial reporting.

[0043] The fifth step involves a simplified rectification process using the non-compliance tracing and rectification module. After receiving the level-three warning information, the module uses a problem-based correlation formula to pinpoint the core issue: the store purchasing staff failed to add the store number to the remarks field as prompted by the system when uploading invoices, rather than a system function or headquarters process issue. It then retrieves rectification solutions for missing remarks from the historical case library. Considering the store's high-frequency, low-amount purchases and the need for rapid processing, a simplified rectification suggestion is selected using a rectification suggestion matching scoring formula. This involves the store purchasing staff adding the store number to the invoice details page in the system. After submission, the system automatically re-triggers compliance checks without additional approval. The task is pushed to the store purchasing staff's to-do list. This suggestion requires no complex process and can be completed by the purchasing staff within 5 minutes, ensuring rectification efficiency and not affecting the store's daily purchasing and invoice accounting progress.

[0044] Step six: The compliant accounting push module completes automated accounting and archiving. After the store purchasing staff adds notes, the system re-enters the invoice data into the multi-source data fusion module and the AI ​​compliance detection module. After compliance is determined, the compliant accounting push module automatically generates standardized accounting vouchers according to the company's preset financial rules, with office supplies corresponding to management expenses - office expenses and cleaning supplies corresponding to management expenses - material consumption. These vouchers include debit and credit accounts, amounts, number of attached documents, and the information of the preparer. The number of attached documents is the total number of invoices and purchase requisitions. The vouchers are then automatically pushed to the company's UFIDA financial software for accounting. This process avoids the repetitive work of store finance staff manually entering more than 500 small invoices, saving approximately 80 hours of manpower per month. At the same time, the system uses blockchain notarization technology to archive accounting vouchers, invoices, and purchase requisitions. The headquarters finance department can access the invoice accounting data of each store in real time, meeting the financial needs of centralized management and decentralized operation of chain enterprises. Furthermore, blockchain technology ensures that the data is tamper-proof, improving the credibility of financial data and facilitating quarterly financial reviews and annual audits.

[0045] In summary, this embodiment demonstrates the system's advantages in efficient adaptation and centralized management within the high-frequency, low-value invoice processing scenario of chain hotels. Addressing the characteristics of numerous stores, large invoice volume, and diverse invoice types, the invoice information collection module achieves unified identification and standardized processing of electronic and paper invoices; the multi-source data fusion module filters high-quality data to support compliance detection; and the AI ​​compliance detection module accurately identifies low-risk issues such as missing remarks. Tiered early warning systems prevent information overload at headquarters, the non-compliance tracing and rectification module provides simplified rectification solutions to meet the efficiency requirements of high-frequency, low-value business, and the compliance accounting push module significantly reduces manual data entry workload. The system not only meets the timeliness requirements of invoice processing under the decentralized operation of chain enterprises but also achieves centralized management through blockchain evidence storage and real-time retrieval by headquarters, effectively reducing financial costs and adapting to the digital financial management needs of service industry chain enterprises.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. An intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion, characterized in that, The system includes: Invoice information collection module: used to receive various types of invoices, extract structured information and normalize formats, and output basic information related to the credibility of invoice data; Multi-source data fusion module: used to connect industrial and commercial big data with enterprise internal ERP data, integrate invoice information, industrial and commercial qualification information and business document information, quantify the reliability of fused data through multi-source data fusion credibility formula, and screen high-quality fused datasets; AI Compliance Detection Module: Based on a high-quality fusion dataset, it uses a comprehensive compliance score formula to quantitatively assess the consistency of invoices across the three flows and compliance risks, and determines whether invoices are compliant. Anomaly warning module: used to trigger tiered warnings for non-compliant invoices identified by the AI ​​compliance detection module, and record the type of non-compliance and related business information; Non-compliance tracing and rectification module: Receives non-compliance judgment results, locates the core business node that caused the invoice non-compliance through the problem tracing correlation formula, and then selects the best rectification suggestions that are suitable for the enterprise scenario from the historical case library through the rectification suggestion adaptation score formula, and pushes them to the corresponding person in charge; Compliance accounting push module: Receives compliance assessment results, generates standardized accounting vouchers according to the company's preset financial rules, automatically pushes them to the company's financial software to complete the accounting entry, and links them to the original documents for archiving.

2. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The invoice information collection module adopts a deep learning OCR engine based on the ResNet+CRNN architecture, which supports the recognition of 12 types of invoices, including VAT special invoices, VAT general invoices, electronic invoices, and motor vehicle sales unified invoices. It includes image preprocessing functions, which process blurred and wrinkled invoices through grayscale conversion, noise reduction, and tilt correction, and map the invoice information into standardized fields. The standardized fields include invoice code, invoice number, buyer and seller tax number, buyer and seller name, name of transaction object, amount, and invoice date, and the fields are normalized in format.

3. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The multi-source data fusion reliability calculation formula in the multi-source data fusion module is: F=ω1·D f +ω2·D i +ω3·D b Where F represents the final credibility of the merged data; ω1 represents the weight of invoice data, ω2 represents the weight of ERP data, ω3 represents the weight of business registration data; D f For the reliability of invoice data; D i For ERP data reliability; D b To ensure the credibility of business registration data.

4. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The formula for calculating the overall compliance score in the AI ​​compliance detection module is: C = F·(0.6S + 0.4R) -1 ), where C is the comprehensive score for invoice compliance; F is the final credibility of the fused data; S is the semantic similarity of the consistency of the three flows; and R is the compliance risk coefficient.

5. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 4, characterized in that, The formula for calculating the semantic similarity of the three-stream consistency is: S = 0.4S t +0.3S m +0.3S p Where S is the three-stream consistency semantic similarity; S t S represents the semantic similarity of the targets; m For monetary matching degree; S p To ensure consistency between the buyer and seller.

6. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The anomaly warning module has three levels of tiered warnings. Level 1 warnings correspond to high-risk scenarios such as no corresponding purchase order, no corresponding contract, transaction subject exceeding the business scope, and invoices marked as abnormal by the tax system. Level 2 warnings correspond to medium-risk scenarios such as amount deviation greater than 0.1% but not exceeding 1% and purchase order approval not being completed. Level 3 early warning corresponds to low-risk scenarios such as missing invoice remarks or mismatched specifications and models of the subject matter.

7. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The formula for calculating the correlation degree of problem tracing in the non-compliance tracing and rectification module is as follows: Where T represents the correlation between the problem and the business node; C min C represents the minimum threshold for compliance score; C represents the comprehensive compliance score of invoices output by the AI ​​compliance detection module; and P represents the process verification coefficient.

8. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The formula for calculating the adaptation score of rectification suggestions in the non-compliance tracing and rectification module is: A = T·(0.7S) c +0.3A e ), where A is the score for the fit of the rectification suggestions; T is the relevance of the problem to its source; S c For historical case similarity; A e For enterprise adaptability.

9. The intelligent invoice compliance detection and automatic accounting system based on multi-dimensional feature fusion according to claim 1, characterized in that, The standardized accounting vouchers generated by the compliant accounting push module include debit and credit accounts, amounts, number of attached documents, and information of the preparer. The number of attached documents is the total number of corresponding invoices, orders, and contracts. The accounting vouchers and corresponding original documents are archived using blockchain evidence storage technology, and all original documents can be accessed with one click by voucher number.

Images