Template and signature consistency intelligent quality inspection method and system for contract approval

Abstract

The invention discloses an intelligent quality inspection method and system for consistency of a template and a signature for contract approval, and particularly relates to the technical field of intelligent quality inspection for contract approval. The problems that contract templates are wrongly selected, pages are missing or repeated, signing and sealing are incomplete, contract numbers and work order numbers are inconsistent, only manual check one by one can be relied on, efficiency is low, and the rate of missing detection and false detection is high are solved. An approval work order is taken as a unified entrance, an asynchronous queue, contract image preprocessing and identification, template knowledge base matching, page-level mapping and signature number joint inspection are introduced, and template suitability, page number integrity and signature consistency are converged into an automatic quality inspection link, so that batch contract pre-quality inspection is realized on the premise that an existing approval process is not transformed, and the quality inspection efficiency is improved. Missing detection and false detection are reduced; and examination and approval efficiency and stability are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent quality inspection technology for contract approval, specifically to an intelligent quality inspection method and system for the consistency of contract approval templates and signatures. Background Technology

[0002] In the context of group contract approval within telecommunications operators, existing systems typically rely on the group's service hall or unified business acceptance platform. Account managers initiate approval work orders and upload scanned copies or photos of the contracts. The backend primarily performs formal checks based on field completeness and simple rules, such as verifying whether required fields are filled, whether the amount and validity period fall within the specified range, and whether the number of attachments meets requirements. For verifying whether the contract template matches the specified business type, whether the actual number of pages matches the template, whether all parties' signatures and seals are complete, and whether the contract number matches the work order number, manual review of the contract image page by page is still the primary method. Reviewers manually check against standard templates based on experience. This approach is sustainable when the number of contracts is small, but in peak scenarios such as centralized signing of group clients, centralized adjustment of packages, and annual bulk procurement, manual quality inspection becomes labor-intensive, standards are difficult to standardize, and omissions and misjudgments are prone to occur. Contracts with missing signatures, seals, or incorrect numbers are difficult to intercept in a timely manner during the approval process.

[0003] To reduce manual workload, some existing solutions attempt to introduce optical character recognition (OCR) and rule engines to perform text recognition on contract images and make simple comparisons based on key fields, or to use general semantic models to perform compliance analysis on contract terms. However, these solutions typically treat contracts as plain text and fail to systematically link "business template version, contract page structure, signature and seal positions, and contract number coding rules": template selection often relies on manual input or manual selection in work orders; page count checks are mostly limited to comparing the total number of pages, failing to establish a stable mapping between template page order and physical contract pages; signature and seal detection is mostly coarse-grained identification of whether a seal is missing, making it difficult to determine whether it is a company seal, authorized signature, or a seal across the binding; number verification is mostly direct string comparison, lacking structured decomposition of internal coding segments such as region, year, business, and transaction history. This makes it difficult for existing technologies to conduct timely and reliable integrated and consistent pre-quality checks on the correct use of contract templates, the completeness of page counts, the completeness of various signatures, and the consistency of contract numbers with work orders in high-concurrency approval scenarios.

[0004] Against this backdrop, in the group contract approval process of telecommunications operators, how can we construct an intelligent quality inspection link that uses the approval work order as the entry point and is oriented towards a template knowledge base? This would enable contract images to simultaneously undergo template compatibility verification, page number integrity verification, signature and seal completeness verification, and contract number and work order number consistency verification under unified rules. In a real business environment with batch processing and high concurrency, this would reduce the reliance on manual verification and experience-based judgment, and lower the risk of missing signatures, seals, and incorrect numbers being overlooked during the approval process. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent quality inspection method and system for ensuring consistency between contract approval templates and signatures, thereby resolving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent quality inspection method for the consistency of contract approval templates and signatures, comprising: S1. Receive the approval work order carrying contract attachments, generate a quality inspection task containing the work order identifier, business type, and attachment path, and write it to the asynchronous queue. S2. Retrieve the quality inspection task from the asynchronous queue, perform preprocessing on the contract image, and call the optical character recognition model and layout parsing model to obtain the text content layout features and clause structure. S3. Based on the business type, filter the candidate template set from the contract template knowledge base, calculate the matching degree based on the layout features and semantic features of the clauses, and select the standard template that exceeds the threshold. S4. Based on the standard template page number information and chapter layout, map the physical pages of the contract to the template pages, and deduce the coordinate range of the expected signature area number retrieval area according to the template signature layout rules. S5. Perform seal detection, signature detection and feature filtering in the expected signature area, and determine whether the company seal authorization signature is missing according to the template signature rules to form a signature consistency result. S6. Extract the number text from the number retrieval area and standardize it. Compare the work order numbers in segments according to preset fields. Combine the template matching results to generate a quality inspection conclusion based on page integrity and signature consistency. Write the conclusion into the quality inspection result database and write it back to the approval work order.

[0007] Furthermore, S1 includes: The group service hall will send the work order identifier, business type, contract number, and contract attachment location to the work order robot; The work order robot verifies the integrity rules and generates a quality inspection record and writes it to the quality inspection result database when it detects any of the following situations: missing business type or missing contract attachment. A fingerprint of the attachment content is generated based on the work order identifier and the content of the contract attachment, and an idempotent key is formed with the work order identifier. The above information, configuration version identifier, and idempotent key are encapsulated into a quality inspection task and written into an asynchronous queue. The asynchronous queue retains only one quality inspection task with the same idempotent key and schedules them according to the submission order.

[0008] Furthermore, S2 includes: The quality inspection service side generates contract images based on the attachment path during the work process. Preprocessing the contract image includes cropping blank edges and tilt correction to obtain a preprocessed contract image. The preprocessed contract image is input into the optical character recognition model to obtain the text entries and signature area text content, and then input into the layout analysis model to obtain the layout analysis information. The clause structure is constructed and layout features are generated based on chapter titles and clause numbers; Write the clause structure and layout features, along with the version identifier of the optical character recognition model, the version identifier of the layout parsing model, and the time stamp, into the quality inspection temporary storage space.

[0009] Furthermore, S3 includes: A contract template knowledge base is pre-established, and contract template entries are set in the contract template knowledge base. Each contract template entry includes a template identifier and a set of applicable business types. The contract template also includes standard page count, chapter layout, list of key terms, signature layout rules, numbering layout rules, and rule version identifiers; Based on the business type field in the quality inspection task, the quality inspection service side filters the applicable business type set from the contract template knowledge base, and the contract template entries containing the business type field are used as the candidate template set.

[0010] Furthermore, the quality inspection service side uses layout features to calculate the degree of layout similarity with the standard page numbers and chapter layouts in the candidate template entries; Calculate the coverage ratio and semantic proximity of key clauses by utilizing the clause structure and the semantic features of items in the key clause list; The matching score is obtained by combining the similarity of the layout, the coverage ratio of key clauses, and the semantic similarity of clauses according to the weight scheme set in the configuration version. Based on the relationship between the matching score and the pass threshold and suspected threshold, a standard template is determined, and the template identifier, matching score, and threshold version identifier are written into the quality inspection temporary storage space.

[0011] Furthermore, S4 includes: After obtaining the standard template, the quality inspection service side reads the number of physical pages and compares it with the standard page number information to obtain the page integrity status. Establish a contract page mapping relationship based on the chapter identification information of the physical pages of the contract and the chapter layout of the standard template; Based on the contract page mapping relationship and in conjunction with the template signature layout rules and numbering layout rules, the expected signature area and numbering retrieval area are generated, and both are represented by standardized coordinate quadruples. Page integrity status, contract page mapping relationship, expected signature area, and number retrieval area are written to the quality inspection temporary storage space.

[0012] Furthermore, S5 includes: The quality inspection service reads the expected signature area from the quality inspection temporary storage space and calls the visual recognition component to extract candidate graphics and candidate trajectories within the expected signature area. Candidates for official seals, authorized signatures, and seals across pages are selected based on color characteristics, outline characteristics, trajectory characteristics, and cross-page imprints, respectively. The system checks the above candidates in the signature area of ​​the last page and the signature area across the seam according to the signature layout rules, generates a signature consistency status, and writes the signature consistency status, missing item details, visual recognition component version identifier and signature layout rule version identifier into the quality inspection temporary storage space.

[0013] Furthermore, S6 includes: The quality inspection service side reads the number retrieval area from the quality inspection temporary storage space, and obtains and standardizes the number text through optical character recognition within the number retrieval area. The work order number is segmented into regional segment, year segment, business segment and serial segment according to the field. The work order number is processed according to the same rule to obtain the segment number. The number comparison result is obtained by comparing the corresponding fields. The quality inspection rule engine generates quality inspection conclusions based on template matching results, page completeness, signature consistency, and number comparison results. Write the quality inspection conclusions and explanations of the differences between the signatures and numbers, along with the template identifier and rule version identifier, into the quality inspection result database.

[0014] On the other hand, the present invention provides an intelligent quality inspection system for the consistency of contract approval templates and signatures, comprising: The quality inspection task management module is used to receive approval work orders with contract attachments, generate quality inspection tasks containing work order identifiers, business types and attachment paths, and write the quality inspection tasks to an asynchronous queue. The image analysis module is used to retrieve quality inspection tasks from the asynchronous queue, perform preprocessing on the contract images, and call the optical character recognition model and layout analysis model to obtain the text content, layout features and clause structure. The template matching module is used to filter candidate template sets from the contract template knowledge base according to business type, input the layout features and clause semantic features into the matching algorithm to calculate the matching degree, and select the templates with matching degree exceeding the threshold as standard templates. The page mapping and region derivation module is used to map the physical pages of the contract to the page sequence of the template based on the page number information and chapter layout of the standard template, and derive the coordinate range of the expected signature area and number retrieval area according to the template signature layout rules. The signature detection module is used to perform seal detection, signature detection and feature filtering in the expected signature area. Based on the template signature layout rules, it determines whether the company seal, authorized signature and cross-seam seal are missing, and forms a signature consistency result. The number comparison and conclusion generation module is used to extract the number text from the number retrieval area and perform normalization processing. It compares the work order numbers in segments according to preset fields, combines the number comparison results with the template matching results, page number integrity and signature consistency to generate quality inspection conclusions, writes the quality inspection conclusions into the quality inspection result database and writes them back to the approval work order.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By using the approval work order as a unified entry point, and combining asynchronous queue scheduling, contract image preprocessing and optical character recognition, template knowledge base-driven format and clause matching, mapping between physical contract pages and template pages, derivation of expected signature areas and number retrieval areas, and joint verification of company seals, authorized signatures, cross-sealing seals, and contract numbers, the contract template adaptability, page completeness, signature completeness, and number consistency are converged into a continuous automated quality inspection chain. This achieves consistent intelligent quality inspection of batch enterprise contracts before execution without modifying the existing group service hall approval process, reducing reliance on personal experience for manual page-by-page verification, reducing the risk of missed or false inspections, and improving the overall processing efficiency and quality stability of the contract approval process.

[0016] 2. By introducing configuration version identifiers, template rule version identifiers, and model version identifiers into quality inspection tasks, and using work order identifiers and attachment content fingerprints to construct idempotent keys to control the queuing and overriding of quality inspection tasks, the entire process of template matching status, page number integrity status, signature consistency results, number difference explanations, and corresponding timestamps is recorded in the quality inspection temporary storage space and quality inspection result database. The quality inspection conclusions are divided into three levels: pass, fail, and require manual review, and are fed back to the approval work order interface. This achieves the effect of traceable rule caliber, verifiable model updates, and accurate location and hierarchical handling of abnormal contracts in large-scale concurrent quality inspection scenarios, which facilitates the continuous optimization of quality inspection strategies by operating units and meets the requirements of internal risk control and external compliance audits. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the intelligent quality inspection method for consistency between contract approval templates and signatures according to the present invention. Figure 2 This is a schematic diagram of the structure of an intelligent quality inspection system for the consistency of contract approval templates and signatures according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: An intelligent quality inspection method for the consistency between contract approval templates and signatures, comprising: S1. Receive the approval work order carrying contract attachments, generate a quality inspection task containing the work order identifier, business type, and attachment path, and write it to the asynchronous queue. The specific implementation is as follows: In the context of contract approval for enterprise customers of telecom operators, account managers initiate approval work orders through the group service hall. An approval work order refers to a business record within the group service hall used to drive the approval process of an enterprise contract. It includes at least the following fields: work order identifier, business type, contract number, enterprise identifier, and contract attachment location. The business type field identifies the product category corresponding to the contract, preferably set to codes such as cloud dedicated line, government and enterprise dedicated line, or voice dedicated line. The contract number field identifies the unique number registered by the operator. The enterprise identifier field identifies the enterprise customer. The contract attachment location field indicates the storage location of contract attachments in the file storage system. Contract attachments refer to a set of contract photo files or scanned contract files bound to the approval work order, preferably using a uniform image format or layout file format to ensure that subsequent recognition stages can retrieve them according to a unified specification.

[0020] A work order robot refers to an automated business component deployed within the operator's operations team. It retrieves the location of approval work orders and contract attachments from the group's service hall and generates subsequent quality inspection tasks. A quality inspection task is a smart contract quality inspection unit encapsulated around an approval work order, carrying fields such as work order identifier, business type, contract number, attachment path, and configuration version identifier, used for tracking throughout the quality inspection process. The configuration version identifier is a version marker of the currently used template rules, signature rules, and recognition model combination, used to identify the rule criteria used for this quality inspection.

[0021] Asynchronous queues refer to message queue services used to temporarily store quality inspection tasks. They employ a first-in-first-out (FIFO) scheduling mechanism to rhythmically retrieve tasks from multiple quality inspection service processes, thereby decoupling pre-contract approval quality inspection from backend identification capabilities.

[0022] In this implementation, when an account manager submits an approval work order in the group service hall, the group service hall transmits the aforementioned fields to the work order robot according to a preset format. The work order robot first checks the work order identifier, business type, contract number, enterprise identifier, and contract attachment location item by item according to the field integrity rules. For approval work orders that do not include contract attachments or lack a business type, a quality inspection record is directly generated and written to the quality inspection result database. The quality inspection result database refers to a storage system that centrally stores the conclusions, reasons, and version identifiers of each quality inspection. At the same time, a prompt message is returned to the group service hall to guide the account manager to complete the missing information before initiating the quality inspection, avoiding idle processing that occupies backend resources. When the fields are complete and the attachment location is valid, the work order robot generates an attachment content fingerprint based on the work order identifier and the content of the contract attachment file. The attachment content fingerprint refers to a mark obtained by summarizing the content of the contract attachment, which is used to identify the situation where the same contract is submitted repeatedly in a short period of time. The work order identifier and the attachment content fingerprint together form an idempotent key. The idempotent key is used to identify multiple triggers that are considered to be the same quality inspection object within the same time window. Preferably, the time window can be set to several minutes. If the same idempotent key appears multiple times within this time window, only the latest trigger is retained.

[0023] The work order robot generates a quality inspection task identifier based on this information. It encapsulates the work order identifier, business type, contract number, attachment path, configuration version identifier, and idempotent key combination into a task message and writes it to the asynchronous queue. When writing, it preferably adopts a first-in-first-out order to ensure that work orders in the same business batch are queued in the order of submission as much as possible. The asynchronous queue retains only one in-transit task for the same idempotent key within the same time window. The previous quality inspection task that has not yet been completed is marked as expired, and the reason for being overwritten is recorded in the quality inspection result database to prevent multiple contradictory results.

[0024] In this way, without changing the original interface of the group's service hall, approval work orders and contract attachments are stably transformed into quality inspection tasks that can be undertaken by the downstream intelligent quality inspection link. This realizes a unified entry point and applicable boundary for the intelligent quality inspection method of contract approval template and signature consistency in the on-site process, providing a clear and traceable preliminary link for subsequent contract image recognition, template matching, signature inspection and number verification. In the preferred implementation, the time window can be set to several minutes in the provincial operator's production system, and the number of concurrent consumption in the asynchronous queue can be set to several. This ensures that when several hundred to several thousand enterprise contract approval work orders enter the quality inspection link every day during peak hours, the front-end submission action maintains a response time of seconds, the queuing time of the quality inspection task in the queue is kept in the tens of seconds range, and the proportion of work order records with missing attachments or missing business types in the quality inspection result library is controlled within a low proportion range of the total number of work orders. This ensures that the overall process has both large-scale processing capabilities and meets the timeliness and compliance requirements of contract approval.

[0025] S2. Retrieve the quality inspection task from the asynchronous queue, perform preprocessing on the contract image, and call the optical character recognition model and layout parsing model to obtain the text content layout features and clause structure. The specific implementation is as follows: After a contract enters the intelligent quality inspection stage, the quality inspection service-side process deployed within the operator retrieves quality inspection tasks sequentially according to the order in the asynchronous queue. The quality inspection service-side process refers to the contract quality inspection service unit that resides on the server. It is used to complete the contract image parsing and template matching preparation work corresponding to a quality inspection task within a limited time limit. Preferably, each quality inspection task can be set to complete this step within a time limit of several seconds to several tens of seconds. The attachment path refers to the storage location identifier of the contract attachments written into the quality inspection task by the work order robot in the previous stage. It can be a path mark in the file storage system or an access mark in object storage. The contract image refers to a set of single-page images obtained by uniformly converting the contract photo file or contract scan file according to the attachment path. Each page of the contract image is associated with a unique page number and a time mark of the generation time to support subsequent page sequence alignment and traceability.

[0026] In actual operation, to ensure the quality of subsequent recognition, the quality inspection service first performs preprocessing on each page of the contract image. Preprocessing refers to the image quality improvement process of cropping obvious blank edges, reducing image noise, correcting tilt angles, unifying resolution, and unifying color space on the original contract image. Cropping is used to remove scan borders and invalid edges, noise reduction is used to reduce scan noise and compress artifacts, tilt angle correction is used to adjust the direction of the main text of the contract to near horizontal, unifying resolution is used to scale contract images from different sources to a uniform pixel density, and unifying color space is used to make a uniform agreement between grayscale mode and color mode. The order and parameter range of the above actions are locked through configuration version, which is represented by the configuration version identifier in the aforementioned quality inspection task, to ensure that the same batch of contracts uses the same preprocessing standards.

[0027] The preprocessed single-page contract image is sequentially fed into an optical character recognition (OCR) model and a layout analysis model. The OCR model is a component used to identify Chinese characters, numbers, and symbols from the contract image, outputting the main text entries, number strings, and text content near the signature area on each page. The main text entries refer to the paragraphs or entries in the contract terms, the number strings refer to the consecutive character sequences that may be the contract number, and the signature area refers to the row where the signature and seal are expected to appear in the standard template and several rows nearby. The layout analysis model is a layout analysis component used to analyze the spatial distribution of text blocks, header areas, footer areas, and table areas in the contract image. Text blocks refer to visual blocks composed of several characters arranged within the same line spacing, the header area refers to the area located at the top edge of the contract image, mainly containing title and number information, the footer area refers to the area located at the bottom edge of the contract image, mainly containing page numbers and the signature area, and the table area refers to the area composed of table lines and grids.

[0028] After obtaining the main text and layout information of the multi-page contract, the quality inspection service process organizes the text of each page using chapter titles and clause numbers as anchors. Chapter titles refer to the title text used to divide the contract into chapters, and clause numbers refer to the contract clause numbers marked with serial numbers. By identifying these chapter titles and clause numbers, the multi-page text is organized into a clause structure according to the chapter hierarchy and clause order. The clause structure refers to the hierarchical relationship record describing the division of the entire contract into chapters and the order of clauses. At the same time, the text block distribution, header area height, footer area height, and table area position in the page layout analysis results are organized to form layout features. Layout features refer to the set of features used to describe the overall form of the contract in terms of page count, text layout, and header and footer structure.

[0029] The clause structure and layout features, along with the version identifiers of the optical character recognition model, the layout parsing model, and the generation time stamp, are all written into the quality inspection temporary storage space. The quality inspection temporary storage space refers to the working storage area on the quality inspection service side used to store an intermediate result of a quality inspection task. It can adopt relational storage or key-value storage format to provide a unified foundation for subsequent template knowledge base matching, signature area deduction, and number position deduction. In actual deployment, for ease of maintenance and upgrade, the optical character recognition component and the layout parsing component can preferably be designed as replaceable modules. As long as the replaced component can still obtain the three types of information—text items, layout features, and clause structure—on the contract image and satisfy the above-mentioned clause structure and layout feature organization method, it is considered an equivalent implementation under the same concept.

[0030] In a preferred embodiment, for contract photos with low resolution or large shooting angle deviations, the upper limit of the correction tilt angle in the preprocessing stage can be set to a range of several degrees, the resolution can be unified to a few thousand pixels on the short side, and the preprocessing parameters, model version, and timestamps can be recorded in the quality inspection temporary storage space. This allows the operator to statistically identify the accuracy rate, inspection time, and failure rate based on the sampling quality inspection records. After confirming that the indicators have reached the predetermined level, the configuration version can be solidified to support the operator in obtaining stable text content, layout features, and clause structure results when hundreds to thousands of enterprise contracts enter this step each month under different collection conditions. This provides a reproducible implementation basis for subsequent template matching and signature consistency checks.

[0031] S3. Based on the business type, filter candidate template sets from the contract template knowledge base, calculate the matching degree using layout features and semantic features of clauses, and select standard templates that exceed the threshold. The specific implementation is as follows: In the context of contract approval by telecom operators, in order to achieve reusable template-driven quality inspection capabilities across different business types and contract versions, operators pre-build a contract template knowledge base. The contract template knowledge base refers to a structured storage collection formed by organizing various standard contract texts, with each contract template entry corresponding to a controlled contract version.

[0032] Each contract template entry must be bound to at least a template identifier, a set of applicable business types, a standard number of pages, a chapter layout, a list of key terms, a signature layout rule, a numbering layout rule, and a rule version identifier. The template identifier is used to uniquely identify the contract template, the set of applicable business types is used to limit which business types of approval work orders the template can be used for, the standard number of pages is used to record the range of pages that the contract corresponding to the template should contain, and the chapter layout is used to describe the distribution of the chapter titles in the template on the logical page order.

[0033] The list of key terms refers to the set of terms that are considered to be required in the template. Each key term stores the title, the relative position of the term in the template, and the semantic features of the term in the list. The title is used to locate the corresponding term in the identification results, the relative position of the term is used to check the approximate page location of the term when mapping subsequent pages, and the semantic features of the term refer to the semantic representation of the full text of the key terms generated based on the internal language model component. It is used to measure the degree of similarity between the meaning of the terms in the identification results and the meaning of the terms in the template.

[0034] The signature layout rules are used to record the arrangement requirements of the company seal, authorized signature, and cross-sealing seal in the logical page order and page area as agreed in the template, providing a basis for subsequent signature consistency checks; the numbering layout rules are used to record the pages and areas where the contract number is expected to appear in the template, for the derivation of the numbering retrieval area in the subsequent process; the rule version identifier is used to mark the current signature layout rule and numbering layout rule version used for the template entry, so as to form an evidence chain of rule standards during the quality inspection process.

[0035] In the previous step, the quality inspection service side's work process has recorded the identified format features and clause structure for each quality inspection task in the quality inspection temporary storage space. The format features refer to the set of morphological features that reflect the entire contract in terms of page count, text block distribution, header and footer height, and table area position. The clause structure refers to the hierarchical relationship of the contract clauses arranged according to chapters and clause order.

[0036] In this step, the quality inspection service first reads the business type field carried by the quality inspection task, and then selects contract template entries containing the business type from the contract template knowledge base to form a candidate template set. The candidate template set refers to the subset of templates that participate in the matching calculation under the current quality inspection task, which is used to avoid cross-interference between templates from different business lines.

[0037] Subsequently, the quality inspection service side compared the layout features with the standard page numbers and chapter layouts in the candidate template entries. By quantifying the relationship between the actual number of pages in the contract and the standard number of pages, and the alignment of the distribution of chapter titles on the physical pages with the template chapter layout, a layout similarity score was obtained for each candidate template. The layout similarity score is a numerical indicator used to measure how close the overall layout of the contract is to the layout of a certain template.

[0038] Meanwhile, the quality inspection service extracts clause content corresponding to the item titles in the key clause list of each template from the clause structure. This clause content is converted into clause semantic features through an internal language model component. These features are then compared with the pre-stored item semantic features in the key clause list of the template. The key clause coverage ratio is obtained based on the proportion of key clauses that can be successfully aligned to the total number of key clauses. The semantic proximity of the clauses is obtained based on the closeness between the semantic features of each corresponding clause. The key clause coverage ratio and the semantic proximity of the clauses together reflect the consistency between the contract clause content and the template clause content.

[0039] The system uses a weighted combination of layout similarity, key clause coverage ratio, and clause semantic similarity to form a matching score based on the preset weighting scheme in the current configuration version. The matching score refers to the comprehensive similarity value given to a certain template item, which is used to determine whether the template can be used as the standard template for this quality inspection.

[0040] To address the complexity of different business lines and templates, the configuration version pre-sets a pass threshold and a suspected threshold for each type of business. The pass threshold refers to the threshold at which the contract can be considered sufficiently close to the template in terms of structure and content after the matching score is reached. The suspected threshold refers to the lower limit of the threshold range that is below the pass threshold but still has a certain degree of similarity.

[0041] The quality inspection service calculates a matching score for each contract template item in the candidate template set. Within a single quality inspection task, the template with the highest matching score that is not lower than the threshold is selected as the standard template. Any candidate template whose matching score does not reach the suspected threshold is marked as an unmatched standard template, which is used to directly determine template inconsistency in the future.

[0042] The template identifier, matching score, threshold version identifier of the threshold and suspected threshold obtained above are recorded together with the aforementioned layout features and clause structure in the quality inspection temporary storage space. In this step, the quality inspection temporary storage space serves as a transit area that provides a fixed reference for subsequent page mapping and rule execution steps.

[0043] In actual deployment, this step supports adjustments to the internal language model components and weighting schemes without altering the contract template knowledge base structure and matching process. As long as the definitions of format similarity, key clause coverage ratio, clause semantic proximity, and matching score remain stable, it can be considered an equivalent implementation of this method. In a preferred embodiment, different pass thresholds and suspected thresholds can be set for different business lines based on historical quality inspection results. For example, the pass threshold for complex government and enterprise dedicated line contracts can be set to a higher value, and the weights of key clause coverage ratio and clause semantic proximity can be appropriately increased to ensure that the matched standard templates have high consistency in both page format and key clause content. Thus, when operators process hundreds to thousands of enterprise contracts each month, they can automatically select suitable standard templates and identify contracts with significant structural differences as unmatched standard templates and prompt for manual review, meeting the accuracy and traceability requirements of intelligent quality inspection scenarios for contract approval template signature consistency.

[0044] S4. Based on the standard template page number information and chapter layout, map the physical pages of the contract to the template pages, and deduce the coordinate range of the expected signature area number retrieval area according to the template signature layout rules. The specific implementation is as follows: After completing template matching and obtaining the standard template, the quality inspection service process performs page-level alignment and key area derivation of the current contract based on the structural information in the standard template entries. The standard template refers to the contract template selected in the aforementioned matching stage that is closest to the current contract in terms of layout and clause content. This template has standard page number information and chapter layout pre-registered in the contract template knowledge base; the standard page number information refers to the number of pages that the contract corresponding to this template should contain and the chapter scope carried by each logical page, and the chapter layout refers to the distribution position of each chapter title on the logical page order.

[0045] A physical page of a contract refers to a collection of single-page contract images formed after image preprocessing and layout recognition. Each physical page of a contract has a continuous page number and chapter identification information. The chapter identification information is marked by the recognition results of the chapter title and clause number in the previous step, which is used to indicate the chapter to which the page belongs or the beginning position of the chapter.

[0046] The quality inspection service first reads the number of physical pages corresponding to the current quality inspection task in the contract. This number is compared with the number of pages in the standard page number information to obtain the page number integrity status. The page number integrity status is used to mark whether the contract is consistent with the standard template in terms of page number. When it is found that the number of physical pages in the contract is less than the standard number of pages or there is obvious duplicate content, the current contract is recorded as incomplete in terms of page number. At the same time, subsequent actions are reserved, and the chapter information continues to try to establish logical alignment. Specifically, the quality inspection service finds the chapter paragraphs corresponding to the chapter titles in the chapter layout of the standard template based on the chapter identification information on the physical pages of the contract. The physical pages containing the first lines of these chapter paragraphs are aligned to the logical pages containing the corresponding chapters in the standard template, thereby constructing the correspondence between the physical page number of the contract and the page number of the template.

[0047] The contract page mapping relationship refers to the mapping result formed during the alignment process mentioned above. It can be represented as a pair of records of physical page numbers and template page numbers, which are used to accurately project the layout rules in the template onto the contract image plane in subsequent steps.

[0048] After obtaining the contract page mapping relationship, the system derives the expected signature area of ​​this contract based on the template signature layout rules pre-registered in the standard template. The template signature layout rules refer to the signature and seal arrangement requirements agreed upon in the template, including at least the logical position descriptions of the last page signature area, the parties' signature areas, and the signature area across the seam. The last page signature area refers to the reserved area on the logical last page of the contract for the company seal and authorized signature. The parties' signature areas refer to the areas where all participating parties in a multi-party contract sign and seal. The signature area across the seam refers to the signature area that spans the edges of adjacent pages for preventing splitting.

[0049] Based on the contract page mapping relationship, the quality inspection service converts the page order and edge description of these logical areas in the template into geometric areas in the physical pages of the contract. Preferably, these areas are given in the form of rectangular boundaries or cross-page rectangular boundaries. Each expected signature area can be described by four boundary coordinates. The coordinate quadruple formed by the four boundary coordinates refers to the numerical combination used to locate the upper boundary, lower boundary, left boundary, and right boundary of the rectangular area in a certain coordinate system. To adapt to contract images of different resolutions, the coordinate quadruple is preferably represented in a standardized coordinate ratio form, that is, each boundary position is converted into a proportional value between zero and one according to the page width and page height. Subsequently, the actual coordinates can be obtained by scaling the actual image according to the current pixel width and height.

[0050] The numbering layout rules refer to the provisions in the standard template regarding the location of contract numbers. They must include at least the logical page where the number is located and the area of ​​the relative page edge. For example, the contract number may be specified to be located in the upper right area of ​​the cover or in the middle of the header of the first page. In this step, the quality inspection service side also uses the contract page mapping relationship to convert the numbering layout rules into a rectangular area on the physical page of the contract, forming a number retrieval area. The number retrieval area refers to the coordinate range on the contract image used to find the number string.

[0051] The expected signature area and number retrieval area are recorded in the form of coordinate quadruples or standardized coordinate ratios, and written into the quality inspection temporary storage space together with the source template identifier and rule version identifier. The quality inspection temporary storage space refers to the aforementioned working storage area used to store the intermediate results of the current quality inspection task. The subsequent signature detection and number comparison stages directly read these area definitions without parsing the template text again.

[0052] This step-by-step mapping from the standard template to the physical pages of the contract ensures that the signature and numbering rules in the template accurately fall into the appropriate positions on the specific contract image. This is suitable for contracts with a completely identical number of standard pages, as well as contracts with a few missing or duplicate pages where the chapter structure can still be aligned. In a preferred embodiment, for a cloud leased line contract template with a standard number of pages, when the number of physical pages is exactly the same and the chapter titles are fully identified, the contract page mapping relationship presents a one-to-one correspondence. The signature area on the last page is deduced as one or more rectangular areas at the lower edge of the last physical page, overlapping the seam. The signature area is deduced as a rectangular area spanning two adjacent pages on the outer edge, and the number retrieval area is deduced as a rectangular area in the upper right corner of the physical first page. In another scenario where the physical pages of a contract are one page short, while recording the incomplete page count, the system can still construct a partial mapping by aligning chapter titles, ensuring that subsequent signature detection and number comparison at least cover the complete content of the chapter structure. The operating unit can then provide applicable adjudication boundaries for the combination of incomplete page count and signature results in the quality inspection rules, thereby ensuring that the overall process remains executable and auditable even in real-world scenarios where the page count is not fully regulated.

[0053] S5. Perform seal detection, signature detection, and feature filtering in the expected signature area. Determine whether the company's official seal, authorization signature, or cross-sealing seal is missing according to the template signature rules to form a signature consistency result. The specific implementation is as follows: After completing the contract page mapping and derivation of the expected signature area, the quality inspection service process reads the set of expected signature areas corresponding to the current quality inspection task from the quality inspection temporary storage space. Within each expected signature area, seal detection, signature detection, and feature filtering are performed sequentially. The expected signature area refers to the rectangular or cross-page rectangular area derived on the physical page of the contract according to the signature layout rules of the standard template, used to hold the company seal, authorized signature, and binding seal. Seal detection refers to identifying graphic objects with the shape of a seal within the expected signature area, and signature detection refers to identifying trajectory objects with the shape of a handwritten signature within the expected signature area. The company seal refers to the contract seal used to indicate the will of the company, generally presented with a circular or polygonal boundary and containing the company name text inside. The authorized signature refers to the name or signature trajectory written by the authorized representative of the contract within the signing area, often appearing as a continuous curve. The binding seal refers to the seal arranged across the edges of adjacent physical pages of the contract, used to prevent the contract from being torn apart or replaced.

[0054] The quality inspection service uses a visual recognition component to obtain a set of candidate graphics and candidate trajectories in each expected signature area. The visual recognition component refers to an image recognition unit that can mark suspected seal areas and suspected signature areas on the contract image, preferably in the form of several rectangular boxes to provide candidate positions. The system calculates the color concentration within the candidate graphics area, whether the color is concentrated in the red or dark range, the degree of contour closure, and the internal text shape. Candidate graphics with contours close to circles or polygons and closed boundaries, and containing ring or radial text inside, are marked as official seal candidates. Candidate graphics that cross the page edge and have continuous imprints in corresponding positions on two adjacent pages are marked as cross-stitch seal candidates. Candidate graphics with discrete color distribution, severely fragmented contours, or no internal text structure are eliminated. For candidate trajectories, signature candidates are selected based on curve continuity, line width variation, and stroke overlap. Regular horizontal lines, vertical lines, or standard printed fonts are eliminated, and trajectories with natural writing characteristics are preserved as much as possible.

[0055] After feature filtering is completed, the system checks the number of official seals and authorized signatures in the signature area of ​​the last page item by item according to the pre-registered signature layout rules in the standard template. The signature area of ​​the last page refers to the reserved area on the logical last page of the template for affixing the company seal and filling in the authorized signature. When the number of official seal candidates is less than the rule requirement or the authorized signature candidate is missing, the page is marked as missing the company seal or missing the authorized signature, and the missing type and physical page number are recorded. For the expected signature area recorded as the cross-page signature area, the system checks whether there are cross-page seal candidates on the edge of the page. When no continuous imprint is found in the cross-page area, the page pair involved is marked as missing the cross-page seal.

[0056] The above inspection results are categorized and organized according to missing items of company seal, missing items of authorized signature, and missing items of seal across the seam, and summarized into a signature consistency result. The signature consistency result indicates whether the current contract meets the standard template signature layout rules in terms of signature and seal. When all expected signature areas have the corresponding company seal, authorized signature, or seal across the seam, the signature consistency result is marked as consistent. When any missing item exists, the signature consistency result is marked as inconsistent. Preferably, a mark requiring manual review can be added to the rules to remind the approver to reconfirm in conjunction with the original image.

[0057] The quality inspection service writes the signature consistency status, missing item details, visual recognition component version identifier, and signature layout rule version identifier into the quality inspection temporary storage space. The visual recognition component version identifier records the currently used signature recognition capability version, and the signature layout rule version identifier records the rule version used for this judgment, thus forming a traceable signature evidence chain. In the actual production system, an upper limit can be set on the number of candidate graphics for the expected signature area on a single page by the visual recognition component to control computational overhead. For cases where the image is blurry or severely occluded, resulting in generally low confidence levels for candidate objects, the page can preferably be recorded as having an uncertain signature status and classified into the signature inconsistency range. At the same time, the reason for insufficient recognition signal quality should be marked in the missing item details for manual review. The replacement of the visual recognition component can be carried out as long as it still... Being able to output candidate official seals, candidate authorized signatures, and candidate cross-sealing seals that meet the above-mentioned color features, contour features, and trajectory features within the expected signature area, and supporting the formation of signature consistency results under the constraints of signature layout rules, can be regarded as an equivalent implementation without changing the core idea of ​​the present invention. In a preferred embodiment, the operating unit can adjust the relevant thresholds of color concentration, contour closure, and curve continuity based on the sampling results of several hundred signed contracts, so that the proportion of contracts that are automatically judged to be signature-consistent is actually missing signatures or seals is controlled within a preset range, while keeping the signature detection time of a single contract within an acceptable range, thereby ensuring that the intelligent quality inspection method for consistency between contract approval templates and signatures is feasible and reusable under different contract scanning quality and different signing habits.

[0058] S6. Extract and standardize the number text from the number retrieval area, compare the work order numbers in segments according to preset fields, and generate a quality inspection conclusion based on the page number integrity and signature consistency of the joint template matching results. Write the conclusion into the quality inspection result database and write it back to the approval work order. The specific implementation is as follows: After completing template matching, page number mapping, and signature verification, the quality inspection service side process reads the number retrieval area, template matching results, page number integrity status, and signature consistency status for the current quality inspection task record from the quality inspection temporary storage space. It then extracts the numbered text from the number retrieval area and completes subsequent comparisons and conclusion generation.

[0059] The number retrieval area is the coordinate range of the contract number derived from the aforementioned number layout rules. It is recorded in the quality inspection temporary storage space in the form of standardized coordinates to limit number recognition to only within this area. The number text refers to one or more continuous strings obtained by optical character recognition within the number retrieval area, which contains the contract number and possible prefix text.

[0060] To avoid the impact of different writing habits on the results, the system prioritizes the standardization of the numbered text. Standardization means unifying the character forms in the numbering without changing the meaning of the number. Full-width characters and half-width characters are unified into a single form, hyphens, slashes, periods and other separators are unified into conventional symbols or removed directly, extra spaces are deleted, and the letter forms are uniformly processed, thereby forming a standard numbering string that can be directly used for field splitting.

[0061] Preset field segmentation refers to dividing the standard number string into several segments according to the numbering rules established internally by the operator. These segments include at least a regional segment, a year segment, a business segment, and a serial number segment. The regional segment is used to identify the region to which the contract belongs, the year segment is used to identify the year of signing, the business segment is used to identify the business category corresponding to the contract, and the serial number segment is used to identify the sequence number under that region and business. The work order number refers to the contract number field entered by the account manager in the work order approval process in the group service hall. The standard numbering structure on the work order side is obtained through the same standardization and field segmentation.

[0062] Template matching result refers to the template identifier and matching status formed in the third step, which is used to explain the standard template and matching quality on which this quality inspection is based; Page completeness refers to the status obtained by comparing the physical pages of the contract recorded in the fourth step with the standard page number; Signature consistency refers to the status of whether the signature is consistent or not formed in the fifth step.

[0063] The quality inspection service will extract and standardize the number text from the number retrieval area and segment it according to preset fields. The work order number will be segmented according to the same rules, and the regional segment, year segment, business segment, and serial number segment will be compared one by one. If there is a difference in the regional segment or business segment, the comparison result of this contract number will be directly recorded as a number inconsistency. If the regional segment, year segment, and business segment are consistent, and there is only a few digit difference in the serial number segment and the difference is within the allowable range set in the rules, the number comparison result will be recorded as a suspected data entry error to prompt the approver to check it carefully. When all four segments are consistent, the number comparison result will be recorded as a number consistency.

[0064] The quality inspection rule engine makes decisions based on a comprehensive analysis of template matching results, page completeness status, signature consistency status, and number comparison results in the current configuration version. The quality inspection rule engine refers to the set of rules and execution units used to merge the above multiple statuses into a unified quality inspection conclusion according to the business side agreement. Under typical configuration, it gives a pass conclusion for contracts that have passed template matching and have complete page count, consistent signatures, and consistent numbers; it gives a fail conclusion for contracts with obviously inconsistent numbers or inconsistent signatures; and it gives a conclusion that manual review is required for contracts with template matching in a suspected state or numbers in a suspected data entry error state.

[0065] The quality inspection service generates a quality inspection record based on this. The quality inspection record includes at least the quality inspection conclusion status, a description of the location of missing signatures (recording the corresponding page number and missing type when there are missing signatures), a description of numbering differences (recording the difference paragraphs and difference forms when there are inconsistent numbers or suspected input errors), template identifier, rule version identifier, model version identifier of the participating identification components, and time stamp. The quality inspection record is then written into the quality inspection result database.

[0066] The quality inspection result database refers to a persistent storage area that centrally stores the conclusions and evidence chains of each quality inspection task, supporting index queries based on work order identifiers or quality inspection task identifiers. The group service hall initiates queries to the quality inspection result database via service interfaces based on work order identifiers to obtain the conclusion status and prompt information from the corresponding quality inspection records. The approval work order interface displays "passed," "failed," or "requires manual review" markings. When a work order fails to pass or requires manual review, it displays explanations of missing signature locations and number discrepancies, guiding account managers or approvers to take corrective measures. For cases where attachments are missing, configuration versions are not available, or asynchronous queue capacity is limited, preventing entry into the normal quality inspection chain, the quality inspection service side returns a separate status identifier and explains the reason in the prompt information. The interface exchange fields are preferably kept within the minimum set of work order identifier, quality inspection task identifier, conclusion status, and prompt text to reduce system coupling and facilitate subsequent internal implementation replacement.

[0067] Through the above process, without changing the existing approval path, template matching results, page completeness, signature consistency, and numbering consistency can be unified into a single quality inspection conclusion, realizing the standardization and traceability of pre-approval quality inspection for contracts. In the optimal working condition of a provincial operation center, the number of daily quality inspection work orders can be set to a medium scale, and several parallel task slots can be configured on the asynchronous queue side. The template matching pass threshold can be set in a higher range, and the suspected threshold can be set in a middle range. Then, combined with the sampling inspection results of automatically approved contract samples during the operation period, the allowable difference in the number of digits and the tolerance boundary for missing signatures can be adjusted to keep the proportion of missing signatures and incorrect numbers in automatically approved contracts within a preset range. At the same time, the time from work order submission to the quality inspection conclusion being written into the quality inspection result database and back to the approval work order can be kept in the tens of seconds range, thereby ensuring that this method has feasibility and stable performance in high-concurrency enterprise contract approval scenarios.

[0068] Example 2: An intelligent quality inspection system for the consistency of contract approval templates and signatures, comprising: The quality inspection task management module is used to receive approval work orders with contract attachments, generate quality inspection tasks containing work order identifiers, business types and attachment paths, and write the quality inspection tasks to an asynchronous queue. The image analysis module is used to retrieve quality inspection tasks from the asynchronous queue, perform preprocessing on the contract images, and call the optical character recognition model and layout analysis model to obtain the text content, layout features and clause structure. The template matching module is used to filter candidate template sets from the contract template knowledge base according to business type, input the layout features and clause semantic features into the matching algorithm to calculate the matching degree, and select the templates with matching degree exceeding the threshold as standard templates. The page mapping and region derivation module is used to map the physical pages of the contract to the page sequence of the template based on the page number information and chapter layout of the standard template, and derive the coordinate range of the expected signature area and number retrieval area according to the template signature layout rules. The signature detection module is used to perform seal detection, signature detection and feature filtering in the expected signature area. Based on the template signature layout rules, it determines whether the company seal, authorized signature and cross-seam seal are missing, and forms a signature consistency result. The number comparison and conclusion generation module is used to extract the number text from the number retrieval area and perform normalization processing. It compares the work order numbers in segments according to preset fields, combines the number comparison results with the template matching results, page number integrity and signature consistency to generate quality inspection conclusions, writes the quality inspection conclusions into the quality inspection result database and writes them back to the approval work order.

[0069] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A smart quality inspection method for the consistency between contract approval templates and signatures, characterized in that, include: S1. Receive the approval work order carrying contract attachments, generate a quality inspection task containing the work order identifier, business type, and attachment path, and write it to the asynchronous queue. S2. Retrieve the quality inspection task from the asynchronous queue, perform preprocessing on the contract image, and call the optical character recognition model and layout parsing model to obtain the text content layout features and clause structure. S3. Based on the business type, filter the candidate template set from the contract template knowledge base, calculate the matching degree based on the layout features and semantic features of the clauses, and select the standard template that exceeds the threshold. S4. Based on the standard template page number information and chapter layout, map the physical pages of the contract to the template pages, and deduce the coordinate range of the expected signature area number retrieval area according to the template signature layout rules. S5. Perform seal detection, signature detection and feature filtering in the expected signature area, and determine whether the company seal authorization signature is missing according to the template signature rules to form a signature consistency result. S6. Extract the number text from the number retrieval area and standardize it. Compare the work order numbers in segments according to preset fields. Combine the template matching results to generate a quality inspection conclusion based on page integrity and signature consistency. Write the conclusion into the quality inspection result database and write it back to the approval work order.

2. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S1 includes: The group service hall will send the work order identifier, business type, contract number, and contract attachment location to the work order robot; The work order robot verifies the integrity rules and generates a quality inspection record and writes it to the quality inspection result database when it detects any of the following situations: missing business type or missing contract attachment. A fingerprint of the attachment content is generated based on the work order identifier and the content of the contract attachment, and an idempotent key is formed with the work order identifier. The above information, configuration version identifier, and idempotent key are encapsulated into a quality inspection task and written into an asynchronous queue. The asynchronous queue retains only one quality inspection task with the same idempotent key and schedules them according to the submission order.

3. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S2 include: The quality inspection service side generates contract images based on the attachment path during the work process. Preprocessing the contract image includes cropping blank edges and tilt correction to obtain a preprocessed contract image. The preprocessed contract image is input into the optical character recognition model to obtain the text entries and signature area text content, and then input into the layout analysis model to obtain the layout analysis information. The clause structure is constructed and layout features are generated based on chapter titles and clause numbers; Write the clause structure and layout features, along with the optical character recognition model version identifier, the layout parsing model version identifier, and the time stamp, into the quality inspection temporary storage space.

4. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S3 includes: A contract template knowledge base is pre-established, and contract template entries are set in the contract template knowledge base. Each contract template entry includes a template identifier and a set of applicable business types. The contract template also includes standard page count, chapter layout, list of key terms, signature layout rules, numbering layout rules, and rule version identifiers; Based on the business type field in the quality inspection task, the quality inspection service side filters the applicable business type set from the contract template knowledge base, and the contract template entries containing the business type field are used as the candidate template set.

5. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 4, characterized in that: The quality inspection service side calculates the similarity of the layout by comparing the layout features with the standard page numbers and chapter layouts in the candidate template entries; Calculate the coverage ratio and semantic proximity of key clauses by utilizing the clause structure and the semantic features of items in the key clause list; The matching score is obtained by combining the similarity of the layout, the coverage ratio of key clauses, and the semantic similarity of clauses according to the weight scheme set in the configuration version. Based on the relationship between the matching score and the pass threshold and suspected threshold, a standard template is determined, and the template identifier, matching score, and threshold version identifier are written into the quality inspection temporary storage space.

6. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S4 include: After obtaining the standard template, the quality inspection service side reads the number of physical pages and compares it with the standard page number information to obtain the page integrity status. Establish a contract page mapping relationship based on the chapter identification information of the physical pages of the contract and the chapter layout of the standard template; Based on the contract page mapping relationship and in conjunction with the template signature layout rules and numbering layout rules, the expected signature area and numbering retrieval area are generated, and both are represented by standardized coordinate quadruples. Page integrity status, contract page mapping relationship, expected signature area, and number retrieval area are written to the quality inspection temporary storage space.

7. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S5 include: The quality inspection service reads the expected signature area from the quality inspection temporary storage space and calls the visual recognition component to extract candidate graphics and candidate trajectories within the expected signature area. Candidates for official seals, authorized signatures, and seals across pages are selected based on color characteristics, outline characteristics, trajectory characteristics, and cross-page imprints, respectively. The system checks the above candidates in the signature area of ​​the last page and the signature area across the seam according to the signature layout rules, generates a signature consistency status, and writes the signature consistency status, missing item details, visual recognition component version identifier and signature layout rule version identifier into the quality inspection temporary storage space.

8. The intelligent quality inspection method for consistency between contract approval templates and signatures as described in claim 1, characterized in that, S6 include: The quality inspection service side reads the number retrieval area from the quality inspection temporary storage space, and obtains and standardizes the number text through optical character recognition within the number retrieval area. The work order number is segmented into regional segment, year segment, business segment and serial segment according to the field. The work order number is processed according to the same rule to obtain the segment number. The number comparison result is obtained by comparing the corresponding fields. The quality inspection rule engine generates quality inspection conclusions based on template matching results, page completeness, signature consistency, and number comparison results. Write the quality inspection conclusions and explanations of the differences between the signatures and numbers, along with the template identifier and rule version identifier, into the quality inspection result database.

9. A smart quality inspection system for the consistency of contract approval templates and signatures, used to implement the smart quality inspection method for the consistency of contract approval templates and signatures as described in any one of claims 1-8, characterized in that, include: The quality inspection task management module is used to receive approval work orders with contract attachments, generate quality inspection tasks containing work order identifiers, business types and attachment paths, and write the quality inspection tasks to an asynchronous queue. The image analysis module is used to retrieve quality inspection tasks from the asynchronous queue, perform preprocessing on the contract images, and call the optical character recognition model and layout analysis model to obtain the text content, layout features and clause structure. The template matching module is used to filter candidate template sets from the contract template knowledge base according to business type, input the layout features and clause semantic features into the matching algorithm to calculate the matching degree, and select the templates with matching degree exceeding the threshold as standard templates. The page mapping and region derivation module is used to map the physical pages of the contract to the page sequence of the template based on the page number information and chapter layout of the standard template, and derive the coordinate range of the expected signature area and number retrieval area according to the template signature layout rules. The signature detection module is used to perform seal detection, signature detection and feature filtering in the expected signature area. Based on the template signature layout rules, it determines whether the company seal, authorized signature and cross-seam seal are missing, and forms a signature consistency result. The number comparison and conclusion generation module is used to extract the number text from the number retrieval area and perform normalization processing. It compares the work order numbers in segments according to preset fields, combines the number comparison results with the template matching results, page number integrity and signature consistency to generate quality inspection conclusions, writes the quality inspection conclusions into the quality inspection result database and writes them back to the approval work order.

Images