A financial scanning accessory and method of verifying association with a credential

Abstract

This invention relates to the field of financial data processing technology and discloses a method for verifying the association between scanned financial attachments and vouchers. The method first acquires the page and page sequence information of continuously scanned financial attachments and voucher entry data. Based on the attachment boundaries and the continuity of adjacent pages, the range of attachments and corresponding pages is determined. The number of attachments and pages are recorded respectively. Candidate relationships are established by extracting the transaction entity, invoice type, amount, date, and page order. Then, the data is filtered according to the transaction entity, invoice type, total amount, and page and entry order. Cases of non-unique correspondence or inconsistencies are identified and recorded based on page quality, attachment quantity, amount differences, and voucher field differences. This achieves accurate association between financial attachments and voucher entries and enables anomaly verification.

Description

Technical Field

[0001] This invention relates to the field of financial data processing technology, specifically to a method for verifying the association between scanned financial attachments and vouchers. Background Technology

[0002] In the process of digital archiving of financial vouchers, paper attachments such as invoices, bank receipts, contracts, acceptance forms, and approval documents are usually scanned to form electronic images, and then a corresponding relationship is established with the corresponding accounting vouchers or reimbursement data. Existing technologies mainly use text recognition, voucher numbers, file indexes, or structured fields to identify and associate scanned images. For example, authorized Chinese invention patent application CN114495085B discloses a reimbursement method for online invoice recognition and management across multiple platforms, which extracts invoice information through text recognition and associates invoice images with reimbursement data. Another example is authorized Chinese invention patent application CN109636553B, which discloses a voucher management method, device, computer equipment, and storage medium, which compares relevant parameters in the voucher with transaction parameters by calling transaction logs to complete voucher verification. However, during continuous batch scanning of financial attachments, a single scanned page may contain multiple invoices, and a contract, schedule, or other attachment may occupy multiple pages consecutively. Therefore, the number of scanned pages does not correspond to the actual number of attachments. Existing methods mainly establish correspondences based on identified text fields, voucher numbers, or existing document indexes. When there are multiple attachments on one page, an attachment spanning multiple pages, missing pages, incorrect pages, or changes in the scanning order, it is easy to cause errors in the correspondence between scanned images and voucher data, and these errors may continue to affect the association results of subsequent pages. In addition, image quality issues such as reflections, creases, and incomplete cropping during the scanning process, as well as missing attachments and errors in the entry of voucher fields, may all result in a mismatch between the image and the voucher data. Existing methods can usually only identify whether there is an anomaly in the correspondence, but it is difficult to further determine whether the anomaly is caused by the quality of the scanned image, missing attachments, or errors in the voucher data. At the same time, when the file name, voucher number, or external index changes due to data migration, file conversion, etc., the original correspondence is also difficult to verify again. Therefore, existing technologies still have technical problems such as the easy occurrence of errors in the correspondence between scanned pages, actual attachments and voucher data under continuous scanning conditions, and the difficulty in accurately determining the cause of anomalies after they occur. These problems affect the accuracy of batch archiving of financial attachments and the reliability of subsequent verification. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a method for verifying the association between financial scanning attachments and vouchers, thus solving the problems mentioned in the background section.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for verifying the association between scanned financial attachments and vouchers includes: Obtain the continuously scanned financial attachment pages, page sequence information, and voucher entry data, and organize the pages according to page sequence; Identify independent attachments and cross-page attachments based on attachment boundaries and the continuity of content on adjacent pages, and record the page range corresponding to each attachment; Record the number of attachments and the number of pages respectively, extract the transaction entity, invoice type, amount, date and page order, and establish candidate relationships with the corresponding fields of the voucher entries; First, filter candidate relationships by transaction entity and bill type. Then, verify that the total amount of the attachments associated with the entry is consistent with the entry amount. When an attachment is associated with multiple entries, the total amount of each allocation is consistent with the attachment amount. Continue filtering by page order and entry order. For cases where multiple correspondences or verification inconsistencies still exist, anomaly detection is performed based on image quality, number of attachments, amount gaps, and differences in voucher fields, and the correspondences and verification records are saved.

[0005] Preferably, the process involves acquiring continuously scanned financial attachment pages, page sequence information, and voucher entry data, and organizing the pages according to page sequence, including: Record the batch identifier, page identifier, paper feed sequence number or device page number, acquisition time and acquisition resolution for each page; When the paper feed sequence number is broken, record the page gap. When the paper feed sequence number is missing, arrange the pages according to the acquisition time. When there is a duplicate paper feed sequence number or a page order conflict, determine the page order based on the page content, acquisition time, device cache records and retransmission records. If it cannot be determined, record the corresponding page order as abnormal. Page data and voucher entry data are saved separately when they arrive in sequence. Once both types of data are obtained, the data format of amount, date, and currency is standardized and the original values ​​are retained.

[0006] Preferably, independent attachments and cross-page attachments are identified based on attachment boundaries and the continuity of content on adjacent pages, including: Obtain the horizontal and vertical dimensions, tilt angle, binding hole position, and paper feed marks of each page; compare the changes in size, tilt angle, and relative position of binding holes of adjacent pages to determine the range of consecutive pages. The attachment area is determined based on the edges, light and dark variations, and local shadows between the attachment paper and the carrier paper. When the edge of the attachment is interrupted, the attachment boundary is determined based on the extension direction of the two sides of the interruption, the table separator line, the fixed title position, the QR code position, and the position of the main text area. When attachments overlap, the scope of the attachments is determined based on the visible edges, the actual size of similar documents, and the extension positions of the edges on both sides of the obscured area.

[0007] Preferably, the page range corresponding to each attachment is recorded, including: Determine the page continuity of multiple attachments based on the titles, page numbers, table boundaries, header positions, footer positions, fixed printing areas, transaction entities, and page specifications of adjacent pages; When a new document title appears on the next page, the transaction entity changes, the layout structure changes, or a new independent attachment starts, the page continuity of the previous attachment is terminated. For double-sided documents, the two pages are determined to belong to the same attachment based on the fixed format of the front and back sides, the invoice number, the transaction entity, and the continuous collection relationship. The attachment identifier, starting page, ending page, corresponding page, and position within the page are recorded.

[0008] Preferably, the number of attachments and the number of pages are recorded separately, and the transaction entity, invoice type, amount, date, and page order are extracted, including: Attachments that are independent of each other on the same page are counted separately. Attachments that are composed of multiple consecutive pages or are composed of front and back pages are counted as one attachment. For each attachment, record the attachment identifier, corresponding page, first appearance page order, and number of pages occupied. Obtain the transaction entity, bill type, amount, and date based on the transaction entity field, bill title, fixed format, amount field, and date field. Mark fields that cannot be reliably read as unreadable, and save the attachment data and the original data of the voucher entries separately.

[0009] Preferably, candidate relationships are established with the fields corresponding to the voucher entries, including: Standardize the character format of the transaction entity's name while retaining the original name, and filter the voucher entries based on the correspondence between the expense items and attachment types configured in the business system; Retain journal entries with consistent transaction entities. If the transaction entity field in the appendix cannot be reliably read, do not exclude journal entries based on this field. Exclude journal entries where the type of invoice does not match the expense item. Screen candidate journal entries according to the date range stipulated in the financial system. Compare the actual number of attachments with the number of attachments recorded on the voucher, and verify the quantity discrepancies according to the actual registration methods of multi-page contracts, double-sided receipts, and multi-copy documents.

[0010] Preferably, candidate relationships are first filtered by transaction entity and type of bill, and then the total amount of the attachments associated with the journal entry is verified to be consistent with the journal entry amount. Furthermore, when multiple journal entries are associated with an attachment, the total amount of each allocated entry is consistent with the attachment amount, including: Verify the transaction entities in the attachments and voucher entries, and filter candidate relationships based on the correspondence between expense items and attachment types; If the transaction parties or types of negotiable instruments listed in the appendix cannot be reliably obtained, the corresponding candidate relationships shall be retained. Using the journal entry as the object, verify the total amount of its corresponding attachments. Using the attachment as the object, verify the total amount of each journal entry corresponding to it, and check whether the amount of the same attachment is used repeatedly in different journal entries.

[0011] Preferably, further filtering is performed by page order and entry order, including: The order of attachment pages is determined by the position of the first page of attachments in the current scanning batch, the order of voucher entries is determined by the original entry number of the voucher, and the order of voucher entries is determined by the line number of the reimbursement details when the business system records the line number of the reimbursement details. When multiple candidate relationships exist, compare the relative order of the attachment page order with the order of the voucher entries. If no business record indicates a change in order, exclude candidate relationships that are opposite to the relative order. When the order of a partial page cannot be determined, continue filtering based on the determined page positions before and after that partial area, combined with the transaction entity, bill type, and amount; If multiple candidate relationships still exist after filtering, save the attachment identifier, page position, and voucher entry information corresponding to each candidate relationship.

[0012] Preferably, for cases where multiple correspondences or verification inconsistencies still exist, anomaly detection is performed based on image quality, number of attachments, amount discrepancies, and differences in voucher fields, including: For candidate relationships that cannot be uniquely determined or have verification discrepancies, check whether reflections, shadows, creases, tilts, and cropping on the page affect the reading of the amount, transaction entity, date, and invoice type; Compare the actual number of attachments, the number of attachments recorded on the voucher, the amount not corresponding, and the types of attachments that should be included, and check whether any attachments are missing; Compare the fields that can be read in the attachment with the original fields on the voucher, and check whether there are any differences between the two. When there are two or more types of anomalies, such as image quality abnormalities, missing attachments, and differences in voucher fields, count the number of judgment conditions that have been met for each type of anomaly and the number of conditions that can be used for judgment.

[0013] Preferably, the corresponding relationship and verification records are saved, including: For the confirmed correspondence between attachments and voucher entries, record the batch identifier, attachment identifier, page range, voucher location, entry location, number of attachments, number of pages, transaction entity, type of invoice, amount, date, unreadable fields, amount differences, and page order anomalies; When the correspondence cannot be uniquely determined, save the attachments, pages, and voucher entries involved in each candidate relationship separately; After re-collecting the pages or manually confirming the correspondence, only the affected attachments and their corresponding voucher entries are re-verified.

[0014] Compared with the prior art, the present invention provides a method for verifying the association between financial scanned attachments and vouchers, which has the following beneficial effects: 1. This invention, by organizing the sequential order of continuously scanned pages, identifies multiple attachments on the same page and attachments occupying multiple consecutive pages based on attachment boundaries and the continuity of content on adjacent pages. It records the number of attachments and the number of pages, and establishes candidate correspondences between attachments and voucher entries based on the transaction entity, document type, amount, date, and page order. Furthermore, it verifies the total amount of attachments corresponding to voucher entries, the total amount of each voucher entry corresponding to an attachment, and the sequential relationship between pages and entries. When an anomaly is found, it combines page image quality, attachment quantity, non-corresponding amount, and voucher field differences to make a judgment and save the verification record. This reduces errors in attachment-voucher correspondence caused by multiple items on one page, multiple pages for one item, missing pages, incorrect pages, and changes in scanning order, and improves the reliability of anomaly identification and subsequent review.

[0015] 2. This invention, by separately saving the original data of attachments and voucher entries, preserves the original state of fields that cannot be reliably read, and records the page range, page order, amount differences, and candidate correspondences corresponding to the attachments; when re-collecting pages, manually confirming correspondences, or reviewing after archiving, the affected pages, attachments, and voucher entries are determined based on the saved records, and only the relevant data is re-checked; when the file name, voucher number, or external index changes, the verification can still be performed based on the corresponding records between attachments, pages, and voucher entries, thereby reducing the duplicate processing of the entire batch of data caused by local anomalies, facilitating subsequent retrieval of attachments and verification of historical processing records, and improving the efficiency and completeness of the review of financial attachments after archiving. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of a method for verifying the association between financial scanning attachments and vouchers according to the present invention; Figure 2 This is a schematic diagram illustrating the continuous scanning page acquisition and page order organization of the present invention; Figure 3 This is a schematic diagram illustrating the identification of independent attachments and cross-page attachments of this invention; Figure 4 This is a schematic diagram illustrating the extraction of appendix information and the establishment of candidate voucher entries in this invention; Figure 5 This is a schematic diagram illustrating the two-way amount and order verification of the appendix and voucher entries of this invention; Figure 6 This is a schematic diagram illustrating the anomaly detection and local verification process of the present invention. Detailed Implementation

[0017] 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.

[0018] Example 1: Figure 1 - Figure 6 A method for verifying the association between financial scan attachments and vouchers is presented, including: Obtain the continuously scanned financial attachment pages, page sequence information, and voucher entry data, and organize the pages according to page sequence; Identify independent attachments and cross-page attachments based on attachment boundaries and the continuity of content on adjacent pages, and record the page range corresponding to each attachment; Record the number of attachments and the number of pages respectively, extract the transaction entity, invoice type, amount, date and page order, and establish candidate relationships with the corresponding fields of the voucher entries; First, filter candidate relationships by transaction entity and bill type. Then, verify that the total amount of the attachments associated with the entry is consistent with the entry amount. When an attachment is associated with multiple entries, the total amount of each allocation is consistent with the attachment amount. Continue filtering by page order and entry order. For cases where multiple correspondences or verification inconsistencies still exist, anomaly detection is performed based on image quality, number of attachments, amount gaps, and differences in voucher fields, and the correspondences and verification records are saved.

[0019] This embodiment provides a method for verifying the association between scanned financial attachments and vouchers. It is applicable to enterprise financial shared service centers, financial departments of administrative and public institutions, accounting archive digitization processing sites, and financial voucher processing environments equipped with reimbursement platforms, accounting platforms, or integrated business and finance platforms. The materials to be processed include at least paper financial attachments such as invoices, bank receipts, contracts, acceptance documents, and expense details that are stored with accounting vouchers. The paper financial attachments can be converted into electronic data through sheet-fed scanning devices, flatbed scanning devices, high-speed scanning devices, or other acquisition devices capable of generating page images. The voucher entry data is obtained from the accounting platform, reimbursement platform, or integrated business and finance platform.

[0020] This method can be executed by a server, financial terminal, edge processing device, or other electronic device with image reading, data processing, data storage, and information interaction functions. The processed data includes at least the financial attachment page, page sequence information indicating the order of page collection, and voucher entry data corresponding to the current processing task. The voucher entry data includes at least the fields of entry number, entry amount, posting date, transaction subject, expense item, and number of attachments. When the business system can provide accounting subject, debit / credit direction, currency, posting exchange rate, or reimbursement detail line number, the corresponding fields can also be read simultaneously for subsequent correlation verification.

[0021] When existing information is insufficient to uniquely determine the correspondence between pages, financial attachments, and voucher entries, the original data and corresponding abnormal states are retained. Missing information is not supplemented through speculation, nor are arbitrarily determined results. The page size difference, tilt angle difference, binding position difference, occlusion ratio, layout position difference, number of consecutive abnormal pages, task waiting time, date range, allowable amount difference, and image quality judgment parameters involved in this embodiment can be determined based on actual equipment performance, page and binding characteristics, historical acquisition samples, and the actual operating rules of the business system. Parameters related to scanned images can be determined based on scanning resolution, equipment repetition positioning, and normal and abnormal page samples. Parameters related to paper, binding, and layout can be determined based on size changes of similar pages, binding position changes, and layout position changes during repeated acquisition. Parameters related to amount, date, and number of attachments can be determined based on the accounting accuracy of the business system, attachment registration methods, and current financial processing rules. When using different acquisition devices, page specifications, or business systems, the corresponding parameters can be re-determined according to the corresponding equipment and business environment.

[0022] Specifically, such as Figure 2 As shown: When a batch of paper financial attachments is continuously collected, the collection device records the batch identifier, page identifier, paper feed sequence number or device page number, collection time, and collection resolution of the current collection task while forming the page image. For scanning devices with continuous paper feeding capability, the paper feed sequence number generated when the paper passes through the paper feed channel can be used to determine the page sequence. For flatbed scanning devices or high-speed document scanners that cannot stably provide paper feed sequence numbers, the continuous shooting sequence number or collection time can be used to determine the page sequence. When using the collection time, each page in the same batch is recorded using the same device clock or a time-synchronized task clock, so that the collection time of each page can be directly compared.

[0023] The scanning resolution can be selected as 300 dpi, which means 300 pixels per inch. Since 1 inch is 25.4 millimeters, at 300 dpi, the actual length corresponding to one pixel is approximately 0.085 millimeters. This correspondence is obtained by dividing 25.4 millimeters by 300, and can be used to convert pixel distances such as page size, binding position, and layout position into actual physical length. 300 dpi can meet the needs of capturing text, table lines, and paper edges in general financial documents. When using other resolutions, the conversion is performed according to the same relationship based on the actual resolution recorded by the device.

[0024] After receiving the page, the page sequence information is checked for continuity. When the paper feed sequence number increases continuously, the pages are arranged according to the paper feed sequence number. When there is a gap in the number, the original paper feed sequence number is retained and the missing page position is recorded. For example, when the paper feed sequence numbers 1, 2, and 4 are obtained in sequence, the page gap between sequence number 2 and sequence number 4 is recorded. The device buffer record, page reception record, and acquisition time are combined to check whether there are any pages that have not been received. The original paper feed sequence number is not renumbered.

[0025] When duplicate paper feed sequence numbers appear in the same batch, compare the acquisition time and page image content of the corresponding pages; if the content of two pages with the same number is different and the acquisition time is different, retain the two pages respectively, determine their order based on the actual acquisition time, and record the abnormal paper feed sequence number; if the content of two pages is the same, and it can be confirmed from the device retransmission record or data reception record that one of the pages is a duplicate transmission, record the duplicate page as a duplicate acquisition page, and do not count it in the subsequent attachment quantity.

[0026] When the paper feed sequence number is missing but the acquisition time is complete, the pages can be arranged from earliest to latest according to the acquisition time. The recording accuracy of the acquisition time should be such that adjacent pages that are acquired consecutively can be distinguished. When the chronological relationship reflected by the paper feed sequence number and the acquisition time is inconsistent, the page generation time, device retransmission record, or cache write order can be further checked. If it can be confirmed that the data transmission order has changed, the pages should be arranged according to the actual acquisition order of the device. If it is still uncertain, the relevant pages should be retained and their partial order should be recorded. When verifying the correspondence between the attachments and the voucher entries based on the page order later, the candidate relationship should not be directly excluded using this partial order.

[0027] The voucher entry data corresponding to the current data collection task is obtained from the accounting platform, reimbursement platform, or business-finance integration platform. The voucher entry data includes at least the following fields: entry number, entry amount, posting date, transaction subject, expense item, and number of attachments. When the business system can provide currency, posting exchange rate, accounting subject, debit / credit direction, or reimbursement details, the corresponding fields can be read simultaneously, and the original values ​​of each field can be retained.

[0028] Scanned pages and voucher entry data can arrive sequentially. If a complete batch of scanned pages has been formed but the voucher entry data has not yet been obtained, the sequence of the formed pages and any abnormal records are saved, and the voucher entry data is awaited. If the voucher entry data arrives first but the page has not yet been fully collected, the original voucher entry data is saved, and the page is awaited to be fully collected. During the waiting period, the data that has not yet been obtained is not replaced with the default amount, default transaction entity, or temporary entry.

[0029] The task waiting time can be determined based on the page collection completion time and the voucher entry data arrival time in the business system's operation record. For example, when the same financial archiving task can usually complete the preparation of two types of data within 15 minutes under normal operating conditions, the waiting period can be set to 15 minutes, and the data arrival status can be checked every 5 minutes, for a total of 3 checks within a waiting period. The 15 minutes is set according to the normal task completion time, the 5 minutes is set according to the data update cycle and query load of the business system, and the 3 checks correspond to the query interval within a 15-minute waiting period. When the actual data preparation time of the business system is different, the waiting period and check interval can be adjusted according to its historical operation record. If the data on the other side is not obtained after reaching the waiting limit, the current task is paused and processing continues after the data is complete.

[0030] After obtaining both page data and voucher entry data, a unified storage format is used for data such as amount, date, and currency from different data sources, while retaining the original values. For foreign currency amounts, the original currency amount, currency type, and the actual accounting exchange rate used in the voucher are saved, and the original amount is not directly changed in the current processing stage. This forms a sequence of financial attachment pages with a clear page sequence and abnormal status, and obtains the voucher entry data corresponding to this batch, providing a data foundation for subsequently determining the range of consecutive pages, identifying financial attachments, and establishing the correspondence between attachments and voucher entries.

[0031] Specifically, such as Figure 3 As shown: For pages that have been arranged according to the order of acquisition, first determine the boundary between the paper carrier and the scanning background, and obtain information such as the horizontal dimension, vertical dimension, tilt angle, and identifiable binding hole position and paper feed marks of the page; the pixel distance in the page is converted into physical length according to the actual resolution recorded by the acquisition device, so that the page size and position data obtained at different resolutions have a uniform comparison scale.

[0032] Whether adjacent pages belong to consecutive pages corresponding to the same document can be determined based on changes in page size, tilt angle, and relative position of binding holes. Relevant parameters can be obtained by repeatedly collecting consecutive pages that have been manually confirmed to belong to the same document using the same scanning device, and then verifying them in conjunction with adjacent pages that have been manually confirmed to belong to different documents. Specifically, the width difference, height difference, tilt angle difference, and positional change of the same binding hole relative to a fixed corner point of the page between normal pages can be recorded, and the stable range of changes formed under normal collection conditions can be used as a reference for judgment.

[0033] For example, the size difference between adjacent pages can be selected as 1.5 mm, the tilt angle difference as 0.6 degrees, and the relative position difference of the binding holes as 2 mm. When at least two of the above three items exceed the corresponding reference values, the adjacent positions can be recorded as page boundary positions that need further confirmation. Among them, 1.5 mm is determined based on the size variation of the same specification paper and the actual fluctuation of the page edge position during normal scanning; 0.6 degrees is determined based on the normal paper feeding deviation of the paper feeding equipment and the tilt angle change after repeated scanning of the same page; 2 mm is determined based on the normal positioning deviation of the punching or binding equipment and the actual change of the page hole position under the same binding conditions. When using other paper specifications, scanning equipment, or binding equipment, the corresponding reference values ​​can be obtained again in the same way.

[0034] For loose-page documents without binding holes, the binding hole position can be disregarded for determination. Instead, the page range can be determined by combining page size, page tilt, paper feed marks, and the loading continuity within the same data acquisition task. When the same document uses different paper sizes, page size changes are not used as a separate page boundary condition but are used in conjunction with other continuous information.

[0035] After determining the range of consecutive pages corresponding to the same voucher, the actual financial attachments are identified within the corresponding pages. For attachments such as invoices, tickets, and receipts pasted on A4 backing paper or other carrier paper, the attachment area can be determined based on the edges formed between the attachment paper and the carrier paper, the changes in brightness, and the local shadows caused by the paper thickness. When an area has attachment edges that can be connected to each other, and the document title, fixed format, or main field position can be identified within the area, the area can be recorded as an independent financial attachment.

[0036] When there are multiple separate attachment areas on the same page, each attachment is recorded separately; for example, if there are 3 non-overlapping transportation tickets on the same scanned page, 3 financial attachments are recorded, while the scanned page is still recorded as 1 page, thus preserving the actual correspondence between the number of attachments and the number of pages.

[0037] When the edge of an attachment is interrupted due to a stamp, crease, or partial obstruction, it can be determined whether it belongs to the same attachment by combining the extension direction of the two sides of the interruption edge, the table separator line, the position of the fixed title, the position of the QR code, the position of the stamp, and the positional relationship of the main text area. If the existing information can support the continuity of the interruption edge and the supplemented attachment area does not conflict with other attachment areas, it can be treated as the same attachment. If it cannot be reliably determined, the current area and boundary cannot be confirmed, and the attachment area is not supplemented based on experience.

[0038] When attachments are overlapped, the visible range of the attachment can be determined based on the visible edges and the overlapping positions of the upper and lower layers of paper. If the complete range of the attachment can be determined based on the visible edges, the actual size of similar documents, or the edge extension positions on both sides of the obscured area, the proportion of the obscured area can be further determined. If the complete range cannot be determined, the attachment identification conclusion should not be made directly based on the obscuration ratio, but the corresponding area should be reserved for subsequent verification.

[0039] For example, the occlusion ratio for initiating manual review can be selected as 30%. This value can be determined by overlaying samples of similar bills, that is, collecting bills with different degrees of occlusion, checking whether the attachment boundaries and key information such as amount, transaction entity, and bill type can still be stably obtained, and determining the reference ratio based on the degree of occlusion corresponding to when the above information begins to be obviously missing. 30% corresponds to about one-third of the attachment area and can be used as an initial value. The distribution of key information on specific bills is different, so even if the occlusion ratio is less than 30%, when the occluded area completely covers the areas that must be read, such as amount and transaction entity, the corresponding fields should be recorded as unreliable.

[0040] For attachments such as contracts, expense details, bank reconciliation documents, and acceptance documents that may occupy multiple pages, the titles, page numbers, table boundaries, header and footer positions, fixed printing areas, transaction entities, and page specifications of adjacent pages can be compared according to the actual collection order of the pages. When the titles are consistent and the page numbers are consecutive, the content continuity between pages can be determined accordingly. When the page numbers cannot be reliably read, the fixed table boundaries, column widths, header positions, and fixed printing area sizes can be further compared.

[0041] For multi-page documents with a fixed format, normal pages of the same format can be repeatedly sampled in advance to record the actual changes in the position of the fixed dividing line and the size of the fixed printed area. For example, the position difference of the fixed dividing line can be selected as 0.3 mm, and the deviation of the height ratio of the fixed printed text can be selected as 3%, as an auxiliary condition for judging the continuity of the format. Among them, 0.3 mm is determined based on the normal position change of the fixed line when the same page is repeatedly sampled, and is used to accommodate the repeated positioning deviation of the equipment and slight paper movement; 3% is determined based on the normal size change of the fixed printed area when the same format document is repeatedly sampled, and is used to accommodate slight paper expansion and contraction and changes in the sampling ratio. When using new sampling equipment or other document formats, the above values ​​can be re-determined using corresponding normal samples.

[0042] When the title, page number, or other content information or fixed layout information of adjacent pages can confirm a continuous relationship, and the subsequent page does not show a clear starting characteristic of another independent attachment, the subsequent page can continue to be classified as the previous attachment; for double-sided bank receipts or other double-sided documents, the fixed format of the front and back sides, the document number, the transaction entity, and the continuous collection relationship can be combined to determine whether the two pages belong to the same attachment; when it is confirmed that they belong to the same attachment, record that 1 attachment corresponds to 2 pages.

[0043] When a new document title, a different transaction entity, a significantly different layout structure, or a new independent attachment starting area appears on the next page, the page continuity of the previous attachment is terminated, and a new attachment record is created from the current page.

[0044] When a page cannot obtain continuous information about the attachment boundaries or layout due to strong reflections, severe creases, incomplete cropping, or abnormal scanning background, the image quality of the corresponding page is recorded as abnormal. The number of consecutive abnormal pages can be determined using historical samples whose attachment start and end ranges have been manually confirmed. Specifically, when 1, 2, 3, or more consecutive pages are missing boundary or layout information, it can be checked whether the attachment range can be recovered based on the normal pages before and after the abnormal page. The automatic processing range is determined based on the maximum number of consecutive missing pages that can still stably restore the attachment relationship.

[0045] For example, the number of consecutive abnormal pages can be selected as 3 pages; when 3 consecutive pages cannot provide enough information to determine the attachment boundary or cross-page relationship, and the normal pages before and after the abnormal area are also insufficient to recover the attachment range, the relevant pages are re-collected; 3 pages can be determined based on the maximum stable number of pages in historical samples that can still recover the attachment start and end relationship after the loss of consecutive page information; if the actual sample shows that stable recovery is not possible after 2 consecutive missing pages, or if other available information can support the judgment of a continuous relationship of more than 3 pages, the number of consecutive abnormal pages is adjusted according to the corresponding sample.

[0046] After the above processing is completed, each financial attachment shall record at least the attachment identifier, corresponding page range, starting page, ending page, actual pages involved and their position within the page; one page may correspond to multiple financial attachments, and one financial attachment may correspond to multiple pages. The resulting attachment and page correspondence records are used for subsequent attachment quantity statistics, page quantity statistics, and extraction of information such as transaction entity, bill type, amount and date.

[0047] Specifically, such as Figure 4 As shown: The number of attachments and the number of pages are recorded separately, and the correspondence between each attachment and its page is saved; for example, when three independent invoices are pasted on the same page, the number of attachments is recorded as 3 and the number of pages as 1; when a three-page contract is a complete attachment, the number of attachments is recorded as 1 and the number of pages as 3; when a double-sided receipt consists of two pages, the number of attachments is recorded as 1 and the number of pages as 2; by recording the two types of quantities separately, the correspondence between the actual number of attachments and the number of scanned pages can be maintained in mixed, multi-page, and double-sided documents.

[0048] For each attachment, at least the attachment identifier, corresponding page, order of first appearance on the page, and number of pages occupied should be recorded. Fields such as transaction entity, invoice type, amount, and date should be obtained from the attachment for voucher association verification. The transaction entity can be obtained from fields such as seller, buyer, payee, payer, bank counterparty, or contract counterparty, depending on the attachment type. The invoice type can be determined based on the invoice title, fixed format, QR code location, table structure, and readable text content. The amount should preferably use fields that can represent the actual amount of the current attachment, such as total price including tax, payment amount, receipt amount, or contract amount. The date should be a date with clear business meaning, such as issuance date, transaction date, or signing date, depending on the attachment type.

[0049] The data obtained from the attachments is stored separately from the original data in the voucher entries; if the amount, transaction subject, or date in the attachments cannot be reliably obtained, the corresponding field in the record cannot be read, and the corresponding field in the voucher entries is not used to fill it in; if a field in the voucher entry is missing, the content of the attachments is not used to directly rewrite the original data of the voucher; thus, the attachment data and the voucher data maintain their respective original sources, providing an independent data basis for subsequent comparisons.

[0050] When text in an attachment cannot be fully read due to reasons such as stamps, creases, reflections, or cutting, the corresponding field area can be determined according to the fixed format. For example, the amount reading area of ​​an invoice can be determined based on the position of the fixed title and amount area, and the corresponding field of a bank statement can be obtained based on the position of the account information area, transaction amount area, and transaction date area. Fields that cannot be reliably obtained are left as unreadable and are not inferred from the content of adjacent attachments, historical values, or voucher data.

[0051] Before comparison, the names of the transaction entities can be standardized in format while retaining the original names. Standardization of format includes at least removing extra spaces, unifying full-width and half-width characters, and standardizing the writing style of common organizational forms according to the existing enterprise name processing rules of the business system. This process is only used to reduce the impact of differences in writing format on name comparison and will not identify different transaction entities as the same entity due to similarity in some parts of the name.

[0052] Based on the expense items in the voucher entries, the corresponding rules for attachment types configured in the accounting or reimbursement platform can be read to determine the acceptable attachment types for different expense items. For example, travel expenses can correspond to railway tickets, air transport tickets, taxi receipts, or accommodation invoices. This correspondence is derived from historical vouchers that have been configured in the business system or confirmed manually. It is used to exclude attachments that are obviously unrelated to the current expense items and is not used as a condition to determine the final correspondence between attachments and voucher entries.

[0053] When establishing candidate correspondences, each attachment is processed as an object, and entries with the same transaction subject are retained in the current batch of voucher entries. If the transaction subject of the attachment cannot be reliably obtained, the subject field is not used to exclude entries. Subsequently, entries that are obviously inconsistent are excluded according to the correspondence between the type of invoice and the expense item, forming a range of voucher entries that the attachment may correspond to. At this time, it is not required that the amount of the attachment be equal to the amount of a single voucher entry, because multiple attachments can correspond to one voucher entry, and the amount of one attachment can also correspond to multiple voucher entries according to the actual accounting content.

[0054] Dates can be used as supplementary filtering information; the dates of the attached business transactions should fall within the actual allowed accounting or reimbursement time range of the business system; for example, if the current financial system of an organization stipulates that the regular reimbursement period is 90 days, and it can be extended to 180 days under special circumstances such as annual carry-over, 90 days and 180 days can be used as the corresponding time ranges respectively; if the actual implementation period is 60 days or other periods, then the time range stipulated by the actual system shall be used; therefore, 90 days, 180 days or 60 days are all derived from the accounting or reimbursement rules actually implemented by the corresponding organization, and the date is only used to exclude candidate relationships that obviously do not meet the time requirements, and is not used as the sole determining condition.

[0055] While forming candidate correspondences, the actual number of identified attachments is compared with the number of attachments recorded on the voucher. If the numbers are inconsistent, the system first checks for multi-page contracts, double-sided receipts, and multi-copy documents according to the attachment registration criteria actually used in the business system. For example, if a 3-page contract is registered as 1 attachment in the business system, the number of pages is 3 while the number of voucher attachments is 1, which is a normal counting difference and is not treated as a missing attachment.

[0056] If the quantity discrepancy cannot be explained according to the actual registration criteria, record the difference between the actual number of attachments and the number of document attachments; if the actual number of attachments is less than the number of document attachments, retain possible reasons such as missing attachments or incomplete identification of overlapping attachments; if the actual number of attachments is greater than the number of document attachments, retain the extra attachments and include the possible error in recording the number of document attachments as subsequent verification content, instead of directly deleting the already identified attachments.

[0057] After the above processing is completed, each attachment will form a record that includes at least the corresponding page range, page order, transaction entity, type of bill, amount, date, and candidate voucher entry. At the same time, the number of attachments, number of pages, original voucher data, and quantity differences that cannot be explained by existing counting rules will be saved for subsequent determination of the correspondence between attachments and voucher entries according to amount and page order.

[0058] Specifically, such as Figure 5 As shown: The existing correspondence between attachments and voucher entries is filtered in sequence according to the transaction entity, invoice type, amount, and page order; the information used for comparison on the attachment side and the voucher entry side includes at least the fields of transaction entity, invoice type, expense item, amount, page order, and entry order.

[0059] When comparing transaction entities, if the transaction entity in the attachment is consistent with the transaction entity in the voucher entry, or if it can be confirmed that they belong to the same entity after the aforementioned name format standardization, the corresponding candidate relationship is retained; if the transaction entity in the attachment cannot be reliably obtained, the candidate relationship is not excluded using this field; if it can be confirmed that the transaction entity in the attachment is inconsistent with the transaction entity in the voucher entry, the corresponding candidate relationship is excluded; when there are situations such as payment on behalf, unified group settlement, or centralized payment, the judgment is made based on the actual transaction entity, payment entity, or settlement relationship already recorded in the business system, and the corresponding entity relationship is not temporarily changed based on manual experience.

[0060] Then, the invoice type and expense item are compared; if the attached invoice type belongs to the attachment type allowed for the corresponding expense item in the business system, the candidate relationship is retained; if it can be confirmed that the invoice type and expense item do not match, the corresponding candidate relationship is excluded; if the invoice type cannot be reliably determined, the candidate relationship is retained and the amount and order are verified.

[0061] For the candidate relationships retained after filtering, each voucher entry is used as the object to check whether the total amount of different attachment combinations is consistent with the entry amount. The specific amounts below are only used to illustrate the amount verification method and do not represent actual business data. For example, if a voucher entry amount is 680 yuan, and the amounts of the attachments with candidate relationships are 180 yuan, 200 yuan, 300 yuan, and 500 yuan, the total amount of the three attachments of 180 yuan, 200 yuan, and 300 yuan is 680 yuan, then the combination meets the entry amount condition. If the total amount of 500 yuan and 180 yuan is also 680 yuan, then there are two candidate combinations that can meet the amount condition. The final correspondence cannot be determined solely based on the amount; the page order and entry order must also be considered for further judgment.

[0062] The allowable difference in amount is determined based on the actual amount recording precision used by the business system. When the RMB business system uses fen as the smallest amount recording unit and retains two decimal places in yuan, the allowable difference in amount can be selected as 0.01 yuan. 0.01 yuan corresponds to 1 fen, which is consistent with the smallest amount change that the above system can record, so it can be used to handle differences caused by the smallest amount unit. When the business system requires the amounts on both sides to be completely consistent, the allowable difference in amount can be 0. When using other currencies or other amount recording precision, it is determined according to the smallest amount unit that the corresponding business system can record. For foreign currency attachments, the amount is compared after converting them to the same accounting standard according to the currency and accounting exchange rate actually recorded in the voucher.

[0063] In addition to verifying the total amount of attachments based on the journal entries, the method also checks how the amount of each attachment is used in different journal entries. When an attachment corresponds to only one journal entry, the amount of the corresponding journal entry should be consistent with the amount of the attachment. When an attachment corresponds to multiple journal entries according to the actual accounting content, the amount corresponding to each journal entry should not be reused, and the total amount should be consistent with the amount of the attachment.

[0064] For example, in a purchase invoice with a total amount of 1200 yuan, 700 yuan corresponds to the office expense entry and 500 yuan corresponds to the maintenance expense entry. The total amount of 700 yuan and 500 yuan should be 1200 yuan. This can be used to verify whether the amount of the attachment corresponding to each voucher entry is complete, and also to check whether the same attachment amount is used repeatedly or whether there is a remaining amount that is not matched with the voucher entry.

[0065] When existing candidate attachments cannot form a combination consistent with the amount of the voucher entry, save the amount difference of the corresponding entry, as well as the attachments and entry positions involved; when some attachment amounts are still unused after all voucher entries have been verified, save the corresponding attachments and remaining amounts; in the above cases, the candidate relationship with the closest amount will not replace the relationship that does not actually meet the amount condition.

[0066] If the amount condition still cannot make the candidate relationship unique, further compare the order of the attachment pages and the order of the voucher entries; the order of the attachment pages is the position of the page where the attachment first appears in the current scan batch, and the order of the voucher entries is the original entry number of the voucher; if the business system can provide the reimbursement details line number, the reimbursement details line number can also be used to represent the order of the voucher data.

[0067] The order verification adopts a relative sequence relationship and does not require that the page number of the attachment and the entry number of the voucher be exactly the same in value. For example, if attachment A appears before attachment B and entry 1 appears before entry 2, the reverse correspondence between attachment A and entry 2 and attachment B and entry 1 is not retained unless there are other business records indicating a change in order. For local areas where the page order is recorded as uncertain, candidate relationships are not directly excluded using the local page order, but the page positions before and after it that can be confirmed and other available fields are used to continue the judgment.

[0068] The order of use of transaction entities, invoice types, amounts, and sequences can be adjusted based on the actual distinctions made in the current data. For example, when there are many similar invoices for the same transaction entity, the candidate range can be narrowed down first by using the amount relationship. When there are clear reimbursement detail lines in the business system, the relative order can be determined first by using the detail lines. Regardless of the order used, the final correspondence is confirmed by two or more independent types of information, not just by a single entity name, a single invoice type, or a single amount.

[0069] When only one set of non-conflicting candidate relationships exists after filtering, a corresponding record is formed between the attachment and the voucher entry; when no candidate relationship meets the existing conditions, the unmet amount, field, or order position is recorded; when there are two or more candidate relationships that simultaneously meet the existing conditions, each candidate relationship is retained, and the differences in attachments, pages, and voucher entries between different candidate relationships are recorded; the above results are used to subsequently determine image quality, missing attachments, or abnormal voucher fields, and the correspondence that has been confirmed is saved.

[0070] Specifically, such as Figure 6 As shown: For the correspondence between attachments and voucher entries that cannot be uniquely determined, and for situations where there are inconsistencies in the number of attachments, amounts, or voucher fields, the cause of the anomaly can be determined based on the page image quality, differences in the number of attachments, differences in amounts, and differences in voucher fields. The confirmed correspondence between attachments and voucher entries and the verification information can then be saved.

[0071] For page image quality abnormalities, check whether the page has strong reflections, obvious shadows, severe creases, abnormal tilt, incomplete cropping, insufficient local contrast, or indeterminate attachment boundaries, and determine whether the abnormal area covers the information area that needs to be read, such as amount, transaction entity, date, and invoice type.

[0072] For strong reflections, the normal brightness range can be determined by using the unreflected areas of the paper in the same batch. When a continuous bright area appears on the page, and its brightness is significantly higher than the surrounding normal paper and close to the upper limit of the brightness that the image can record, the area is determined as a highlight area, and the area ratio of the highlight area in the current information area is further determined.

[0073] For example, the initial check value for the highlight area ratio can be 5%. This value can be determined by sampling normal pages and pages with reflections. That is, by statistically analyzing whether information such as amount, transaction subject, and date can be stably read under different highlight area ratios, and using the highlight area ratio corresponding to the point where the readability of the field begins to decline significantly as the initial check value. 5% is used to determine whether further page quality checks are needed, but does not determine whether the page should be re-collected. When the highlight area ratio is less than 5%, but the highlight area completely covers the information that must be read, such as the amount, it is still recorded that the corresponding field cannot be reliably obtained. When the highlight area ratio exceeds 5%, but the highlight is entirely located in an irrelevant blank area, it is not considered that the entire page is unusable.

[0074] When an attachment page with a discrepancy in amount simultaneously exhibits glare, shadows, or cropping anomalies, and the abnormal area covers the currently unreadable amount area, while the number of attachments, document type, and other available information are not missing, the page can be listed as a target for re-collection. During re-collection, the lighting direction can be adjusted, the paper can be flattened, or the effective collection range can be expanded to re-obtain the corresponding page.

[0075] After the data is re-collected, only the attachments and voucher entries affected by the new page are reprocessed. Specifically, the boundaries and corresponding fields of the attachments on the page can be redefined, and the voucher entries that have a candidate correspondence with the attachments can be re-verified. Other records that do not have an attachment correspondence with the re-collected page and do not involve the same candidate voucher entries remain unchanged, thus avoiding duplicate processing of the entire batch.

[0076] For situations where attachments may be missing, a judgment can be made by combining the actual number of attachments identified, the number of attachments recorded on the voucher, and the amount that is missing. For example, if a voucher records 5 attachments, and 4 attachments can be identified on the scanned page, and there are no obvious image quality issues on the pages of the 4 attachments, but there is still a difference of 300 yuan in the relevant voucher entries, and a type of attachment corresponding to the current expense item is not present in the scanned data, one attachment may be missing, along with the corresponding amount difference, for verification when supplementing attachments later. The above 5 items, 4 items, and 300 yuan are used to illustrate the relationship between the joint judgment of quantity difference and amount difference. In actual processing, the actual quantity and amount obtained from the current voucher and attachments are used.

[0077] For situations where voucher fields may contain errors, information that can be consistently obtained from multiple attachments can be compared with the original fields on the voucher. For example, if multiple attachments all yield the same transaction entity, but the corresponding voucher entry only records the transaction entity field as another entity with a similar name, while the amount, date, and document type all match, then the difference in the transaction entity field can be recorded. Similarly, if the amount in the attachments consistently yields 1230 yuan, while the voucher entry records 1320 yuan, and other information matches, then the difference in the amount field can be recorded. The above amounts are used to illustrate the method for judging field differences; in actual applications, the actual amounts obtained from the attachments and vouchers are used. During processing, the original fields on the voucher are not directly modified, but at least the field names, original voucher values, values ​​obtained from attachments, relevant pages, and corresponding judgment information are saved for review by personnel with appropriate permissions.

[0078] When two or more of the following categories of anomalies—image quality abnormalities, missing attachments, and voucher field errors—have corresponding judgment information, the number of judgment conditions that have been met for each type of anomaly, the total number of judgment conditions that can be used, and the ratio between the two can be calculated. Image quality abnormalities can be judged based on conditions such as whether key fields are covered by reflections, shadows, cropping, or creases, and whether the boundaries of attachments can be determined. Missing attachments can be judged based on conditions such as whether the number of attachments is insufficient, whether there are unexplained differences in amount, and whether the types of attachments that should appear are missing. Voucher field errors can be judged based on conditions such as whether multiple attachments produce consistent results for the same field, and whether only the corresponding field in the voucher is inconsistent with the attachment information.

[0079] For example, a 15 percentage point difference between the highest and second-highest percentages can be used as the initial reference value for entering a single anomaly handling method; when the percentage of one type of anomaly meeting the conditions is 80% and the percentage of another type of anomaly is 60%, the difference between the two is 20 percentage points, and the anomaly corresponding to the former can be handled first; when the percentages corresponding to the two types of anomalies are 75% and 65% respectively, the difference between the two is 10 percentage points, then both anomaly possibilities are retained and the case is transferred to manual review.

[0080] The 15 percentage point can be determined based on historical samples where the cause of the anomaly has been manually confirmed. Specifically, samples confirmed to be image quality anomalies, missing attachments, and incorrect voucher fields can be selected, and the proportion of each sample meeting the corresponding judgment conditions can be calculated. Then, the proportion difference between the correct anomaly type and other anomaly types can be compared. When determining the reference value, the rework impact caused by incorrect re-collection, incorrect request for supplementary attachments, or incorrect modification of vouchers can also be considered, and the proportion difference that can relatively stably distinguish historically confirmed anomalies can be selected as the initial reference value. When not enough historical anomaly samples have been accumulated, the cause of the anomaly is not automatically determined based on the 15 percentage point. If two or more anomaly causes cannot be ruled out, the case is transferred to manual review.

[0081] When no anomalies are found in the number of attachments, amount, and voucher fields, but there are still two or more attachments and voucher entries that meet the existing conditions, the correspondence is recorded as not being uniquely determined, and the differences in attachments, pages, and voucher entries between different correspondences are saved. After the specific correspondence is manually confirmed, only the attachment amount and voucher entry amount affected by the confirmation result are re-verified, and data unrelated to the confirmation result is not reprocessed.

[0082] The corresponding records, whether automatically processed or manually verified, should include at least the following fields: batch identifier, attachment identifier, corresponding page range of attachments, voucher and entry location, number of attachments, number of pages, transaction entity, type of invoice, amount, and date. They should also save the currently used parameter version, fields that cannot be reliably read, amount discrepancies, page order anomalies, and anomaly handling information. For fields that cannot be reliably obtained, their status should be saved, and speculative content should not be used to replace the original data.

[0083] If the accounting or archiving platform allows the writing of related information, the correspondence between attachments and voucher entries, the attachment page range, and the verification time can be written into the related fields of the business system; if the original scanned image cannot be modified, the corresponding information can be saved in an independent database or related record file.

[0084] When the archiving system allows for the setting of codes for locating associated records in blank areas of scanned images, the code can be placed in a location that does not cover the main text, signatures, or other original information, and is not located within the subsequent cropping range. For example, if there is a continuous blank area at the bottom of the page, and the scanning and subsequent document cropping will not cover this area, the code location can be a blank area 5 mm to 15 mm from the bottom edge of the page. This range can be determined based on the blank width at the bottom of the actual voucher page, the edge removal range of the scanning device, and the document cropping location. If there is main text, signatures, or other original information within the 5 mm to 15 mm range, then no code should be set in this area. When the archiving system does not allow modification of the original image, the corresponding information should be stored in an independent database or associated record file.

[0085] When reviewing archived records, the corresponding pages, attachments, and key financial fields can be retrieved and compared with the original saved content. If the number of pages decreases, missing pages can be further identified. If the number of pages remains unchanged but the number of attachments changes, the attachment area can be checked for replacement, re-cutting, or changes in the identification range. If the number of pages and attachments remains unchanged, but the amount, transaction entity, or type of invoice changes, the corresponding attachment can be located based on the corresponding records of the attachments and pages.

[0086] When an abnormal interruption occurs during the result saving process, the current attachment that has not been fully saved is used as the scope of recovery; the previous attachments and corresponding voucher entries that have been fully saved are kept in their original records, and the temporary content written in the current attachment that has not yet been completed can be undone. After recovery, it will be reprocessed from the current attachment to avoid the same attachment forming a duplicate correspondence; when the scanning equipment malfunctions, only the page that needs to be re-captured and the attachments and voucher entries related to that page are paused, and other records that have been confirmed and are not related to the malfunctioning page remain unchanged.

[0087] Through the above processing, when attachments and voucher entries cannot uniquely correspond or when there are inconsistencies in quantity, amount, and fields, it is possible to distinguish between abnormal page images, possible missing attachments, and possible errors in voucher fields. Re-collection, attachment supplementation, or field review can be limited to the actual affected pages, attachments, and voucher entries. This can reduce the spread of local anomalies to the correspondence of other attachments in the same batch and improve the stability of batch financial attachment association verification and the reliability of subsequent review.

[0088] Example 2: Based on Example 1, the specific application process of a method for verifying the association between financial scanned attachments and vouchers is further explained: This embodiment uses the actual processing of a batch of financial archived materials as an example to further illustrate the method for verifying the association between scanned financial attachments and vouchers.

[0089] The documents to be archived should include at least transportation tickets, procurement documents, bank receipts, and invoices, as well as other financial attachments. Paper documents should be continuously scanned in the actual binding order. The scanning resolution can be selected as 300 dpi. When capturing each page, information such as batch identifier, page identifier, paper feed number, and capture time should be recorded simultaneously. 300 dpi means that 300 pixels are captured per inch. Since 1 inch is 25.4 mm, the actual length corresponding to one pixel is approximately 0.085 mm. Based on this, the pixel distances such as page size, binding hole position, and fixed layout position can be converted into the actual physical length.

[0090] This batch yielded 7 scanned pages, numbered 1 to 7 sequentially. The acquisition time of each page corresponds to the order reflected by the paper feed number, so the pages are arranged according to the paper feed number. At the same time, the corresponding voucher entry data for this batch was obtained from the accounting platform. The voucher entry data includes at least the fields of entry number, entry amount, posting date, transaction entity, expense item, and number of attachments, and the corresponding original values ​​are retained.

[0091] Scanned page data and voucher entry data are allowed to arrive sequentially. For example, based on the business system's operation records, if the same archiving task can typically complete page acquisition and voucher data preparation within 15 minutes, the waiting period can be set to 15 minutes, with the data arrival status checked every 5 minutes, for a total of 3 checks within one waiting period. The 15 minutes is determined based on the normal task completion time, the 5 minutes is determined based on the business system's data update cycle and query load, and the 3 checks correspond to the check interval within a 15-minute waiting period. In this embodiment, both types of data are acquired within the normal waiting period.

[0092] The actual contents of the 7 scanned pages are as follows: Page 1 is an A4 backing paper with 3 independent transportation tickets pasted on it; Pages 2 to 4 are the same 3 consecutive pages of procurement documents; Pages 5 and 6 are the front and back of the same double-sided bank receipt; Page 7 is a separate invoice.

[0093] When processing the pages, first determine the boundary between the paper carrier and the scanned background, and obtain information such as the page width, height, tilt angle, and the position of the identifiable binding holes. This batch uses A4 carrier paper with uniform punching and binding, so the page size, tilt angle, and relative position of the binding holes can be combined to determine whether consecutive pages still belong to the page range corresponding to the same voucher.

[0094] For example, the size difference between adjacent pages can be selected as 1.5 mm, the tilt angle difference can be selected as 0.6 degrees, and the relative position difference of the binding holes can be selected as 2 mm. When at least two of the above three items exceed the corresponding values, the adjacent positions are recorded as page boundary positions that need further confirmation. Among them, 1.5 mm is determined based on the size change after normal scanning of the same size paper; 0.6 degrees is determined based on the normal paper feed skew when scanning the same page repeatedly; and 2 mm is determined based on the actual change of the page hole position under the same binding conditions. In this batch, pages 1 to 7 did not have more than two of the corresponding values ​​exceeding the corresponding values ​​at the same time, so the actual financial attachments will continue to be identified within the current continuous page range.

[0095] For page 1, based on the paper edges, light and dark variations formed between the transportation ticket and the A4 backing paper, as well as the ticket layout, three separate attachment areas can be identified, and are therefore recorded as the first attachment, the second attachment, and the third attachment, respectively; the number of pages for page 1 is 1, while the actual number of attachments is 3.

[0096] The first, second, and third attachments all contain information such as the transaction entity, type of negotiable instrument, amount, and date. The amount in the first attachment is 180 yuan, the amount in the second attachment is 200 yuan, and the amount in the third attachment is 300 yuan. Each field is obtained from the corresponding negotiable instrument and retains its original content.

[0097] For pages 2 to 4, we determine whether the three pages belong to the same attachment based on the document title, page number, and fixed layout. The three pages have the same title, consecutive page numbers, and consistent fixed table structure. When we further compare the fixed separators and fixed printed areas on each page, the positional difference of the fixed separators does not exceed 0.3 mm, and the deviation of the height ratio of the fixed printed text does not exceed 3%. Therefore, pages 2 to 4 are recorded as the same financial attachment, namely the fourth attachment.

[0098] Among them, 0.3 mm is determined based on the normal changes in the fixed line position when the same version of the document is repeatedly collected, and is used to accommodate the positional deviation caused by repeated positioning of the scanning device and slight movement of the paper; 3% is determined based on the normal size changes of the fixed printing area when the same version of the document is repeatedly collected, and is used to accommodate slight expansion and contraction of the paper and changes in the normal collection ratio; the fourth attachment has 1 attachment, corresponding to 3 pages, and can obtain information such as the transaction subject, document type, date and amount; in this embodiment, the amount of the fourth attachment is 1200 yuan.

[0099] Pages 5 and 6 have corresponding fixed formats for the front and back of bank receipts, and the transaction entities and collection order can be matched. Therefore, the two pages are recorded as the fifth attachment; the fifth attachment has 1 attachment and 2 corresponding pages.

[0100] Page 7 has complete invoice boundaries and a fixed format, so it is recorded as the sixth attachment. The transaction entity, invoice type and date in page 7 can be obtained normally, but the amount area has obvious reflection when it is collected for the first time, so the amount field cannot be read reliably. The state that this field cannot be read is retained, and the amount in the voucher entry is not used to fill it in.

[0101] Therefore, the 7 scanned pages in this batch actually formed 6 financial attachments, with page 1 corresponding to 3 attachments, pages 2 to 4 jointly corresponding to the fourth attachment, pages 5 and 6 jointly corresponding to the fifth attachment, and page 7 corresponding to the sixth attachment; the number of attachments and the number of pages are saved separately, and the actual page range corresponding to each attachment is recorded.

[0102] For each attachment, at least the following fields should be recorded: attachment identifier, corresponding page, order of first appearance on page, transaction entity, type of bill, amount, and date. The page order of the attachments is determined by their first appearance position. Therefore, the first to third attachments first appear on page 1, the fourth attachment first appears on page 2, the fifth attachment first appears on page 5, and the sixth attachment first appears on page 7. When there are multiple attachments on the same page, their relative order can also be preserved according to their actual arrangement position on the page.

[0103] The data obtained from the attachments is saved separately from the original data in the voucher entries; for example, the amounts of 180 yuan, 200 yuan and 300 yuan in the first to third attachments are all obtained from the corresponding vouchers themselves; when the amount in the sixth attachment cannot be read due to reflection, only the actual state of the field is recorded, and the attachment data is not supplemented by using the voucher amount.

[0104] Before comparing the names of the trading entities, extra spaces can be removed, full-width and half-width characters can be standardized, and common writing differences can be standardized according to the enterprise name processing rules already adopted by the business system. The above processing is only used to reduce the impact of format differences and will not identify different trading entities as the same entity because some characters are the same.

[0105] Based on the expense items in the voucher entries, read the attachment type rules already configured in the business system; for example, travel expenses can correspond to transportation tickets and accommodation invoices, and procurement expenses can correspond to procurement invoices, contracts, or settlement documents; first, exclude obviously inconsistent voucher entries based on the transaction entity and the type of invoice to form the range of voucher entries that may correspond to each attachment.

[0106] In this embodiment, the first attachment, the second attachment, and the third attachment are all travel-related documents, forming a candidate relationship with a travel expense entry of 680 yuan; the total amount of the three attachments is 180 yuan plus 200 yuan plus 300 yuan, which is 680 yuan, consistent with the amount of the voucher entry. Therefore, the attachment combination meets the entry amount condition.

[0107] When the RMB business system uses fen as the minimum amount recording unit and retains two decimal places in yuan, the allowable difference in amount can be selected as 0.01 yuan; 0.01 yuan corresponds to 1 fen, which is consistent with the minimum amount change that the business system can record; in this embodiment, the total amount of the 3 attachments is exactly equal to the 680 yuan entry amount, and the amount difference is 0.

[0108] If multiple candidate combinations have the same total amount, then other information should be used to differentiate them. For example, if there is another attachment of 500 yuan in the candidate range, the total amount of 500 yuan and 180 yuan is also 680 yuan. In this case, the corresponding relationship cannot be uniquely determined based on the amount alone. The selection should be further combined with the type of invoice, page order, and journal entry order. If it is still impossible to determine, the corresponding candidate relationship should be retained, and one of the groups should not be selected arbitrarily.

[0109] The fourth attachment is for 1200 yuan, corresponding to a 700 yuan office expense entry and a 500 yuan maintenance expense entry in the voucher. Based on the actual accounting entries, the fourth attachment can correspond to the two entries mentioned above. The total amount of 700 yuan and 500 yuan is 1200 yuan, which is consistent with the amount in the fourth attachment. Since each part of the amount is used only once in the fourth attachment, there is no duplication or remaining amount.

[0110] During reverse verification, the attachment amount corresponding to the 700 yuan office expense entry is 700 yuan, and the attachment amount corresponding to the 500 yuan repair expense entry is 500 yuan, both consistent with their respective entry amounts. By verifying the total attachment amount corresponding to the voucher entry and the total amount of each entry corresponding to the attachment, it is possible to simultaneously identify situations such as duplicate use of attachment amounts, incomplete correspondence between entry amounts, and remaining amounts in attachments.

[0111] The fifth attachment compares the transaction entity, transaction amount, transaction date, and page order in the bank receipt with the corresponding voucher entries. If the bank receipt serves as proof of payment for the transaction corresponding to the fourth attachment, its relationship with the corresponding voucher is preserved according to the attachment registration method actually used by the business system, and the same transaction amount is not recorded again as an independent purchase amount.

[0112] For attachments that still have multiple candidate relationships after amount verification, further compare the attachment page order with the voucher entry order; for example, if attachment A appears before attachment B, and entry 1 also appears before entry 2, in the absence of other business records indicating a change in order, the reverse relationship of attachment A corresponding to entry 2 and attachment B corresponding to entry 1 is not retained; if the local page order cannot be determined, the candidate relationship is not directly excluded using the local page order, but is further judged by combining the amount, transaction entity, type of bill, and the page positions that can be confirmed before and after.

[0113] For the sixth appendix, since the amount area is covered by reflections in the initial page capture, the image quality of the page needs to be further checked. The normal brightness range can be determined by using the paper area of ​​the same batch that is not affected by reflections. When a continuous bright area appears in page 7, and its brightness is significantly higher than the surrounding paper and close to the upper limit of image brightness, the corresponding area is identified as a highlight area, and the overlap between the area and the amount field is checked.

[0114] For example, the initial check value for the highlight area ratio can be 5%; this value can be determined based on the highlight area ratio corresponding to when the main fields in the normal page and reflective page samples cannot be read stably; 5% is only used to initiate further checks and does not determine whether to re-collect the page; when the highlight ratio is less than 5%, but the highlight area completely covers the amount field, the amount is still recorded as unreliable; when the highlight ratio exceeds 5%, but is entirely located in irrelevant blank areas, the entire page is not deemed unusable.

[0115] In this embodiment, the highlight area of ​​page 7 covers the amount field, so page 7 is re-captured; during re-capture, the position of the page and the light source is adjusted and the paper is flattened so that the amount area is readable again; after the new page is obtained, only the relevant fields of the sixth attachment are re-read, and the relationship between the sixth attachment and its candidate voucher entries is re-verified; the correspondences that have been confirmed in the first to fifth attachments remain unchanged.

[0116] If the amount in the sixth attachment can be obtained normally after re-collection and is consistent with the amount in the corresponding voucher entry, then record that the original anomaly was due to a page image quality issue, and save the re-collected page and the verification result.

[0117] In another scenario, if the voucher records 5 attachments, but only 4 attachments can actually be confirmed, and none of the 4 attachments have obvious image quality issues, the corresponding voucher entry still has an unexplained amount difference of 300 yuan. At the same time, a type of attachment typically required for the current expense item is not present in the scanned data. Therefore, it is possible to record that 1 attachment is missing and the amount difference is 300 yuan. The above 5, 4, and 300 yuan are used to illustrate the process of jointly judging the difference in the number and amount of attachments. In actual application, the actual number and amount obtained from the current voucher and attachments are used.

[0118] If there are no obvious abnormalities in the number of attachments and image quality, but the fields that can be reliably obtained from the attachments are inconsistent with the original fields of the voucher, then the voucher fields should be further verified. For example, if the amount in the attachments can be consistently obtained as 1230 yuan, while the amount recorded in the voucher entry is 1320 yuan, and the transaction subject, date, and type of invoice all correspond, then the difference in the voucher amount field can be recorded, and the attachment amount of 1230 yuan, the original voucher amount of 1320 yuan, and the relevant pages can be saved for subsequent review, without directly modifying the original voucher data.

[0119] When there are two or more types of abnormal page images, missing attachments, and abnormal voucher fields that simultaneously have corresponding judgment information, the proportion of the number of judgment conditions that have been met for each type of abnormality to the total number of currently available judgment conditions can be counted separately.

[0120] For example, a difference of 15 percentage points between the highest and second-highest percentages can be used as the initial reference value for entering a single anomaly handling method; when the corresponding percentages of the two types of anomalies are 80% and 60% respectively, the difference between the two is 20 percentage points, and the anomaly cause corresponding to 80% can be handled first; when the corresponding percentages of the two types of anomalies are 75% and 65% respectively, the difference between the two is 10 percentage points, then both anomaly possibilities are retained and manual review is performed.

[0121] The 15 percentage point can be determined based on historical samples where the cause of the anomaly has been manually confirmed. Specifically, the percentage of each judgment condition met in the samples of image quality anomalies, missing attachments, and voucher field anomalies can be statistically analyzed. The percentage difference between the correct anomaly type and other anomaly types can be compared, and the corresponding value can be determined in combination with the rework caused by erroneous re-collection, erroneous request for supplementary attachments, or erroneous verification of vouchers. When there are not enough historical anomaly samples, the cause of the anomaly cannot be directly determined based on this value.

[0122] After confirming the relationship between all attachments and voucher entries, save the associated verification record for this batch. The saved content should include at least the batch identifier, attachment identifier, page range corresponding to the attachment, voucher and entry location, number of attachments, number of pages, transaction entity, type of invoice, amount and date, and fields that cannot be reliably read, page order anomalies, amount differences, parameters used, and anomaly handling.

[0123] After processing in this embodiment, page 1 corresponds to the first to third attachments, pages 2 to 4 correspond to the fourth attachment, pages 5 and 6 correspond to the fifth attachment, and page 7 corresponds to the sixth attachment. At the same time, the correspondence between each attachment and the corresponding voucher entry is saved after verification of the transaction subject, type of voucher, amount, and order. When page 7 is re-collected, the abnormal state of the original page and the verification result after re-collection are also saved so as to review the formation process of the correspondence in the future.

[0124] When the accounting platform or archiving platform allows the writing of related information, the correspondence between attachments and voucher entries, as well as the attachment page range, can be written into the corresponding business system; when modification of the original scanned images is not allowed, the above information can be saved in an independent database or related record file.

[0125] When the archiving system allows for the setting of codes for locating associated records in blank areas of scanned images, such as when there is a continuous blank area at the bottom of the page and this area does not involve text, signatures, or subsequent cropping, the code position can be selected in the blank area 5 mm to 15 mm from the bottom edge of the page; the 5 mm to 15 mm range is determined based on the actual width of the blank area at the bottom of the page, the edge trimming range of the scanning device, and the cropping position of the document; when the area contains text, signatures, or other original information, or when the archiving system does not allow modification of the original scanned image, the corresponding code should not be set on the page.

[0126] When reviewing this batch later, the page and attachment data can be retrieved again based on the saved correspondence between attachments and pages; if the number of pages decreases, the missing pages can be located; if the number of pages remains the same but the number of attachments changes, the attachment area can be checked for replacement, re-cutting, or changes in the recognition range; if the number of pages and attachments remains the same, but fields such as amount, transaction entity, or bill type change, the corresponding attachment can be located and verified based on the saved records.

[0127] In this embodiment, 7 scanned pages correspond to 6 actual financial attachments, including cases where one page corresponds to multiple attachments and cases where one attachment corresponds to multiple pages. By saving the page, attachment, and voucher entry information separately, and then combining the transaction entity, invoice type, amount, and the relative order between the page and the entry, the correspondence is determined. When there are abnormal image quality, differences in the number of attachments, or differences in voucher fields, only the pages, attachments, and voucher entries that are actually affected are re-collected or reviewed, thereby reducing the impact of local anomalies on the correspondence of other attachments in the same batch and improving the reliability of the batch association results of financial attachments.

[0128] 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.

[0129] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for verifying the association between financial scanned attachments and vouchers, characterized in that, include: Obtain the continuously scanned financial attachment pages, page sequence information, and voucher entry data, and organize the pages according to page sequence; Identify independent attachments and cross-page attachments based on attachment boundaries and the continuity of content on adjacent pages, and record the page range corresponding to each attachment; Record the number of attachments and the number of pages respectively, extract the transaction entity, invoice type, amount, date and page order, and establish candidate relationships with the corresponding fields of the voucher entries; First, filter candidate relationships by transaction entity and bill type. Then, verify that the total amount of the attachments associated with the entry is consistent with the entry amount. When an attachment is associated with multiple entries, the total amount of each allocation is consistent with the attachment amount. Continue filtering by page order and entry order. For cases where multiple correspondences or verification inconsistencies still exist, anomaly detection is performed based on image quality, number of attachments, amount gaps, and differences in voucher fields, and the correspondences and verification records are saved.

2. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Retrieve continuously scanned financial attachment pages, page sequence information, and voucher entry data, and organize the pages according to page sequence, including: Record the batch identifier, page identifier, paper feed sequence number or device page number, acquisition time and acquisition resolution for each page; When the paper feed sequence number is broken, record the page gap. When the paper feed sequence number is missing, arrange the pages according to the acquisition time. When there is a duplicate paper feed sequence number or a page order conflict, determine the page order based on the page content, acquisition time, device cache records and retransmission records. If it cannot be determined, record the corresponding page order as abnormal. Page data and voucher entry data are saved separately when they arrive in sequence. Once both types of data are obtained, the data format of amount, date, and currency is standardized and the original values ​​are retained.

3. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Identify independent attachments and cross-page attachments based on attachment boundaries and the continuity of content on adjacent pages, including: Obtain the horizontal and vertical dimensions, tilt angle, binding hole position, and paper feed marks of each page; compare the changes in size, tilt angle, and relative position of binding holes of adjacent pages to determine the range of consecutive pages. The attachment area is determined based on the edges, light and dark variations, and local shadows between the attachment paper and the carrier paper. When the edge of the attachment is interrupted, the attachment boundary is determined based on the extension direction of the two sides of the interruption, the table separator line, the fixed title position, the QR code position, and the position of the main text area. When attachments overlap, the scope of the attachments is determined based on the visible edges, the actual size of similar documents, and the extension positions of the edges on both sides of the obscured area.

4. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Record the page range corresponding to each attachment, including: Determine the page continuity of multiple attachments based on the titles, page numbers, table boundaries, header positions, footer positions, fixed printing areas, transaction entities, and page specifications of adjacent pages; When a new document title appears on the next page, the transaction entity changes, the layout structure changes, or a new independent attachment starts, the page continuity of the previous attachment is terminated. For double-sided documents, the two pages are determined to belong to the same attachment based on the fixed format of the front and back sides, the invoice number, the transaction entity, and the continuous collection relationship. The attachment identifier, starting page, ending page, corresponding page, and position within the page are recorded.

5. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Record the number of attachments and the number of pages respectively, and extract the transaction entity, bill type, amount, date, and page order, including: Attachments that are independent of each other on the same page are counted separately. Attachments that are composed of multiple consecutive pages or are composed of front and back pages are counted as one attachment. For each attachment, record the attachment identifier, corresponding page, first appearance page order, and number of pages occupied. Obtain the transaction entity, bill type, amount, and date based on the transaction entity field, bill title, fixed format, amount field, and date field. Mark fields that cannot be reliably read as unreadable, and save the attachment data and the original data of the voucher entries separately.

6. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Establish candidate relationships with fields corresponding to voucher entries, including: Standardize the character format of the transaction entity's name while retaining the original name, and filter the voucher entries based on the correspondence between the expense items and attachment types configured in the business system; Retain journal entries with consistent transaction entities. If the transaction entity field in the appendix cannot be reliably read, do not exclude journal entries based on this field. Exclude journal entries where the type of invoice does not match the expense item. Screen candidate journal entries according to the date range stipulated in the financial system. Compare the actual number of attachments with the number of attachments recorded on the voucher, and verify the quantity discrepancies according to the actual registration methods of multi-page contracts, double-sided receipts, and multi-copy documents.

7. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, First, filter candidate relationships by transaction entity and bill type. Then, verify that the total amount of the attachments associated with the journal entry matches the journal entry amount, and that when multiple journal entries are associated with an attachment, the total amount of each allocation matches the attachment amount, including: Verify the transaction entities in the attachments and voucher entries, and filter candidate relationships based on the correspondence between expense items and attachment types; If the transaction parties or types of negotiable instruments listed in the appendix cannot be reliably obtained, the corresponding candidate relationships shall be retained. Using the journal entry as the object, verify the total amount of its corresponding attachments. Using the attachment as the object, verify the total amount of each journal entry corresponding to it, and check whether the amount of the same attachment is used repeatedly in different journal entries.

8. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Continue filtering by page order and entry order, including: The order of attachment pages is determined by the position of the first page of attachments in the current scanning batch, the order of voucher entries is determined by the original entry number of the voucher, and the order of voucher entries is determined by the line number of the reimbursement details when the business system records the line number of the reimbursement details. When multiple candidate relationships exist, compare the relative order of the attachment page order with the order of the voucher entries. If no business record indicates a change in order, exclude candidate relationships that are opposite to the relative order. When the order of a partial page cannot be determined, continue filtering based on the determined page positions before and after that partial area, combined with the transaction entity, bill type, and amount; If multiple candidate relationships still exist after filtering, save the attachment identifier, page position, and voucher entry information corresponding to each candidate relationship.

9. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, For cases where multiple correspondences or verification inconsistencies still exist, anomaly detection is performed based on image quality, number of attachments, amount discrepancies, and differences in voucher fields, including: For candidate relationships that cannot be uniquely determined or have verification discrepancies, check whether reflections, shadows, creases, tilts, and cropping on the page affect the reading of the amount, transaction entity, date, and invoice type; Compare the actual number of attachments, the number of attachments recorded on the voucher, the amount not corresponding, and the types of attachments that should be included, and check whether any attachments are missing; Compare the fields that can be read in the attachment with the original fields on the voucher, and check whether there are any differences between the two. When there are two or more types of anomalies, such as image quality abnormalities, missing attachments, and differences in voucher fields, count the number of judgment conditions that have been met for each type of anomaly and the number of conditions that can be used for judgment.

10. The method for verifying the association between financial scanned attachments and vouchers according to claim 1, characterized in that, Save the corresponding relationships and verification records, including: For the confirmed correspondence between attachments and voucher entries, record the batch identifier, attachment identifier, page range, voucher location, entry location, number of attachments, number of pages, transaction entity, type of invoice, amount, date, unreadable fields, amount differences, and page order anomalies; When the correspondence cannot be uniquely determined, save the attachments, pages, and voucher entries involved in each candidate relationship separately; After re-collecting the pages or manually confirming the correspondence, only the affected attachments and their corresponding voucher entries are re-verified.

Images