Data auditing method, device, system, processor and machine readable storage medium

Abstract

The embodiment of the application provides a kind of data auditing method, device, system, processor and machine readable storage medium, belong to computer technical field.The method comprises: obtaining the voucher image of the to-be-audited bill, the voucher image is matched with at least one preset voucher slice template, the voucher slice template that can be matched with the voucher image is determined as the voucher slice template corresponding to the voucher image, and the element slice of the voucher image is extracted according to the voucher slice template corresponding to the voucher image;And in the case where the voucher image cannot be matched with the preset voucher slice template, the element slice of the voucher image is extracted according to the semantic segmentation model;Based on image recognition, extract the element information in element slice, and audit the element information in element slice.The application can effectively improve the efficiency and accuracy of bill auditing, and reduce the labor cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, in particular to a data auditing method, a data auditing device, a data auditing system, a processor and a machine readable storage medium. BACKGROUND

[0002] Data auditing is an important work content of each financial institution, and through comprehensive checking of business tickets generated after transactions, it can timely find whether there is irregular operation in the business handling process and prevent financial risks. At present, the auditing method for business tickets is usually to manually arrange and check a large amount of data of the documents and reports retained in the business handling process, which is not only low in efficiency but also prone to errors. SUMMARY

[0003] The embodiments of the present application provide a data auditing method, a data auditing device, a data auditing system, a processor and a machine readable storage medium to solve the above problems.

[0004] In order to achieve the above-mentioned purpose, the first aspect of the present application provides a data auditing method, comprising:

[0005] Obtaining a voucher image including to-be-audited data;

[0006] Matching the voucher image with at least one preset voucher slice template, determining the voucher slice template capable of matching the voucher image as the voucher slice template corresponding to the voucher image, extracting the element slice of the voucher image according to the voucher slice template corresponding to the voucher image, and in the case that the voucher image cannot match the preset voucher slice template, extracting the element slice of the voucher image according to a semantic segmentation model;

[0007] Extracting element information in the element slice based on image recognition, and auditing the element information in the element slice.

[0008] Optionally, the voucher image includes a first identifier for representing a first category of the to-be-audited data and a second identifier for representing a second category of the to-be-audited data, and the voucher slice template includes a third identifier for identifying the first category of the to-be-audited data and a fourth identifier for identifying the second category of the to-be-audited data; determining the voucher slice template capable of matching the voucher image as the voucher slice template corresponding to the voucher image comprises:

[0009] Determining the voucher slice template in which the first identifier matches the third identifier and the second identifier matches the fourth identifier as the voucher slice template corresponding to the voucher image.

[0010] Optionally, the element slice of the voucher image is extracted according to the semantic segmentation model, comprising:

[0011] The element slice of the voucher image is output by the semantic segmentation model taking the voucher image as input, wherein the semantic segmentation model is obtained by training a convolutional neural network with historical voucher images including labeled data of different element information, and the labeled data includes position information of element information and description information of element information.

[0012] Optionally, the element information in the element slice includes transaction fields and transaction data, and the image recognition includes OCR recognition and ICR recognition; the element information in the element slice is extracted based on image recognition, and the element information in the element slice is audited, comprising:

[0013] The transaction fields in the element slice are recognized by OCR, and the transaction data in the element slice are recognized by ICR, wherein the transaction fields include bill name and bill code, and the transaction data include transaction account number, transaction serial number, transaction amount and transaction date;

[0014] The corresponding business serial data is obtained according to the transaction fields or transaction data in the element slice, and the transaction fields and transaction data in the element slice are audited based on the business serial data.

[0015] Optionally, the element slice of the voucher image is extracted according to the voucher slice template corresponding to the voucher image, comprising:

[0016] The position of element information in the voucher image is determined according to the voucher slice template corresponding to the voucher image, and the element slice of the voucher image is extracted according to the position of element information in the voucher image;

[0017] The voucher slice template further includes position information of different element information;

[0018] The element slice of the voucher image is extracted according to the position of element information in the voucher image, comprising: taking the position information of each element information in the voucher slice template corresponding to the voucher image as the position information of element information in the voucher image;

[0019] The slice area of element information in the voucher image is determined based on the position information of element information in the voucher image;

[0020] The element slice of the voucher image is extracted according to the slice area of element information in the voucher image.

[0021] Optionally, the first identifier matches the third identifier, comprising:

[0022] obtaining a number of pixel points where the first mark coincides with the third mark;

[0023] In a case where the number of pixel points where the first mark coincides with the third mark and the percentage of the number of pixel points of the first mark or the third mark respectively reaches a threshold value, it is determined that the first mark matches the third mark.

[0024] Optionally, the second mark matches the fourth mark, comprising:

[0025] obtaining position information of the fourth mark;

[0026] determining a second category recognition area of the credential image based on the position information of the fourth mark;

[0027] In a case where it is determined based on image recognition that the second category recognition area has a second mark matching the fourth mark, it is determined that the second mark matches the fourth mark.

[0028] The second aspect of the present application provides a data auditing device, comprising:

[0029] an image obtaining module configured to obtain a credential image of data to be audited;

[0030] an element slicing module configured to match the credential image with at least one preset credential slicing template, determine a credential slicing template capable of matching the credential image as a credential slicing template corresponding to the credential image, extract an element slice of the credential image according to the credential slicing template corresponding to the credential image, and in a case where the credential image is unable to match the preset credential slicing template, extract an element slice of the credential image according to a semantic segmentation model;

[0031] an auditing module configured to extract element information in the element slice based on image recognition, and perform auditing on the element information in the element slice.

[0032] The third aspect of the present application provides a data auditing system, comprising:

[0033] an image scanning device configured to scan data to be audited and generate a credential image of the data to be audited; and the above-mentioned data auditing device.

[0034] The fourth aspect of the present application provides a processor configured to execute the above-mentioned data auditing method.

[0035] The fifth aspect of the present application provides a machine readable storage medium having instructions stored thereon, the instructions causing the processor to be configured to execute the above-mentioned data auditing method when executed by the processor.

[0036] The sixth aspect of the present application provides a computer program product comprising a computer program, characterized in that the computer program, when executed by a processor, implements the above-mentioned data auditing method.

[0037] The present application can effectively improve the efficiency and accuracy of bill auditing by obtaining the voucher image of the data to be audited, extracting the element slice of the element information in the voucher image based on the position of the element information in the voucher image, and performing image recognition on the obtained element slice to recognize the element information in the element slice, and auditing the recognized element information.

[0038] Other features and advantages of the embodiments of the present application will be described in detail in the subsequent specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0039] The accompanying drawings are included to provide a further understanding of the embodiments of the present application, and constitute a part of the specification, and are used to explain the embodiments of the present application together with the following specific embodiments, but do not constitute a limitation on the embodiments of the present application. In the drawings:

[0040] Figure 1 a method flowchart of a data auditing method is schematically shown;

[0041] Figure 2 a comparison flowchart of element information is schematically shown;

[0042] Figure 3 a slice template schematic diagram is schematically shown;

[0043] Figure 4 a schematic block diagram of a data auditing device is schematically shown;

[0044] Figure 5 a schematic block diagram of a data auditing system is schematically shown. DETAILED DESCRIPTION

[0045] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. It should be understood that the specific embodiments described herein are only used to illustrate and explain the embodiments of the present application, and are not used to limit the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0046] It should be noted that the acquisition, storage, use, processing and the like of data in the technical solutions of the present application comply with the relevant provisions of national laws and regulations. The technical solutions of various embodiments of the present application can be combined with each other, but must be based on the realization of ordinary technical personnel in the art. When the combination of technical solutions contradicts each other or cannot be realized, it should be considered that the combination of technical solutions does not exist and is not within the protection scope required by the present application.

[0047] To solve the above problems, as shown in an embodiment of the present application, a data auditing method is provided, comprising: Figure 1

[0048] S100, acquiring a voucher image including data to be audited;

[0049] S200, matching the voucher image with at least one preset voucher slice template, determining the voucher slice template that can match the voucher image as the voucher slice template corresponding to the voucher image, extracting the element slice of the voucher image according to the voucher slice template corresponding to the voucher image; and in the case that the voucher image cannot match the preset voucher slice template, extracting the element slice of the voucher image according to the semantic segmentation model;

[0050] S300, extracting element information in the element slice based on image recognition, auditing the element information in the element slice.

[0051] In this way, the present embodiment acquires the voucher image of the data to be audited, extracts the element slice of the element information in the voucher image based on the position of the element information in the voucher image, and performs image recognition on the obtained element slice to recognize the element information in the element slice, and audits the recognized element information, which can effectively improve the efficiency and accuracy of bill auditing and reduce the labor cost.

[0052] In the present embodiment, data auditing includes but is not limited to bill auditing, wherein the data to be audited includes but is not limited to transaction-generated data to be audited bills with fixed templates and data to be audited bills without fixed templates, such as handwritten bills; the voucher image includes but is not limited to bill image, and the voucher slice template includes but is not limited to bill slice template. The following describes the application of the method of the present embodiment to bill auditing of the financial system as an example. It can be understood that in the bill auditing of the financial system, the data to be audited is the bill to be audited, the voucher image is the bill image, and the voucher slice template is the bill slice template.

[0053] ​Specifically, by setting a scanning point in the business front desk, the generated financial bill is scanned to obtain the bill image of the bill to be audited. The business front desk uploads the obtained bill image to the audit department system after each scanning, or the business front desk can batch upload the scanned bill image to the audit system of the audit department at a set time every day. The audit system processes the received bill image in batches in real time or at a set time through a preset audit model. Since the elements in the same type and nature of financial bill / instrument usually have a fixed layout, the corresponding element slice template can be defined in advance, and the financial bill / instrument with a fixed template is referred to as a preset bill / instrument. However, in the process of business handling, in order to solve business problems, some same type and nature of financial bill / instrument as the preset bill may be generated, but the layout is different. This type of financial bill / instrument is referred to as a non-preset bill / instrument. Since this type of bill does not have a corresponding predefined template, the element information of the non-preset bill / instrument cannot be extracted by template matching. In order to solve this problem, in the embodiment, for the preset bill, the element slice is extracted by preset bill slice template matching, and for the non-preset bill, the element slice is extracted by the semantic segmentation model constructed based on R-CNN. Therefore, for each bill image, it is first necessary to determine whether there is an element slice template that can match the obtained bill image in the preset bill slice template. If there is an element slice template that can match the bill image, it indicates that the bill is a preset bill, and the element slice template is determined as the element slice template corresponding to the bill image. Then, according to the bill slice template, the position of the element information to be audited in the bill image is determined, and the region including each element information is cut out based on the position of each element information recognized to obtain the element slice of each element information. If there is no element slice template that can match the bill image, it indicates that the bill is a non-preset bill, and the element slice cannot be extracted by template matching. In this case, the semantic segmentation model is used to extract the element slice of the bill image. Finally, the element information in the element slice is extracted based on image recognition, so as to audit the element information. It can be understood that the element slice includes at least one element information. For example, a certain financial bill includes bill name, transaction amount, bill code, and transaction date element information. Each element slice finally extracted includes one of the bill name, transaction amount, bill code, and transaction date element information.

[0054] In the creation of the element slice template corresponding to the financial instrument of different categories and properties, the business experts familiar with the field can manually mark and frame the slice area of each element information on the instrument image of the financial instrument of different categories and properties, so as to determine the area of the element information in the instrument that needs to be audited. In the matching of the element slice, the coincidence degree of the pre-marked frame such as the red frame on the template and the original frame such as the black frame on the instrument image can be used to identify whether the instrument image and the element slice template match. When the coincidence degree is higher than a threshold, it is considered that the instrument image and the current instrument element slice template match, and then the element information on the instrument image can be cropped according to the pixel coordinate values of each element information marked on the current instrument element slice template, to obtain the element slice including each element information. It can be understood that in the embodiment, the manual marking and framing can also be marking the area of the element information by using a drawing tool or software.

[0055] In the embodiment, the instrument image includes a first identifier for representing a first category of the instrument to be audited and a second identifier for representing a second category of the instrument to be audited, and the instrument slice template includes a third identifier for identifying the first category of the instrument to be audited and a fourth identifier for identifying the second category of the instrument to be audited. The instrument slice template that can match the instrument image is determined as the instrument slice template corresponding to the instrument image, including: determining the instrument slice template with the matched first identifier and third identifier and the matched second identifier and fourth identifier as the instrument slice template corresponding to the instrument image.

[0056] In one specific embodiment, whether the instrument to be audited is a preset instrument is determined by sequentially matching the instrument image with the element slice templates in the template library. When a certain element slice template in the template library has a third identifier and a fourth identifier that can match the first identifier and the second identifier of the instrument image respectively, the element slice template is taken as the instrument slice template corresponding to the instrument image.

[0057] In another specific embodiment, the bill slice template corresponding to the bill to be audited can be identified according to the first identifier and the second identifier on the bill. For example, the first identifier and the second identifier on the bill image can be identified, and the bill slice template library is queried to find a bill slice template matching the first identifier and the second identifier. The positions of the element information can be determined by the preset slice template, and the slice template includes the area identifier of each element information. For example, when uploading the bill image in the front desk, the bill category of the bill to be audited is associated, such as the bill category of the bill to be audited is a savings passbook. After the bill image is received by the auditing department, the bill category of the bill to be audited is first identified as a savings passbook, and then the savings passbook slice template corresponding to the savings passbook is obtained from the template library. The positions of the element information in the bill image are determined by matching the bill image with the preset element information area identifier in the savings passbook slice template. According to the element information area identifier in the savings passbook slice template, each element information area in the bill image is cut into an element slice. The text and character content of the element information in each element slice are identified and extracted based on image recognition, such as ICR or OCR. When the received bill needs to be audited by the auditing department, the transaction information is obtained according to the rules, such as the transaction flow data in a specified time period. The received bill image is retrieved according to the transaction flow data, and each identified element information is compared with the corresponding transaction flow data one by one, so as to realize the automatic auditing of the bill. It can be understood that in order to accurately compare the element information, the transaction flow number corresponding to the bill image can also be associated when uploading the bill image in the front desk.

[0058] In this embodiment, the element slice of the bill image is extracted according to the semantic segmentation model, including: taking the bill image as input, outputting the element slice of the bill image through the semantic segmentation model, wherein the semantic segmentation model is obtained by training the convolutional neural network with historical bill images including different element information annotation data, and the annotation data includes position information of the element information and description information of the element information. In this embodiment, the element slice extraction of the non-pre-set bill can be performed by the pre-trained semantic segmentation model. The semantic segmentation model is obtained based on the training of the R-CNN neural network. Before constructing the semantic segmentation model, the R-CNN neural network is trained with historical bill images with annotation data as training samples, wherein the annotation data includes position annotation of each element information on the bill, such as a box representing the region of the element information, and a description of the element information, such as the element name of the element information being "transaction amount" or "bill name", etc. The R-CNN neural network is trained with the training samples as input to predict the region and element name of each element information, and the parameters of the R-CNN neural network are adjusted according to the prediction result. Finally, under the condition that the R-CNN neural network meets the convergence condition, the trained semantic segmentation model is obtained. When the element slice of the bill image is extracted according to the semantic segmentation model, the acquired bill image is input into the trained semantic segmentation model to detect and classify the element information region of the bill image, and the semantic segmentation is performed according to the detection result of the element information region, and finally the predicted element slice including the element information and the description of the element information are output, so as to realize the extraction of the element slice of the non-pre-set bill. The training and prediction process of the semantic segmentation model based on the R-CNN neural network is a prior art, which will not be described here.

[0059] As shown in Figure 2 To further ensure the accuracy of the element information comparison, in this embodiment, when the automatic comparison confirms that the element information in the bill image is consistent with the element information of the transaction flow data, the comparison result is marked as "consistent" and sent to the next node such as the audit monitoring module of the audit system, so that the administrator can perform subsequent checking and auditing work. If the automatic comparison result is inconsistent, the comparison task is assigned to the first collection personnel, who manually checks according to the bill image. If the manual checking result is consistent, the comparison result is submitted to the audit department administrator. If the manual checking result is inconsistent, the comparison task is assigned to the second collection personnel, who manually checks according to the bill image. If the manual checking result is consistent, the comparison result is submitted to the audit monitoring module. If the manual checking result is inconsistent, the comparison result is submitted to the audit monitoring module.

[0060] It can be understood that when the non-preset bill, that is, the template recognition fails, the element slicing extraction is performed using the region-based semantic segmentation model, and then the sliced elements are subjected to OCR recognition to identify the voucher name, element name, and element value of the to-be-audited bill, and the like, and then compared with the element information of the transaction stream. If they are consistent, the audit is consistent, and if they are inconsistent, the comparison task is assigned to the first collection personnel for manual checking according to the bill image. If the manual checking result is consistent, the comparison result is submitted to the audit department administrator, and if the manual checking result is inconsistent, the comparison task is assigned to the second collection personnel for manual checking according to the bill image. If the manual checking result is consistent, the comparison result is submitted to the audit monitoring module, and if the manual checking result is inconsistent, the comparison inconsistent result is submitted to the audit monitoring module. Or, if the automatic comparison result is inconsistent, the comparison task is assigned to the first collection personnel for checking and inputting the element information of the to-be-audited bill item by item to obtain the first collection result. The first collection result is compared with the element information in the transaction stream information. If they are consistent, the comparison result is submitted to the audit monitoring module, and if they are inconsistent, the comparison inconsistent result is submitted to the audit monitoring module. Then, the task is assigned to the second collection personnel for inputting the element information of the to-be-audited bill to obtain the second collection result, and the second collection result is compared with the first collection result. If the first collection result and the second collection result are consistent, the comparison result is submitted to the audit monitoring module, and if the first collection result and the second collection result are inconsistent, the comparison inconsistent result is submitted to the audit monitoring module. Through the comparison process of the present embodiment, the audit range of the non-preset scanning voucher is expanded, the errors of the intelligent recognition model are effectively prevented, the business data fed back to the audit business department has more audit value, and the audit effect is improved.

[0061] The element information in the element slice includes transaction fields and transaction data, and the image recognition includes OCR recognition and ICR recognition. The element information in the element slice is extracted based on image recognition, and the element information in the element slice is audited, including: the transaction fields in the element slice are recognized by OCR, and the transaction data in the element slice are recognized by ICR. The transaction fields include the bill name and the bill code, and the transaction data include the transaction account number, the transaction serial number, the transaction amount, and the transaction date. The corresponding business stream data is obtained according to the transaction fields or the transaction data in the element slice, and the transaction fields and the transaction data in the element slice are audited based on the business stream data.

[0062] ICR recognition is a classification model trained by a large number of historical voucher image, mainly used for recognizing handwritten characters by recognizing the features of the collected image, which can accurately recognize the preset voucher handwritten characters; OCR is an optical character recognition method based on image optical character recognition, mainly used for recognizing printed characters on the ticket, including voucher title, element name and element value. In particular, if the printed characters of the to-be-audited ticket are blocked, for example, the "current deposit certificate" is blocked at the top and recognized by ICR as "frozen deposit certificate", the OCR model can also be used to calculate the similarity between the recognized characters and the preset voucher name, and if the similarity is greater than the preset threshold, it can be considered as "current deposit certificate". Therefore, for the preset ticket, the element information value can be recognized by ICR, that is, the recognition of handwritten information, for non-pre-set tickets, the element name information can be recognized by OCR, and the handwritten information after the element name can be recognized by ICR. In this embodiment, since the transaction field, that is, the ticket name, ticket code, element name and the like are usually printed characters, after the element slice of the to-be-audited ticket is obtained, the transaction field of the element information is recognized by OCR; and the transaction data is usually handwritten, for example, transaction amount and date, therefore, the transaction data is recognized by ICR, so that the text or character content of each element information is recognized and extracted for comparison with the corresponding element information in the business flow data. For example, in a specific example, the transaction serial number in the ticket image is recognized to obtain the corresponding business flow data.

[0063] In step S200, the element slice of the ticket image is extracted according to the ticket slice template corresponding to the ticket image, including: determining the position of the element information in the ticket image according to the ticket slice template corresponding to the ticket image, and extracting the element slice of the ticket image according to the position of the element information in the ticket image; the ticket slice template further includes: position information of different element information; the element slice of the ticket image is extracted according to the position of the element information in the ticket image, including: taking the position information of each element information in the ticket slice template corresponding to the ticket image as the position information of the element information in the ticket image; determining the slice area of the element information in the ticket image based on the position information of the element information in the ticket image; extracting the element slice of the ticket image according to the slice area of the element information in the ticket image.

[0064] Typically, different types of financial instruments have templates with fixed information layouts, and the corresponding information elements vary. For example, in a savings certificate, the information elements typically include: voucher number, account number, account name, amount (uppercase), amount (lowercase), principal withdrawal amount, withdrawal date, and deposit date. In transfer checks and cash checks, the information elements typically include: voucher number, payment account number, amount in uppercase, amount in lowercase, issue date, and payee name. In wire transfer vouchers and special transfer vouchers, the information elements typically include: transaction date, payment account number, payer name, payee bank, payee name, payee bank, amount in uppercase, amount in lowercase, issue date, and payee name. The layout of each element in different types of financial instruments also varies significantly.

[0065] like Figure 3 As shown, in this embodiment, the first identifier is an image identifier that can identify whether the bill to be audited belongs to a certain category among savings certificates, transfer checks or wire transfer vouchers. For example, if the bill to be audited is a transfer check, the first identifier can be the black border of the RMB (uppercase) column in the transfer check. Since the position of the element information is different in different categories of financial bills, for example, the position of the amount data in the bill is different in the transfer check and the savings certificate, the position of their borders is also different. By identifying the position of the border, the preliminary category of the bill to be audited can be determined. The second identifier can clearly represent the characteristics of the bill to be audited, for example, the name of the bill. It can be understood that the second identifier can be a separate identification identifier or one of the element information. For example, for a transfer check, when the amount data frame of a certain slicing template successfully matches the amount data frame of the bill to be audited, it is necessary to further confirm the category of the bill to be audited by matching the second identifier to ensure the accuracy of recognition. For example, the bill name of the slicing template can be matched with the name of the bill to be audited to determine whether they overlap to identify whether the second identifier and the fourth identifier match. When both the amount data frame and the bill name can be matched successfully, it is determined that the category of the current bill slicing template is consistent with the category of the bill to be audited, and then the element position of the bill to be audited in the bill image can be determined by the element information position preset on the current bill slicing template. It can be understood that the element information preset on the bill slicing template can be pre-marked in the form of a box.

[0066] Specifically, the matching of the first identifier and the third identifier includes: obtaining a number of pixel points that coincide between the first identifier and the third identifier; and in a case where the number of pixel points that coincide between the first identifier and the third identifier and a percentage of the number of pixel points of the first identifier or the third identifier respectively reach a threshold value, determining that the first identifier and the third identifier match. Taking a transfer check as an example, when the bill slice template and the bill image are matched, whether the two match is determined by calculating the coincidence degree of the pixel points of the amount data frame on the bill image of the bill to be audited and the amount data frame on the slice template. When the coincidence degree reaches the threshold value, it is considered that the bill to be audited and the slice template both belong to the transfer check type bill.

[0067] The matching of the second identifier and the fourth identifier includes: obtaining position information of the fourth identifier; determining a second type recognition area of the bill image based on the position information of the fourth identifier; and in a case where the second type recognition area determined based on the image recognition exists the second identifier matching the fourth identifier, determining that the second identifier and the fourth identifier match. Similarly, taking the transfer check as an example, in the case of confirming the matching of the amount data frame, the corresponding position on the bill image of the bill to be audited is identified according to the region position of the bill name "transfer check" on the slice template. If the text content of the region on the bill image of the bill to be audited is "transfer check" through ICR or OCR identification, it is determined that the second identifier and the fourth identifier match, thereby determining that the bill to be audited is a transfer check type bill.

[0068] As shown in Figure 4 The second aspect of the present application provides a data auditing device, which comprises:

[0069] An image acquisition module configured to acquire a voucher image of data to be audited;

[0070] An element slice module configured to match the voucher image with at least one preset voucher slice template, determine the voucher slice template that can match the voucher image as the voucher slice template corresponding to the voucher image, extract the element slice of the voucher image according to the voucher slice template corresponding to the voucher image, and in a case where the voucher image cannot match the preset voucher slice template, extract the element slice of the voucher image according to a semantic segmentation model;

[0071] An auditing module configured to extract element information in the element slice based on image recognition, and audit the element information in the element slice.

[0072] As shown in Figure 5 The third aspect of the present application provides a data auditing system, which comprises:

[0073] An image scanning device configured to scan the data to be audited and generate a voucher image of the data to be audited; and the data auditing device as described above.

[0074] The fourth aspect of the present application provides a processor configured to execute the data auditing method as described above. The processor includes a core, and the core is configured to call a corresponding program unit from a memory. The core can be one or more, and the core is configured to implement the information pushing method based on multi-modal feature fusion by adjusting core parameters. The memory can include a non-permanent memory in a computer readable medium, a random access memory (RAM), and / or a non-volatile memory such as a read-only memory (ROM) or a flash memory (flash RAM), and the memory includes at least one memory chip.

[0075] The fifth aspect of the present application provides a machine readable storage medium having instructions stored thereon, the instructions, when executed by a processor, cause the processor to be configured to execute the data auditing method as described above.

[0076] The machine readable storage medium includes permanent and non-permanent, removable and non-removable media, and can be implemented by any method or technology to store information. The information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. According to the definition herein, the computer readable medium does not include transitory computer readable media, such as modulated data signals and carriers.

[0077] The sixth aspect of the present application provides a computer program product, including a computer program, characterized in that the computer program, when executed by a processor, implements the data auditing method as described above.

[0078] Those skilled in the art will appreciate that embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0079] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0080] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0081] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks Figure 1 means for functionally implementing the steps listed in the flowchart block or blocks.

[0082] It should be further understood that the terms "comprise", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a" does not, without more constraints, foreclose the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0083] The application has been described in an embodiment thereof. It should be understood, however, that embodiments might be made without departing from the spirit and essential characteristics thereof, and that changes might be made in the details and in the arrangement of parts thereof without departing from its scope. The application is thus to be construed as not limited to the embodiment described above, but to encompass any changes that come within the scope and spirit of the following claims.

Claims

1. A data audit method, characterized in that: include: Obtaining a voucher image including the data to be audited; matching the voucher image with at least one preset voucher slice template, determining the voucher slice template that matches the voucher image as the voucher slice template corresponding to the voucher image, and extracting element slices of the voucher image based on the voucher slice template corresponding to the voucher image; and extracting element slices of the voucher image based on a semantic segmentation model if the voucher image cannot match the preset voucher slice template; Extracting element information from the element slice based on image recognition, where the element information in the element slice includes transaction fields and transaction data, and obtaining corresponding business flow data based on the transaction fields or transaction data in the element slice; If it is confirmed that the element information in the element slice is consistent with the element information of the business flow data, the comparison result is confirmed to be consistent, and the current comparison result is submitted to the audit monitoring module. If it is confirmed that the comparison result is inconsistent, the comparison task is assigned to the first collection personnel; If the comparison result of the first collector is consistent, the current comparison result is submitted to the audit monitoring module; if the comparison result of the first collector is inconsistent, the comparison task is assigned to the second collector; If the comparison result of the second collection personnel is consistent, or if the comparison result of the second collection personnel is inconsistent, submit the current comparison result to the audit monitoring module; The voucher image includes a first identifier for representing a first category of the data to be audited and a second identifier for representing a second category of the data to be audited, and the voucher slice template includes a third identifier for identifying the first category of the data to be audited and a fourth identifier for identifying the second category of the data to be audited; Determining a voucher slice template that can match the voucher image as the voucher slice template corresponding to the voucher image includes: Determine the voucher slicing template in which the first identifier matches the third identifier and the second identifier matches the fourth identifier as the voucher slicing template corresponding to the voucher image; The first identifier is an image identifier, and the first identifier matches the third identifier, including: Obtaining the number of pixels where the first identifier and the third identifier overlap; When the percentage of the number of overlapping pixels of the first identifier and the third identifier to the number of pixels of the first identifier or the third identifier reaches a threshold, it is determined that the first identifier matches the third identifier.

2. The data auditing method according to claim 1, characterized in that: Extracting element slices of the voucher image based on a semantic segmentation model includes: The voucher image is used as input, and the semantic segmentation model outputs element slices of the voucher image, wherein the semantic segmentation model is obtained by training a convolutional neural network using historical voucher images with annotated data including different element information, and the annotated data includes location information of the element information and description information of the element information.

3. The data audit method according to claim 1, characterized in that: The image recognition includes OCR recognition and ICR recognition; the extracting element information from the element slice based on image recognition includes: The transaction field in the element slice is identified by OCR, and the transaction data in the element slice is identified by ICR. The transaction field includes the bill name and bill code, and the transaction data includes the transaction account number, transaction serial number, transaction amount and transaction date.

4. The data auditing method according to claim 1, characterized in that: Extracting element slices of the voucher image according to a voucher slice template corresponding to the voucher image includes: Determining the position of element information in the voucher image according to a voucher slice template corresponding to the voucher image, and extracting element slices of the voucher image according to the position of the element information in the voucher image; The voucher slice template also includes: location information of different element information; The extracting element slices of the voucher image according to the position of the element information in the voucher image includes: Using the position information of each element information in the voucher slice template corresponding to the voucher image as the position information of the element information in the voucher image; determining a slice area of ​​the element information in the voucher image based on position information of the element information in the voucher image; Extracting element slices of the voucher image according to a slice area of ​​element information in the voucher image.

5. The data auditing method according to claim 1, characterized in that: The second identifier matches the fourth identifier, including: Obtaining location information of the fourth marker; determining a second category identification area of ​​the credential image based on the position information of the fourth identifier; In a case where it is determined based on image recognition that a second identifier matching the fourth identifier exists in the second category identification area, it is determined that the second identifier matches the fourth identifier.

6. A data auditing device, characterized in that: include: an image acquisition module configured to acquire a voucher image including data to be audited; a feature slice module configured to match the voucher image with at least one preset voucher slice template, determine the voucher slice template that matches the voucher image as the voucher slice template corresponding to the voucher image, extract feature slices of the voucher image based on the voucher slice template corresponding to the voucher image; and, if the voucher image cannot match the preset voucher slice template, extract feature slices of the voucher image based on a semantic segmentation model; an audit module configured to extract element information from the element slice based on image recognition, the element information in the element slice including transaction fields and transaction data, and obtain corresponding business flow data based on the transaction fields or transaction data in the element slice; If it is confirmed that the element information in the element slice is consistent with the element information of the business flow data, the comparison result is confirmed to be consistent, and the current comparison result is submitted to the audit monitoring module. If it is confirmed that the comparison result is inconsistent, the comparison task is assigned to the first collection personnel; If the comparison result of the first collector is consistent, the current comparison result is submitted to the audit monitoring module; if the comparison result of the first collector is inconsistent, the comparison task is assigned to the second collector; If the comparison result of the second collection personnel is consistent, or if the comparison result of the second collection personnel is inconsistent, submit the current comparison result to the audit monitoring module; The voucher image includes a first identifier for representing a first category of the data to be audited and a second identifier for representing a second category of the data to be audited, and the voucher slice template includes a third identifier for identifying the first category of the data to be audited and a fourth identifier for identifying the second category of the data to be audited; Determining a voucher slice template that can match the voucher image as the voucher slice template corresponding to the voucher image includes: Determine the voucher slicing template in which the first identifier matches the third identifier and the second identifier matches the fourth identifier as the voucher slicing template corresponding to the voucher image; The first identifier is an image identifier, and the first identifier matches the third identifier, including: Obtaining the number of pixels where the first identifier and the third identifier overlap; When the percentage of the number of overlapping pixels of the first identifier and the third identifier to the number of pixels of the first identifier or the third identifier reaches a threshold, it is determined that the first identifier matches the third identifier.

7. A data audit system, characterized in that: include: An image scanning device configured to scan the data to be audited and generate a voucher image of the data to be audited; And the data auditing device according to claim 6.

8. A processor, characterized in that: The method is configured to execute the data auditing method according to any one of claims 1 to 5.

9. A machine-readable storage medium having instructions stored thereon, characterized in that: When the instruction is executed by a processor, the processor is configured to execute the data auditing method according to any one of claims 1 to 5.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, it implements the data audit method according to any one of claims 1 to 5.

Images