Invoice identification method

Abstract

One or more embodiments of the invention provide an invoice identification method, and the method comprises the steps: obtaining a target invoice uploaded by a user; identifying whether the target invoice is of a known invoice type; if yes, adopting a preset target AI model combination to carry out invoice identification operation on the target invoice so as to extract invoice information; and if not, performing identification operation on the target invoice by adopting a plurality of preset AI model combinations, and extracting invoice information from a plurality of identification results corresponding to the plurality of AI model combinations.

Description

Technical Field

[0001] One or more embodiments of this specification relate to the field of terminal technology, and in particular, to an invoice recognition method. Background Art

[0002] Processing overseas invoices presents numerous challenges. Unlike domestic invoices, which have standardized templates, invoices from overseas suppliers vary widely, often in multiple languages ​​and versions. The current process relies entirely on manual labor, which not only incurs significant labor costs but also places high demands on the cognitive abilities of operators.

[0003] Under manual processing methods, invoice processing efficiency is low. On average, it takes 15-20 minutes to process an invoice throughout its entire life cycle. In addition, there is a lack of effective online management methods, and the entire chain processing relies on manual work. Summary of the Invention

[0004] In view of this, one or more embodiments of this specification provide the following technical solutions:

[0005] According to the first aspect of one or more embodiments of this specification, an invoice identification method is proposed, which includes: obtaining a target invoice uploaded by a user; identifying whether the target invoice is a known invoice type; if so, using a preset target AI model combination to perform an invoice identification operation on the target invoice to extract invoice information; if not, using a preset multiple AI model combination to perform identification operations on the target invoice respectively, and extracting the invoice information from multiple identification results corresponding to the multiple AI model combinations.

[0006] According to the second aspect of one or more embodiments of this specification, an invoice identification device is proposed, including: an invoice uploading module, used to obtain a target invoice uploaded by a user; an invoice identification module, used to identify whether the target invoice is a known invoice type; if so, using a preset target AI model combination to perform an invoice identification operation on the target invoice to extract invoice information; if not, using a preset multiple AI model combination to perform identification operations on the target invoice respectively, and extracting the invoice information from multiple identification results corresponding to the multiple AI model combinations.

[0007] According to a third aspect of one or more embodiments of this specification, an electronic device is proposed, comprising: a processor; a memory for storing processor-executable instructions; wherein the processor implements the steps of the method described in the first aspect by running the executable instructions.

[0008] According to a fourth aspect of one or more embodiments of this specification, a computer-readable storage medium is provided, on which computer instructions are stored. When the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0009] According to a fifth aspect of one or more embodiments of this specification, a computer program product is proposed, comprising a computer program / instruction, which implements the steps of the method described in the first aspect when executed by a processor.

[0010] As can be seen from the above examples, this specification uses a combination of preset target AI models for identification of known invoice types, leveraging the collaborative work between models to accurately locate and extract invoice information. For unknown invoice types, a combination of multiple AI models is used for separate identification, and information is then extracted from the multiple identification results. This approach fully leverages the advantages of different models, not only ensuring the accuracy of identifying known invoice types, but also significantly improving the accuracy of identifying unknown invoice types through the integrated use of multiple models. This greatly enhances the efficiency, accuracy, and applicability of invoice identification, effectively meeting the recognition needs of various invoice types. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 It is a structural diagram of an invoice identification service system provided by an exemplary embodiment.

[0012] Figure 2 This is one of the flow charts of an invoice identification method provided by an exemplary embodiment.

[0013] Figure 3 This is the second flowchart of an invoice identification method provided by an exemplary embodiment.

[0014] Figure 4 This is the third flowchart of an invoice identification method provided by an exemplary embodiment.

[0015] Figure 5 It is a structural diagram of an electronic device provided by an exemplary embodiment.

[0016] Figure 6 It is a block diagram of an invoice recognition device provided by an exemplary embodiment. DETAILED DESCRIPTION

[0017] The user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this manual are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or refuse.

[0018] There are many difficulties in overseas invoice processing in terms of style, processing flow and management. Although existing technologies have certain means, it is difficult to effectively deal with complex overseas invoice processing scenarios. Unlike domestic invoices that have standard templates, the invoice styles of overseas suppliers (Content Provider, CP) vary, with one CP having one style, and covering multiple languages ​​such as Chinese, Russian, English, and French. This diversity greatly increases the difficulty of invoice processing, and traditional fixed pattern recognition methods are difficult to apply. The processing process relies on manual labor: It relies entirely on manual labor to identify overseas invoices, with huge labor costs and high requirements on the cognitive ability of the operators. During the processing process, it is necessary to manually download the invoice information one by one and identify them one by one offline. On average, it takes 15-20 minutes to process the entire life cycle of an invoice, which is extremely inefficient.

[0019] Lack of effective online management: The process cannot be carried out online, and manual identification methods are difficult to realize online process, resulting in difficulty in improving overall management efficiency and unable to adapt to the needs of efficient operation of modern enterprises; there is a risk of repeated entry and use of invoices, and only basic audits are performed, and it is impossible to determine whether the invoices are duplicated, and whether the information of the buyer and seller is consistent, which brings potential risks to the company's financial management; tax declaration data is inaccurate and difficult, and tax declaration relies on tax personnel to manually distinguish between value-added tax (VAT) invoices and manually organize the data. This method is prone to errors, resulting in inaccurate tax declaration data, increasing the difficulty and risk of tax processing; accounting processing is not closed-loop: the input tax amount has not been temporarily estimated and converted to input tax processing, resulting in a missing accounting link, affecting the integrity and accuracy of financial data, and is not conducive to the standardized management of corporate finances.

[0020] Existing invoice OCR recognition technologies use fixed-position marking, making it difficult to accurately identify and locate key information in the diverse international invoice formats. Overseas invoice formats are not standardized, with varying field positions and styles. Traditional methods are unable to adapt to this diversity, resulting in poor recognition results. While some products support OCR for overseas pro forma invoices, each has its own flaws. For example, ABBYY FlexiCapture requires certain upfront configuration and template creation, and may have limitations with special fonts or handwriting. Alibaba Cloud OCR may require parameter configuration and algorithm adjustments when processing special scenarios or complex invoice formats, and may be limited by computing resources when processing large amounts of data. Klippa OCR's recognition accuracy decreases for complex or irregular invoice formats, requiring continuous data training and model optimization. IronOCR, based on the open-source Tesseract OCR engine, lacks the same recognition accuracy as professional commercial OCR software for complex invoice formats and special fonts, and requires advanced technical expertise for secondary development and optimization.

[0021] In view of this, this specification proposes an invoice recognition method that uses an AI model combination consisting of multiple AI models to perform invoice recognition.

[0022] During implementation, the target invoice uploaded by the user is obtained; whether the target invoice is a known invoice type is identified; if so, the preset target AI model combination is used to perform an invoice recognition operation on the target invoice to extract the invoice information; if not, the preset multiple AI model combinations are used to perform recognition operations on the target invoice respectively, and the invoice information is extracted from the multiple recognition results corresponding to the multiple AI model combinations.

[0023] In the above technical solution, for known invoice types, a preset target AI model combination is used for identification, and the collaborative work between models is used to accurately locate and extract invoice information. For unknown invoice types, a combination of multiple AI models is used to identify them separately, and then information is extracted from the multiple identification results. This approach fully utilizes the advantages of different models, not only ensuring the accuracy of identifying known invoice types, but also significantly improving the recognition accuracy of unknown invoice types through the comprehensive application of multiple models. This greatly improves the efficiency, accuracy, and applicability of invoice recognition, effectively meeting the recognition needs of various types of invoices.

[0024] Figure 1 This is a schematic diagram of the architecture of an invoice identification service system provided by an exemplary embodiment. Figure 1 As shown, the system may include a server 11, a network 12, and several electronic devices, such as a PC (Personal Computer) 13, a mobile phone 14, and the like.

[0025] The server 11 may be a physical server containing an independent host, or a virtual server hosted by a host cluster. During operation, the server 11 may run a server-side program of an application to implement the relevant functions of the application. For example, when the server 11 runs a program for an invoice recognition service, it may be implemented as a corresponding invoice recognition service platform.

[0026] PC 13 and mobile phone 14 are only some types of electronic devices that users can use. In fact, users can obviously also use electronic devices such as the following types: tablet devices, laptops, PDAs (Personal Digital Assistants), wearable devices (such as smart glasses, smart watches, etc.), etc., and one or more embodiments of this specification do not limit this. During operation, the electronic device can run the client-side program of a certain application to implement the relevant functions of the application. For example, when the electronic device runs the program of the invoice recognition service, it can be implemented as the client of the invoice recognition service. Among them, the client application of the above-mentioned invoice recognition service can be started and run on the electronic device. The client-side program can be a native application installed on the electronic device, or the client-side program can be a small program, a quick application or other similar forms. Of course, when using web technologies such as HTML5 or similar, the relevant functions can be implemented through the page displayed by the browser. The browser here can be an independent browser application or a browser module embedded in certain applications.

[0027] Regarding the network 12 for interaction between electronic devices such as PC 13 and mobile phone 14 and server 11, communication can be achieved using a wired or wireless network based on the communication methods supported by the corresponding electronic devices, and this specification does not limit this. For example, if PC 13 supports both wired and wireless communication, then communication can be achieved using a wired or wireless network as needed, while mobile phone 14 generally only supports wireless communication and thus can achieve communication using a wireless network.

[0028] In order to enable people skilled in the art to better understand the technical solutions in this application, the technical solutions in this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this application.

[0029] See Figure 2 , Figure 2 An exemplary embodiment provides an invoice identification method, which can be executed by an invoice management server or an invoice management platform. The method may include the following steps:

[0030] Step S210: Obtain the target invoice uploaded by the user.

[0031] In one embodiment, a user (eg, supplier, relevant business processing personnel) may upload the target invoice to be identified to the server in a specified manner, for example, through an invoice upload interface on a client.

[0032] In one embodiment, multiple invoices can be uploaded simultaneously.

[0033] In one embodiment, the uploaded invoice may include pictures, documents (such as PDF, Word, etc.), and other forms.

[0034] It should be noted that the invoice in the embodiment of the present application may specifically refer to an overseas invoice.

[0035] Step S220: Identify whether the target invoice is a known invoice type. If so, perform an invoice recognition operation on the target invoice using a preset target AI model combination to extract invoice information. If not, perform recognition operations on the target invoice using multiple preset AI model combinations, and extract invoice information from multiple recognition results corresponding to the multiple AI model combinations.

[0036] In one embodiment, the invoice management server can provide multiple AI models for invoice recognition and combine them into multiple AI model combinations. The AI ​​models can include a layout analysis model, a text recognition model, and a structured parsing model.

[0037] Layout analysis models can be used to identify the layout of invoices, locating key areas within the complex invoice layout, such as the invoice header, item details, and amount, which contain various invoice information. Leveraging image recognition algorithms and deep learning techniques, layout analysis models intelligently analyze the overall structure of invoices. Even for extremely complex invoice formats, they can accurately locate invoice information, providing a solid foundation for subsequent text recognition. Examples include the Surya model and the LayoutLMv3 model.

[0038] Text recognition models can be used to identify text on invoices. These models utilize optical character recognition technology combined with deep learning algorithms, resulting in powerful character recognition capabilities. They can accurately recognize text in different languages, diverse fonts, and uncertain handwriting. Examples include the PaddleOCR model and the Tesseract OCR model.

[0039] Structured parsing models can be used to structure recognized text information. These models can classify, organize, and summarize the identified, disjointed text according to specific rules and formats, transforming it into structured invoice information. By accurately parsing and associating various invoice data, such as invoice number, date, amount, and buyer / seller information, they create a clear logical structure, facilitating subsequent storage, querying, and analysis. Examples include the OmniParser model and the Parseur model.

[0040] By combining the various AI models mentioned above and collaborating with each other, invoice recognition can be achieved. Each AI model combination can include one or more of the above AI models, and can also be combined with other traditional OCR models.

[0041] In one embodiment, for a known invoice type, a target AI model combination can be determined through early in-depth learning and optimization. For example, a large amount of historical invoice sample data of the same invoice type can be used to train and optimize a machine learning algorithm to determine the target AI model combination.

[0042] Among them, the classification of invoices can be divided according to dimensions such as country and region, behavior type, invoice purpose, and supplier name.

[0043] In one embodiment, different target AI model combinations may be determined for different invoice types, or the same target AI model combination may be used.

[0044] In one embodiment, after receiving the target invoice, a preliminary identification may be performed on the target invoice to determine whether the invoice is of a known invoice type.

[0045] Initial identification of the target invoice can include identifying information such as the country or region, language, and format of the target invoice. This information can be entered by the user when uploading the invoice, or by extracting and analyzing key textual information on the invoice, such as the invoice header, specific behavioral terms, and tax-related identifiers. By identifying this information, it is possible to quickly determine whether the target invoice is a known invoice type.

[0046] In one embodiment, if the target invoice is determined to be a known invoice type through preliminary identification, a preset target AI model combination is used to perform an invoice recognition operation on the target invoice to accurately extract the invoice information of the target invoice.

[0047] In one embodiment, if the target invoice is determined to be an unknown invoice type after preliminary identification, since it is impossible to determine whether it is applicable to the target AI model, a combination of multiple AI models can be used to identify the target invoice separately; and then the invoice information of the target invoice is determined from the identification results of the multiple AI model combinations.

[0048] In the above embodiment, for known invoice types, a preset combination of target AI models is used for identification, leveraging the collaborative work between models to accurately locate and extract invoice information. For unknown invoice types, a combination of multiple AI models is used for separate identification, and information is then extracted from the multiple identification results. This approach fully leverages the strengths of different models, not only ensuring accurate recognition of known invoice types but also significantly improving the accuracy of unknown invoice types through the integrated use of multiple models. This significantly enhances the efficiency, accuracy, and applicability of invoice recognition, effectively meeting the recognition needs of various invoice types.

[0049] In one embodiment, the target AI model combination includes: a layout analysis model, a text recognition model, and a structured parsing model.

[0050] After determining that the target invoice is of a known invoice type, the layout analysis model can be used to identify the target invoice's layout and identify the areas containing invoice information. These areas can be identified, such as the invoice header area and the item details area. By labeling each area, the invoice information likely to be contained within it can be clearly identified. For example, the invoice header area might contain information such as the invoice type, number, and invoice date, while the item details area might contain information such as the port or service name, specifications, quantity, unit price, and amount.

[0051] After the regions containing the invoice information are determined, a text recognition model can be used to perform text recognition on the invoice information in the regions to obtain the invoice information. Based on the identifiers corresponding to the regions, the invoice information corresponding to the identifiers is extracted from the regions.

[0052] Then, a structured parsing model is used to process the extracted invoice information and convert it into structured invoice information. The structured parsing model uses natural language processing techniques and parsing algorithms to classify, organize, and summarize the invoice information. Key data is extracted from the text content of the invoice information and structured and stored according to pre-set formats and rules. For example, this storage may utilize a JSON structure or a key-value structure. This structured parsing model makes the invoice information organized and clear, facilitating subsequent data analysis, storage, and interaction with other business systems.

[0053] In the above example, by sequentially processing the target invoice using a layout analysis model, a text recognition model, and a structured parsing model, the Surya model's powerful layout analysis capabilities are leveraged to precisely identify the invoice information area; the PaddleOCR model's excellent multilingual support and efficient text recognition capabilities are leveraged to accurately capture invoice information; and the OmniParser model's outstanding structured information extraction capabilities are leveraged to convert invoice information into structured data. This complete process makes the extraction of invoice information more comprehensive, accurate, and standardized, greatly facilitating subsequent data analysis, storage, and interaction with other business systems, significantly improving the quality and efficiency of invoice processing.

[0054] In one embodiment, for target invoices of unknown types, a parallel processing approach can be adopted, where multiple AI models are combined to identify the target invoice. After the multiple AI models have completed the identification of the target invoice, multiple recognition results are obtained. Through a voting mechanism, the invoice information is determined from the multiple recognition results.

[0055] Specifically, the voting mechanism involves statistically analyzing the recognition results of each AI model combination for each key information field on the invoice information, such as the invoice number, invoice date, amount, currency, etc., and using the result with the highest frequency as the final key information. For example, if the recognition results of the three AI models for the amount are "1000," "1000," and "100," respectively, the voting mechanism can determine "1000" as the final recognition result for the amount.

[0056] In one embodiment, after determining the invoice information of the unknown invoice type, the invoice information of the unknown invoice type can be further learned and trained, and the target AI model combination can be optimized. For example, the layout analysis model can learn the layout rules of the invoice type to improve the recognition accuracy of each area; the text recognition model will optimize the specific fonts, language types, etc. that may appear in the invoice type to enhance the recognition ability of these texts; the structured parsing model can learn the structural characteristics of the invoice information of the invoice type and optimize the parsing rules to extract the required structured information more accurately and completely. By continuously performing adaptive optimization training on the target AI model combination, it has the ability to accurately identify the invoice type. At this point, the invoice type can be determined as a new known invoice type.

[0057] In the above-mentioned embodiment, the method of extracting invoice information from multiple recognition results corresponding to a combination of multiple AI models through a voting mechanism comprehensively considers the recognition strengths of multiple models. When faced with complex invoice recognition tasks, different models may excel in certain aspects. The voting mechanism can integrate these strengths and select the most highly recognized recognition result as the final invoice information. This approach increases the probability of obtaining accurate invoice information from complex and diverse recognition results, enhances the reliability and credibility of invoice recognition results, reduces the potential errors caused by a single model, and improves overall recognition effectiveness.

[0058] See Figure 3 , Figure 3 An exemplary embodiment provides an invoice identification method, which may include the following steps:

[0059] Step S310: Obtain the target invoice uploaded by the user.

[0060] Step S320: perform invoice recognition based on the AI ​​model.

[0061] Among them, steps S310 and S320 can be implemented as follows Figure 2 In the method embodiment of steps S210 and S220,

[0062] Step S330: Verify the identified invoice information; if the verification fails, execute step S340; if the verification succeeds, execute step S350.

[0063] In one embodiment, the confidence level of the identified invoice information can be verified; if the confidence level obtained through verification is high, reaching or exceeding a preset confidence threshold, it indicates that the accuracy and reliability of the invoice information are effectively guaranteed, i.e., the verification is successful; if the confidence level obtained through verification is low, lower than the preset confidence threshold, it indicates that there is a certain degree of uncertainty and error risk in the invoice information, i.e., the verification fails.

[0064] Step S340: manual processing.

[0065] In one embodiment, a manual intervention mechanism will be triggered, and the target invoice and the identified invoice information may be sent to an operation and maintenance personnel for manual processing, and the operation and maintenance personnel will carefully review and verify the target invoice.

[0066] Step S350: Export the invoice information for subsequent tax declaration, financial accounting, and file management processes.

[0067] In one embodiment, if the verification is successful, the invoice information can be used in subsequent tax declaration, financial accounting, and file management processes.

[0068] In one embodiment, during the invoice information verification process, the fields to be verified can be first determined based on the target invoice type. Different invoice types contain different key information and format requirements, so the invoice information that can be extracted from the invoice may vary. For example, for a VAT invoice, the fields to be verified may include the invoice code, invoice number, invoice date, the buyer's and seller's taxpayer identification numbers, the amount, the tax amount, and the tax rate. For an ordinary invoice, the fields to be verified may include the invoice header, item content, and the amount.

[0069] Then, the fields to be verified in the invoice information are verified; if the verification fails, the target invoice is sent to the operation and maintenance personnel for manual processing; if the verification succeeds, the invoice information is used for subsequent processes.

[0070] In one embodiment, the verification of the invoice information may include an integrity check and a consistency check of the invoice information.

[0071] In one embodiment, the fields to be checked may include a first key field for integrity verification. The first key field may be a core component of the invoice information and can be used to determine the validity and usability of the invoice. For example, in a value-added tax invoice, basic information such as the invoice number, invoice date, and amount are typically listed as the first key field. The completeness of the invoice information can be determined by verifying whether the first key field is included in the invoice information. If the completeness reaches or exceeds a completeness threshold (e.g., 100% or 90%), the completeness verification is successful. If the completeness is below the completeness threshold, the completeness verification fails.

[0072] In one embodiment, the field to be verified may include a second key field for consistency verification. The second key field may be a portion of the invoice information that is closely related to the transaction information and can be used to determine the authenticity and compliance of the invoice. For example, information related to the seller and buyer can be used to verify the consistency of the second key field in the invoice information with the transaction information.

[0073] For different invoice types, the corresponding first key field and second key field are shown in the following table:

[0074]

[0075]

[0076] It should be noted that the first key field and the second key field may be the same, partially the same, or completely different.

[0077] In the above embodiment, a comprehensive invoice information quality assurance mechanism is established by determining the fields to be verified corresponding to the target invoice type, verifying these fields within the invoice information, and transferring verification failures to manual processing. During the invoice processing process, automatic verification can promptly detect potential errors, screen out problematic invoices, and hand them over to operations and maintenance personnel for targeted processing. This effectively prevents erroneous invoice information from entering subsequent business processes, ensures the accuracy and reliability of invoice information, reduces business risks caused by invoice errors, and improves the rigor and standardization of invoice processing.

[0078] See Figure 4 , Figure 4 An exemplary embodiment provides an invoice identification method, such as Figure 4 As shown, the Surya model is used as the layout analysis model, the PaddleOCR model is used as the text recognition model, and the OmniParser model is used as the structured parsing model. The method includes the following steps:

[0079] S410: The user uploads the invoice.

[0080] Users can upload the invoice to be identified in the form of images or documents to the mobile server.

[0081] S420, image preprocessing module.

[0082] The invoice images uploaded by users can be pre-processed in advance, such as image denoising, grayscale, binarization, tilt correction, image enhancement and other operations.

[0083] S430, intelligent scheduling module.

[0084] Based on invoice characteristics (such as invoice type and complexity) and system resource availability, the system determines whether to use a serial or parallel processing mechanism. It serves as a decision-making hub, rationally allocating processing tasks to achieve efficient invoice information extraction. For example, if the invoice type is known, serial processing is used; if the invoice type is unknown, parallel processing is used.

[0085] S440, serial processing mechanism.

[0086] A preset target AI model combination can be used for invoice recognition. The target AI model combination includes: Surya model, PaddleOCR model and OmniParser model.

[0087] Surya model: Identifies the invoice layout, locates several areas containing various invoice information, and marks them.

[0088] PaddleOCR model: Based on optical character recognition technology, it recognizes the text information in each located area and extracts text information.

[0089] OmniParser model: performs structured parsing on the text information extracted by the previous model, organizes the scattered text information into meaningful invoice information fields according to the business logic of the invoice, and finally outputs structured invoice information.

[0090] S450, parallel processing mechanism.

[0091] Based on the system's resources, the available engines can be determined. Each engine corresponds to a specific AI model combination. For example, Engine 1 corresponds to the Surya model + PaddleOCR model, Engine 2 corresponds to the PaddleOCR model + OmniParser model, and Engine 3 corresponds to the Surya model + PaddleOCR model + OmniParser model. Multiple engines simultaneously perform invoice recognition on invoices, obtaining recognition results from different perspectives. Then, through a pre-set invoice setting mechanism, invoice information is determined from the multiple recognition results.

[0092] S460, large language module.

[0093] Perform semantic understanding, further information integration, and error correction on the invoice information output by serial or parallel processing mechanisms. Leverage the natural language processing capabilities of large language models to ensure that invoice information conforms to business logic and language expression standards.

[0094] S470: Confidence verification.

[0095] Using specific algorithms and rules, the reliability and accuracy of invoice information is evaluated, and a confidence value is assigned. The confidence value is then determined to be above a preset confidence threshold to determine whether the invoice information meets the requirements.

[0096] S480, manual processing.

[0097] If the confidence check result falls below the preset threshold, it indicates that there may be errors or uncertainty in the invoice information. In this case, the invoice and identification information will be sent to human operators for manual verification and correction to ensure the invoice information is accurate.

[0098] S490, model optimization.

[0099] After obtaining accurate invoice information, whether through automated or manual processing, this data can be used to train and optimize previous models (such as the Surya model and PaddleOCR model). This continuously improves the model's ability to recognize and process invoices to accommodate a wider range of invoice types and more complex business scenarios.

[0100] Figure 5 This is a schematic structural diagram of a device provided by an exemplary embodiment. Figure 5 At the hardware level, the device includes a processor 502, an internal bus 504, a network interface 506, a memory 508, and a non-volatile memory 510. Of course, it may also include hardware required for other functions. One or more embodiments of this specification can be implemented based on software, such as the processor 502 reading the corresponding computer program from the non-volatile memory 510 into the memory 508 and then running it. Of course, in addition to software implementation, one or more embodiments of this specification do not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0101] Please refer to Figure 6 , the invoice recognition device can be used for Figure 5 The invoice recognition device may include: an invoice uploading module 601 and an invoice recognition module 602.

[0102] The invoice upload module 601 is used to obtain the target invoice uploaded by the user; the invoice identification module 602 is used to identify whether the target invoice is a known invoice type; if so, the preset target AI model combination is used to perform an invoice identification operation on the target invoice to extract the invoice information; if not, the preset multiple AI model combinations are used to perform identification operations on the target invoice respectively, and the invoice information is extracted from the multiple identification results corresponding to the multiple AI model combinations.

[0103] Furthermore, the target AI model combination includes: a layout analysis model, a text recognition model and a structured parsing model; the invoice recognition module 602 is used to use the layout analysis model to identify the layout of the target invoice, so as to determine several areas containing invoice information from the target invoice; use the text recognition model to perform text recognition on the invoice information in the several areas to obtain the invoice information; use the structured parsing model to convert the invoice information into structured invoice information.

[0104] Furthermore, the invoice recognition module 602 is used to extract invoice information from multiple recognition results corresponding to the multiple AI model combinations through a voting mechanism.

[0105] Furthermore, the layout analysis model includes a Surya model; the text recognition model includes a PaddleOCR model; and the structured parsing model includes an OmniParser model.

[0106] Furthermore, the invoice identification module 602 is further configured to determine a field to be verified corresponding to the invoice type of the target invoice;

[0107] Verify the fields to be verified in the invoice information; if the verification fails, send the target invoice to the operation and maintenance personnel for manual processing.

[0108] Furthermore, the field to be checked includes a first key field for integrity verification; the invoice identification module 602 is used to verify whether the invoice information contains the first key field.

[0109] Furthermore, the field to be checked includes a second key field for consistency verification; the invoice identification module 602 is used to obtain transaction information corresponding to the target invoice; and verify whether the second key field in the invoice information is consistent with the transaction information.

[0110] Based on the same concept as the above method, this specification also provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor implements the steps of the method described in any of the above embodiments by running the executable instructions.

[0111] Based on the same concept as the above method, this specification also provides a computer-readable storage medium on which computer instructions are stored. When the instructions are executed by a processor, the steps of the method described in any of the above embodiments are implemented.

[0112] Based on the same concept as the above method, this specification also provides a computer program product, including a computer program / instruction, which implements the steps of the method described in any of the above embodiments when executed by a processor.

Claims

1. A method for identifying invoices, characterized in that: The method comprises: Get the target invoice uploaded by the user; Identify whether the target invoice is a known invoice type; if so, use the preset target AI model combination to perform invoice recognition operations on the target invoice to extract invoice information; if not, use the preset multiple AI model combinations to perform recognition operations on the target invoice respectively, and extract the invoice information from the multiple recognition results corresponding to the multiple AI model combinations.

2. The method according to claim 1, characterized in that The target AI model combination includes: a layout analysis model, a text recognition model, and a structured parsing model; the invoice recognition operation is performed on the target invoice using the preset target AI model combination to extract invoice information, including: Using a layout analysis model to identify the layout of the target invoice, so as to determine a number of areas containing invoice information from the target invoice; Using a text recognition model, performing text recognition on the invoice information in the plurality of regions to obtain the invoice information; A structured parsing model is used to convert the invoice information into structured invoice information.

3. The method according to claim 1, characterized in that Extracting invoice information from the multiple recognition results corresponding to the multiple AI model combinations includes: Through a voting mechanism, invoice information is extracted from multiple recognition results corresponding to the combination of the multiple AI models.

4. The method according to claim 2, characterized in that The layout analysis model includes the Surya model; the text recognition model includes the PaddleOCR model; and the structured parsing model includes the OmniParser model.

5. The method according to claim 1, wherein The method further comprises: Determining a field to be verified corresponding to the invoice type of the target invoice; Verify the fields to be verified in the invoice information; if the verification fails, send the target invoice to the operation and maintenance personnel for manual processing.

6. The method according to claim 5, characterized in that The field to be verified includes a first key field for integrity verification; and verifying the field to be verified in the invoice information includes: Verify whether the invoice information contains the first key field.

7. The method according to claim 5, characterized in that The field to be verified includes a second key field for consistency verification; and verifying the field to be verified in the invoice information includes: Obtaining transaction information corresponding to the target invoice; Verify whether the second key field in the invoice information is consistent with the transaction information.

8. An electronic device, characterized in that: include: processor; A memory for storing processor-executable instructions; wherein the processor implements the steps of the method according to any one of claims 1 to 7 by executing the executable instructions.

9. A computer-readable storage medium, characterized in that Computer instructions are stored thereon, and when the instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7.

Images