Integrated financial intelligent terminal acquirer

Abstract

The application provides an integrated financial intelligent terminal acquirer, comprising: a self-interchange component receiving reimbursement documents, scanning and identifying the reimbursement documents, and extracting first key information of the reimbursement documents; an information processing component extracting second key information of the reimbursement documents, and processing the first key information and the second key information through a data processing engine; the processed first key information and the second key information are simultaneously electronically archived, and corresponding approval face sheets and accounting vouchers are automatically printed and archived; a function interface component provides new login authentication, receives reimbursement document interface authentication, and synchronizes the state of the reimbursement document interface. The application realizes full-process automation from receiving reimbursement applications, information processing, to data storage and state feedback; not only improves the efficiency of the reimbursement process, but also reduces manual intervention, reduces the error rate, and provides convenience in information management.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information processing terminals, in particular to an integrated financial intelligent terminal acquirer. BACKGROUND

[0002] With the development of information technology, the reimbursement system and the acquirer have become indispensable tools in enterprise operation. The existing technology includes a logistics type financial reimbursement acquirer, which realizes fast and accurate processing of reimbursement documents by integrating scanning, identification, data transmission and other modules. The device is usually composed of a scanner, a processor, a display and a printer, and the workflow covers document scanning, information identification, data processing and reimbursement confirmation. However, in the prior art, the reimbursement system and the acquirer operate independently, lacking effective data sharing and interaction mechanism. After the acquirer completes the delivery, the transaction data generated is not effectively connected with the reimbursement system, and there are still problems such as complicated process, unclear document responsibility, document rejection, low reimbursement efficiency, etc., which cannot realize the intelligent goal of the whole process of automatic acquirer, scanning, auditing, refunding and archiving.

[0003] The existing technology 1, application number: CN202310427650.3 discloses a power company financial reimbursement data processing system and method, which is used for the power reimbursement personnel to log in to the financial reimbursement data processing system through the company intranet for online financial reimbursement, without the need to face the power financial personnel for reimbursement, saving time for the power reimbursement personnel to perform financial reimbursement, greatly improving the efficiency of financial reimbursement. The power reimbursement personnel port is used for the power reimbursement personnel to log in to the financial reimbursement data processing system for financial reimbursement, the power financial personnel port is used for the power financial personnel to log in to the financial reimbursement data processing system for initial audit of financial reimbursement, and the power leader personnel port is used for the power leader personnel to log in to the financial reimbursement data processing system for final audit of financial reimbursement. Although a large amount of manual work is not required, the workload is small, the efficiency of financial reimbursement is greatly improved, and it is beneficial to fast financial reimbursement. However, it does not involve the connection relationship between the reimbursement system and the acquirer, resulting in complex and tedious reimbursement process and low reimbursement efficiency.

[0004] Prior art 2, application number CN202211535247.4, discloses a method and system for processing financial reimbursement forms. The method includes: receiving an original form submitted by the reimbursement applicant, the original form containing the applicant's name, department information, and form ID; writing the data from the original form into the corresponding financial intermediate node table at a preset node in the form processing flow; and, upon receiving a summary request submitted by the summary initiator, retrieving data from the financial intermediate node table to generate a financial summary table. While this can improve the efficiency of reimbursement form summarization and meet the content requirements of different levels of personnel for financial summary reports, the fact that reimbursement forms are still collected separately leads to unclear document responsibilities and a high document rejection rate.

[0005] Prior art three, application number: CN 202210380038.0, discloses a method, apparatus, equipment, and medium for processing financial reimbursements. The financial reimbursement system, based on a reimbursement request, obtains the credit rating of the employee to be reimbursed from the financial shared service center platform, corresponding to the employee's identity identifier in the reimbursement request; determines the financial reimbursement process corresponding to the employee's credit rating, and processes the reimbursement according to the document information in the reimbursement request; analyzes the reimbursement processing of the employee to be reimbursed, obtains the employee's indicators and corresponding scores, and reports these indicators and scores to the financial shared service center platform so that the platform can update the employee's credit rating based on these indicators and scores. While this solves the problem of incomplete employee credit score determination potentially leading to uncontrollable reimbursement quality during the reimbursement process, it is not a definitive solution. However, the reimbursement system is still separate from the acquiring machine, which provides limited help in standardizing reimbursement form filling and reimbursement processes, and further improvements are needed in reimbursement efficiency.

[0006] Current technologies 1, 2, and 3 suffer from cumbersome reimbursement processes, inaccurate form completion, high rejection rates, and low levels of automation across the entire reimbursement process. Therefore, this invention provides an integrated intelligent financial terminal acquiring machine, effectively combining the reimbursement system with the acquiring machine. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides an integrated intelligent financial terminal acquiring machine, comprising:

[0008] The self-service document submission component is responsible for receiving expense reports, scanning and recognizing them, and extracting the first key information from the expense reports, including the amount, date, and recipient.

[0009] The information processing component is responsible for extracting the second key information of the reimbursement document, and processing the first key information and the second key information through the data processing engine, which includes automatic verification, classification and arrangement; at the same time, the first key information and the second key information after processing are electronically archived, and the corresponding approval face sheet and accounting voucher are automatically printed and archived;

[0010] The functional interface component is responsible for providing new login authentication, receiving reimbursement document interface authentication and synchronizing reimbursement document state interface; wherein, the new login authentication includes Three Gorges line cloud code scanning authentication, face recognition authentication and card swiping authentication.

[0011] Optionally, the self-interchange component includes:

[0012] The document receiving module is responsible for receiving the completed reimbursement document and placing it in the scanning area for temporary storage of the reimbursement document.

[0013] The image processing module is responsible for triggering the image capture process when the reimbursement document is placed in the scanning area and the visual sensor detects the existence of the reimbursement; the image of the reimbursement document is captured using the visual sensor, and the image is preprocessed to obtain a preprocessed image.

[0014] The result output module is responsible for identifying the size, direction, quality and first key information of the reimbursement document from the preprocessed image; at the same time, a user interaction interface is provided to display operation instructions and feedback information through a touch screen or a display screen; wherein, the preprocessing includes denoising and contrast enhancement.

[0015] Optionally, the result output module includes:

[0016] The image selection submodule is responsible for generating multiple candidate images each time the image is automatically captured, and selecting an image as the main image for subsequent processing from the multiple candidate images after preprocessing.

[0017] The physical property submodule is responsible for determining the size of the reimbursement document according to the width and height of the main image, determining the orientation of the long side and the short side, determining the horizontal or vertical direction, and automatically adjusting the image direction according to the orientation; according to the set quality standard, it is judged whether the main image meets the processing requirements; if it meets, the next step is continued; if it does not meet, the user is prompted to reposition or rescan the document;

[0018] The information extraction submodule is responsible for extracting the first key information from the main image that meets the processing requirements, and feeding back the result to the user interface in real time to display the extraction result for user confirmation.

[0019] Optionally, the physical property submodule includes:

[0020] An edge detection unit is responsible for converting the main image into a grayscale image after the main image is selected, binarizing the image by using an adaptive Gaussian threshold, smoothing the image by using a Gaussian filter, and performing edge detection according to an edge intensity gradient by using an edge detection algorithm, and identifying the outer frame of the document in the image;

[0021] A size acquisition unit is responsible for applying polygon approximation to the extracted contour, smoothing the contour and simplifying the edge, and detecting the shape features thereof, calculating the minimum circumscribed rectangle of the contour, and calculating the width and height of the image by performing geometry on the extreme points of the contour;

[0022] An orientation adjustment unit is responsible for judging the orientation of the document according to the ratio of the width to the height, recording the document as horizontal if the width is greater than the height, otherwise recording the document as vertical, and automatically rotating the image by 90 degrees or 180 degrees according to the orientation information, and resampling the rotated image by using a bilinear interpolation algorithm or a nearest neighbor interpolation algorithm.

[0023] Optionally, the edge detection unit comprises:

[0024] A convolution operation subunit is responsible for performing edge detection on the main image after Gaussian smoothing, using two 3x3 convolution kernels to perform twice convolution operation on the original image to obtain the gradient of the main image, and calculating the gradient amplitude and direction of each pixel.

[0025] A gradient calculation subunit is responsible for comparing each pixel with its neighborhood in the gradient direction according to the calculated gradient direction; if the amplitude of the pixel is greater than the amplitude of the neighborhood pixel, the pixel is retained; otherwise, the intensity of the pixel is set to zero.

[0026] A threshold comparison subunit is responsible for setting a high threshold and a low threshold, judging the intensity of each pixel after non-maximum suppression, and tracing the connected edges in the image by using the connectivity of the pixels marked as strong edges, marking any pixel connected to the strong edge as an edge to form a complete edge map.

[0027] Optionally, the information extraction sub-module comprises:

[0028] A feature merging unit is responsible for labeling the main image that meets the processing requirements, determining the position and format of the first key information, selecting a pre-trained model to fine-tune on the labeled main image containing the reimbursement document, constructing an image pyramid to generate multiple versions of input images from different scales, extracting features from each scale of the input image, merging the deep features from different scales, and converting the extracted features into a fixed-length feature vector through a fully connected layer.

[0029] The text acquisition unit is responsible for positioning the key region using the bounding box, clarifying the relative position of the key region in the main image, using an OCR engine to recognize characters in the extracted key region, and acquiring text information. The recognized text information is subjected to rule judgment to ensure the rules of the first key information.

[0030] The information checking unit is responsible for analyzing the extracted text information in the context, judging that the amount is located in a specific area and the invoice number is in the header by setting a pattern matching rule, and using a named entity recognition algorithm to analyze the text to extract the first key information with more semantic context.

[0031] Optionally, the information processing component includes:

[0032] The content extraction module is responsible for extracting the second key information, including departure time, arrival time, departure location, arrival location, passenger, seat level, and ticket amount, from the scanned image of the reimbursement document by using the data processing engine OCR.

[0033] The engine classification module is responsible for automatically checking, classifying, and organizing the first key information and the second key information using an optical character recognition engine, an intelligent verification engine, a classification engine, and a data organization engine. At the same time, the user directly extracts the reimbursement document for processing. If the reimbursement document is rejected, the user takes the single and changes or supplements the attachments according to the prompt and then submits the reimbursement document again.

[0034] The archiving processing module is responsible for electronically archiving the approved reimbursement document, printing the approved reimbursement document, and archiving the accounting voucher.

[0035] Optionally, the engine classification module includes:

[0036] The engine setting submodule is responsible for independently setting each engine, and the reimbursement document to be processed is sent to the input queue for asynchronous processing. Each engine has an independent work queue for receiving tasks to be processed, and the message queue service distributes the tasks to be processed to the corresponding engine.

[0037] The queue service submodule is responsible for quickly converting the information of the reimbursement document submitted by the user into a processing task and sending it to the input queue when the user submits the reimbursement document. The message queue service takes the task from the input queue and distributes it to each work queue. Each engine independently obtains the task from the queue and processes it by listening to its work queue. At the same time, the engines can run concurrently, waiting for the next task, and after processing is completed, the results are written to the result storage or sent to the next link.

[0038] Result aggregation sub-module, responsible for aggregating the processing results of all engines in the data collation engine, performing overall data collation, storage and return.

[0039] Optionally, the engine classification module further comprises:

[0040] Data extraction sub-module, responsible for using the OCR engine to perform optical character recognition on the scanned reimbursement document image, converting the pixel data in the image into machine-readable text data, including image preprocessing, feature extraction, and character recognition; the output is a text data structure;

[0041] Preliminary verification sub-module, responsible for performing structured preliminary verification on the text data extracted by the OCR engine, including verifying whether the text format conforms to the pre-set pattern, checking whether the required fields exist, and identifying characters or strings that cannot be accurately recognized;

[0042] Logical verification sub-module, responsible for performing advanced logical verification on the data that passes the preliminary verification, cross-verified by pre-defined rules and business logic;

[0043] Abnormality processing sub-module, responsible for automatically marking all abnormal data found in the logical verification stage as objects that need attention, generating an error report that lists each non-compliance item found in detail; at the same time, relevant personnel will be notified of the abnormal situation so that they can intervene in a timely manner;

[0044] Feedback and correction sub-module, responsible for presenting the documents marked as abnormal in the user interface, allowing users to access the error report and view specific error information; a user-friendly interface will be provided to assist users in revising and supplementing abnormal data; after correction, the user can submit the revised document back to the system for re-verification;

[0045] Final confirmation sub-module, responsible for marking the document data that meets the requirements as valid after all verification steps are completed; at this time, the audited and confirmed document data will be stored in the database and marked as ready for the subsequent approval process.

[0046] Optionally, the functional components include:

[0047] State synchronization module, responsible for synchronizing the reimbursement document state to the collecting system when the reimbursement document is rejected for approval, the user withdraws, the user submits for approval, and the document is archived; for rejected or withdrawn reimbursement documents, the user performs the single processing through the self-service interface, and for archived documents, the physical archiving operation is performed; in addition, the reimbursement document in the submitted for approval state will not allow the user to perform the self-service single operation;

[0048] The recognition transformation module is responsible for upgrading and transforming the execution of reimbursement documents, and the transformation will realize multi-document recognition function to support parallel processing of different types of documents.

[0049] The data access control module is responsible for calling the authentication interface to obtain the access token according to the clientID and clientSecret obtained from the acquirer system; the token needs to be included in the HTTP request header in subsequent requests;

[0050] The image processing and conversion module is responsible for sending all image data to the reimbursement system for optical character recognition and intelligent audit after the acquirer completes the scanning of all reimbursement documents; according to the state recognition of the reimbursement documents, the specified reimbursement documents will be automatically printed, and each reimbursement document can be physically printed one by one.

[0051] The autonomous submission component of the present application receives and processes the reimbursement documents submitted by the user, performs scanning and recognition functions to extract the information of the reimbursement documents in digital form; automatically extracts and recognizes the first key information (such as amount, date and name). Significance: improves the efficiency of the reimbursement process, making the submission process more convenient and efficient for users; ensures accurate extraction of key data, reduces the risk of human input errors; realizes digital storage of information, laying a good foundation for subsequent data processing. The information processing component extracts the second key information of the reimbursement document, including possible project categories, expense types, etc.; the first key information and the second key information are checked, classified and arranged through the data processing engine; complete electronic archiving, automatically store the processed information, and print the approval face sheet and accounting voucher. Significance: through automated data processing, the processing efficiency and accuracy of the reimbursement document are improved, reducing the need for manual operation; provides a systematic archiving and auditing process, which helps enterprises better manage finances and generate reports; ensures tracking and integration of all reimbursement data, providing convenience for compliance audits. The functional interface component provides a login authentication interface for user identity verification, ensuring that only authorized users can access the system; handles the interface authentication of the reimbursement document to ensure data security and compliance of access; provides an interface to synchronize the status of the reimbursement document to provide real-time feedback on the progress of the reimbursement audit. Significance: through secure identity authentication and interface authentication, the privacy and data security of users are protected, preventing unauthorized access; real-time update of the status of the reimbursement document improves user experience, allowing users to keep abreast of the progress of the reimbursement, increasing transparency; it is beneficial to integrate with other systems, improving the coherence and efficiency of the overall process and achieving greater business value.

[0052] Other features and advantages of the present application will be set forth in the description that follows, and in part will be apparent from the description, or can be learned by practice of the application. The purposes and other advantages of the present application will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.

[0053] The technical solutions of the present application are described in further detail below with the help of the accompanying drawings and examples. BRIEF DESCRIPTION OF DRAWINGS

[0054] The accompanying drawings are included to provide a further understanding of the present application and are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and together with the description serve to explain the present application. In the drawings:

[0055] Figure 1 It is an integrated financial intelligent terminal acquirer block diagram in the embodiment 1 of the present application;

[0056] Figure 2 It is a self-service component block diagram in the embodiment 2 of the present application;

[0057] Figure 3 It is a result output module block diagram in the embodiment 3 of the present application;

[0058] Figure 4 It is a physical property sub-module block diagram in the embodiment 4 of the present application;

[0059] Figure 5 It is an edge detection unit block diagram in the embodiment 5 of the present application;

[0060] Figure 6 It is an information extraction sub-module block diagram in the embodiment 6 of the present application;

[0061] Figure 7 It is an information processing component block diagram in the embodiment 7 of the present application;

[0062] Figure 8 It is an engine classification module block in the embodiment 8 of the present application Figure 1 ;

[0063] Figure 9 It is an engine classification module block in the embodiment 9 of the present application Figure 2 ;

[0064] Figure 10 It is an engine classification module block in the embodiment 10 of the present application Figure 3 ;

[0065] Figure 11 It is a function interface component block diagram in the embodiment 11 of the present application. DETAILED DESCRIPTION

[0066] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0067] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the embodiments of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0068] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0069] Example 1: As Figure 1 As shown, this embodiment of the invention provides an integrated financial intelligent terminal acquiring machine, comprising:

[0070] The self-service document submission component is responsible for receiving expense reports, scanning and recognizing them, and extracting the primary key information from the reports, including the amount, date, and recipient's name.

[0071] The information processing component is responsible for extracting the second key information from the expense report and processing the first and second key information through the data processing engine. The processing includes automatic verification, classification and organization. At the same time, the processed first and second key information is electronically archived and the corresponding approval form and accounting voucher are automatically printed for archiving.

[0072] The functional interface component is responsible for providing interfaces for adding login authentication, receiving expense reimbursement documents, and synchronizing expense reimbursement document status; among them, adding login authentication includes Three Gorges Cloud QR code authentication, facial recognition authentication, and employee card swiping authentication.

[0073] The working principle and beneficial effects of the above technical solution are: the autonomous submission component of the embodiment receives the reimbursement documents, and simultaneously scans and identifies the reimbursement documents and extracts the first key information of the reimbursement documents; the first key information includes amount, date, and name, etc.; the information processing component extracts the second key information of the reimbursement documents, and processes the first key information and the second key information through the data processing engine, including automatic verification, classification, and arrangement; the first key information and the second key information after processing are electronically archived, and the corresponding approval face sheet and accounting voucher are automatically printed and archived; the functional interface component provides new login authentication, receives reimbursement document interface authentication, and synchronizes the reimbursement document state interface. The autonomous submission component of the above scheme receives and processes the reimbursement documents submitted by the user, performs scanning and identification functions, and extracts the information of the reimbursement documents in a digital form; the first key information (such as amount, date, and name) is automatically extracted and identified. Significance: It improves the efficiency of the reimbursement process, making the submission process of the user more convenient and fast; it ensures the accurate extraction of key data and reduces the risk of human input errors; it realizes the digital storage of information and lays a good foundation for subsequent data processing. The information processing component extracts the second key information of the reimbursement documents, including possible project categories and expense types; the first key information and the second key information are verified, classified, and arranged through the data processing engine; electronic archiving is completed, the processed information is automatically stored, and the approval face sheet and accounting voucher are printed. Significance: Through automated data processing, it improves the processing efficiency and accuracy of the reimbursement documents, reduces the need for manual operation; it provides a systematic archiving and auditing process, which helps enterprises better manage finances and generate reports; it ensures the tracking and integration of all reimbursement data, making it easier for compliance audits. The functional interface component provides a login authentication interface for user identity verification, ensuring that only authorized users can access the system; it processes the interface authentication of the reimbursement documents to ensure data security and compliance of access; it provides an interface for synchronizing the state of the reimbursement documents to provide real-time feedback on the audit progress of the reimbursement. Significance: Through secure identity authentication and interface authentication, the privacy and data security of users are protected, preventing unauthorized access; real-time updates of the reimbursement document state improve user experience, allowing users to keep abreast of the reimbursement progress and increase transparency; it is beneficial for integration with other systems, improving the coherence and efficiency of the overall process and achieving greater business value.

[0074] In summary, the entire integrated financial intelligent terminal acquirer of the embodiment realizes full-process automation from reimbursement application reception, information processing, to data storage and state feedback through the cooperative work of the autonomous submission component, the information processing component, and the functional interface component. Not only does it improve the efficiency of the reimbursement process, but it also reduces manual intervention and reduces the error rate, while providing convenience in information management. It helps enterprises achieve higher accuracy, efficiency, and compliance in financial management, improving overall operational efficiency.

[0075] The embodiment constructs a data sharing mechanism, uses a micro-service architecture, and uses service governance to ensure smooth cooperation and communication between services, including service registration, discovery, configuration, and routing. Service deployment pursues automation and independence, and containerization technology is the key. Service fault tolerance guarantees stability through timeout, retry, and circuit breaker strategies. Image file transfer is used, and encoding converts image information into a format that can be stored, transmitted, and processed in a computer system into a dream database. Data transmission uses JSON format for encapsulation, and the interface uses OAuth 2.0 authentication mechanism for Three Gorges Row Cloud single sign-on control. In addition, the interface supports the HTTPS protocol, uses SSL encryption technology to ensure the security of data transmission, meets the data security control of the acquirer system background management system, and realizes the effective combination of the reimbursement system and the acquirer.

[0076] Embodiment 2: as shown in Figure 2 Based on the embodiment 1, the autonomous delivery component provided by the embodiment of the application comprises:

[0077] A document receiving module is responsible for receiving the completed reimbursement document and placing it in the scanning area for temporary storage of the reimbursement document.

[0078] An image processing module is responsible for triggering the image capture process when the reimbursement document is placed in the scanning area and the visual sensor detects the existence of the reimbursement. The image processing module captures the image of the reimbursement document using the visual sensor, pre-processes the image, and obtains the pre-processed image.

[0079] A result output module is responsible for identifying the size, direction, quality, and first key information of the reimbursement document from the pre-processed image. The result output module also provides a user interaction interface that displays operation instructions and feedback information through a touch screen or a display screen. The pre-processing includes denoising and contrast enhancement.

[0080] The working principle and beneficial effects of the above technical solution are: the document receiving module of the embodiment receives the filled reimbursement document and places it within the scanning area for temporary storage of the reimbursement document; the image processing module uses a visual sensor to capture the image of the reimbursement document, pre-processes the image to obtain a pre-processed image; the result output module identifies the size, direction, quality and first key information of the reimbursement document from the pre-processed image; a user interaction interface is also provided to display operation instructions and feedback information through a touch screen or a display screen; the pre-processing includes denoising and contrast enhancement, etc. The document receiving module of the above scheme ensures the physical management of the reimbursement document before scanning, preventing disordered placement and loss of the document through a standardized receiving process. Significance: It provides a clear physical interface, making it easy for users to submit reimbursement documents and reducing the risk of misplacing or losing the documents; through the temporary storage function, the efficiency of subsequent image processing is improved, ensuring that the required documents are properly handled and placed before scanning; it provides a stable and reliable source of documents for subsequent processing modules, improving the stability of the overall system. The image processing module uses a visual sensor to capture the image of the reimbursement document, ensuring that the image quality is high enough for subsequent processing; the captured image is pre-processed, including denoising, contrast enhancement, etc., to obtain a clear and easily identifiable pre-processed image. Significance: Image processing improves image quality, providing accurate and clear input for the recognition module, thereby improving the accuracy of subsequent data extraction; denoising and contrast enhancement can eliminate environmental light and printing quality interference, ensuring the stability and consistency of recognition; the effect of pre-processing is crucial for the quality of subsequent data extraction and analysis. The result output module extracts the size, direction, quality and first key information (such as amount, date, and name) of the reimbursement document from the pre-processed image; a user interaction interface is provided to display operation instructions and feedback information through a touch screen or a display screen, including confirmation of the scanning result or suggestion for re-scanning. Significance: The ability to extract key data directly affects the degree of automation and accuracy of the reimbursement process, reducing manual intervention and improving efficiency; through the user interaction interface, user experience is improved, allowing users to visually see operation prompts and feedback, enhancing the controllability of the operation; a feedback mechanism is provided, allowing users to quickly respond when the recognition result is not satisfactory, improving subsequent operations and ensuring the final accuracy of the data.

[0081] In summary, the document receiving module, image processing module and result output module in the autonomous submission assembly of the embodiment work together to achieve efficient and accurate processing of reimbursement documents. The document receiving module ensures the standardization of physical management, the image processing module improves the clarity of the image, and the result output module is responsible for extracting key information and interacting with the user. This modular design not only improves the functionality and user experience of the system, but also lays a solid foundation for the automation of the entire reimbursement process.

[0082] Example 3: As Figure 3 As shown, based on Embodiment 2, the result output module provided in this embodiment of the invention includes:

[0083] The image selection submodule is responsible for generating multiple candidate images for each automatic capture, and selecting the image as the main image for subsequent processing from the multiple preprocessed candidate images.

[0084] The physical characteristics submodule is responsible for measuring the width and height of the main image to determine the size of the expense report; determining the orientation of its long and short sides to determine whether it is horizontal or vertical, and automatically adjusting the image orientation accordingly; judging whether the main image meets the processing requirements according to the set quality standards; if it does, proceed to the next step; if it does not, prompt the user to reposition or rescan the document.

[0085] The information extraction submodule is responsible for extracting the first key information from the main image that meets the processing requirements. For each key information extracted, the result is immediately fed back to the user interface and displayed for user confirmation.

[0086] The working principle and beneficial effects of the above technical solution are as follows: In this embodiment, the image selection submodule generates multiple candidate images for each automatic capture. From these pre-processed candidate images, it selects the image as the primary image for subsequent processing. The physical characteristics submodule measures the width and height of the primary image to determine the size of the expense report; it determines the orientation of its long and short sides to determine whether it is horizontal or vertical, and automatically adjusts the image orientation accordingly; based on the set quality standards, it judges whether the primary image meets the processing requirements; if it does, it continues to the next step; if it does not, it prompts the user to reposition or rescan the document; the information extraction submodule extracts the first key information from the primary image that meets the processing requirements. For each key information extracted, the result is immediately fed back to the user interface, displaying the extraction result for user confirmation. The image selection submodule of the above solution generates multiple candidate images for image preprocessing, selects the best-performing primary image for subsequent processing, and automatically compares the images using multiple evaluation criteria (such as sharpness and contrast) to ensure that the selected image contains valid information. Significance: Improves the accuracy and reliability of information extraction by filtering unsuitable candidate images, ensuring subsequent operations are based on high-quality images; reduces the error rate, increases the system's automation and intelligence, and reduces the need for manual intervention; provides a better user experience because the information ultimately confirmed by the user comes from multiple filtered candidate images, rather than randomly generated images. The physical characteristics submodule measures the width and height of the reimbursement document in real time based on the selected main image, calculates and determines its size; determines the image orientation, identifies the long and short sides, and automatically adjusts the image orientation to ensure consistency; and performs image quality assessment, judging whether it meets processing requirements based on pre-set standards (such as brightness, contrast, and sharpness). Significance: Ensures the accuracy of subsequent information processing by clearly determining size and orientation, avoiding information extraction errors caused by incorrect image orientation; through the quality assessment mechanism, it can promptly identify images that do not meet processing standards, providing feedback to the user to prevent the spread of erroneous data and effectively reducing the risk of reimbursement processing; improves the system's intelligent recognition capabilities, increases automation, reduces user operation steps, and makes the user experience more user-friendly. The information extraction submodule extracts key information, such as reimbursement amount, date, and recipient details, from the main images that meet the processing requirements. It provides real-time feedback of the extraction results to the user interface, allowing users to instantly see the system's processing outcome and make confirmations and corrections. Significance: It achieves efficient data extraction, helping users quickly obtain the necessary key information and improving the overall speed of reimbursement processing; through real-time feedback, it enhances user engagement and operational transparency, allowing users to confirm extraction results in real time and increasing their trust in the system; through the information extraction module, data is promptly embedded into the system's workflow, making subsequent reimbursement approval and financial management processes smoother and more efficient.

[0087] To sum up, in the result output module, the image selection sub-module, the physical property sub-module and the information extraction sub-module respectively achieve the purpose of accurate extraction of reimbursement document information through their respective technical effects. Through the cooperative work of each sub-module, the entire module not only improves the intelligence and efficiency of the system, but also significantly improves user experience, data accuracy and processing speed, and achieves the best practice of automatic reimbursement processing.

[0088] Embodiment 4: as shown in Embodiment 3, on the basis of Embodiment 3, the physical property sub-module provided by the embodiment of the application comprises: Figure 4

[0089] The edge detection unit is responsible for converting the main image into a gray-scale image after the main image is selected, binarizing the image by using an adaptive Gaussian threshold, smoothing the image by using a Gaussian filter, performing edge detection according to an edge intensity gradient by using an edge detection algorithm, and identifying the outer frame of the document in the image.

[0090] The size acquisition unit is responsible for applying polygon approximation to the extracted contour, smoothing the contour and simplifying the edge, and detecting the shape characteristics; calculating the minimum circumscribed rectangle of the contour, performing geometry by finding the extreme points of the contour, and calculating the width and height of the image.

[0091] The orientation adjustment unit is responsible for judging the orientation of the document according to the ratio of the width and the height; if the width>the height, recording as horizontal; otherwise, recording as vertical; automatically rotating the image by 90 degrees or 180 degrees according to the orientation information, and resampling the rotated image by using a bilinear interpolation algorithm or a nearest neighbor interpolation algorithm.

[0092] ​The working principle and beneficial effects of the technical solution are as follows: the edge monitoring unit of the embodiment converts the main image into a grayscale image after the main image is selected, binarizes the image by using an adaptive Gaussian threshold; smoothes the image by using a Gaussian filter, detects the edge according to the edge intensity gradient, and detects the edge by using a more efficient edge detection algorithm; identifies the outer frame of the document in the image; the size acquisition unit applies polygon approximation to the extracted contour, smoothes the contour and simplifies the edge, and detects the shape feature; calculates the minimum circumscribed rectangle of the contour, performs geometry by finding the extreme points of the contour, and calculates the width and height of the image; the orientation adjustment unit judges the orientation of the document according to the ratio of the width and height; if the width>height, it is recorded as horizontal; otherwise, it is recorded as vertical; according to the orientation information, the image is automatically rotated by 90 degrees or 180 degrees, and the resampling of the rotated image is performed by using a bilinear interpolation or a nearest neighbor interpolation algorithm. The edge monitoring unit of the above scheme converts the main image into a grayscale image, reduces the data complexity, and lays a foundation for subsequent processing; based on the local pixel brightness feature, the image under different lighting conditions can accurately extract the edge and enhance the feature contrast; effectively removes the noise in the image while maintaining the clarity of the edge, and improves the accuracy of subsequent edge detection; application of advanced edge detection algorithm improves the accuracy of identifying the outer frame of the document. Significance: Ensures that clear and accurate contour information is obtained in the subsequent processing stage, improving the reliability of the entire processing process; through adaptive processing and optimization algorithm, the system can adapt to various environmental conditions, realizing stronger universality, which is crucial for the light changes that may occur in actual use scenarios; reduces the risk of misidentification and ensures that the extracted edge information accurately reflects the true shape of the document. The size acquisition unit simplifies the extracted contour, reduces unnecessary complexity, and maintains the features of the shape, making the edge processing more efficient; the extreme points of the contour are found by geometric calculation, the complex contour is simplified to a rectangle, and the width and height are calculated. Significance: Ensures accurate identification of the size of the reimbursement document, and provides necessary data support for subsequent processing steps; polygon approximation reduces the calculation pressure and improves the processing speed, ensuring data accuracy while improving overall efficiency; through effective geometric calculation, the actual size of the image is accurately judged, providing a more solid foundation for information extraction and subsequent data processing. The orientation adjustment unit automatically judges the orientation (horizontal or vertical) of the document by analyzing the ratio of the width and height, and judges based on geometric features; if the orientation is detected to be incorrect, the system can automatically perform 90-degree or 180-degree rotation and resampling by using a bilinear interpolation or a nearest neighbor interpolation algorithm to restore the image quality.Significance: Automated orientation recognition and adjustment minimizes user intervention, making system operation smoother and more efficient, and improving user experience; it ensures that subsequent information extraction and data processing are based on a consistent format, avoiding information errors and processing delays caused by image orientation issues; and through high-quality interpolation technology, it ensures that the rotated image does not exhibit significant distortion, maintaining the clarity and accuracy of the extracted information.

[0093] In summary, the collaborative operation of the edge detection unit, size acquisition unit, and orientation adjustment unit within the physical characteristics submodule in this embodiment enables the system to extract clear and accurate size and orientation information from images. This not only enhances the efficiency and accuracy of data processing but also improves the overall user experience, playing a crucial role in intelligent and automated processing. The entire system can efficiently process expense reports, ensuring accurate information extraction and smooth subsequent data processing.

[0094] Example 5: Figure 5 As shown, based on Embodiment 4, the edge detection unit provided in this embodiment of the invention includes:

[0095] The convolution operation subunit is responsible for edge detection on the Gaussian-smoothed main image. It uses two 3x3 convolution kernels to perform two convolution operations on the original image to obtain the gradient of the main image; it calculates the gradient magnitude and direction of each pixel.

[0096] The gradient calculation subunit is responsible for comparing each pixel with its neighborhood along the calculated gradient direction. If the magnitude of a pixel is greater than the magnitude of its neighboring pixels, the pixel is retained; otherwise, its intensity is set to zero.

[0097] The threshold comparison subunit is responsible for setting a high threshold and a low threshold. After non-maximum suppression, it judges the intensity of each pixel. For pixels marked as strong edges, it uses their connectivity to trace the connected edges in the image. Any pixel connected to a strong edge is marked as an edge, forming a complete edge map.

[0098] If the intensity is greater than the high threshold, it is marked as a strong edge;

[0099] If the intensity is less than the low threshold, it is marked as non-edge;

[0100] If the intensity is between the low and high thresholds, then whether it is an edge is determined by connectivity; pixels connected to strong edges are marked as edges.

[0101] The working principle and beneficial effects of the technical solution are as follows: the convolution operator unit of the embodiment performs edge detection on the main image after Gaussian smoothing, uses two 3x3 convolution kernels to perform twice convolution operation on the original image to obtain the gradient of the main image; the gradient amplitude and direction of each pixel are calculated; the gradient calculation subunit compares each pixel with its neighborhood in the gradient direction according to the calculated gradient direction; if the amplitude of the pixel is greater than that of the neighborhood pixel, the pixel is retained; otherwise, the intensity of the pixel is set to zero; the threshold comparison subunit sets a high threshold and a low threshold to judge the intensity of each pixel after non-maximum suppression; for the pixels marked as strong edges, the connected edges in the image are tracked by using the connectivity, any pixel connected with the strong edge is marked as an edge to form a complete edge map; if the intensity is greater than the high threshold, the pixel is marked as a strong edge; if the intensity is less than the low threshold, the pixel is marked as a non-edge; if the intensity is between the low threshold and the high threshold, whether the pixel is an edge is determined according to the connectivity, and the pixel connected with the strong edge is marked as an edge. The convolution operator unit applies two 3x3 convolution kernels to the image after Gaussian smoothing, and the convolution operation can extract the edge information of the image. Specifically, the convolution kernels focus on the horizontal and vertical gradients respectively; after convolution, the gradient amplitude and direction of each pixel are calculated by using the obtained horizontal and vertical gradients. Significance: the image information is converted into gradient information, so that the subsequent processing can focus on the part with strong changes (i.e. the edge) in the image, and important basic data are provided for the non-maximum suppression and edge connection; by calculating the gradient of each pixel, the intensity and direction of different features in the image can be effectively obtained, thereby providing accurate basis for subsequent edge detection. The gradient calculation subunit compares the amplitude of each pixel with the amplitude of its neighborhood pixels (adjacent pixels along the gradient direction) according to the calculated gradient direction. If the current pixel is a local maximum, the intensity of the pixel is retained; otherwise, the intensity of the pixel is set to zero. Through non-maximum suppression, the edge width is reduced, and only the pixels representing the edge are retained, so that the edge performance is refined. Significance: by eliminating the non-edge pixels, the accuracy and precision of edge detection are significantly improved, and the false detection points are reduced; only the local maximum is retained, which is helpful to form clear and detailed edge lines; the contour of each edge in the finally obtained edge map is accurate and clear, and high-quality data basis is provided for the subsequent processing links (such as thresholding and edge connectivity analysis). The threshold comparison subunit sets a high threshold and a low threshold to classify pixels and distinguish strong edges, non-edges and weak edges (pixels between the two). By analyzing the connectivity between strong edges and weak edges, the weak edge pixels connected with the strong edges are marked as edges, so that complete edge connection is formed.Significance: Ensure that only pixels connected to strong edges are identified as edges in the final output edge image, which effectively eliminates isolated noise points and improves the integrity and accuracy of the image; By using the connectivity method, it allows local edges (weak edges) to form effective connections with strong edges, ensuring the continuity of important features during image segmentation or subsequent identification process, greatly enhancing the quality of edge detection.

[0102] In summary, the various sub-units of the edge detection unit of the present embodiment work together to effectively convert from image to edge information. The convolution operation sub-unit extracts edge features and calculates gradients through edge detection; the gradient calculation sub-unit refines the edges to ensure that only true edge pixels are retained; the threshold comparison sub-unit further enhances the edge continuity and overall quality of the image. The edge detection process not only improves the accuracy and clarity of edge recognition, but also lays a solid foundation for subsequent image processing and information extraction.

[0103] Embodiment 6: As shown in the embodiment 3, on the basis of the embodiment 3, the information extraction sub-module provided by the present embodiment comprises: Figure 6

[0104] The feature merging unit is responsible for labeling the main image that meets the processing requirements and determining the position and format of the first key information; a pre-trained model is selected to fine-tune on the labeled main image containing the reimbursement document; an image pyramid is constructed to generate multiple versions of input images from different scales, and feature extraction is performed on each scale of input image; depth features from different scales are merged; the extracted features are converted into fixed-length feature vectors through a fully connected layer;

[0105] The text acquisition unit is responsible for positioning the key area using the bounding box to clarify the relative position of the key area in the main image; using an OCR engine, character recognition is performed on the extracted key area to obtain text information; the recognized text information is subjected to rule judgment to ensure the rules of the first key information;

[0106] Date format: For example, check if it is in the format of YYYY-MM-DD or MM / DD / YYYY, etc.

[0107] Amount format: Check if the amount contains a currency symbol (such as $ or ¥) and appropriate decimal places (such as 123.45);

[0108] Invoice number format: Check the specific structure of the invoice number (such as the length of the combination of letters and numbers);

[0109] ​The information checking unit is responsible for analyzing the extracted text information in context, judging that the amount is located in a specific area and the invoice number is in the header by setting pattern matching rules, and using named entity recognition algorithm to analyze the text to extract the first key information with more semantic context.

[0110] The working principle and beneficial effects of the above technical solution are: the feature merging unit of the embodiment labels the main image meeting the processing requirements, determines the position and format of the first key information, selects a pre-trained model, fine-tunes it on the labeled main image containing the reimbursement document, constructs an image pyramid, generates multiple versions of input images from different scales, extracts features from each scale of input image, merges deep features from different scales, converts the extracted features into fixed-length feature vectors through a fully connected layer, the text acquisition unit uses a bounding box to locate the key area and clarify the relative position of the key area in the main image, uses an OCR engine to recognize characters in the extracted key area and obtain text information, performs rule judgment on the recognized text information to ensure the rules of the first key information, such as date format: check if it is in the format of YYYY-MM-DD or MM / DD / YYYY, amount format: check if the amount contains a currency symbol (such as $ or ¥) and appropriate decimal places (such as 123.45), invoice number format: check the specific structure of the invoice number (such as the length of the combination of letters and numbers), and the information checking unit analyzes the extracted text information in the context, judges that the amount is located in a specific area and the invoice number is in the header through setting pattern matching rules, and uses named entity recognition algorithm to analyze the text to extract the first key information with more semantic context. The feature merging unit of the above scheme labels the main image, determines the position and format of the key information, selects a pre-trained model for fine-tuning to adapt to specific types of reimbursement documents, constructs an image pyramid to generate input images of different scales to ensure that different sizes and inclined texts can be handled, merges the features extracted from different scales to form a comprehensive feature vector, and through labeling, the position of the key information (such as invoice number, amount and date) in the image can be determined clearly, improving the accuracy of subsequent extraction, fine-tuning the pre-trained model to adapt to specific tasks, improving the effect and accuracy of feature extraction, constructing an image pyramid to obtain multi-scale features to enhance the robustness of the model to images, and effectively recognizing texts of different sizes and directions, and merging multi-level features into a unified feature vector to improve the expression ability of the recognition model. Significance: It lays a good foundation for subsequent text extraction and realizes efficient and accurate information extraction on specific types of reimbursement documents, which greatly promotes the automation and intelligent business process. The text acquisition unit can accurately locate the key information through bounding box analysis, making the information extraction more targeted, applies OCR technology to quickly convert the text in the image into a machine-readable text format, improves the extraction efficiency, and performs rule checking on the extracted text to ensure that it meets certain format standards (such as date, amount, etc.). Significance: It ensures the accuracy and effectiveness of information extraction, thereby reducing the burden of data cleaning and processing in the later stage and providing reliable data input in the automated reimbursement process.The information checking unit can more intelligently understand the relationship between information through context analysis, improve the accuracy of extraction, check the extracted information with the expected format and position relationship, effectively exclude incorrect information and context inconsistency, and extract more meaningful context data through NER to enrich the reimbursement information and improve the completeness of the information. Meaning: further improve the correctness and intelligence of the extracted information, provide higher quality basic data for subsequent data processing and decision-making, and enhance the transparency and credibility of the automatic reimbursement system.

[0111] The pre-training model acquisition method in this embodiment is trained on a large-scale data set (such as ImageNet) and has good feature extraction capability. The following are the steps for obtaining the pre-training model: select multiple commonly used deep learning architectures such as ResNet, EfficientNet, VGG, and MobileNet; use the standard library in the popular deep learning framework (such as PyTorch or TensorFlow) to directly download the pre-training model from the model library; and further train on the labeled data set containing the reimbursement documents. The last few layers of the model (such as the fully connected layer) need to be modified to adapt to specific tasks, such as outputting different categories. The training steps are: prepare the data loader to input the training data into the model; set the loss function and optimizer; start the training process, and update the model parameters through backpropagation and optimization algorithms.

[0112] The steps for constructing the image pyramid are as follows: assuming that there is an input image, first find out its original size (width and height); scale (usually between 0.5 and 1.0); for each scaling ratio, use the image scaling function to generate the corresponding image; for each pyramid level, different size images are extracted to generate corresponding feature representations; the features of each scale will be input to the subsequent network layer.

[0113] The merging of multi-scale features can enhance the comprehensive feature representation capability of the model. In the fine-tuned deep learning model, a feature vector is generated for each scale of image input. The features can be obtained through the output of the last few convolutional layers; at the same time, the intermediate layers of feature extraction can also be designed in the model to obtain the feature output; concatenation: the feature vectors from each scale are concatenated column by column (or row by row) to form a larger feature vector; finally, the merged feature vector is input to the fully connected layer to generate a fixed-length output feature vector, which provides input for subsequent classification or regression tasks;

[0114] To sum up, through the cooperation of the feature merging unit, the text acquisition unit and the information checking unit, the information extraction sub-module can extract key information from the complex structure of the reimbursement form with high accuracy, quickly and intelligently, reduce manual intervention and improve the automation degree of the process. The human cost is reduced, the error rate is reduced, the data quality is improved, and timely and reliable data support is provided for the financial management and business flow of enterprises or organizations.

[0115] Embodiment 7: as shown in Embodiment 1, on the basis of the information processing component provided by the embodiment of the application, the information processing component comprises: Figure 7 The content extraction module is responsible for extracting second key information from the scanned image of the reimbursement form by the data processing engine OCR, and the second key information includes departure time, arrival time, departure place, arrival place, passenger, seat level and ticket amount, etc.

[0116] The engine classification module is responsible for automatically checking, classifying and arranging the first key information and the second key information by using an optical character recognition engine, an intelligent checking engine, a classification engine and a data arrangement engine; at the same time, the user directly extracts the reimbursement form for processing, if the reimbursement form is rejected, the user takes the form by himself / herself, and according to the prompt, replaces or supplements the attachments, and then delivers the reimbursement form again.

[0117] The archiving processing module is responsible for electronically archiving the approved reimbursement form, printing the approved reimbursement form, and archiving the accounting voucher.

[0118]

[0119] ​The working principle and beneficial effects of the above technical solution are: the content extraction module of the embodiment extracts second key information from the scanned image of the reimbursement document through the data processing engine OCR, which includes departure time, arrival time, departure place, arrival place, passenger, seat level, and ticket amount; the engine classification module uses optical character recognition engine, intelligent verification engine, classification engine, and data arrangement engine to automatically verify, classify, and arrange the first key information and the second key information; at the same time, the user directly extracts the reimbursement document for processing, if the reimbursement document is rejected, the user takes the document independently, and according to the prompt, replaces or supplements the attachments, and then delivers the reimbursement document again; the archiving processing module archives the approved reimbursement document electronically, prints the approval form, and archives the accounting voucher. The content extraction module of the above scheme can quickly identify and extract key information from the scanned image through OCR technology, reducing the time of manual input and verification; combined with OCR and pre-training model fine-tuning, it can process various fonts, formats and layouts, thereby improving the recognition accuracy; it can process a large number of reimbursement documents, improving the overall approval and management efficiency. Significance: significantly reduces the artificial workload, speeds up the information processing speed, and enables enterprises or organizations to more efficiently manage expense reimbursement; as it can automatically process complex expense reimbursement information, it reduces the error rate and improves the quality and reliability of the data. The engine classification module uses the intelligent verification engine to verify the extracted information according to rules and logic, ensuring the validity and consistency of the information; the classification engine classifies the information according to predetermined standards for subsequent processing and archiving; provides a function for users to directly extract and process reimbursement documents, dynamically optimizes through a feedback mechanism (such as rejection reason prompt), and users can change or supplement the attachments according to system feedback. Significance: improves the flexibility of information processing, enabling rapid response to different types of reimbursement documents; enhances the transparency and user experience of the reimbursement process, allowing users to promptly understand the processing status and improving overall satisfaction. The archiving processing module stores the compliantly approved reimbursement documents in the electronic system, making information query and management more convenient, avoiding paper document loss and damage; provides paper support for approved reimbursement documents and vouchers, facilitating subsequent financial audit and inspection; systematizes all approved reimbursement information, further improving data traceability and manageability. Significance: through electronic means, it enhances the security and efficiency of data management, reduces storage costs and space requirements, and improves information retrieval speed; supports financial reimbursement compliance and audit, reducing the risks associated with manual management.

[0120] In this embodiment, the OCR (Optical Character Recognition) engine is responsible for converting text information in scanned or photographed document images into editable and searchable digital text; it supports the recognition of various fonts, layouts, and formats to ensure accurate information extraction. The intelligent verification engine automatically verifies the data extracted by the OCR engine, checking its accuracy, completeness, and compliance; this may include numerical range verification (e.g., whether the amount is within a reasonable range) and logical verification (e.g., whether the date is valid, whether the context is consistent), etc. The classification engine automatically classifies documents according to preset rules or machine learning algorithms; classification criteria may include document type (e.g., travel expenses, office supplies expenses), department, and expense nature, etc. The data processing engine formats and standardizes the verified and classified data to ensure data consistency and readability; this may include removing redundant information, correcting erroneous data, and generating standard reports, etc.

[0121] In summary, this embodiment integrates multiple functional modules to form an automated and intelligent expense reimbursement document processing system. This helps reduce processing time and labor costs; improves the transparency of the reimbursement process and user self-service capabilities; enhances information accuracy and reduces errors, improving data quality; and improves the convenience and security of data management through electronic archiving, promoting internal management efficiency. Through these automated and intelligent processing methods, organizations can achieve more efficient financial management, improving operational efficiency while ensuring compliance and audit requirements.

[0122] Example 8: As Figure 8 As shown, based on Embodiment 7, the engine classification module provided in this embodiment of the invention includes:

[0123] The engine settings submodule is responsible for setting up each engine independently. Expense receipts to be processed are sent to the input queue for asynchronous processing. Each engine has its own work queue to receive tasks to be processed and distributes these tasks to the appropriate engine through a message queue service.

[0124] The queue service submodule is responsible for quickly converting the relevant information of the expense report into processing tasks when a user submits an expense report, and sending them to the input queue. The message queue service retrieves tasks from the input queue and distributes them to various work queues. Each engine independently retrieves tasks from its work queue and processes them by listening to it. At the same time, each engine can run concurrently, waiting for the next task. After processing, the results are written to the result storage or sent to the next stage.

[0125] The results aggregation submodule is responsible for aggregating the processing results of all engines in the data processing engine, performing overall data processing, storage, and return.

[0126] The working principle and beneficial effects of the above technical solution are as follows: the engine setting submodule of the embodiment sets each engine independently, and the to-be-processed reimbursement document is sent to an input queue for asynchronous processing; each engine has an independent work queue for receiving to-be-processed tasks, and the to-be-processed tasks are distributed to the corresponding engines through a message queue service; when a user submits a reimbursement document, the related information of the reimbursement document is quickly converted into a processing task and sent to the input queue; the message queue service takes the task from the input queue and distributes the task to each work queue; each engine independently obtains the task from the queue and processes it by listening to its work queue; at the same time, the engines can run concurrently, wait for the next task, and write the result to the result storage or send it to the next link after processing is completed; the result aggregation submodule aggregates the processing results of all engines in the data arrangement engine for overall data arrangement, storage, and return. The engine setting submodule of the above scheme can set each engine independently according to specific requirements, such as adjusting the recognition parameters of OCR and the rules of intelligent verification; an interface or API is provided to enable users to quickly adapt to new processing requirements (such as reimbursement documents of different formats); the to-be-processed reimbursement document is sent to the input queue to ensure that the task distribution and processing can be performed asynchronously, improving the overall efficiency. Significance: The modularity and flexibility of the system are improved, enabling each engine to adjust to different business requirements, thereby improving processing efficiency; the system's dependence on a single engine is reduced, and the fault tolerance of processing is increased, so that if a certain engine fails, other engines can continue to work normally. The queue service submodule converts the reimbursement document into a processing task and uses a queue (such as a message queue) to achieve asynchronous processing, avoiding long waiting times; it ensures that multiple engines can process multiple tasks concurrently at the same time, improving the overall system throughput; through the message queue service, the distribution, retry, and failure management of tasks are effectively handled. Significance: The response speed of the system is significantly improved, and users will not experience delays when submitting reimbursement, improving user experience; the system load is reduced, and it can better manage peak task submissions by decoupling the pressure on the entire system from any single module. The result aggregation submodule integrates the processing results of different engines to ensure data consistency and integrity; after data processing is completed, the results are returned to the user or the downstream system, providing necessary feedback information (such as processing results, error information, etc.); all engine processing results are effectively managed and stored for subsequent analysis and auditing. Significance: The transparency of data processing is improved, enabling users to clearly understand the processing results, thereby enhancing trust; it provides basic data support for subsequent data analysis, report generation, and decision-making, improving the overall management efficiency of financial reimbursement.

[0127] In summary, the engine classification module of the present embodiment, by dividing the reimbursement document processing flow into several independent but interdependent sub-modules, embodies the following meanings: modular design allows the system to be more easily extended or integrated with new functions in the future; through asynchronous processing and independent engine settings, even if some components fail, the overall system can still operate normally; through fast task conversion and effective feedback mechanism, users will experience smoother operation experience in the use process; the parallel processing capability of each engine enables the system to efficiently utilize computing resources and improve overall processing capacity; the result aggregation sub-module ensures the orderly management of all processing results, facilitating data retrieval and subsequent analysis. The entire engine classification module not only improves the processing efficiency and accuracy of reimbursement documents, but also lays a technical foundation for future business expansion and technical upgrade.

[0128] Embodiment 9: As shown in Embodiment 7, on the basis of Embodiment 7, the engine classification module provided by the present embodiment further comprises: Figure 9

[0129] A data extraction sub-module is responsible for using an OCR engine to perform optical character recognition on the scanned reimbursement document image, converting pixel data in the image into machine-readable text data, including image preprocessing (such as denoising and binarization), feature extraction (identifying the shape and contour of characters), and character recognition (using a trained model to predict character categories); the output is a text data structure containing all extracted recognizable characters in the document and their corresponding position and format information;

[0130] A preliminary verification sub-module is responsible for performing structured preliminary verification on the text data extracted by the OCR engine, including verifying whether the text format conforms to the preset pattern (such as date format "YYYY-MM-DD" and amount format "#,###.00"), checking whether the required fields exist (such as the reimbursing person and the reimbursement amount), and identifying characters or strings that cannot be accurately recognized;

[0131] A logic verification sub-module is responsible for performing advanced logic verification on the data that passes the preliminary verification, cross-verifying through predefined rules and business logic; including:

[0132] Verifying whether the reimbursement amount exceeds the set budget limit, such as comparing the reimbursement amount with the corresponding data field in the budget database;

[0133] Verifying whether the reimbursement date is within the set range, such as comparing the current date and the report date to confirm its reasonableness;

[0134] Verifying the matching of the reimbursing person and his / her department, such as checking the consistency of the reimbursing person information and the department field in the employee information database; the goal of logic verification is to ensure the business reasonableness of the relationship and content between data;​

[0135] Abnormality processing submodule, responsible for all abnormal data found in the logical verification stage will be automatically marked by the system as the object of attention, generating an error report, which lists each non-compliance item found in detail, including specific field name, original data, error type and suggested correction direction; at the same time, the relevant personnel will receive a notification about the abnormal situation, so that they can intervene in time to handle;

[0136] Feedback and correction submodule, responsible for the bill marked as abnormal will be presented in the user interface, users can access the error report to view each specific error information; a user-friendly interface will be provided to assist users in revising and supplementing abnormal data, including text boxes for inputting missing fields, drop-down menus for selecting the correct category, and prompt information to guide users to make compliance processing; after the correction is completed, the user can submit the corrected bill back to the system for re-verification;

[0137] Final confirmation submodule, responsible for all verification steps are completed, the bill data that meets the requirements will be marked as valid; at this time, the bill data that has been audited and confirmed will be stored in the database and marked as ready for the subsequent approval process; including updating the status, archiving the data and generating the necessary transaction log for audit.

[0138] The working principle and beneficial effects of the technical solution are as follows: the data extraction submodule of the embodiment uses an OCR engine to perform optical character recognition on the scanned reimbursement document image, converts the pixel data in the image into machine-readable text data, includes preprocessing of the image (such as denoising and binarization), feature extraction (identifying the shape and contour of the character), and character recognition (using a trained model to predict the character class); the output is a text data structure that contains all the extracted recognizable characters in the document and their corresponding position and format information; the preliminary verification submodule performs structured preliminary verification on the text data extracted by the OCR engine, including verifying whether the text format conforms to the preset mode (such as date format "YYYY-MM-DD", amount format "#,###.00"), checking whether the required fields exist (such as the reimbursing person and the reimbursement amount), and identifying characters or strings that cannot be accurately recognized; the logic verification submodule performs advanced logic verification on the data that passes the preliminary verification, cross-verification through predefined rules and business logic; including: verifying whether the reimbursement amount exceeds the set budget limit, for example, by comparing the reimbursement amount with the corresponding data field in the budget database; verifying whether the reimbursement date is within the set range, for example, by comparing the current date and the report date to confirm its reasonableness; verifying the matching of the reimbursing person and his / her department, for example, by checking the consistency of the reimbursing person information and the department field in the employee information database; the goal of logic verification is to ensure the business reasonableness of the relationship and content between the data; the exception handling submodule automatically marks all abnormal data found in the logic verification stage as objects that need attention, generates an error report that lists each non-compliant item found in detail, including the specific field name, original data, error type, and suggested correction direction; at the same time, relevant personnel will be notified of the abnormal situation so that they can intervene in the processing in a timely manner; the feedback and correction submodule presents the documents marked as exceptions in the user interface, and the user can access the error report to view the specific error information; a user-friendly interface will be provided to assist the user in revising and supplementing the abnormal data, including text boxes for inputting missing fields, drop-down menus for selecting the correct category, and prompt information to guide the user to make compliant processing; after the correction is completed, the user can submit the corrected document back to the system for re-verification; the final confirmation submodule marks the document data that meets the requirements as valid after all verification steps are completed; at this time, the audited and confirmed document data is stored in the database and marked as ready for the subsequent approval process; including updating the status, archiving the data, and generating necessary transaction logs for auditing. The data extraction submodule of the above scheme accurately converts the pixel data in the image into processable text data through optical character recognition (OCR) technology; denoising and binarization operations are applied to improve recognition accuracy and reduce errors and noise interference; a trained model is used to identify the shape and contour of the character to ensure high accuracy of the extraction result.The achieved significance: improve the automation and accuracy of data extraction, reduce manual intervention, and improve processing efficiency; lay a reliable data foundation for subsequent verification and logical processing. The preliminary verification submodule quickly detects the format legality of text data (such as date format, amount format) through regular expressions or pattern matching; automatically checks the existence of mandatory fields to ensure that key data is not missed; accurately identifies characters that cannot be accurately converted in OCR extraction, and cleans the information before entering the subsequent processing link. The achieved significance: through efficient preliminary verification, ensure the data quality entering the subsequent steps, reduce the potential errors in subsequent processing; improve the overall stability and reliability of the entire system, prevent error data from flowing into the subsequent processing process. The logical verification submodule cross-verifies the logical relationship of the data set according to the predefined business logic rules; checks the logical consistency of the data by comparing the customer database, budget database and employee information; ensures that the reimbursement amount, date, and personnel information are reasonable for business operations. The achieved significance: ensure the logical reasonableness between data, reduce the burden of manual intervention and review, and improve data processing reliability; through logical review, risk control can be achieved to prevent unreasonable reimbursement requests and improve the financial compliance of the enterprise. The exception handling submodule automatically marks the exceptions found in logical verification and generates detailed error reports; relevant personnel can receive exception status notifications in a timely manner to quickly take action to solve the feedback errors. The achieved significance: significantly improves the response time to abnormal situations, enabling problems to be quickly identified and addressed, which helps maintain the smoothness of business processes; improves the transparency and traceability of reimbursement review, ensuring that all operations are documented for future audits. The feedback and correction submodule provides a user-friendly interface for users to easily view and correct marked exception data; through text boxes, drop-down menus and prompt information, users can effectively complete the correction. The achieved significance: promotes user participation, improves user satisfaction with reimbursement document correction and acceptance of the system; through the feedback and correction link, improve the accuracy of data to ensure that the final data meets business requirements. The final confirmation submodule confirms the effectiveness of the document data after all verifications and stores it in the database; records the status changes during each exception handling and data confirmation process, which helps audit and problem tracking. The achieved significance: ensures data integrity and accuracy, so that there will be no data errors in subsequent approval processes; improves approval efficiency through clear process control to ensure that each document is processed in a timely manner.

[0139] In summary, through the cooperative work of the above-mentioned submodules, the engine classification module can realize efficient, accurate and automated processing of reimbursement documents. Each submodule not only improves processing speed and quality, but also increases the stability and controllability of the system, providing a efficient reimbursement processing tool for the finance department and promoting the standardization and informatization of enterprise financial management.

[0140] Embodiment 10: As shown in the embodiment 7, on the basis of the embodiment 7, the engine classification module provided by the embodiment of the application further comprises: Figure 10

[0141] An information analysis submodule is responsible for structuring the extracted first key information and the obtained key information of the reimbursement document; performs semantic analysis on the recognized text information, and extracts specific key information by using a natural language processing technology; and comprises:

[0142] Text segmentation: the overall text information is segmented according to a separator or a format rule, and each field is recognized;

[0143] Entity recognition: a natural language processing model that is trained is used to perform named entity recognition (NER) on the text, so as to determine the semantic category of each field, such as extracting a “department” as an entity category and recognizing a corresponding department name from the text;

[0144] Data standardization: the extracted text information is standardized in format, including uniform string case, removal of redundant spaces, conversion into a preset encoding format, and the like;

[0145] A rule matching submodule is responsible for classifying and processing the analyzed information based on predefined classification logic and business rules, and comprises:

[0146] Rule set establishment: a classification rule set is configured in advance, including different expense types, corresponding reimbursement processes and conditions thereof;

[0147] Condition evaluation: for each document information, the analysis result is compared with the rule set members item by item, and a condition judgment (such as “if the expense type is ‘travel expense’, then enter the travel reimbursement process”) is used to determine the belonging category;

[0148] Dynamic allocation: the document information that meets the specified conditions is responsively allocated to the corresponding reimbursement process or category, and detailed information of each decision is recorded, so as to facilitate subsequent tracking and auditing;

[0149] An image storage submodule is responsible for storing the classified and processed reimbursement document images in a preset storage location, and comprises the following operations:

[0150] File arrangement: the image files are arranged by category and date, and a corresponding directory structure is created, so as to ensure the efficiency and convenience of retrieval;

[0151] Data encryption and backup: in the storage process, the security of the document image files is ensured, including encryption processing of the stored data and regular backup, so as to prevent data loss or leakage;

[0152] ​Metadata attachment: When storing image files, relevant metadata (such as classification information, storage time, processing status, etc.) is attached to provide necessary information support for subsequent queries and reviews.

[0153] The working principle and beneficial effects of the above technical solution are: the information analysis submodule of the embodiment performs structured processing on the extracted first key information and the obtained key information of the reimbursement document; semantic analysis is performed on the recognized text information, and specific key information is extracted using natural language processing technology; the rule matching submodule classifies the analyzed information based on predefined classification logic and business rules; the image storage submodule stores the classified reimbursement document images in a preset storage location. The information analysis submodule of the above scheme accurately segments the text information in the document into various fields, ensuring information structuring, facilitating subsequent processing and analysis; using a trained natural language processing model, important entities (such as departments and expense types) in the text are accurately identified, ensuring that key data is extracted and identified; by using a unified format and cleaning redundant data, the consistency and readability of the information are improved, ensuring that errors are not caused by inconsistent formats during subsequent processing. The significance achieved: provides basic support for overall information processing, making it easier to implement subsequent classification and rule matching through structured data; reduces the risk of human error, improves the overall efficiency and accuracy of information processing, and improves data quality, laying a good foundation for subsequent processes. The rule matching submodule system configures classification rules in advance to ensure that the classification process meets actual business needs, making the classification stable and flexible; by comparing the parsed information with the rule set item by item, the system can flexibly judge and accurately classify each reimbursement request, reducing the complexity of human judgment; automatically assigns documents that meet the conditions to the designated reimbursement process or category, and records relevant decisions to ensure the traceability of classification. The significance achieved: improves the automation level of reimbursement processing, reduces human intervention, and improves work efficiency; ensures that each reimbursement document can be processed according to actual needs, improving the compliance and transparency of operations, and supporting subsequent audit operations to enhance the overall security of data management. The image storage submodule organizes image files by category and date, creating an effective file storage structure to ensure efficient and convenient information retrieval; implements security controls on data during storage to prevent unauthorized access and data loss, while ensuring the persistent security of data through backups; attaches key information (such as classification, timestamp, processing status, etc.) to each image file to provide necessary context information for subsequent queries and reviews. The significance achieved: by strengthening data security and organization, effective management of reimbursement document data is achieved, thereby reducing potential data risks; through systematic and structured image storage, quick access and review are supported, improving work efficiency and better meeting audit and compliance requirements.

[0154] In summary, the engine classification module aims to improve the automation and intelligence level of reimbursement document processing, while ensuring the accuracy and security of information. This not only improves work efficiency and reduces labor costs, but also helps to meet the requirements of enterprises for financial management compliance and transparency. In summary, it has important practical significance for improving internal management efficiency, supporting decision-making, and reducing economic risks of enterprises.

[0155] Embodiment 11: as Figure 11 indicated, on the basis of embodiment 1, the functional components provided by the embodiment of the application include:

[0156] The state synchronization module is responsible for synchronizing the state of the reimbursement document to the collection system when the actions of rejecting the approval of the reimbursement document, withdrawing the user, submitting the approval of the user, and archiving the document occur. For the rejected or withdrawn reimbursement document, the user performs the order cancellation processing through the self-service interface, and for the archived document, the physical archiving operation is performed. In addition, the reimbursement document in the "submit for approval" state will not allow the user to perform the self-order cancellation operation.

[0157] The identification and modification module is responsible for upgrading and modifying the executed reimbursement document, and at the same time, the modification will realize the multi-document identification function to support the parallel processing of different types of documents.

[0158] The reimbursement document includes borrowing, repayment, business trip application, leave application, personnel traffic record, cadre exchange traffic record, travel expenses, cadre exchange traffic expenses, make-up, business trip and make-up, travel expenses for visiting relatives, conference and consultation fee budget, conference and consultation fee settlement, business entertainment fee budget, business entertainment fee settlement, expense reimbursement, nursing fee, small assets, special fee settlement, trade union fee budget, trade union fee settlement, retirement worker fee and difficulty subsidy.

[0159] The data access control module is responsible for calling the authentication interface to obtain the access token according to the clientID and clientSecret obtained from the collection system. The token needs to be included in the HTTP request header in the subsequent request to ensure the permission of other business interface calls. The validity period of the token is 60 minutes, and it needs to be reacquired after the expiration.

[0160] The image processing and conversion module is responsible for sending all image data to the reimbursement system for optical character recognition and intelligent audit after the collection machine completes the scanning of all reimbursement documents. According to the state recognition of the reimbursement document, the specified reimbursement document will be automatically printed, and each reimbursement document can be physically printed one by one.

[0161] The working principle and beneficial effects of the above technical solution are: the state synchronization module of the embodiment synchronizes the reimbursement document state to the collecting system when the actions of disapproval, user withdrawal, user submission for approval and document archiving of the reimbursement document occur; for the disapproved or withdrawn reimbursement document, the user performs the single withdrawal processing through the self-service interface, and for the archived document, the physical archiving operation is performed; in addition, the reimbursement document in the "submission for approval" state will not allow the user to perform the self-withdrawal operation; the identification and modification module performs the upgrade and modification of the reimbursement document, and at the same time, the modification implements the multi-document identification function to support the parallel processing of different types of documents; the reimbursement document includes borrowing, repayment, business trip application, leave application, personnel traffic record, cadre exchange traffic record, travel expenses, cadre exchange traffic expenses, make-up, business trip and make-up, travel expenses for visiting relatives, conference and consultation fee budget, conference and consultation fee settlement, business entertainment fee budget, business entertainment fee settlement, expense reimbursement, convalescent fee, sporadic assets, special fee settlement, trade union fee budget, trade union fee settlement, retired worker fee and difficulty subsidy; the data access control module calls the authentication interface according to the clientID and clientSecret obtained from the collecting system to obtain the access token; the token needs to be included in the HTTP request header in the subsequent request to ensure the permission of the other business interface call; the validity period of the token is 60 minutes, and it needs to be reacquired after timeout; the image processing conversion module calls the corresponding interface to send all image data to the reimbursement system for optical character recognition and intelligent audit after the collecting machine completes the scanning of all reimbursement documents. According to the state recognition of the reimbursement document, the specified reimbursement document printing operation is automatically performed, and it is ensured that each reimbursement document can be physically printed one by one. The state synchronization module of the above scheme realizes the instant synchronization of the reimbursement document state between different systems, ensures that the collecting system always has the latest document state information; allows the user to perform the single withdrawal of the disapproved or withdrawn document through the self-service interface operation, improves the user experience and the convenience of the system; clearly distinguishes the operation permissions in different states, such as the single withdrawal of the user in the submission for approval state, reduces the risk of operation misuse. The significance achieved: ensures the consistency and accuracy of the data, helps to prevent approval and process errors caused by inconsistent document state information; improves the transparency of the reimbursement process, enables the user to know the document state in real time, enhances the user's trust in the system; through the fault tolerance mechanism and permission control, the safety of the overall process is optimized, and the potential risk is reduced. The identification and modification module upgrades and modifies up to 23 types of reimbursement documents, enhancing the system's support capability for various types of documents; the modification realizes the parallel processing capability of multiple documents, improving the document processing efficiency. The significance achieved: improves the flexibility and adaptability of the reimbursement system, better responds to business demand changes and expansion; optimizes the efficiency of user operation, realizes efficient management of different types of documents, and reduces the operation time and error rate.The data access control module obtains a token through a client ID and a client secret, ensuring that the system has the necessary permission control when interacting with data. In each subsequent interface call, the token is included in the HTTP request header to verify the requester's permissions. This achieves the following: it strengthens system security, ensuring that only verified requests can access sensitive information and perform important operations; it significantly reduces the risk of unauthorized access and data leakage, providing strong support for data protection. The image processing and conversion module can quickly and accurately send image data to the reimbursement system for subsequent processing after scanning is complete. It implements optical character recognition and intelligent audit processes, automatically extracting and auditing document information to improve automation and efficiency. The system can automatically trigger printing based on document status recognition to ensure that each reimbursement document is physically printed. This achieves the following: it improves the efficiency and accuracy of image processing, reduces manual auditing, and improves overall business processing speed. Through automated management and printing mechanisms, it reduces the workload of operators and optimizes resource allocation and workflow.

[0162] In summary, the coordinated operation of various modules in this embodiment enables efficient and automated processing of reimbursement documents in the entire functional component system. Each module not only improves the functionality and reliability of the system, but also significantly enhances user experience and system security. The integration of the design and implementation of this embodiment not only helps to improve work efficiency, but also provides stronger support for enterprises in managing finances and complying with regulations.

[0163] Embodiment 12: Based on embodiments 1-11, the data interaction process between the integrated financial intelligent terminal acquirer's display reimbursement system and the acquirer is provided, which includes:

[0164] The reimbursement system clicks on the travel expense report, and a pop-up window appears asking whether it is a pure business travel document:

[0165] (1) Select 'Yes': the user will not be able to upload the bill, and the document can only be reimbursed as a pure business travel document;

[0166] (2) Select 'No': the user will receive a single number QR code from the acquirer application on the line cloud app, and the user can use this QR code to submit the document to the acquirer;

[0167] Using decoding technology, the pattern is converted back to the original information, and the line cloud app will receive the QR code, which can be used to submit the document to the acquirer;

[0168] After approval and electronic archiving of the reimbursement system, the smart acquirer can automatically print the corresponding approval face sheet and accounting voucher for archiving, completing the full-cycle loop.

[0169] Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. An integrated intelligent financial terminal acquiring machine, characterized in that, Include: The self-service document submission component is responsible for receiving expense reports, scanning and recognizing them, and extracting the first key information from the expense reports, including the amount, date, and recipient. The information processing component is responsible for extracting the second key information from the expense report and processing the first and second key information through the data processing engine. The processing includes automatic verification, classification and organization. At the same time, the processed first and second key information is electronically archived and the corresponding approval form and accounting voucher are automatically printed for archiving. The functional interface component is responsible for providing interfaces for adding login authentication, receiving expense reimbursement documents, and synchronizing expense reimbursement document status; among them, adding login authentication includes Three Gorges Cloud QR code authentication, facial recognition authentication, and employee card swiping authentication. The self-service order submission component includes: The document receiving module is responsible for receiving completed expense reports and placing them within the scanning area for temporary storage. The image processing module is responsible for triggering the image capture process when the vision sensor detects the presence of the expense report when it is placed in the scanning area; it uses the vision sensor to capture the image of the expense report, performs preprocessing operations on the image, and obtains the preprocessed image. The output module is responsible for identifying the size, orientation, quality, and primary key information of the expense report from the preprocessed image; it also provides a user interface to display operation instructions and feedback information via a touch screen or display screen; the preprocessing includes noise reduction and contrast enhancement. The results output module includes: The image selection submodule is responsible for generating multiple candidate images for each automatic capture, and selecting the image from the multiple preprocessed candidate images as the main image for subsequent processing. The physical characteristics submodule is responsible for measuring the width and height of the main image to determine the size of the expense report; determining the orientation of its long and short sides to determine whether it is horizontal or vertical, and automatically adjusting the image orientation accordingly; judging whether the main image meets the processing requirements according to the set quality standards; if it does, proceed to the next step; if it does not, prompt the user to reposition or rescan the document. The information extraction submodule is responsible for extracting the first key information from the main image that meets the processing requirements. For each key information extracted, the result is immediately fed back to the user interface and the extraction result is displayed for user confirmation. The physical properties submodule includes: The edge detection unit is responsible for converting the selected main image to a grayscale image, binarizing the image using an adaptive Gaussian threshold, smoothing the image using Gaussian filtering, performing edge detection based on the edge intensity gradient, and identifying the outer border of documents in the image. The size acquisition unit is responsible for applying polygon approximation to the extracted contour, smoothing the contour and simplifying the edges, and detecting its shape features; calculating the minimum bounding rectangle of the contour, performing geometric calculations by finding the extreme points of the contour, and calculating the width and height of the image. The orientation adjustment unit is responsible for determining the orientation of the file based on the ratio of its width to its height. If the width is greater than the height, it is recorded as horizontal; otherwise, it is recorded as vertical. Based on the orientation information, the image is automatically rotated by 90 degrees or 180 degrees, and the rotated image is resampled using bilinear interpolation or nearest neighbor interpolation algorithms.

2. The integrated financial intelligent terminal acquiring machine as described in claim 1, characterized in that, The edge detection unit includes: The convolution operation subunit is responsible for edge detection on the Gaussian-smoothed main image. It uses two 3x3 convolution kernels to perform two convolution operations on the original image to obtain the gradient of the main image; it calculates the gradient magnitude and direction of each pixel. The gradient calculation subunit is responsible for comparing each pixel with its neighborhood along the calculated gradient direction. If the magnitude of a pixel is greater than the magnitude of its neighboring pixels, then the pixel is retained. Otherwise, set its strength to zero; The threshold comparison subunit is responsible for setting a high threshold and a low threshold. After non-maximum suppression, it judges the intensity of each pixel. For pixels marked as strong edges, it uses their connectivity to trace the connected edges in the image. Any pixel connected to a strong edge is marked as an edge, forming a complete edge map.

3. The integrated financial intelligent terminal acquiring machine as described in claim 1, characterized in that, The information extraction submodule includes: The feature merging unit is responsible for annotating the main images that meet the processing requirements, determining the location and format of the first key information; selecting a pre-trained model and fine-tuning it on the annotated main images containing expense reports; constructing an image pyramid to generate multiple versions of input images from different scales, and extracting features from the input images at each scale; merging deep features from different scales; and converting the extracted features into fixed-length feature vectors through a fully connected layer. The text acquisition unit is responsible for locating key regions using bounding boxes and clarifying the relative positions of key regions in the main image; using the OCR engine to perform character recognition on the extracted key regions to obtain text information; and performing rule judgment on the recognized text information to ensure the regularity of the first key information. The information verification unit is responsible for performing contextual analysis on the extracted text information. By setting pattern matching rules, it determines that the amount is located in a specific area, while the invoice number is at the beginning. It uses named entity recognition algorithms to analyze the text, and the extracted first key information is more semantically contextual.

4. The integrated financial intelligent terminal acquiring machine as described in claim 1, characterized in that, Information processing components, including: The content extraction module is responsible for extracting the second key information from the scanned image of the expense report through the data processing engine OCR. The second key information includes departure time, arrival time, departure point, arrival point, passengers, seat class, and ticket amount. The engine classification module is responsible for automatically verifying, classifying, and organizing the first and second key information using an optical character recognition engine, intelligent verification engine, classification engine, and data processing engine. At the same time, users can directly retrieve expense reports for processing. If an expense report is rejected, the user can retrieve the report independently and resubmit it after replacing or supplementing the attachments as prompted. The archiving module is responsible for electronically archiving approved expense reimbursement documents, printing approval forms, and archiving accounting vouchers.

5. The integrated financial intelligent terminal acquiring machine as described in claim 4, characterized in that, The engine classification module includes: The engine settings submodule is responsible for setting up each engine independently. Expense receipts to be processed are sent to the input queue for asynchronous processing. Each engine has its own work queue to receive tasks to be processed and distributes these tasks to the appropriate engine through a message queue service. The queue service submodule is responsible for quickly converting the relevant information of the expense report into processing tasks when a user submits an expense report, and sending them to the input queue. The message queue service retrieves tasks from the input queue and distributes them to various work queues. Each engine independently retrieves tasks from its work queue and processes them by listening to it. At the same time, each engine can run concurrently, waiting for the next task. After processing, the results are written to the result storage or sent to the next stage. The results aggregation submodule is responsible for aggregating the processing results of all engines in the data processing engine, performing overall data processing, storage, and return.

6. The integrated financial intelligent terminal acquiring machine as described in claim 4, characterized in that, The engine categorization module also includes: The data extraction submodule is responsible for using the OCR engine to perform optical character recognition on the scanned expense report images, converting the pixel data in the image into machine-readable text data. This includes image preprocessing, feature extraction, and character recognition; the output is a text data structure. The preliminary validation submodule is responsible for performing structured preliminary validation on the text data extracted by the OCR engine, including verifying whether the text format conforms to the preset pattern, checking whether the required fields exist, and identifying characters or strings that cannot be accurately recognized. The logic validation submodule is responsible for performing advanced logic validation on data that has passed the initial validation, and performing cross-validation through predefined rules and business logic. The exception handling submodule is responsible for automatically marking all abnormal data found during the logic verification phase as objects requiring attention by the system and generating an error report that details each non-compliance item found. At the same time, relevant personnel will receive notifications about the abnormal situation so that they can intervene and handle it in a timely manner. The Feedback and Correction submodule is responsible for displaying documents marked as abnormal in the user interface. Users can access error reports and view specific error information. It will provide a user-friendly interface to assist users in revising and supplementing abnormal data. After the correction is completed, the user can submit the corrected document back to the system for re-verification; The final confirmation submodule is responsible for marking compliant document data as valid after all verification steps are completed. At this point, the approved document data is stored in the database and marked as ready to enter the subsequent approval process.

7. The integrated financial intelligent terminal acquiring machine as described in claim 1, characterized in that, Functional components, including: The status synchronization module is responsible for synchronizing the status of expense reimbursement documents to the receiving system when the approval process is rejected, the user withdraws the document, the user submits the document for approval, or the document is archived. For rejected or withdrawn expense reimbursement documents, users can process the cancellation through the self-service interface, while for archived documents, physical archiving will be performed. In addition, users will not be allowed to cancel expense reimbursement documents that are in the submission approval status. The identification and transformation module is responsible for upgrading and transforming the expense reimbursement documents. At the same time, the transformation will realize the multi-document recognition function to support the parallel processing of different types of documents. The data access control module is responsible for calling the authentication interface to obtain an access token based on the clientID and clientSecret obtained from the acquiring system; the token must be included in the HTTP request header in subsequent requests; The image processing and conversion module is responsible for sending all image data to the reimbursement system for optical character recognition and intelligent review after the receiving machine has scanned all reimbursement documents. Based on the status recognition of the reimbursement documents, it will automatically perform the printing operation of the specified reimbursement documents and ensure that each reimbursement document can be physically printed one by one.

Images