A document automatic delivery and return processing system of a fusion financial management platform
By integrating the financial management platform, acquiring machine equipment, and interface modules, and utilizing CNN and SVM technologies to optimize the document processing workflow, the problems of low efficiency and poor accuracy in traditional document processing have been solved. This has enabled automated and highly accurate document processing and return management, and improved the integration capabilities of the system with the financial management platform.
Patent Information
- Authority / Receiving Office
- CN Β· China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GENERAL HOSPITAL OF PLA
- Filing Date
- 2026-03-30
- Publication Date
- 2026-06-23
AI Technical Summary
Traditional document processing methods rely on manual operation, which has problems such as low processing efficiency, poor identification accuracy, cumbersome cancellation judgment, imperfect storage and archiving management, and information asynchrony between systems.
Design an automatic document delivery and return processing system that integrates a financial management platform. The system uses the financial management platform, a receiving machine, and an interface module. It uses an OCR recognition module to extract and optimize image features based on a convolutional neural network (CNN), combines a support vector machine (SVM) for return judgment, and uses the interface module to realize real-time data synchronization and storage strategy optimization.
It improves the automation level of document processing and the accuracy of OCR recognition, realizes the automation and efficiency of order cancellation judgment, enhances the integration capability of the system with the financial management platform, and ensures the consistency of document status and processing records and the optimized use of storage space.
Smart Images
Figure CN122264972A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of financial information processing technology, specifically to an automatic document delivery and return processing system that integrates a financial management platform. Background Technology
[0002] In the financial management process of modern enterprises, document processing is a core link, involving the collection, scanning, recognition, review, correction, and archiving of various documents such as invoices, receipts, and contracts. Traditional processing methods rely heavily on manual operation, resulting in problems such as low processing efficiency, poor recognition accuracy, cumbersome return order judgment, imperfect storage and archiving management, and information asynchrony between systems. Specifically, the workload of manual scanning and data entry is large, and document backlog is likely to occur during peak periods; traditional OCR systems have limited ability to recognize documents with low resolution, low contrast, or noise and uneven lighting, resulting in a high text recognition error rate; return order judgment requires manual comprehensive consideration of document type, review results, amount, submission time, and historical return order behavior, which is time-consuming and susceptible to human factors; storage space is limited, documents are prone to misplacement or loss, and there is a lack of a real-time synchronization mechanism with the financial management platform, making it impossible to achieve a closed-loop business process. Therefore, to address the above problems, an automatic document delivery and return order processing system integrating with the financial management platform is proposed. Summary of the Invention
[0003] The purpose of this invention is to provide an automatic document delivery and cancellation processing system that integrates a financial management platform to solve the problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: An automated document delivery and return processing system integrating a financial management platform includes: a financial management platform, a document receiving terminal device, and an interface module. The financial management platform manages document status, provides an interface for connecting with the document receiving terminal device, processes instructions for document delivery, review, return, and archiving, and records historical document processing data. The financial management platform also dynamically triggers scanning, OCR recognition, or return operations on the document receiving terminal device based on document status, achieving real-time status synchronization. The document receiving terminal device includes a scanning module, an OCR recognition module, an information correction module, a return processing module, and a data upload module. The scanning module acquires document images, the OCR recognition module extracts text from the images, the information correction module manually or automatically corrects the recognition results, the return processing module executes return logic, and the data upload module sends the processing results to the financial management platform. The interface module transmits data between the document receiving terminal device and the financial management platform, supports real-time synchronization of instructions and data, and ensures complete consistency between document status and processing records with the financial management platform. The interface module employs a bidirectional communication protocol to ensure status synchronization of all modules and high efficiency in document processing.
[0005] Preferably, the OCR recognition module extracts image features based on a convolutional neural network (CNN) and preprocesses and optimizes the document image by collecting the following image quality factors: resolution factor, obtained by calculating the number of pixels per inch (DPI) of the image, representing the image's clarity and detail; contrast factor, calculated by the image's gray-level variance, representing the difference between text and background; noise level factor, calculated by the signal-to-noise ratio (SNR), representing random noise interference in the image; illumination factor, calculated by the image's average brightness, representing the influence of light source intensity; and distortion factor, calculated by the error value of image geometric transformations (such as affine transformations), representing angular and geometric distortion. Each quality factor is individually standardized to the [0,1] interval, and the recognition parameters of the CNN model (such as convolutional kernel weights and learning rate) are dynamically adjusted through weighted calculations to improve the recognition accuracy of low-quality document images.
[0006] Preferably, the OCR optimization formula is a multi-factor weighted nonlinear formula:
[0007] Wherein, π , πΆ, π, πΏ, and π represent resolution, contrast ratio, noise level, illumination, and distortion factor, respectively. , , , , For each factor, a nonlinear mapping function is used, employing either the sigmoid or tanh function. , , , , The corresponding weight coefficients are obtained through training on historical datasets, with an optimal value range of 0.1 to 0.3. π is the activation function, specifically the ReLU function. The output of the formula is used to dynamically adjust the CNN recognition threshold and convolutional kernel weights.
[0008] Preferably, the OCR recognition module supports using the non-linear weighted result of multiple image quality factors as input to dynamically adjust the CNN model parameters; the adjustment includes optimizing the convolution kernel weights through gradient descent algorithm, and the learning rate is dynamically set according to the weighted result of the quality factors; Automatic optimization processing of low-quality document images is achieved, thereby improving the accuracy and reliability of OCR recognition and its compatibility with the document management process of the financial management platform.
[0009] Preferably, the order cancellation processing module is used to determine the order cancellation based on the document's review results, type, completeness, and business rule factors synchronized with the financial management platform; the business rule factors include the document amount threshold, the handler's historical order cancellation rate, and the document's urgency level; When a document fails to pass the review and its type is "invoice" or "receipt", the system marks the document as "returned" according to the rules and triggers the return operation. At the same time, the system will automatically notify the person in charge to retrieve the document via SMS or email. The system uses time series analysis to analyze user behavior patterns by analyzing historical order cancellation data and processing records from the financial management platform, and automatically optimizes the timeliness of order cancellation processing.
[0010] Preferably, the order cancellation processing module uses a support vector machine (SVM) to determine order cancellations and introduces multi-dimensional feature factors into the determination model, including document type, review result, completeness, submission time, and OCR recognition quality factor. The SVM model uses grid search for hyperparameter optimization and evaluates its performance using 5-fold cross-validation. The formula for determining order cancellation is:
[0011] in, , , , , The feature vector is composed of OCR recognition quality factors, where π is the weight matrix, obtained by training through stochastic gradient descent, π is the bias, and π is the sigmoid activation function; The system dynamically adjusts the sensitivity of order cancellation decisions based on image quality, improving decision-making accuracy.
[0012] Preferably, the order cancellation processing module supports flexible configuration based on different document types, business processes, and custom rules of the financial management platform; the custom rules include order cancellation thresholds, document type priorities, and notification templates; The system uses a rules engine to parse the business logic of the financial management platform, implements enterprise-level custom cancellation strategies, and ensures that they are fully compatible with the business logic of the financial management platform.
[0013] Preferably, the acquiring device includes a temporary storage compartment and a filing box. When the capacity of the temporary storage compartment or the filing box reaches the warning value (80% of the total capacity), the system notifies the administrator through a warning mechanism (SMS or platform notification). The system achieves optimal utilization of document storage space by dynamically adjusting storage strategies and combining them with the archiving status of the financial management platform. The storage strategies include automatically clearing temporary storage cells or compressing image data based on document priority to avoid misplacement or loss of documents.
[0014] Preferably, the system communicates with the financial management platform through an interface module, uses a RESTful API for data transmission, and supports real-time processing of document information, OCR recognition results, and order cancellation instructions; Data transmission uses JSON format and is based on OAuth 2.0 for authentication, while also supporting retry mechanisms and exception logging. The system can provide real-time feedback on document processing status, identification results, and order cancellation information to the financial management platform, achieving seamless integration.
[0015] Preferably, the system optimizes the OCR model and order cancellation judgment algorithm in real time through a dynamic feedback mechanism; the feedback data sources include user-corrected document data, and the model is retrained weekly; The optimization process employs online learning algorithms, automatically adjusting parameters (such as CNN convolution kernel weights and SVM kernel functions) with each data update, ensuring long-term stable operation of the system and consistency with the business rules of the financial management platform.
[0016] Compared with the prior art, the beneficial effects of the present invention are: The system provided by this invention dynamically adjusts the OCR recognition model through a multi-factor weighted nonlinear optimization algorithm. It combines quantitative quality factors such as resolution, contrast, noise level, lighting conditions, and image distortion to dynamically optimize CNN model parameters, achieving high-precision text extraction from low-quality document images. The order cancellation processing module, based on an SVM model, incorporates multi-dimensional features such as document type, review result, completeness, submission time, and OCR quality factors into the judgment, automating order cancellation judgment. It also optimizes the judgment strategy by combining historical data from the financial management platform, amount thresholds, and user behavior patterns. The system ensures optimal utilization of storage space through monitoring the capacity of temporary storage compartments and archive boxes, and implementing dynamic storage strategies (including automatic document priority cleanup and image compression). It achieves real-time data synchronization and command linkage between the acquiring device and the financial management platform through RESTful API interfaces and OAuth 2.0 authentication. Simultaneously, the system has a built-in dynamic feedback mechanism that periodically updates the CNN and SVM models for long-term adaptive optimization. Through the above technical solutions, this invention significantly improves the automation level of document processing, OCR recognition accuracy, and order cancellation judgment efficiency, while enhancing the system's integration capabilities with the enterprise financial management platform. Attached Figure Description
[0017] Figure 1 This is a schematic diagram of the system structure of the present invention. Detailed Implementation
[0018] The following implementation methods are combined Figure 1 The system structure diagram shown illustrates in detail the present invention, "An Automatic Document Delivery and Return Processing System Integrating a Financial Management Platform." This embodiment is merely an example, and those skilled in the art can make appropriate modifications and optimizations without departing from the spirit of the invention.
[0019] This system includes a financial management platform, acquiring terminals, and interface modules, achieving real-time synchronization of data and instructions through a two-way communication protocol. The financial management platform's main functions include: managing document status, such as "pending scanning," "pending approval," "returned," and "archived"; providing API interfaces for acquiring terminals to access, enabling scan triggering, OCR recognition initiation, return decision triggering, and archiving confirmation; recording document processing history, including scan time, recognition results, number of returns, archiving time, and operator operation records; and supporting custom business rules, including document amount thresholds, type priorities, and return notification templates.
[0020] The main functional modules of the acquiring machine are as follows: Scanning module: used to acquire images of paper documents or receipts, supporting color and grayscale scanning, with adjustable resolution (300-600 DPI recommended); OCR recognition module: extracts text from images based on a CNN model and optimizes image quality through multi-factor weighting; Information correction module: performs manual or automatic correction on OCR recognition results, supporting template matching and regular expression verification; Return order processing module: determines whether a document should be returned based on OCR recognition quality, business rules, and an SVM model; Data upload module: synchronizes recognition results, return order information, and status to the financial management platform, and supports a retry mechanism.
[0021] Interface module: It adopts RESTful API and JSON data format, supports OAuth2.0 authentication and exception logging, and achieves highly reliable integration between the system and the financial management platform.
[0022] During image acquisition, the user places the document on the scanning tray, and the scanning module starts acquisition. Scanning parameters: 300 DPI, color mode, output image format PNG or JPEG; the image is temporarily cached in the acquiring device for subsequent processing. Then, the acquired image undergoes the following quantitative analysis, with each factor normalized to the [0,1] interval: Among them, the resolution factor is calculated by DPI, the contrast factor is calculated by gray level variance, the noise level factor is calculated by signal-to-noise ratio (SNR), the illumination factor is calculated by average brightness, and the distortion factor is calculated by affine transformation error.
[0023] Then, the image quality factor is used as input to the CNN model, and the optimized value is calculated using a multi-factor weighted formula: nonlinear mapping function The CNN uses sigmoid and ReLU activation functions; the CNN parameters include kernel weights, number of convolutional layers, and learning rate, based on... Dynamic adjustments are made to optimize the recognition of low-quality images.
[0024] Then, the OCR-recognized text is matched with the template, such as regular expression validation for invoice number and amount fields; abnormal fields are automatically marked, and manual correction can be quickly performed on the touch screen.
[0025] In the order cancellation judgment process, business rule factors include document type (invoice / receipt / contract, etc.), review result (pass / fail), document completeness (field missing or abnormal), submission time (whether it timed out), OCR recognition quality factor vector [R,C,N,L,W], document amount threshold, the handler's historical order cancellation rate, and urgency level.
[0026] The feature vector of the SVM decision model is:
[0027] in Model training uses grid search to optimize hyperparameters and 5-fold cross-validation to evaluate accuracy.
[0028] Formula for order cancellation probability: The threshold is determined by marking the order as cancelled if P(Return) > 0.5, and automatically triggering a notification.
[0029] The system then moves the returned order to a temporary storage box or archive box and sends a text message or email notification to the person in charge; for highly urgent orders, priority can be set for push notification.
[0030] An alert is triggered when the temporary storage capacity reaches 80% of the total capacity, and the same applies to the archive box; the system automatically cleans up the temporary storage cells according to the document priority, and low-priority documents can have their images compressed or be archived later.
[0031] The acquiring device communicates with the financial management platform via a RESTful API; the data format is JSON, containing fields such as document ID, OCR text, cancellation status, and operation time; OAuth2.0 authentication is supported, and failure retries and exception logging are supported to ensure high reliability.
[0032] Finally, the corrected order data from users is collected as training samples, and the CNN and SVM models are updated online weekly. Based on the latest order cancellation judgment results and OCR accuracy, the convolution kernel weights, learning rate, and SVM kernel function parameters are dynamically adjusted. Through time series analysis of order processing history, predictive order cancellation optimization is achieved, such as automatically adjusting notification delays, retry strategies, and order priority sorting.
[0033] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only for the purpose of helping to understand the method and core ideas of the present invention. The above descriptions are only preferred embodiments of the present invention. It should be noted that due to the limitations of textual expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of the present invention, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of the present invention.
Claims
1. A document automatic delivery and return processing system integrating a financial management platform, characterized in that: include: The system comprises a financial management platform, an acquiring machine, and an interface module. The financial management platform manages document status, provides an interface for connecting with the acquiring machine, processes instructions for document delivery, review, return, and archiving, and records historical document processing data. It also dynamically triggers scanning, OCR recognition, or return operations on the acquiring machine based on document status, achieving real-time status synchronization. The acquiring machine includes a scanning module, an OCR recognition module, an information correction module, a return processing module, and a data upload module. The scanning module acquires document images, the OCR recognition module extracts text from the images, the information correction module manually or automatically corrects the recognition results, the return processing module executes return logic, and the data upload module sends the processing results to the financial management platform. The interface module facilitates data transmission between the acquiring machine and the financial management platform, supports real-time synchronization of instructions and data, and ensures complete consistency between document status and processing records with the financial management platform. The interface module employs a bidirectional communication protocol to ensure status synchronization across modules and high efficiency in document processing.
2. The document automatic delivery and return processing system integrating a financial management platform according to claim 1, characterized in that: The OCR recognition module extracts image features based on a convolutional neural network (CNN) and preprocesses and optimizes document images by collecting the following image quality factors: resolution factor, obtained by calculating the number of pixels per inch (DPI), representing the image's sharpness and detail; contrast factor, calculated by the image's gray-level variance, representing the difference between text and background; noise level factor, calculated by the signal-to-noise ratio (SNR), representing random noise interference in the image; illumination factor, calculated by the image's average brightness, representing the influence of light source intensity; and distortion factor, calculated by the error value of image geometric transformations (such as affine transformations), representing angular and geometric distortion. Each quality factor is individually standardized to the [0,1] interval, and the recognition parameters of the CNN model (such as convolutional kernel weights and learning rate) are dynamically adjusted through weighted calculations to improve the recognition accuracy of low-quality document images.
3. The document automatic delivery and return processing system integrating a financial management platform according to claim 2, characterized in that: The OCR optimization formula is a multi-factor weighted nonlinear formula: ; Wherein, π , πΆ, π, πΏ, and π represent resolution, contrast ratio, noise level, illumination, and distortion factor, respectively. , , , , For each factor, a nonlinear mapping function is used, employing either the sigmoid or tanh function. , , , , The corresponding weight coefficients are obtained through training on historical datasets, with an optimal value range of 0.1 to 0.
3. π is the activation function, specifically the ReLU function. The output of the formula is used to dynamically adjust the CNN recognition threshold and convolutional kernel weights.
4. The document automatic delivery and return processing system integrating a financial management platform according to claim 3, characterized in that: The OCR recognition module supports using the non-linear weighted result of multiple image quality factors as input to dynamically adjust the CNN model parameters; the adjustment includes optimizing the convolution kernel weights through gradient descent algorithm, and the learning rate is dynamically set according to the weighted result of the quality factors; Automatic optimization processing of low-quality document images is achieved, thereby improving the accuracy and reliability of OCR recognition and its compatibility with the document management process of the financial management platform.
5. The document automatic delivery and return processing system integrating a financial management platform according to claim 1, characterized in that: The order cancellation processing module is used to determine order cancellation based on the document's review results, type, completeness, and business rule factors synchronized with the financial management platform; the business rule factors include document amount threshold, the handler's historical order cancellation rate, and document urgency level; When a document fails to pass the review and its type is "invoice" or "receipt", the system marks the document as "returned" according to the rules and triggers the return operation. At the same time, the system will automatically notify the person in charge to retrieve the document via SMS or email. The system uses time series analysis to analyze user behavior patterns by analyzing historical order cancellation data and processing records from the financial management platform, and automatically optimizes the timeliness of order cancellation processing.
6. The document automatic delivery and return processing system integrating a financial management platform according to claim 5, characterized in that: The order cancellation processing module uses a support vector machine (SVM) to determine order cancellations and incorporates multi-dimensional feature factors into the determination model, including document type, review result, completeness, submission time, and OCR recognition quality factor. The SVM model uses grid search for hyperparameter optimization and evaluates its performance using 5-fold cross-validation. The formula for determining order cancellation is: ; in, , , , , The feature vector is composed of OCR recognition quality factors, where π is the weight matrix, obtained through stochastic gradient descent, π is the bias, and π is the sigmoid activation function. The system dynamically adjusts the sensitivity of order cancellation decisions based on image quality, improving decision-making accuracy.
7. The document automatic delivery and return processing system integrating a financial management platform according to claim 6, characterized in that: The order cancellation processing module supports flexible configuration based on different document types, business processes, and custom rules of the financial management platform; custom rules include order cancellation thresholds, document type priorities, and notification templates; The system uses a rules engine to parse the business logic of the financial management platform, implements enterprise-level custom cancellation strategies, and ensures that they are fully compatible with the business logic of the financial management platform.
8. The automatic document delivery and return processing system integrating a financial management platform according to claim 1, characterized in that: The acquiring machine includes a temporary storage compartment and a filing box. When the capacity of the temporary storage compartment or filing box reaches the warning value (80% of the total capacity), the system notifies the administrator through a warning mechanism (SMS or platform notification). The system achieves optimal utilization of document storage space by dynamically adjusting storage strategies and combining them with the archiving status of the financial management platform. The storage strategies include automatically clearing temporary storage cells or compressing image data based on document priority to avoid misplacement or loss of documents.
9. The automatic document delivery and return processing system integrating a financial management platform according to claim 1, characterized in that: The system communicates with the financial management platform through an interface module, uses a RESTful API for data transmission, and supports real-time processing of document information, OCR recognition results, and order cancellation instructions. Data transmission uses JSON format and is based on OAuth 2.0 for authentication, while also supporting retry mechanisms and exception logging. The system can provide real-time feedback on document processing status, identification results, and order cancellation information to the financial management platform, achieving seamless integration.
10. The document automatic delivery and return processing system integrating a financial management platform according to claim 9, characterized in that: The system optimizes the OCR model and order cancellation judgment algorithm in real time through a dynamic feedback mechanism; the feedback data sources include user-corrected document data and the model is retrained weekly. The optimization process employs online learning algorithms, automatically adjusting parameters (such as CNN convolution kernel weights and SVM kernel functions) with each data update, ensuring long-term stable operation of the system and consistency with the business rules of the financial management platform.