A compliance intelligent verification method for special fund use of a school cafeteria

Abstract

The application discloses a kind of school canteen special fund use compliance intelligent verification method and system, it is related to financial audit and computer vision technical field.The method includes: obtaining the financial accounting data and business field image data of canteen;Through image recognition technology, the category and component of the dishes distributed by the special fund corresponding window are determined;Combined with the pre-stored recipe database, the theoretical food material consumption data is calculated;Theoretical consumption data is associated and compared with the actual delivery cost recorded in financial accounts;According to the comparison result, compliance verification report is automatically generated.The system includes data acquisition interface, visual recognition unit, recipe database, verification engine and report generation unit for realizing the above method.The application realizes the automation, traceable verification of financial compliance requirements such as "special purpose fund" by fusing multi-modal data and intelligent analysis, changes traditional post-event manual audit into in-process intelligent prevention and control, effectively improves the transparency, accuracy and efficiency of fund supervision.

Description

Technical Field

[0001] This invention relates to the field of information and intelligent technology in financial auditing, and more specifically, to an intelligent verification method and system for the compliance of school canteen special funds, which integrates computer vision and multi-source data analysis. Background Technology

[0002] With the continuous increase in national investment in education and the growing emphasis on students' nutrition and health, the management and use of special fiscal funds for programs such as the "Nutrition Improvement Program" for rural compulsory education students have become a key focus of supervision. Multiple national ministries have successively issued a series of documents, including the "Implementation Measures for the Nutrition Improvement Program for Rural Compulsory Education Students" and the "Guiding Opinions on Strengthening the Supervision of Food Safety and Dietary Funds Management in Primary and Secondary Schools," which clearly require school canteens to maintain separate accounts and conduct independent accounting, ensuring that special funds are used for their designated purposes.

[0003] However, in current technological practices, compliance auditing of canteen financial management faces significant challenges. Current mainstream solutions primarily focus on financial accounting software, namely establishing canteen accounting systems based on the "Government Accounting System" to achieve dual-entry accounting for both financial and budgetary accounting. While such systems can standardize the bookkeeping process and generate financial statements, they are essentially still digital records and processing of manually entered business documents, and thus have the following inherent drawbacks: The separation of accounting records from actual operations, coupled with a lack of effective auditing methods, results in a disconnect between the system-recorded "financial accounts" and the actual "material accounts" and "service accounts" in the cafeteria kitchen. This makes it impossible to automatically verify whether ingredients purchased with "nutrition improvement program funds" are indeed processed into student meals and consumed by the target student group, rather than being diverted to faculty or staff meals or other mealtimes. This "last mile" verification gap leaves room for misappropriation of funds or management oversights.

[0004] Reliance on manual labor leads to low audit efficiency and a high risk of errors: Existing audit methods heavily depend on post-audit sampling of paper or electronic documents, as well as on-site manual inventory checks and inquiries. This method is time-consuming, labor-intensive, has a narrow scope, and is difficult to conduct continuous and thorough reviews of massive amounts of business data, resulting in insufficient objectivity and comprehensiveness in audit conclusions.

[0005] Data silos prevent the formation of a closed-loop chain of evidence: Procurement data, inventory data, financial data, and actual catering service data belong to different systems or ledgers, and are disconnected from each other. When compliance issues arise, it is difficult to quickly and automatically and accurately link cash flow, material entry and exit records, and specific catering activities in a temporal and spatial dimension to form a complete and irrefutable chain of audit evidence.

[0006] Therefore, there is an urgent need for a new technology solution that can break through the limitations of simple accounting, deeply integrate financial management with the operation of real business, and automatically, objectively and traceably complete the compliance verification of the use of special funds, so as to solve the technical problems existing in the current technology, such as difficulty in verifying accounts and actual assets, low audit efficiency, and broken evidence chains. Summary of the Invention

[0007] In a first aspect, this disclosure provides an intelligent verification method for the compliance of special funds used in school canteens, including the following steps: Multi-source data acquisition and alignment: Acquire the financial accounting data and business site image data of the target canteen within the target time period, and align and encapsulate the financial accounting data and the business site image data according to preset rules to generate verification data units; wherein, the financial accounting data includes raw material outbound cost records corresponding to specific special funds, recorded according to preset accounting subjects; Computer vision-based food service behavior recognition: The business site image data in the verification data unit is identified and analyzed to determine the type and portion information of dishes distributed through the service window corresponding to the specific special fund, and a food service behavior recognition list is generated. Theoretical ingredient consumption calculation: Based on the pre-stored recipe database, the theoretical ingredient consumption data is calculated according to the dish categories and portion information in the meal service behavior identification list; Financial-business data correlation verification: The theoretical food consumption data is correlated and compared with the raw material outbound cost records in the verification data unit; Generate and output verification report: Based on the comparison results, generate a verification report on the compliance of the use of the specific special funds during the target period.

[0008] Preferably, the step of identifying and analyzing the on-site image data specifically includes: From the image data of the business site, multiple food serving windows corresponding to different types of funds were located and identified; For the video stream of the food serving window corresponding to the specific special fund, perform frame-by-frame or keyframe analysis; The category of dishes distributed in the window is identified using image classification or object detection models; The portion size of the dish is estimated based on the proportion of the plate in the image.

[0009] Preferably, the step of comparing and correlating theoretical food consumption data with raw material outbound cost records specifically includes: Determine whether all food categories involved in the theoretical food consumption data are included in the raw material outbound cost record; Calculate the total cost of the theoretical food consumption data and compare it with the total amount of the raw material outbound cost record; If the food categories are fully included and the difference is below a preset threshold, then the accounts are deemed to be consistent with the actual situation. If the food categories are not fully included or the difference is higher than a preset threshold, it is determined to be an abnormality in the accounts and an alert is triggered.

[0010] Preferably, after the alert is triggered, it also includes: Automatically extract image data fragments and financial data records related to the aforementioned accounting discrepancies; The hash value of the image data fragment is bound to the financial data record and uploaded to the blockchain network for evidence storage, so as to generate an immutable audit trail.

[0011] Preferably, the method further includes a cost allocation step, which is performed before the acquisition of financial accounting data, including: Obtain the canteen's total procurement data and revenue data corresponding to various funds during the target time period; Based on preset allocation rules, the costs in the total procurement data are allocated to the specific special funds and other fund types to generate raw material outbound cost records corresponding to the specific special funds; wherein, the allocation rules include proportional allocation rules or allocation rules that prioritize specific special funds.

[0012] Secondly, this disclosure also provides an intelligent verification system for the financial compliance of school canteens, used to implement the method described in any one of the intelligent verification methods for the use of special funds for school canteens, the system comprising: The data acquisition and alignment module is used to perform the data acquisition and alignment steps as described in a smart verification method for the compliance of special funds used in school canteens, and to generate verification data units. The visual recognition and analysis module, connected to the data acquisition and alignment module, is used to recognize and analyze the image data in the verification data unit and output a list of food service behavior recognition. The theoretical consumption calculation module is connected to the visual recognition and analysis module and is used to call the recipe database and calculate the theoretical food consumption summary list based on the food service behavior recognition list. The intelligent verification and judgment module is connected to the data acquisition and alignment module and the theoretical consumption calculation module, respectively. It is used to compare and judge the theoretical food consumption summary list with the financial outbound records in the verification data unit and generate a verification conclusion. The report generation and output module is connected to the intelligent verification and judgment module and is used to generate and output a compliance verification report based on the verification conclusion.

[0013] Optionally, the system further includes: The cost allocation preprocessing module is used to calculate the total procurement cost based on preset allocation rules before the data acquisition and alignment module runs, and to provide the allocated financial data to the data acquisition and alignment module.

[0014] Optionally, the system further includes: The audit evidence solidification module is connected to the intelligent verification and judgment module and is used to perform credible evidence storage operation on key evidence information when the verification conclusion is abnormal. The present invention has the following advantages: (1) This invention achieves automated and traceable verification of compliance requirements for "dedicated funds for specific purposes," solving the technical challenge of auditing the separation of accounts and actual expenditures. By intelligently linking and comparing actual meal service data (dishes, portion sizes) identified using computer vision technology with accounting data in the financial accounting system, this invention establishes, for the first time, a direct and objective data link between "dedicated fund expenditures" and "meal service behavior at specific windows." This fundamentally changes the traditional auditing model that relies on post-event random checks of documents, achieving in-process, automated, and data-driven substantive supervision of the compliance of fund usage.

[0015] (2) This invention provides an intelligent auditing method that integrates multi-source heterogeneous data, improving regulatory efficiency and the credibility of evidence. This invention creatively integrates and analyzes three types of heterogeneous data—image streams, business documents, and financial accounts—within the same technical framework, and can solidify and preserve key verification evidence through technologies such as blockchain. This method not only significantly improves the efficiency and coverage of auditing work, but also ensures the integrity and immutability of the audit evidence chain through technical means, providing a highly credible technical basis for regulatory decisions. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. These drawings are incorporated in and constitute a part of this specification. They illustrate embodiments conforming to this disclosure and, together with the specification, serve to explain the technical solutions of this disclosure. It should be understood that the following drawings only show some embodiments of this disclosure and should not be considered as limiting the scope. Those skilled in the art can obtain other related drawings based on these drawings without creative effort.

[0017] Figure 1Data flow diagram of a method for intelligent verification of the compliance of special funds for school canteens provided in an embodiment of the present invention; Figure 2 A flowchart illustrating the steps of an intelligent verification method for the compliance of special funds used in school canteens, provided in an embodiment of the present invention; Figure 3 This is a system structure diagram of an intelligent verification system for the financial compliance of a school canteen, provided as an embodiment of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings to make the technical solution of the present invention clearer and more complete. It should be noted that the described embodiments are for illustrative purposes only and are not intended to limit the present invention. Other implementation methods that can be made by those skilled in the art based on the content of the present invention without creative effort should all fall within the protection scope of the present invention.

[0019] In this application and its claims, unless otherwise expressly stated, the terms "comprising," "including," and similar expressions should be understood to indicate the presence of the listed items without excluding the presence or addition of other items. The words "an," "a," and similar terms should not be construed as limited to the singular in this application, and may also include multiple items.

[0020] Furthermore, the accompanying drawings in this application are merely illustrative and not necessarily drawn to scale. The same reference numerals denote components with the same or similar functions. To clearly illustrate the present invention, specific details are provided in the following embodiments. Those skilled in the art should understand that these details are not essential for implementing the present invention, and other methods may be used to implement it without affecting the basic idea of ​​the invention.

[0021] The overall flow of the method involved in this invention is as follows: Figure 1 As shown, the details are as follows: Example

[0022] S101: Multi-source data acquisition and processing This step involves collecting raw data from two independent data sources—a financial software database and on-site camera footage from the cafeteria—and pairing and packaging them according to preset rules to form a standard data package for analysis in subsequent steps. The specific process is as follows: (1) Receive the task and read the "pairing rules". Upon receiving a verification task instruction (e.g., "Verify the use of nutrition meal funds on November 5, 2025"), the system first accesses a pre-set "configuration table." This table explicitly specifies: ① Which physical location of the camera (represented by the "window number", such as CK001) corresponds to which type of special fund (represented by the "fund code", such as FUND_NP).

[0023] ② Under which accounting subject in the financial software are all the expenses of this special fund recorded (represented by the "subject code", such as 5001.01).

[0024] (2) According to the rules, issue a "fetch data" command at the same time.

[0025] After obtaining the "pairing rules", the system sends instructions in two directions simultaneously: ① Sending a command to the accounting software: The system automatically generates a query command based on the "account code" (e.g., 5001.01) and the date specified in the task (e.g., 2025-11-05), and sends it to the accounting software's database. This command means: "Please send me all raw material outbound details (including product name, quantity, and amount) recorded under account 5001.01 as of November 5, 2025." ② Send a command to the video system: The system automatically generates a command based on the "window number" (e.g., CK001) and the date specified in the task, and sends it to the video management service. This command means: "Please send me all the key images (or video stream clips) captured by the camera installed in window CK001 on November 5, 2025." (3) Receive the raw data and assign "pairing labels" to them.

[0026] After receiving the raw data from the two channels, the system will immediately assign a "tag" to each piece of data for subsequent matching: ① Processing Financial Data: The accounting software returns a structured table. The system automatically appends the "fund code" and "window number" information obtained from the "pairing instructions" to each record in this table. For example, a record of "10 kg of eggs shipped out" is processed as: {Product Name: Eggs, Quantity: 10 kg, Amount: xx yuan, Fund Code: FUND_NP, Window Number: CK001}.

[0027] ② Image Data Processing: The data transmitted from the camera is an image or video stream. When the system receives each frame, it automatically embeds the precise time of capture (timestamp) and the camera's own identifier (which can be associated with the "window number"). For example, an image is packaged as: {Image Data: xxx, Timestamp: 2025-11-05 12:05:30, Window Number: CK001}.

[0028] (4) Pair by tag and package into a standard data package.

[0029] All tagged data is sent to a processing module. This module searches and matches data based on the two key tags: "window number" and "date". It finds all financial records and image data with "window number" CK001 and date 2025-11-05, binds them together, and packages them into a standard "verification data unit".

[0030] The output (i.e., the deliverable of this step) is a well-structured "verification data unit," with the following content paradigm:

[0031] S102: Computer vision-based food service behavior recognition.

[0032] This step receives the "verification data unit" output from step S101 and automatically analyzes the actual image data pointed to by the "on-site image index" to identify "who is at which window, what dishes are being distributed, and the approximate quantity". The specific process is as follows: (1) Use the target detection model to locate the food serving window.

[0033] The system inputs the image files from the "verification data unit" into a pre-trained object detection model. The model's specific structure is a single-stage detection architecture based on a convolutional neural network, and its workflow is as follows: ① Input and Feature Extraction: The original image (e.g., an RGB image with dimensions of 640x640 pixels) first enters the backbone network of the model (e.g., using a CSPDarknet or ResNet structure). This backbone network transforms the image into a series of feature maps at different scales through multiple layers of convolution, normalization, and activation function operations. These feature maps progressively capture abstract visual information from low to high, ranging from edges and textures to object parts.

[0034] ② Target localization and classification: The extracted feature maps are fed into the model's detection head. The detection head typically consists of several convolutional layers, responsible for predicting two types of key information in parallel at each preset location (anchor point) of the feature map: Bounding box: Predicts the offset and scaling of a rectangular box relative to its preset anchor point to locate the possible position of the "food serving window" in the image.

[0035] Category and Confidence: Simultaneously predict the probability (confidence) that the bounding box contains the target "food serving window", and classify whether it is the target object.

[0036] ③ Post-processing and output: The dense predictions output by the model are processed by a non-maximum suppression algorithm to eliminate overlapping detections of the same window. Finally, the system obtains one or more detection boxes, each containing precise pixel coordinates (e.g., [x_min, y_min, x_max, y_max]) and a confidence score (e.g., 0.95) that the location is the "serving window".

[0037] (2) Window area capture and dish recognition.

[0038] The system crops out the corresponding "food serving window" sub-image region from the original image based on the detection box coordinates output by the target detection model.

[0039] ① Invoking the image classification model: Input the sub-image into another pre-trained image classification model. This model typically uses a deep convolutional neural network structure (such as EfficientNet, Vision Transformer, etc.; this embodiment uses EfficientNet). Its core process is as follows: Feature encoding: Sub-images are transformed into a high-dimensional feature vector after passing through multiple convolutional or attention layers of the model. This vector comprehensively represents the visual content of the image.

[0040] Classification decision: The feature vector is passed to the fully connected classification layer of the model (usually one or more linear layers followed by a softmax function). Based on the learned pattern, the classification layer calculates the probability distribution of the image belonging to each preset dish category (such as "scrambled eggs with tomatoes", "braised pork", "rice", etc.).

[0041] ② Output dish category: The system selects the dish category with the highest probability as the result of this recognition. For example, the output is "Braised pork, confidence level 0.88".

[0042] (3) Component estimation based on reference objects.

[0043] To quantify the dish identification results into verifiable physical quantities, the system performs portion estimation: ① Reference Object Recognition and Scale Calibration: The system utilizes standard reference objects with known dimensions present in the image (such as a round dinner plate of a fixed model with a physical diameter of 28.0 cm). Image processing techniques (such as contour detection and template matching) are used to locate the pixel dimensions of the dinner plate in the image.

[0044] ②Pixel-Physical Size Conversion: Calculate the actual length corresponding to a unit pixel in the current image (e.g., 0.1 cm / pixel) based on "actual physical size of the reference object / pixel size of the reference object image".

[0045] ③ Dish Region Analysis and Estimation: For the identified dish region, the area occupied by the dish is roughly segmented using image segmentation algorithms (such as U-Net-based structures) or color threshold analysis. The pixel area of ​​this region is calculated, and then the actual length of the unit pixel obtained in the previous step is used to convert it into the approximate physical area or volume of the dish. Combined with the typical density of the dish (e.g., braised pork is about 1.2 grams per cubic centimeter, which has been preset in the system), the approximate weight of a single serving of the dish is finally estimated (e.g., "about 250 grams").

[0046] (4) Generate structured recognition results.

[0047] After completing the above analysis, this step outputs a structured "Meal Service Behavior Identification List" as an enhancement to the "Verification Data Unit". Its content format is as follows:

[0048] This completes the current step. Through a structured deep learning model and image processing algorithms, it parses the original image into a set of quantitative data records containing spatial location, dish type, and estimated portion size, providing direct "actual consumption" data input for the next step of automated comparison with financial data. S103: Calculation of theoretical food consumption.

[0049] This step receives the "verification data unit" output from step S102, which has been appended with the "meal service behavior identification list". Its core task is to convert the identified "dishes" and "portions" information into specific "raw materials" and "consumption" information, thereby constructing a theoretical consumption list that can be directly compared with the financial "outbound records". The specific process is as follows: (1) Locate and read the standard recipe database.

[0050] The system internally maintains a "standard recipe database". This database is stored in the form of structured data tables, such as a table named recipe_bom, whose core fields include: recipe_id (dish number) recipe_name (dish name, such as "braised pork" or "scrambled eggs with tomatoes") material_id (raw material number) material_name (name of raw material, such as "pork belly", "ginger", "sugar", "egg") standard_weight (standard weight in grams) material_unit (unit of measurement, such as "gram" or "milliliter") A typical record example is as follows:

[0051] (2) Perform dish matching and BOM query The system iterates through the "Meal Service Behavior Identification List" in the "Verification Data Unit" and performs the following operations on each record: ① Matching Dishes: Using the "dish name" (e.g., "braised pork") in the recognition results as the query key, the system performs an exact match in the recipe_bom table to find the corresponding recipe_id and all standard bills of materials for that dish. "If the recognized dish is not in the recipe library, it is marked as 'unknown dish' and manual review is triggered," to demonstrate the system's robustness. ② Query results: The complete record set of the standard BOM for this dish is obtained, which constitutes a list of all the raw materials required for a standard "braised pork" and their theoretical proportions.

[0052] (3) Calculate the scaling ratio based on the identified components.

[0053] Because the portion size identified by S102 is an "estimated portion" (e.g., "approximately 250 grams"), it may differ from the "standard serving size" in a standard recipe (e.g., "one serving of braised pork is 200 grams"). The system needs to perform a proportional conversion. ① Calculate the scaling factor: Divide the "estimated portion" (250 grams) output by S102 by the "standard portion" (200 grams) of the dish in the database to obtain a scaling factor (250 / 200=1.25). This means that the identified dish is 1.25 times the standard portion.

[0054] ② Calculate theoretical raw material consumption: Multiply the standard_weight of each raw material in the standard BOM by this scaling factor to obtain the theoretical raw material consumption corresponding to this identification action. An example of the calculation process is as follows: Theoretical consumption of skinless pork belly = 200g * 1.25 = 250g Theoretical sugar consumption = 15 grams * 1.25 = 18.75 grams Theoretical consumption of cooking wine = 10 ml * 1.25 = 12.5 ml (4) Compile and output a list of theoretical food consumption.

[0055] After completing the above calculations for all records in the "Meal Supply Behavior Identification List", the system merges and summarizes them according to the raw material name to generate a total theoretical consumption list for this verification task and specific window.

[0056] The output is a "Summary List of Theoretical Ingredient Consumption" appended to the continuously expanding "Verification Data Units". Its paradigm is as follows:

[0057] This completes the current step. It successfully transforms and summarizes the visually recognized food service behavior, which is based on "dishes," into theoretical consumption data with clear quantitative values, based on "raw materials." This provides a basis for quantifiable comparison with the financial outbound records obtained in step S101, which are also based on "raw materials." S104: Verification of the correlation between financial and business data.

[0058] This step receives the "verification data unit" that has completed the first three steps. Its core task is to perform automated auditing, comparing the "theoretical food consumption summary list" obtained in S103 with the "financial material list" obtained in S101, and determining the compliance of fund usage based on preset rules. The specific process is as follows: (1) Data alignment and matching.

[0059] The system first performs a precise match between the "Theoretical Ingredient Consumption Summary List" and the "Financial Material List" within the "Verification Data Unit." The matching criteria are the same verification date, the same associated window, and the same raw material name. The system creates an internal comparison workspace, aligning the data from both sides by raw material name, resulting in the following temporary structure:

[0060] (2) Calculate using preset compliance rules The system loads a pre-defined compliance verification rule library. For each aligned "raw material" work item, the system automatically performs calculations, the core of which is comparing the values ​​of both sides: ① Calculate the difference: The system calculates the difference between the financial outbound quantity and the theoretical consumption quantity. For example, the difference for eggs is 1400-1350=+50 grams (50 grams more in the accounts), and the difference for pork belly is 0-3125=-3125 grams (3125 grams less in the accounts).

[0061] ② Application Rules: The system compares the difference values ​​with preset rule thresholds. Rules typically include: Tolerance threshold: For example, "For staple foods such as rice, noodles, and oil, the difference between the recorded amount and the actual amount is allowed to be within ±5%; for fresh meat, the difference is allowed to be within ±8%." If the difference is within the threshold, it is marked as "compliant".

[0062] Zero or Negative Value Warning Rule: If the "Financial Outbound Quantity" is 0 or much less than the "Theoretical Consumption Quantity" (resulting in a large negative value), a higher level of inspection will be triggered. For example, the rule may stipulate: "If the theoretical consumption of high-priced raw materials (such as meat and aquatic products) is >0, but the outbound quantity of its corresponding special fund account is =0, then a warning must be triggered." (3) Generate verification conclusions and evidence chains.

[0063] Based on the results calculated according to the rules, the system generates the final verification conclusion.

[0064] ① Conclusion generation: If the comparison results of all raw materials are "compliant", then the overall conclusion "Accounts and physical inventory match, verification passed" is generated.

[0065] ② Warning Generation: If any raw material triggers the warning rule, a conclusion of "Verification Anomaly, Suspected Cost Misappropriation Found" will be generated. The system will pinpoint the exact problem, for example: "Anomaly: Window CK001, Date 2025-11-05, Raw Material 'Skinless Pork Belly'. Theoretical consumption 3125 grams, financial outflow 0 grams. This raw material is not reflected in the nutrition meal special fund account, posing a risk of being misappropriated by other funds." ③ Constructing a chain of evidence: The system automatically associates and binds all data involved in this verification (including the specific image ID that caused the problem, identification records, BOM calculation process, and financial voucher number) to form a complete and traceable chain of evidence.

[0066] The output is a final "Compliance Verification Report," which serves as the final output format of the "Verification Data Unit." Its paradigm is as follows:

[0067] This completes the current step. Through automated data comparison and rule-based judgment, it achieves intelligent auditing of the compliance of special fund usage, transforming the process of "manually checking accounts and reconciling invoices" into a data-driven, rule-based, and automatically generated technical process that generates conclusions and evidence. S105: Generate and output verification report.

[0068] This step receives the "verification data unit" output from step S104, which already contains the final "verification conclusion" and a complete chain of evidence. Its core task is to automatically convert this structured data into a formal report that is both readable and rigorous, suitable for management review and decision-making, and then archive and distribute it. The specific process is as follows: (1) Select a report template and fill in the data.

[0069] The system calls a preset report template based on the conclusion type (such as "Pass" or "Abnormal") in the "Verification Conclusion". This template defines the report's title, chapter structure (such as overview, conclusion summary, detailed analysis, evidence index), style, and fixed narrative paragraphs.

[0070] (1) Data Mapping: The system automatically fills the key fields in the "Verification Data Unit" into the corresponding placeholders in the template. For example, the verification date and verification window are filled into the report header, and the conclusion is filled into the conclusion summary.

[0071] (2) Natural Language Generation: For core verification findings, the system combines structured comparison data (such as theoretical consumption, financial outbound volume, and difference values) with preset text fragments to automatically generate a descriptive and easy-to-understand conclusion.

[0072] ① Example (Passing Status): "On November 5, 2025, the verification for the nutrition meal window (CK001) was completed. The system identified and calculated the consumption of the main dishes for the day. After comparing with the financial outbound records under the Business Activity Expenses - Nutrition Meal account, the discrepancies between the accounts and actual consumption of the main raw materials such as eggs and rice were all within the reasonable loss range of ±5%. Conclusion: The use of nutrition improvement plan funds for the day was compliant, and the verification passed." ② Example (Abnormal Situation): "Warning Detection: On November 5, 2025, a high-value dish, braised pork belly, was identified at the nutrition meal window (CK001), theoretically consuming 3125 grams of skinless pork belly. However, upon verification, there were no records of any pork raw materials being issued under the 'Business Activity Expenses - Nutrition Meal' account on that day. Judgment Basis: This situation triggered the rule of 'missing expenditures of high-priced raw materials in special fund accounts.' Recommendation: This anomaly indicates that special funds may have been misappropriated. It is recommended to immediately conduct a manual audit of the day's food procurement, requisition, and window meal service process." (2) Embed visual charts and evidence links.

[0073] To present the data more intuitively, the system automatically generates and embeds charts: ① Chart Generation: For example, generate a bar chart to compare the "theoretical consumption" and "financial outbound quantity" of the main raw materials side by side, making the differences immediately apparent. Outliers are highlighted with a prominent color (such as red).

[0074] ② Embedded Chain of Evidence: Key conclusions in the report are linked to via hyperlinks or QR codes. For example, clicking the text "Identified Braised Pork" leads to the original image; clicking "No Outbound Records" links to a blank query results page or relevant logs for that item in the financial system. These links point to the original data and intermediate results stored in the system, ensuring the report's traceability.

[0075] (3) Report formatting, output, and distribution.

[0076] ① Formatting: The system will fill in the complete report content and render it into the specified file format, such as a PDF document or an HTML webpage.

[0077] ② Electronic signature and evidence storage: After the final report file is generated, the system calls the electronic signature service or trusted timestamp service to add a digital signature and timestamp to the report file, ensuring the authenticity and integrity of the report, and can be archived as audit evidence.

[0078] ③ Distribution: According to preset rules, the system automatically sends the report to the designated responsible person (such as the cafeteria manager, school finance director, or auditor) via internal messaging system, email, or integrated office platform. At the same time, the report and the complete "verification data unit" are archived to the historical database for future retrieval and audit review.

[0079] The output is a formal, legally and technically binding electronic verification report, along with archived system logs. The report itself is the final deliverable of the entire technical process, marking the completion of a closed loop for an automated compliance verification task.

[0080] Furthermore, when the verification conclusion is abnormal, the system automatically packages the key information of the warning event (time, window, identified dish, related account summary, and hash value of related image frame), and uploads the hash value of the data packet to the blockchain by calling a third-party blockchain evidence storage service API to generate tamper-proof audit evidence.

[0081] The overall process of the method involved in this invention described above is summarized as follows: Figure 2 As shown. In addition, the present invention also provides a supplementary embodiment, as follows: S100 (Supplementary Example): Cost Allocation Preprocessing As a preferred implementation, to improve the accuracy of verification at the data source, a cost allocation preprocessing step can be added before step S101. This step aims to address the issue that canteen procurement often involves multiple meals and multiple types of funds being purchased in a unified manner, and the financial records are not directly broken down to specific funds.

[0082] The specific process is as follows: ① Data preparation: The system acquires the total procurement and inventory data of the canteen within the target verification period (including all raw material categories, quantities, and total amounts), as well as the actual income data of various funds (such as income from the nutrition improvement program and income from students' self-paid meal fees).

[0083] ② Execute the amortization algorithm: Invoking Allocation Rules: The system uses preset allocation rules. There are two main types of rules: Priority allocation rule: The algorithm first ensures that the "Nutrition Improvement Program Fund" bears the full cost based on its actual income amount, and the remaining cost is then allocated to other funds according to their income ratio. This algorithmically prioritizes and guarantees the full utilization of the special funds.

[0084] Proportional allocation rule: All procurement costs are allocated proportionally according to the revenue share of each type of fund.

[0085] Automatic calculation: The system performs mathematical calculations based on the selected rules. For example, when using proportional allocation, the allocated cost of a certain fund = total procurement cost × (current period revenue of the fund / total current period revenue of the canteen).

[0086] ③ Generate allocation results: The algorithm outputs an allocation details table, clearly listing the specific amount and quantity of each purchase cost that should be attributed to "Nutrition Improvement Program Funds", "Student Self-Funding Funds", etc.

[0087] ④ Data Integration: This "allocation details table" will serve as a more accurate and direct basis for obtaining "financial accounting data" in step S101. The system can directly query or simulate the generation of "raw material outbound cost records" based on the allocation results and detailed by fund type. This ensures that subsequent verification (S104) is based on reasonably segmented and clearly targeted financial data, greatly enhancing the accuracy and reliability of the verification conclusions.

[0088] This S100 step, as an optional pre-enhancement step, reflects the present invention's intervention and control from the source of financial data processing, solidifying compliance requirements through algorithms, and constructing a full-chain, tamper-proof automated supervision technology solution for "procurement-allocation-verification". Example

[0089] like Figure 3 As shown, the present invention also provides an intelligent verification system for the compliance of special funds used in school canteens, for implementing the method described in the foregoing embodiments. The system includes: The data acquisition and alignment module is used to connect the financial accounting software and the canteen on-site camera in parallel through configurable rules, and perform multi-source data acquisition, standardized encapsulation and spatiotemporal alignment operations as described in S101 to generate structured verification data units. The visual recognition and analysis module, connected to the data acquisition and alignment module, is used to perform window positioning, dish recognition and portion estimation based on the target detection and image classification model as described in S102 on the image data in the verification data unit, and generate a list of food service behavior recognition. The theoretical consumption calculation module is connected to the visual recognition and analysis module. It is used to call the pre-stored recipe database and, based on the food service behavior recognition list, perform dish matching and scaling calculation as described in S103 to generate a theoretical ingredient consumption summary list. The intelligent verification and judgment module is connected to the data acquisition and alignment module and the theoretical consumption calculation module, respectively. It is used to compare the theoretical food consumption summary list with the financial outbound records in the verification data unit, and perform the preset compliance rule calculation as described in S104 to generate verification conclusions and abnormal details. The report generation and output module is connected to the intelligent verification and judgment module, and is used to perform the structured report automatic generation, visualization chart embedding and electronic signature distribution operations as described in S105 based on the verification conclusion and anomaly details.

[0090] Optionally, the system further includes: The cost allocation preprocessing module is used to perform automatic allocation calculation of total procurement costs based on priority or proportional allocation rules as described in S100 before the data acquisition and alignment module runs, so as to provide an accurate and detailed financial data base for verification.

[0091] The audit evidence solidification module is connected to the intelligent verification and judgment module. When the verification conclusion is abnormal, it automatically extracts key evidence information and performs blockchain hash storage or trusted timestamp solidification operations by calling the third-party trusted evidence storage service interface. For a detailed description of this embodiment, please refer to the corresponding descriptions in the foregoing embodiments, which will not be repeated here.

[0092] The basic principles of this disclosure have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this disclosure are merely examples and not limitations, and should not be considered as essential features of each embodiment of this disclosure. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the scope of this disclosure to the necessity of employing the aforementioned specific details for implementation.

[0093] In this disclosure, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The block diagrams of devices, apparatuses, devices, and systems involved in this disclosure are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as "comprising," "including," "having," etc., are open-ended terms meaning "including but not limited to," and are used interchangeably with them. The terms "or" and "and" as used herein refer to the terms "and / or," and are used interchangeably with them unless the context clearly indicates otherwise. The term "such as" as used herein refers to the phrase "such as but not limited to," and is used interchangeably with it.

[0094] Additionally, as used herein, the "or" used in a list of items beginning with "at least one" indicates a separate list, such that a list of, for example, "at least one of A, B, or C" means A or B or C, or AB or AC or BC, or ABC (i.e., A and B and C). Furthermore, the word "exemplary" does not imply that the described example is preferred or better than other examples.

[0095] It should also be noted that in the systems and methods of this disclosure, the components or steps can be decomposed and / or recombined. These decompositions and / or recombinations should be considered as equivalent solutions to this disclosure.

[0096] Various changes, substitutions, and modifications can be made to the technology described herein without departing from the teachings defined by the appended claims. Furthermore, the scope of the claims of this disclosure is not limited to the specific aspects of the processes, machines, manufactures, events, means, methods, and actions described above. Currently existing or later-developed processes, machines, manufactures, events, means, methods, or actions that perform substantially the same function or achieve substantially the same result as the corresponding aspects described herein can be utilized. Therefore, the appended claims include such processes, machines, manufactures, events, means, methods, or actions within their scope.

[0097] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of this disclosure. Therefore, this disclosure is not intended to be limited to the aspects shown herein, but rather to be carried out within the widest scope consistent with the principles and novel features disclosed herein.

[0098] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this disclosure to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A method for intelligent verification of the compliance of special funds used in school canteens, characterized in that, Includes the following steps: Multi-source data acquisition and alignment: Acquire the financial accounting data and business site image data of the target canteen within the target time period, and align and encapsulate the financial accounting data and the business site image data according to preset rules to generate verification data units; wherein, the financial accounting data includes raw material outbound cost records corresponding to specific special funds, recorded according to preset accounting subjects; Computer vision-based food service behavior recognition: The business site image data in the verification data unit is identified and analyzed to determine the type and portion information of dishes distributed through the service window corresponding to the specific special fund, and a food service behavior recognition list is generated. Theoretical ingredient consumption calculation: Based on the pre-stored recipe database, the theoretical ingredient consumption data is calculated according to the dish categories and portion information in the meal service behavior identification list; Financial-business data correlation verification: The theoretical food consumption data is correlated and compared with the raw material outbound cost records in the verification data unit; Generate and output verification report: Based on the comparison results, generate a verification report on the compliance of the use of the specific special funds during the target period.

2. The method according to claim 1, characterized in that, The steps for identifying and analyzing on-site image data specifically include: From the image data of the business site, multiple food serving windows corresponding to different types of funds were located and identified; For the video stream of the food serving window corresponding to the specific special fund, perform frame-by-frame or keyframe analysis; The category of dishes distributed in the window is identified using image classification or object detection models; The portion size of the dish is estimated based on the proportion of the plate in the image.

3. The method according to claim 1, characterized in that, The step of correlating and comparing theoretical food consumption data with raw material outbound cost records specifically includes: Determine whether all food categories involved in the theoretical food consumption data are included in the raw material outbound cost record; Calculate the total cost of the theoretical food consumption data and compare it with the total amount of the raw material outbound cost record; If the food categories are fully included and the difference is below a preset threshold, then the accounts are deemed to be consistent with the actual situation. If the food categories are not fully included or the difference is higher than a preset threshold, it is determined to be an abnormality in the accounts and an alert is triggered.

4. The method according to claim 3, characterized in that, After the alert is triggered, it also includes: Automatically extract image data fragments and financial data records related to the aforementioned accounting discrepancies; The hash value of the image data fragment is bound to the financial data record and uploaded to the blockchain network for evidence storage, so as to generate an immutable audit trail.

5. The method according to claim 1, characterized in that, The method further includes a cost allocation step, which is performed before the acquisition of financial accounting data, including: Obtain the canteen's total procurement data and revenue data corresponding to various funds during the target time period; Based on preset allocation rules, the costs in the total procurement data are allocated to the specific special funds and other fund types to generate raw material outbound cost records corresponding to the specific special funds; wherein, the allocation rules include proportional allocation rules or allocation rules that prioritize specific special funds.

6. A smart verification system for the financial compliance of school canteens, characterized in that, The system for implementing the method of any one of claims 1 to 5 comprises: The data acquisition and alignment module is used to perform the data acquisition and alignment steps as described in claim 1 and generate verification data units; The visual recognition and analysis module, connected to the data acquisition and alignment module, is used to recognize and analyze the image data in the verification data unit and output a list of food service behavior recognition. The theoretical consumption calculation module is connected to the visual recognition and analysis module and is used to call the recipe database and calculate the theoretical food consumption summary list based on the food service behavior recognition list. The intelligent verification and judgment module is connected to the data acquisition and alignment module and the theoretical consumption calculation module, respectively. It is used to compare and judge the theoretical food consumption summary list with the financial outbound records in the verification data unit and generate a verification conclusion. The report generation and output module is connected to the intelligent verification and judgment module and is used to generate and output a compliance verification report based on the verification conclusion.

7. The system according to claim 6, characterized in that, The system also includes: The cost allocation preprocessing module is used to calculate the total procurement cost based on preset allocation rules before the data acquisition and alignment module runs, and to provide the allocated financial data to the data acquisition and alignment module.

8. The system according to claim 6, characterized in that, The system also includes: The audit evidence solidification module is connected to the intelligent verification and judgment module and is used to perform credible evidence storage operation on key evidence information when the verification conclusion is abnormal.

Images