Method for processing agricultural insurance business data and related device

Abstract

The application provides a kind of agricultural insurance business data processing method and related equipment, applicable to the field of financial technology business. Including: obtaining target agricultural insurance business data, target agricultural insurance business data includes target accessory material data and corresponding target business metadata;Target accessory material data is extracted by multi-channel feature, to obtain text content feature and image visual feature;Text content feature is analyzed by semantic analysis model, to determine the document category and key business entity of target accessory material data;The authenticity of image visual feature is analyzed, and a tampering risk assessment result is generated;Based on key business entity, target business metadata and tampering risk assessment result, multi-dimensional logic verification is performed, and a plurality of dimensions of verification factor is obtained;According to the verification factor of multiple dimensions, an audit result is generated.The application can automatically audit agricultural insurance business data, improve the efficiency and objectivity of agricultural insurance business data auditing.

Description

Technical Field

[0001] This application applies to the field of financial technology business, and in particular relates to a method and related equipment for processing agricultural insurance business data. Background Technology

[0002] In the agricultural insurance industry, data processing is a core link connecting insurance companies and farmers. With the expansion of agricultural insurance coverage, the volume of business data has surged. How to process this data quickly and accurately directly impacts farmers' production recovery and insurance companies' risk management capabilities.

[0003] In related technologies, the processing of business data in agricultural insurance mainly relies on the human experience of auditors for judgment. For example, in the claims process, auditors need to manually review the reports, on-site photos, or documents submitted by farmers, and combine their own experience to judge the authenticity of the disaster and the extent of the loss. However, on the one hand, human judgment is highly subjective, and different auditors may have different standards for handling the same case; on the other hand, faced with massive amounts of case data, manual review is inefficient and easily leads to a backlog of cases. Summary of the Invention

[0004] The main objective of this application is to propose a method and related equipment for processing agricultural insurance business data, which can automate the review of agricultural insurance business data and improve the efficiency and objectivity of the review.

[0005] To achieve the above objectives, a first aspect of this application proposes a method for processing agricultural insurance business data, the method comprising: Acquire target agricultural insurance business data, which includes target ancillary material data and corresponding target business metadata; Multi-channel feature extraction is performed on the target attachment material data to obtain text content features and image visual features, respectively; The text content features are analyzed using a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data; The visual features of the image are analyzed for authenticity to generate a tampering risk assessment result; Based on the key business entity, the target business metadata, and the tampering risk assessment results, multi-dimensional logical verification is performed to obtain verification factors in multiple dimensions; The audit results for the target agricultural insurance business data are generated based on the verification factors across multiple dimensions.

[0006] In some embodiments, the multi-channel feature extraction of the target attachment material data to obtain text content features and image visual features includes: Optical character recognition is performed on the target accessory material data to obtain a text character sequence and the coordinate information corresponding to the text character sequence, which are used as the text content features; Target detection is performed on the target accessory material data to obtain the target image and the texture and edge features of the target image, which are used as the visual features of the image.

[0007] In some embodiments, the step of performing authenticity analysis on the visual features of the image to generate a tampering risk assessment result includes: The target image is recompressed to obtain the corresponding recompressed image; Calculate the difference distribution between the target image and the recompressed image, and determine the compression artifact differences in local regions based on the difference distribution; The target image is converted to the frequency domain, high-frequency coefficient distribution features are extracted, and the presence of periodic correlated noise fingerprints is detected based on the high-frequency coefficient distribution features. Based on the compression artifact differences and the periodic correlation noise fingerprint, the tampering risk assessment result for the target image is calculated.

[0008] In some embodiments, the semantic analysis model is constructed by fine-tuning a pre-defined corpus in the agricultural insurance field based on a general large language model. The process of parsing the text content features using the pre-constructed semantic analysis model to determine the document category and key business entities of the target attachment material data includes: Construct prompt words, which include step-by-step reasoning paths and preset output format constraints for guiding the semantic analysis model to generate intermediate analysis processes; The prompt word instruction and the text content features are input into the semantic analysis model so that the semantic analysis model can extract the structured key business entities based on the document category determined by the step-by-step reasoning path and the document category.

[0009] In some embodiments, the target business metadata includes geospatial range data and business rule constraint data of the insured land plot; the target ancillary material data includes image data carrying spatial location tags; and the multi-dimensional logical verification based on the key business entity, the target business metadata, and the tampering risk assessment result yields multiple verification factors, including: Perform spatial topology inclusion verification based on the spatial location label and the geospatial range data to obtain a spatial logical verification factor; The time-series features and numerical features are obtained from the key business entities. Business rule matching verification is performed based on the time-series features, the numerical features and the business rule constraint data to obtain the business logic verification factor.

[0010] In some embodiments, generating audit results for the target agricultural insurance business data based on the verification factors across multiple dimensions includes: The tampering risk assessment results, the spatial logic verification factor, and the business logic verification factor are input into a preset fusion decision rule engine; If the tampering risk assessment result indicates that it has not been tampered with, and both the spatial logic verification factor and the business logic verification factor have passed the verification, then an automatically approved audit conclusion will be output. If the tampering risk assessment result indicates the presence of risk, or if any verification factor fails the verification, a manual review conclusion containing the specific risk cause code will be output, and a risk marker will be generated.

[0011] In some embodiments, the method further includes: Obtain the results of manual review of the target agricultural insurance business data; Target agricultural insurance business data whose manual review results are inconsistent with the audit results are marked as difficult samples; The semantic analysis model and image identification model are incrementally trained using the difficult samples.

[0012] To achieve the above objectives, a second aspect of this application provides an apparatus for processing agricultural insurance business data, the apparatus comprising: The acquisition module is used to acquire target agricultural insurance business data, which includes target attachment material data belonging to the multimodal type and corresponding target business metadata. The feature extraction module is used to perform multi-channel feature extraction on the target attachment material data to obtain text content features and image visual features respectively; The parsing module is used to parse the text content features through a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data; The authenticity analysis module is used to perform authenticity analysis on the visual features of the image and generate a tampering risk assessment result. The logic verification module is used to perform multi-dimensional logic verification based on the key business entity, the target business metadata and the tampering risk assessment result, and obtain verification factors in multiple dimensions. The generation module is used to generate audit results for the target agricultural insurance business data based on the verification factors in multiple dimensions.

[0013] To achieve the above objectives, a third aspect of this application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0014] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0015] The method and related equipment for processing agricultural insurance business data proposed in this application include: acquiring target agricultural insurance business data, which includes target ancillary material data and corresponding target business metadata; performing multi-channel feature extraction on the target ancillary material data to obtain text content features and image visual features; parsing the text content features using a pre-built semantic analysis model to determine the document category and key business entities of the target ancillary material data; performing authenticity analysis on the image visual features to generate a tampering risk assessment result; performing multi-dimensional logical verification based on the key business entities, target business metadata, and tampering risk assessment result to obtain multiple-dimensional verification factors; and generating an audit result for the target agricultural insurance business data based on the multiple-dimensional verification factors.

[0016] According to the agricultural insurance business data processing method proposed in this application, the following steps are taken: First, target agricultural insurance business data, including target attachment material data and corresponding target business metadata, is acquired, providing a comprehensive and accurate input data source for subsequent multimodal intelligent analysis. Next, multi-channel feature extraction is performed on the target attachment material data to obtain text content features and image visual features, achieving digital feature decoupling of unstructured attachment data and enabling perception of data content from different dimensions. Then, a pre-built semantic analysis model is used to parse the text content features, determining document categories and key business entities, deeply understanding the business semantics of the attachments, and extracting core information. Simultaneously, image visual features are analyzed for authenticity, generating a tampering risk assessment result, effectively identifying forgery traces that are difficult to detect with the naked eye, ensuring the authenticity of claims materials. Then, multi-dimensional logical verification is performed based on key business entities, target business metadata, and the tampering risk assessment result, obtaining multiple verification factors. Cross-validation of multi-source data ensures the consistency of business logic in time, space, and content. Finally, an audit result is generated based on the multiple verification factors. Through the above steps, this application enables automated review of agricultural insurance business data, improving the efficiency and objectivity of the review process.

[0017] Other features and advantages of this disclosure will be set forth in the following description and will be apparent in part from the description or may be learned by practicing the disclosure. The objectives and other advantages of this disclosure may be realized and obtained by means of the structures particularly pointed out in the description, claims and drawings. Attached Figure Description

[0018] Figure 1 This is a flowchart of the agricultural insurance business data processing method provided in the embodiments of this application; Figure 2 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 3 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 4 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 5 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 6 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 7 This is another flowchart of the method for processing agricultural insurance business data provided in the embodiments of this application; Figure 8 This is a schematic diagram of the structure of the agricultural insurance business data processing device provided in the embodiments of this application; Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0022] In the agricultural insurance industry, data processing is a core link connecting insurance companies and farmers. With the expansion of agricultural insurance coverage, the volume of business data has surged. How to process this data quickly and accurately directly impacts farmers' production recovery and insurance companies' risk management capabilities.

[0023] In related technologies, the processing of business data in agricultural insurance mainly relies on the human experience of auditors for judgment. For example, in the claims process, auditors need to manually review the reports, on-site photos, or documents submitted by farmers, and combine their own experience to judge the authenticity of the disaster and the extent of the loss. However, on the one hand, human judgment is highly subjective, and different auditors may have different standards for handling the same case; on the other hand, faced with massive amounts of case data, manual review is inefficient and easily leads to a backlog of cases.

[0024] Based on this, embodiments of this application provide a method and related equipment for processing agricultural insurance business data, which can automatically review agricultural insurance business data and improve the efficiency and objectivity of the review of agricultural insurance business data.

[0025] The agricultural insurance business data processing method and related equipment provided in this application are specifically described through the following embodiments. First, the agricultural insurance business data processing method in this application embodiment is described.

[0026] The agricultural insurance business data processing method provided in this application is applicable to the fields of fintech, medical services, and artificial intelligence. This method can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the agricultural insurance business data processing method, but is not limited to the above forms.

[0027] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0028] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirection to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data required for the proper functioning of these embodiments acquired.

[0029] Figure 1 This is an optional flowchart of the agricultural insurance business data processing method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S106.

[0030] Step S101: Obtain target agricultural insurance business data, which includes target attachment material data and corresponding target business metadata.

[0031] Step S102: Perform multi-channel feature extraction on the target attachment material data to obtain text content features and image visual features respectively.

[0032] Step S103: The text content features are analyzed using a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data.

[0033] Step S104: Perform authenticity analysis on the visual features of the image to generate a tampering risk assessment result.

[0034] Step S105: Perform multi-dimensional logical verification based on key business entities, target business metadata, and tampering risk assessment results to obtain verification factors in multiple dimensions.

[0035] Step S106: Generate audit results for the target agricultural insurance business data based on multiple dimensions of verification factors.

[0036] Steps S101 to S106 of this embodiment first acquire target agricultural insurance business data, including target attachment material data and corresponding target business metadata, providing a comprehensive and accurate input data source for subsequent multimodal intelligent analysis. Next, multi-channel feature extraction is performed on the target attachment material data to obtain text content features and image visual features, achieving digital feature decoupling of unstructured attachment data and enabling perception of data content from different dimensions. Subsequently, a pre-built semantic analysis model is used to parse the text content features, determining document categories and key business entities, deeply understanding the business semantics of the attachments, and extracting core information. Simultaneously, authenticity analysis is performed on the image visual features, generating a tampering risk assessment result, effectively identifying forgery traces that are difficult to detect with the naked eye, ensuring the authenticity of the claims materials. Then, multi-dimensional logical verification is performed based on key business entities, target business metadata, and the tampering risk assessment result, obtaining multiple verification factors. Cross-validation of multi-source data ensures the consistency of business logic in time, space, and content. Finally, an audit result is generated based on the multiple verification factors. Through the above steps, this application enables automated review of agricultural insurance business data, improving the efficiency and objectivity of the review process.

[0037] In step S101 of some embodiments, target agricultural insurance business data is acquired, which serves as the input to the entire processing flow. Specifically, this involves receiving a data packet containing multimodal information through an API interface with the core insurance business system. The target supporting documentation data refers to original evidence documents collected during the underwriting or claims process. This data typically exists in the form of unstructured images (such as JPEG or PNG format on-site inspection photos or ID card photos) or documents (such as PDF format land transfer contracts or scanned invoices). These images usually contain spatial location tags to be verified (such as latitude and longitude in EXIF ​​information). The corresponding target business metadata refers to structured benchmark data already entered into the business system. This data serves as the truth standard for subsequent verification and specifically covers the geographical boundary data of the insured land, the insurance liability period stipulated in the policy, the maximum insured amount of the insured object, and related business rule constraint data, such as deductibles and premium factors.

[0038] In step S102 of some embodiments, multi-channel feature extraction is performed on the target attachment material data. This aims to transform the human-readable attachment into feature vectors that can be processed by computer algorithms, employing a strategy of parallel processing of text and visual channels. In the text channel, an Optical Character Recognition (OCR) model is invoked to scan the attachment, not only recognizing the content of text symbols but also extracting the absolute coordinate information of each character or text line on the page, thereby generating a text character sequence containing positional attributes, i.e., text content features. In the visual channel, a target detection algorithm (such as YOLO or Faster R-CNN) is used to perform a full-image scan of the attachment, locating target image regions with business significance (e.g., official seal area, signature area, close-up area of ​​damaged crops). Subsequently, feature encoding is performed on these target images to extract image visual features reflecting image texture roughness, edge gradient distribution, and color histogram, providing a data foundation for subsequent anti-counterfeiting analysis.

[0039] Please see Figure 2 In some embodiments, step S102 may include, but is not limited to, steps S201 to S202.

[0040] Step S201: Perform optical character recognition on the target attachment material data to obtain the text character sequence and the coordinate information corresponding to the text character sequence, which are used as text content features.

[0041] Step S202: Target detection is performed on the target accessory material data to obtain the target image and the texture and edge features of the target image as image visual features.

[0042] In step S201 of some embodiments, the optical character recognition engine receives preprocessed target attachment material data, such as images of land transfer contracts or VAT invoices for seed purchases, and locates text regions in the images. For each located line or individual text block, the recognition network converts it into a corresponding computer-encoded text character sequence, such as recognizing specific character content like "2025", "3000 yuan", or "corn affected by disaster". Simultaneously, the OCR engine outputs pixel-level coordinate information for each text character sequence in the original image. This coordinate information can be presented as the coordinates of the four surrounding vertices (x, y) or the center point coordinates plus width and height, clearly defining the spatial position of the text on the page. These text character sequences and their corresponding coordinate information together constitute the text content features, preserving both the semantic content and the page layout information of the document. This provides a structured data foundation for subsequently distinguishing key-value pair relationships in forms (such as the proximity of the text "amount" to the specific number "3000").

[0043] In step S202 of some embodiments, the target detection algorithm (such as the YOLO series or Faster R-CNN) performs a full-image scan of the target attachment material data, identifies and locates regions of interest with specific business significance based on pre-trained category labels, and crops these regions from the original background to form the target image. For example, it locates and crops the red official seal area from a scanned contract, or locates and crops specific disaster-stricken and collapsed areas from a large-scale aerial photograph of farmland. Subsequently, the target image undergoes low-level visual feature calculation: texture features reflecting the surface roughness, graininess, and regularity of the image are extracted through gray-level co-occurrence matrix or convolutional neural network filters; edge features reflecting drastic changes in image grayscale are extracted using edge detection operators. These texture features and edge features together constitute the image visual features, describing the visual attributes of the target object at the physical pixel level.

[0044] Through steps S201 to S202 described above, this embodiment of the application utilizes optical character recognition technology to accurately extract text attributes from attachments, achieving digital reconstruction of document layout information. Simultaneously, it employs object detection and low-level feature calculation technologies to extract core business objects from the background and quantify their physical visual attributes. This dual-channel feature extraction mechanism effectively decouples unstructured multimodal attachments into structured text layout features and "image physical features," providing high-quality feature inputs for subsequent in-depth semantic understanding and accurate authenticity verification, avoiding information aliasing or loss caused by single feature extraction.

[0045] In step S103 of some embodiments, the text content features are parsed using a pre-built semantic analysis model. This semantic analysis model is built on a general large model and fine-tuned on an agricultural insurance-specific corpus, possessing a deep understanding of agricultural insurance terminology. The extracted text content features are input into the model along with preset prompts. The prompts guide the model to work along a step-by-step reasoning path: first, the overall semantics of the text are analyzed to determine the document category; then, based on the determined document category, the model locates and extracts structured key business entities from the text sequence, such as extracting the invoice date and amount from an invoice, or extracting the insured's name and land parcel code from a contract, thereby completing the structured reconstruction of unstructured information.

[0046] Please see Figure 3 In some embodiments, the semantic analysis model is constructed by fine-tuning instructions on a pre-defined corpus of agricultural insurance based on a general large language model. Step S103 may include, but is not limited to, steps S301 to S302.

[0047] Step S301: Construct prompt word instructions. The prompt word instructions include step-by-step reasoning paths and preset output format constraints used to guide the semantic analysis model to generate intermediate analysis processes.

[0048] Step S302: Input the prompt word instructions and text content features into the semantic analysis model so that the semantic analysis model can extract structured key business entities based on the document category determined by the step-by-step reasoning path and the document category.

[0049] In step S301 of some embodiments, a prompt instruction containing Chain-of-Thought (CoT) logic and structured constraints is constructed. The prompt instruction is a set of instructions that encapsulates the task context, reasoning logic paradigm, and output specifications. The step-by-step reasoning path defines the cognitive order in which the model processes the input text. For example, the instruction model might be: "Step 1: Scan all keywords in the text (e.g., 'invoice stamp', 'Party A and Party B') to determine the document's category; Step 2: Load the corresponding field extraction template based on the determined category; Step 3: Locate the specific field value in the text and perform format cleaning." Simultaneously, preset output format constraints define the structure of the model's output, explicitly limiting the standardized naming rules for keys and the data type of values ​​(e.g., they must be in "YYYY-MM-DD" format), ensuring that the generated results of the large model strictly adapt to the parsing requirements of the downstream system.

[0050] In step S302 of some embodiments, the constructed prompt word instruction is concatenated with the text content features extracted in the previous step (i.e., the character sequence recognized by OCR) and input into the semantic analysis model after instruction fine-tuning. This semantic analysis model utilizes its domain knowledge learned from the agricultural insurance corpus (covering massive amounts of non-standard place names, common names of crops, and handwritten correction data) to perform inference tasks following the input prompt word instruction. The model first performs a document classification task, dividing attachments into preset categories such as "land transfer contract," "inspection image," and "purchase voucher" based on the semantic features in the text. Next, the model activates an attention mechanism bound to the document category, focusing on specific text fragments for entity extraction; for example, under the purchase voucher category, it extracts "unit price" and "total amount," ultimately outputting a parsed result containing document category labels and structured key-value pairs.

[0051] Through steps S301 to S302 above, this embodiment of the application utilizes thought chain prompts to effectively guide the reasoning direction of the large language model, avoiding attention divergence and illusion problems when the model processes complex and long documents. At the same time, combined with a dedicated model fine-tuned with data from the agricultural insurance field, it achieves accurate identification and dynamic extraction of document categories and key entities in specific business scenarios. This not only solves the problem of insufficient understanding of agricultural professional terms by general models, but also ensures, through standardized output constraints, that unstructured text can be losslessly converted into structured data that can be directly called by downstream rule engines, thereby improving the accuracy and robustness of semantic analysis.

[0052] In step S104 of some embodiments, an authenticity analysis is performed on the visual features of the image. The purpose is to build a technical anti-counterfeiting barrier. In agricultural insurance, cases of tampering with the time and location of photos or using online images to defraud insurance are common. This step does not focus on the content of the image, but rather on analyzing whether the physical attributes of the image are natural and reasonable. By analyzing subtle features such as the distribution of noise, compression marks, or lighting consistency hidden in the visual features of the image, it is determined whether the image has been tampered with or synthesized, and a quantitative tampering risk assessment result is output, such as the probability of tampering or the risk level. This provides an objective anti-counterfeiting basis for subsequent review decisions and makes up for the shortcoming that digital tampering traces are difficult to detect with the naked eye.

[0053] Please see Figure 4 In some embodiments, step S104 may include, but is not limited to, steps S401 to S404.

[0054] Step S401: Perform recompression processing on the target image to obtain the corresponding recompressed image.

[0055] Step S402: Calculate the difference distribution between the target image and the recompressed image, and determine the compression artifact differences in the local area based on the difference distribution.

[0056] Step S403: Convert the target image to the frequency domain space, extract the high-frequency coefficient distribution features, and detect the presence of periodic correlated noise fingerprints based on the high-frequency coefficient distribution features.

[0057] Step S404: Based on the compression artifact differences and periodic correlation noise fingerprint, calculate the tampering risk assessment result for the target image.

[0058] In step S401 of some embodiments, a lossy recompression operation is performed on the target image (such as a close-up image reflecting details of crop damage) located and cropped by the target detection algorithm. Specifically, a preset JPEG quantization table is loaded, a fixed compression quality factor is set, such as 95% or 90%, and the target image in its original format or original compression state is re-encoded and saved to generate a recompressed image. This recompressed image maintains a high degree of consistency with the original image in terms of visual content, but introduces a uniform quantization noise baseline across the entire image at the pixel level. This provides a reference for subsequent comparison of differences to reveal inconsistencies in the compression history (i.e., tampering traces) of different image regions.

[0059] In step S402 of some embodiments, pixel-level difference calculations are performed between the original target image and the newly generated recompressed image to generate a difference map of the same size as the original image. To make minute pixel differences machine-readable, the difference map can be brightness-enhanced to obtain the compression artifact differences in local areas. In specific business scenarios, if a photo is an unaltered original image, its error level distribution should be uniform; however, if layers of "flames" or "withered crops" are artificially pasted into the photo, the tampered area, due to coming from different source images, has undergone compression times and quantization matrices inconsistent with the background. Therefore, in the difference distribution map, this local area will present abnormally bright patches or noise clusters that are significantly brighter or darker than the background, thereby identifying the boundary where the tampering occurred.

[0060] In step S403 of some embodiments, a Fast Fourier Transform (FFT) algorithm can be invoked to map the target image from the spatial domain to the frequency domain. In the frequency domain spectrum, low-frequency components representing smooth regions are ignored, and the focus is on extracting and analyzing the distribution features of high-frequency coefficients representing edge variations and noise details. For the "copy-and-move" forgery methods commonly used in agricultural insurance (i.e., using cloning tools to copy a small patch of lodged corn into a large area), this resampling and interpolation operation disrupts the random noise distribution of the natural image in the frequency domain, leaving behind spikes or grid-like textures with specific patterns. By detecting the presence of such unnatural periodic correlated noise fingerprints in the spectrum, the presence of pixel interpolation or region copying behavior can be identified.

[0061] In step S404 of some embodiments, a normalized value (probability value from 0 to 100%) is calculated by weighted logic combining the features of compressed artifact difference data (characterizing compression history consistency) from the spatial domain and periodic correlation noise fingerprint data (characterizing resampling traces) from the frequency domain. This value serves as the tampering risk assessment result for the target image. For example, when the analysis shows local highlighting and strong periodic correlation is detected in the frequency domain, an extremely high tampering risk score is output, and the corresponding tampering type is marked (such as "stitching and synthesis" or "cloning and forgery"). This result directly serves as a rejection indicator for judging the validity of evidence in subsequent logical verification steps.

[0062] Through steps S401 to S404 described above, this application embodiment constructs a dual image forensics mechanism combining spatial and frequency domains. Utilizing the irreversible principle of digital image processing, it reveals pixel-level tampering traces imperceptible to the naked eye. This mechanism effectively identifies sophisticated forgery methods such as splicing, compositing, and cloning using Photoshop software, and provides the entire review process with objective authenticity evidence based on mathematical characteristics, independent of subjective judgment.

[0063] In step S105 of some embodiments, multi-dimensional logical verification is performed based on key business entities, target business metadata, and tampering risk assessment results. This integrates fragmented information into a complete chain of evidence, placing the key business entities (such as invoice dates), target business metadata (such as policy validity periods), and tampering risk assessment results obtained in the aforementioned steps on the same logical plane for cross-comparison. This verification is multi-dimensional, including dimensions such as time (whether the invoice date is within the insurance period), content (whether the contract name matches the insured), and compliance (whether the image is the original). Through this comprehensive comparison, a series of verification factors are output, clearly indicating whether each logical verification passes or fails; for example, the time logic verification passes, while the authenticity verification fails.

[0064] Please see Figure 5 In some embodiments, the target business metadata includes geospatial range data and business rule constraint data of the insured land plot; the target attachment material data includes image data carrying spatial location tags, and step S105 may include, but is not limited to, steps S501 to S502.

[0065] Step S501: Perform spatial topological inclusion verification based on spatial location labels and geospatial range data to obtain spatial logical verification factors.

[0066] Step S502: Obtain time-series features and numerical features from key business entities, and perform business rule matching verification based on time-series features, numerical features and business rule constraint data to obtain business logic verification factors.

[0067] In step S501 of some embodiments, firstly, spatial location tags are extracted from the EXIF ​​metadata of the target attachment material data (specifically, on-site inspection photos or verification images). These tags typically contain longitude, latitude, and altitude information. To eliminate coordinate deviations between different acquisition devices or map systems, coordinate system standardization processing can be performed on the extracted coordinates. Subsequently, the geographic boundary data of the insured land parcel pre-stored in the target business metadata is retrieved. This data is typically a sequence of polygon vertices generated by a GIS surveying system, accurately describing the shape of the land parcel. Based on these two sets of data, a spatial topological inclusion check is performed using a point-plane relationship algorithm in computational geometry. That is, it is calculated whether the standardized shooting point coordinates fall within the polygonal range of the insured land parcel or its preset buffer zone (e.g., extending 50 meters outward). If the calculation result is true, it indicates that the photo was indeed taken on the insured land parcel, and a signal indicating that the check has passed is generated; otherwise, a signal indicating that the check has failed is generated. This logical judgment result is encapsulated as a spatial logical verification factor, quantifying the compliance of the evidence materials in geographic space.

[0068] In step S502 of some embodiments, specific temporal features (such as invoice issuance date and contract signing date) and numerical features (such as invoice amount, seed purchase quantity, and disaster-affected area) are first selected from the key business entities extracted in the preceding step S103. Next, business rule constraint data from the target business metadata is read. This data defines the effective boundaries of the insurance contract, including the start and end dates of insurance liability, the maximum insured amount per acre of the insured object, and the deductible threshold. The extracted features are compared with the constraint data: it is determined whether the temporal features are strictly within the insurance liability period; and whether the numerical features are within a reasonable threshold range (e.g., whether the total amount of seeds purchased exceeds the theoretical maximum demand for the corresponding insured area). This comparison aims to detect violations such as retroactive insurance purchases after an accident, excessive claims, or false invoices. Based on the results of each comparison, a comprehensive "business logic verification factor" is generated to qualitatively characterize the case risk from a business compliance perspective.

[0069] Through steps S501 to S502, this embodiment transforms the experience-based judgments such as "whether the location is correct," "whether the time is compliant," and "whether the amount matches," which are difficult to quantify precisely in traditional manual review, into mathematical calculations. By performing automated verification on two core dimensions—spatial topology and business rules—it can quickly capture clues such as remote shooting or logical contradictions, ensuring that every review conclusion is based on objective data consistency, thereby greatly improving the accuracy and automation level of agricultural insurance risk control.

[0070] In step S106 of some embodiments, based on the multiple verification factors generated above, and according to a preset review strategy, these factors are comprehensively evaluated to generate a final conclusion. Based on the state combination of each verification factor, it is determined whether the current case meets the conditions for automatic approval, or whether specific rejection or review rules have been triggered. For example, when all verification factors indicate normality, an approval review conclusion is generated; while when any verification factor in any dimension indicates anomaly (such as information mismatch or risk of tampering), an review conclusion containing warning information is generated.

[0071] Please see Figure 6 In some embodiments, step S106 may include, but is not limited to, steps S601 to S603.

[0072] Step S601: Input the tampering risk assessment results, spatial logic verification factors, and business logic verification factors into the preset fusion decision rule engine.

[0073] Step S602: If the tampering risk assessment result indicates that it has not been tampered with, and both the spatial logic verification factor and the business logic verification factor have passed the verification, then an automatically approved audit conclusion will be output.

[0074] Step S603: If the risk assessment result indicates that there is a risk, or any verification factor fails the verification, then output the manual review conclusion containing the specific risk cause code and generate a risk mark.

[0075] In step S601 of some embodiments, the tampering risk assessment results, spatial logic verification factors, and business logic verification factors generated in the preceding steps are used as decision-making bases and uniformly input into a preset fusion decision rule engine. This engine internally maintains a set of deterministic business compliance judgment logic, which is used to comprehensively weigh and logically calculate the multi-dimensional input data to determine the final flow status of the business data.

[0076] In step S602 of some embodiments, when the input tampering risk assessment result is detected as untampered or in a low-risk range, and both the spatial logic verification factor and the business logic verification factor are shown as passed, the engine determines that the case meets all compliance requirements. At this time, an automatically approved review conclusion is directly output, allowing business data to automatically flow to the next stage without manual intervention.

[0077] In step S603 of some embodiments, once the risk assessment result indicating tampering is detected as having a forgery risk, or any logical verification factor shows a verification failure, the engine determines that the case has a compliance issue. At this time, a manual review conclusion containing the specific risk reason code is immediately output, and a visual report with risk markers is generated to clearly point out the problem for manual reviewers to focus on.

[0078] Through the above steps S601 to S603, this embodiment of the application utilizes a rule engine to automatically fuse and judge multi-dimensional verification data, replacing tedious manual decision-making while effectively avoiding subjective misjudgments and improving the efficiency and objectivity of agricultural insurance business data review.

[0079] Please see Figure 7 In some embodiments, the method provided in this application may further include, but is not limited to, steps S701 to S703.

[0080] Step S701: Obtain the manual review results for the target agricultural insurance business data.

[0081] Step S702: Mark the target agricultural insurance business data whose manual review results are inconsistent with the audit results as difficult samples.

[0082] Step S703: Incremental training of the semantic analysis model and the image identification model is performed using difficult samples.

[0083] In step S701 of some embodiments, the manual review results for the target agricultural insurance business data are directly obtained through an interactive terminal or API interface. The manual review results originate from a secondary assessment by a claims specialist with advanced privileges of cases intercepted by the system or randomly sampled. Specifically, when the intelligent review outputs a rejection or pending verification conclusion due to high tampering risk or logical verification failure, the claims specialist reviews the original supporting materials and marked risk points on the review interface and enters the final confirmation conclusion. This review result includes not only the qualitative review status but also information on the specialist's correction of erroneous data. For example, manually correcting incorrect invoice amounts recognized by OCR or removing false alarms caused by lighting issues during photography. This manually confirmed data constitutes a high-confidence truth label.

[0084] In step S702 of some embodiments, when a substantial conflict is detected between the two conclusions—that is, the machine judges it as non-compliant but the human judges it as compliant, or the machine judges it as compliant but the human judges it as non-compliant—then the target agricultural insurance business data and its associated intermediate feature vector are locked and marked as a difficult sample. For example, a genuine electronic invoice that is misjudged by the model as a screenshot, or a contract whose semantic parsing fails due to illegible handwriting, will be classified, marked, and stored in a special difficult example database, distinguishing it from ordinary consistent historical data.

[0085] In step S703 of some embodiments, the original data (text or image) of the difficult samples is used as input, the results of manual review are used as supervision signals, the loss function is calculated using the backpropagation algorithm, and the model parameters are updated so that the model adjusts the decision boundary to adapt to these boundary situations where misjudgments have occurred. The new version of the model, which has completed training and validation, will be automatically deployed to replace the old version of the model, thereby enabling the algorithm to handle specific scenarios.

[0086] Through steps S701 to S703 above, this embodiment of the application constructs an active learning mechanism based on feedback loop. It uses manual review to correct single business deviations and transforms the implicit experience of experts into explicit data samples to feed back to the algorithm model. This mechanism enables the review system to achieve self-evolution by continuously learning from difficult samples, effectively solving the problem of static models failing when facing complex and ever-changing agricultural insurance business scenarios. Thus, while maintaining a high automation rate, it continuously enhances the robustness of the system, enabling automated review of agricultural insurance business data and improving the efficiency and objectivity of agricultural insurance business data review.

[0087] Please see Figure 8 This application also provides an agricultural insurance business data processing apparatus, which can implement the above-mentioned agricultural insurance business data processing method. The apparatus includes: The acquisition module is used to acquire target agricultural insurance business data, which includes target attachment material data belonging to multimodal types and corresponding target business metadata. The feature extraction module is used to perform multi-channel feature extraction on the target attachment material data to obtain text content features and image visual features respectively; The parsing module is used to parse the text content features through a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data; The authenticity analysis module is used to perform authenticity analysis on the visual features of images and generate tampering risk assessment results. The logic verification module is used to perform multi-dimensional logic verification based on key business entities, target business metadata, and tampering risk assessment results, and obtain verification factors in multiple dimensions. The generation module is used to generate audit results for target agricultural insurance business data based on multiple dimensions of verification factors.

[0088] The specific implementation of the agricultural insurance business data processing device is basically the same as the specific implementation of the above-mentioned agricultural insurance business data processing method, and will not be described again here.

[0089] This application also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the above-mentioned method for processing agricultural insurance business data. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0090] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes: The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application. The memory 902 can be implemented as a read-only memory (ROM), static storage device, dynamic storage device, or random access memory (RAM). The memory 902 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called by the processor 901 to execute the agricultural insurance business data processing method of the embodiments of this application. The input / output interface 903 is used to implement information input and output; The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904); The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0091] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for processing agricultural insurance business data.

[0092] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0093] The method and related equipment for processing agricultural insurance business data proposed in this application include: acquiring target agricultural insurance business data, which includes target ancillary material data and corresponding target business metadata; performing multi-channel feature extraction on the target ancillary material data to obtain text content features and image visual features; parsing the text content features using a pre-built semantic analysis model to determine the document category and key business entities of the target ancillary material data; performing authenticity analysis on the image visual features to generate a tampering risk assessment result; performing multi-dimensional logical verification based on the key business entities, target business metadata, and tampering risk assessment result to obtain multiple-dimensional verification factors; and generating an audit result for the target agricultural insurance business data based on the multiple-dimensional verification factors.

[0094] According to the agricultural insurance business data processing method proposed in this application, the following steps are taken: First, target agricultural insurance business data, including target attachment material data and corresponding target business metadata, is acquired, providing a comprehensive and accurate input data source for subsequent multimodal intelligent analysis. Next, multi-channel feature extraction is performed on the target attachment material data to obtain text content features and image visual features, achieving digital feature decoupling of unstructured attachment data and enabling perception of data content from different dimensions. Then, a pre-built semantic analysis model is used to parse the text content features, determining document categories and key business entities, deeply understanding the business semantics of the attachments, and extracting core information. Simultaneously, image visual features are analyzed for authenticity, generating a tampering risk assessment result, effectively identifying forgery traces that are difficult to detect with the naked eye, ensuring the authenticity of claims materials. Then, multi-dimensional logical verification is performed based on key business entities, target business metadata, and the tampering risk assessment result, obtaining multiple verification factors. Cross-validation of multi-source data ensures the consistency of business logic in time, space, and content. Finally, an audit result is generated based on the multiple verification factors. Through the above steps, this application enables automated review of agricultural insurance business data, improving the efficiency and objectivity of the review process.

[0095] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0096] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0097] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0098] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0099] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0100] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0101] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0102] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0103] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0104] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0105] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. A method for processing agricultural insurance business data, characterized in that, The method includes: Acquire target agricultural insurance business data, which includes target ancillary material data and corresponding target business metadata; Multi-channel feature extraction is performed on the target attachment material data to obtain text content features and image visual features, respectively; The text content features are analyzed using a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data; The visual features of the image are analyzed for authenticity to generate a tampering risk assessment result; Based on the key business entity, the target business metadata, and the tampering risk assessment results, multi-dimensional logical verification is performed to obtain verification factors in multiple dimensions; The audit results for the target agricultural insurance business data are generated based on the verification factors across multiple dimensions.

2. The method for processing agricultural insurance business data according to claim 1, characterized in that, The multi-channel feature extraction of the target attachment material data yields text content features and image visual features, including: Optical character recognition is performed on the target accessory material data to obtain a text character sequence and the coordinate information corresponding to the text character sequence, which are used as the text content features; Target detection is performed on the target accessory material data to obtain the target image and the texture and edge features of the target image, which are used as the visual features of the image.

3. The method for processing agricultural insurance business data according to claim 2, characterized in that, The step of performing authenticity analysis on the visual features of the image and generating a tampering risk assessment result includes: The target image is recompressed to obtain the corresponding recompressed image; Calculate the difference distribution between the target image and the recompressed image, and determine the compression artifact differences in local regions based on the difference distribution; The target image is converted to the frequency domain, high-frequency coefficient distribution features are extracted, and the presence of periodic correlated noise fingerprints is detected based on the high-frequency coefficient distribution features. Based on the compression artifact differences and the periodic correlation noise fingerprint, the tampering risk assessment result for the target image is calculated.

4. The method for processing agricultural insurance business data according to claim 1, characterized in that, The semantic analysis model is constructed based on a general large language model and fine-tuned on a pre-defined corpus in the agricultural insurance field. The pre-constructed semantic analysis model is used to parse the text content features and determine the document category and key business entities of the target attachment material data, including: Construct prompt words, which include step-by-step reasoning paths and preset output format constraints for guiding the semantic analysis model to generate intermediate analysis processes; The prompt word instruction and the text content features are input into the semantic analysis model so that the semantic analysis model can extract the structured key business entities based on the document category determined by the step-by-step reasoning path and the document category.

5. The method for processing agricultural insurance business data according to claim 1, characterized in that, The target business metadata includes geographic spatial range data and business rule constraint data of the insured land plot; the target attachment material data includes image data with spatial location tags. The multi-dimensional logical verification is performed based on the key business entity, the target business metadata, and the tampering risk assessment results to obtain multiple verification factors, including: Perform spatial topology inclusion verification based on the spatial location label and the geospatial range data to obtain a spatial logical verification factor; The time-series features and numerical features are obtained from the key business entities. Business rule matching verification is performed based on the time-series features, the numerical features and the business rule constraint data to obtain the business logic verification factor.

6. The method for processing agricultural insurance business data according to claim 5, characterized in that, The step of generating an audit result for the target agricultural insurance business data based on the verification factors across multiple dimensions includes: The tampering risk assessment results, the spatial logic verification factor, and the business logic verification factor are input into a preset fusion decision rule engine; If the tampering risk assessment result indicates that it has not been tampered with, and both the spatial logic verification factor and the business logic verification factor have passed the verification, then an automatically approved audit conclusion will be output. If the tampering risk assessment result indicates the presence of risk, or if any verification factor fails the verification, a manual review conclusion containing the specific risk cause code will be output, and a risk marker will be generated.

7. The method for processing agricultural insurance business data according to claim 1, characterized in that, The method further includes: Obtain the results of manual review of the target agricultural insurance business data; Target agricultural insurance business data whose manual review results are inconsistent with the audit results are marked as difficult samples; The semantic analysis model and image identification model are incrementally trained using the difficult samples.

8. A data processing device for agricultural insurance business, characterized in that, The device includes: The acquisition module is used to acquire target agricultural insurance business data, which includes target attachment material data belonging to the multimodal type and corresponding target business metadata. The feature extraction module is used to perform multi-channel feature extraction on the target attachment material data to obtain text content features and image visual features respectively; The parsing module is used to parse the text content features through a pre-built semantic analysis model to determine the document category and key business entities of the target attachment material data; The authenticity analysis module is used to perform authenticity analysis on the visual features of the image and generate a tampering risk assessment result. The logic verification module is used to perform multi-dimensional logic verification based on the key business entity, the target business metadata and the tampering risk assessment result, and obtain verification factors in multiple dimensions. The generation module is used to generate audit results for the target agricultural insurance business data based on the verification factors in multiple dimensions.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the agricultural insurance business data processing method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, characterized in that, when the computer program is executed by a processor, it implements the method for processing agricultural insurance business data as described in any one of claims 1 to 7.

Images