Claim settlement risk control method and device for aquaculture insurance, storage medium and computer equipment

Abstract

The application relates to the technical field of insurance claim and business security, and discloses a claim risk control method and device for breeding industry insurance, a storage medium and computer equipment. The method comprises the following steps: in response to a claim request for breeding industry insurance, obtaining claim condition information of the claim request for breeding industry insurance, identifying whether the claim request for breeding industry insurance satisfies quick claim conditions based on the claim condition information; if the quick claim conditions are satisfied, obtaining user-side field evidence data and policy data corresponding to the claim request for breeding industry insurance, automatically identifying preset key information of the user-side field evidence data to obtain key information identification data; performing logical consistency verification on the claim request for breeding industry insurance according to the user-side field evidence data, the policy data and the key information identification data, and performing damage assessment when the logical consistency verification is passed; and performing quick claim risk control processing on the claim request for breeding industry insurance based on the damage assessment result. The method can be applied to a financial technology scene.

Description

Technical Field

[0001] This application relates to the fields of insurance claims and business security technology, and in particular to a claims risk control method, device, storage medium and computer equipment for aquaculture insurance. Background Technology

[0002] As an important policy tool for mitigating agricultural production risks and ensuring stable income for farmers, livestock insurance has long faced industry-wide challenges such as "difficulty in assessing losses, slow claims processing, and high costs." This problem is particularly prominent in livestock insurance.

[0003] On the one hand, livestock insurance claims (such as breeding sows, fattening pigs, and sheep) require the harmless disposal of dead animals within a legally stipulated timeframe. The window for preserving on-site evidence is extremely short, and once evidence is lost, it is difficult to trace, creating inherent time pressure on claims investigation and loss assessment. On the other hand, a very high percentage of livestock insurance claims involve small numbers of animals lost and low payouts. Traditional on-site investigation methods require dedicated investigators to travel to and from the site for each small claim, and the transportation, manpower, and time costs of a single investigation often exceed the payout amount itself, resulting in a severe cost imbalance. Simultaneously, the frequent entry and exit of investigators into livestock farms increases the risk of transmission of major animal diseases such as African swine fever, posing a threat to livestock biosecurity.

[0004] Therefore, there is an urgent need for an innovative solution for automated and timely insurance claims risk control in small-scale livestock farming. Summary of the Invention

[0005] In view of this, embodiments of this application provide a method, apparatus, storage medium, and computer equipment for claims risk control in aquaculture insurance.

[0006] According to one aspect of this application, a claims risk control method for aquaculture insurance is provided, the method comprising: In response to a livestock insurance claim request, obtain the claim conditions information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the conditions for fast claim processing based on the claim conditions information; If the conditions for fast claims settlement are met, the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request are obtained, and the pre-trained recognition model is called to automatically identify the preset key information in the user-side on-site evidence data to obtain key information recognition data. Based on the on-site evidence data from the user terminal, the policy data, and the key information identification data, a logical consistency check is performed on the aquaculture insurance claim. If the logical consistency check is passed, a loss assessment analysis is performed on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model. Based on the loss assessment results, the aquaculture insurance claim requests are processed quickly for risk control.

[0007] According to another aspect of this application, a claims risk control device for aquaculture insurance is provided, the device comprising: The judgment module is used to respond to a livestock insurance claim request, obtain the claim condition information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the fast claim conditions based on the claim condition information; The identification module is used to obtain the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request if the conditions for fast claims are met, and to call the pre-trained identification model to automatically identify the preset key information in the user-side on-site evidence data to obtain key information identification data. The loss assessment module is used to perform logical consistency verification on the aquaculture insurance claim based on the on-site evidence data from the user terminal, the policy data, and the key information identification data. When the logical consistency verification is passed, the module performs loss assessment analysis on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model. The execution module is used to perform rapid claims risk control processing on the aquaculture insurance claim requests based on the loss assessment analysis results.

[0008] According to another aspect of this application, a storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the aforementioned claims risk control method for aquaculture insurance.

[0009] According to another aspect of this application, a computer device is provided, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the program to implement the aforementioned claims risk control method for aquaculture insurance.

[0010] By utilizing the above technical solution, the present application provides a method, device, storage medium, and computer equipment for claims risk control in aquaculture insurance. First, it responds to a farmer's aquaculture insurance claim request and obtains claim condition information. Based on this, it identifies whether the request meets preset fast claims conditions. Only after meeting the fast claims conditions is the corresponding user-end on-site evidence data and policy data obtained. A pre-trained recognition model automatically extracts key information such as ear tag numbers, number of deaths, and weight of the insured animal from the on-site evidence. Subsequently, a logical consistency check is performed based on a cross-comparison between the on-site evidence data, policy data, and the identified key information. Only after passing the check is the data allowed to enter the loss assessment model to complete the loss assessment. Finally, based on the loss assessment analysis results, it determines whether to execute fast claims, achieving fully automated processing of small-amount cases from reporting to decision-making. This allows eligible small-amount cases to automatically complete loss assessment and determination after effective evidence collection without the need for an investigator to be present, shortening the claims cycle, reducing labor costs, and reducing the risk of disease transmission from investigators entering and leaving the farm.

[0011] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0012] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 A flowchart illustrating a claims risk control method for aquaculture insurance provided in an embodiment of this application is shown. Figure 2 This paper illustrates a flowchart of another claims risk control method for aquaculture insurance provided in an embodiment of this application. Figure 3 This illustration shows a structural schematic diagram of a claims risk control device for aquaculture insurance provided in an embodiment of this application; Figure 4 A schematic diagram of the device structure of a computer device provided in an embodiment of this application is shown. Detailed Implementation

[0013] The present application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present application can be combined with each other.

[0014] This embodiment provides a claims risk control method for aquaculture insurance, such as... Figure 1As shown, the method includes: Step 101: In response to a livestock insurance claim request, obtain the claim conditions information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the fast claim conditions based on the claim conditions information.

[0015] The claims condition information refers to a multi-dimensional data set that reflects whether a livestock insurance claim qualifies for the fast-track claims process. This information includes, but is not limited to, information about the insurance institution in the farmer's region, the type of insured object, the number of claims in a single incident, and the pre-payment amount. The completeness of this information directly determines the accuracy of fast-track claims condition identification. Fast-track claims conditions refer to a set of pre-configured admission rules by the insurance institution. These rules include multiple judgment dimensions such as institution activation status, insurance category classification, head count threshold, and amount threshold. Their purpose is to divert cases that do not meet the fast-track standards at the entry point, avoiding waste of subsequent processing resources.

[0016] Specifically, when a farmer initiates a livestock insurance claim through a client application (which can be a smartphone app, WeChat mini-program, or webpage), the client responds to the request by first obtaining the claim conditions information filled in by the farmer. This can be achieved by the client providing a claim conditions information input interface, where the farmer sequentially selects or fills in their region, type of insured property, number of items damaged, and estimated loss amount; the client then encapsulates this information into a claim request data packet and sends it to the server.

[0017] For example, the claim conditions information includes at least one of the following: the region where the farmer is located, the insured object, the number of claims, and the estimated amount of compensation. Identifying whether the livestock insurance claim request meets the conditions for fast claims based on the claim conditions information includes at least one of the following: identifying whether the region where the farmer is located has activated fast claims service; identifying whether the insured object belongs to a preset fast claims insurance catalog; identifying whether the number of claims is less than or equal to a preset threshold for the number of fast claims per incident; and identifying whether the estimated compensation amount is less than or equal to a preset upper limit threshold for fast claims.

[0018] After receiving the claim request data packet, the server parses and extracts the claim conditions information. The server pre-stores a fast claim conditions configuration table, which is configured and maintained by insurance institution administrators through a backend management interface. During the identification process, the server compares each extracted claim condition item with the rules in the fast claim conditions configuration table: Regarding the identification of the institution dimension, the server queries whether the corresponding insurance institution in the region where the farmer is located has been marked as "enabled" for fast claims service in the configuration table. If the institution has not enabled it, the claim request is identified as not meeting the fast claims conditions.

[0019] Regarding the identification of insurance types, the server checks whether the insurance type belongs to the preset fast-claim insurance category in the fast-claim condition configuration table based on the insured object type information. If it does not belong to the category, the claim request is identified as not meeting the fast-claim conditions. This fast-claim insurance category can be configured differently according to the breeding structure and insurance product differences in different regions. For example, fattening pigs can be included in the fast-claim category in major pig-producing areas, and sheep can be included in the fast-claim category in pastoral areas.

[0020] Regarding the identification of the quantity dimension, the server compares the number of claims with the preset fast claim head count threshold in the configuration table. If the number of claims exceeds the threshold, the claim request is identified as not meeting the fast claim conditions.

[0021] Regarding the identification of the amount dimension, the server compares the pre-payment amount information with the preset fast claim amount upper limit threshold in the configuration table. If the pre-payment amount exceeds the threshold, the claim request is identified as not meeting the fast claim conditions.

[0022] The identification order of the above four dimensions can be preset by the system or dynamically adjusted according to actual business needs. When all claim conditions meet the conditions for fast claim processing, the claim request is identified as meeting the fast claim processing conditions, and the system allows it to enter the fast claim processing process; when any dimension does not meet the fast claim processing conditions, the claim request is identified as not meeting the fast claim processing conditions, and the system transfers it to the regular claim processing queue, where it is processed by a human surveyor according to the regular process. The system also displays a prompt message indicating that the claim has been transferred to regular claim processing and instructions for subsequent processing to the farmer on the client side.

[0023] By identifying claims conditions, a large number of cases that do not meet the criteria for fast claims are effectively diverted at the entry point, avoiding resource consumption in subsequent data collection and model invocation stages. At the same time, cases that meet the criteria can smoothly enter the automated processing channel, laying the foundation for improving overall claims efficiency.

[0024] Step 102: If the conditions for fast claims settlement are met, obtain the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request, call the pre-trained recognition model to automatically identify the preset key information in the user-side on-site evidence data, and obtain key information recognition data.

[0025] The user-side on-site evidence data refers to a collection of multimedia evidence files directly related to the loss of the insured object, taken and uploaded by farmers at the scene of the accident via a client application. This collection includes, but is not limited to, group photos of the deceased insured objects, ear tag photos, weighing photos, and photos of the harmless disposal process. The client-embedded watermark camera automatically embeds the time of capture, latitude and longitude coordinates of the capture location, and the unique identifier of the capturing device into the image files. This watermark information is bound to the image content and used for subsequent verification of the evidence's authenticity. Policy data refers to structured data related to the policy involved in the claim, stored in the insurance institution's business system. This data includes, but is not limited to, a list of insured objects (including ear tag numbers for each insured object), initial weight data for each object entered at the time of underwriting, and coordinates of the insured's geographical location.

[0026] In one optional embodiment, the user-end on-site evidence data includes photos of the deceased targets taken and uploaded via the user-end's camera, photos of ear tags, photos of weighing, and photos of harmless disposal; the preset key information includes the quantity information, ear tag number information, and weight information of the deceased targets; the pre-trained recognition model includes a pre-trained ear tag number optical character recognition model, a pre-trained intelligent counting model, and a pre-trained intelligent counting model; the step of calling the pre-trained recognition model to automatically recognize the preset key information in the user-end on-site evidence data to obtain key information recognition data includes: calling the pre-trained ear tag number optical character recognition model to recognize the ear tag number in the ear tag photo to obtain the ear tag number recognition result; calling the pre-trained intelligent counting model to recognize the quantity of deceased targets in the photos of the deceased targets to obtain the quantity recognition result; and calling the pre-trained intelligent counting model to recognize the scale reading in the weighing photo and estimate the weight of the target based on the scale reading to obtain the target weight recognition result.

[0027] Specifically, after identifying that the claim request meets the conditions for fast claims processing, the server sends an evidence upload instruction to the farmer's client. Upon receiving the instruction, the client displays an evidence upload guide page on its interface. This guide page displays the shooting specifications for photos of the deceased animal, including photos of the animal with its ears, photos of its weighing, and photos of its harmless disposal, in the form of images, text, or videos. The farmer follows the instructions to complete the shooting and uploading of each type of photo in sequence.

[0028] In one implementation, regarding the taking and identification of photos of the dead targets, farmers place all the dead targets from the recent incident on a flat surface, ensuring that none of them obstruct each other. Under sufficient lighting conditions, they take a group photo using a client-side camera and upload it. Upon receiving the photo, the server invokes a pre-trained intelligent point counting model to identify the number of dead targets in the photo. During the identification process, the intelligent point counting model first performs scale normalization on the photo image; then, it extracts multi-scale feature maps of the image through its backbone network; next, it densely samples on the feature maps to generate candidate regions; finally, it uses a classification sub-network and a bounding box regression sub-network to perform target confidence scoring and location refinement on each candidate region. Candidate regions with confidence scores exceeding a preset detection threshold are identified as dead targets. The number of all identified dead targets is counted as the death count identification result, and the bounding box coordinates of each dead target in the image are output. As another alternative implementation, when the number of death targets is large enough to cause partial occlusion between targets in the group photo, the intelligent counting model can also adopt an attention-based occlusion handling strategy. By analyzing the local features of the body that are not occluded (such as legs, torso edges, etc.), the existence of the occluded target can be inferred to ensure counting accuracy.

[0029] In another implementation, regarding the photographing and recognition of ear tags, farmers take and upload photos of each dead ear tag individually, ensuring the ear tag is horizontally facing the camera lens, the ear tag number area is fully visible in the frame, and the lighting is uniform. Upon receiving each ear tag photo, the server calls a pre-trained optical character recognition (OCR) model to identify the ear tag number in the photo. During the recognition process, the OCR model first preprocesses the ear tag photo image, including scaling the image to the model input size, performing grayscale processing to reduce lighting interference, and using an adaptive thresholding method for binarization to enhance the contrast between the characters and the background. The preprocessed image is then input into a convolutional neural network to extract the visual feature sequence of the character region. This visual feature sequence is then input into a sequence decoding network, which autoregressively generates each character in the ear tag number string until an end marker is generated, finally outputting the complete ear tag number recognition result and the overall recognition confidence score. As an alternative implementation, when farmers use RFID electronic ear tags, the client also utilizes the near-field communication (NFC) function of the mobile device to directly read the stored ear tag number from the electronic ear tag. This reading result is compared and verified with the OCR recognition result. If they match, the ear tag number is confirmed with high confidence; if they do not match, manual verification is triggered. Alternatively, when the OCR recognition confidence level of the ear tag number falls below a preset threshold due to wear, dirt, or other reasons, the server sends a prompt to the client to retake the image, requiring the farmer to clean the ear tag surface and adjust the shooting angle to obtain a clearer image.

[0030] In another implementation, regarding the taking and identification of weighing photos, farmers use a scale to weigh each dead target individually. After the scale reading stabilizes, they take a photo containing both the scale reading display area and the dead target in the same frame and upload it. Upon receiving the weighing photo, the server invokes a pre-trained intelligent weighing model to identify the weight of the target. During the identification process, the intelligent weighing model first determines the scale type: if the scale is electronic, the scale reading identification submodule locates the LCD or digital display area of ​​the electronic scale and reads the digital reading on the display using optical character recognition; if the scale is mechanical, the scale reading identification submodule reads the reading by detecting the position of the scale lines and the pointer deflection angle. Based on the identified scale reading, combined with the proportional relationship between the length or height of the target in the weighing photo and a known reference object (such as the scale platform size), the estimated mass of the target is calculated.

[0031] In another implementation, regarding the taking and uploading of photos of harmless treatment, after farmers complete the treatment according to the harmless treatment procedures stipulated by the local animal husbandry authorities, they take photos of the treatment site and upload them. These photos are used for subsequent verification of compliance with regulations.

[0032] After various photo recognitions are completed, the server summarizes the ear tag identification results, mortality count results, and target weight identification results to form key information identification data. Through the intelligent counting model's automatic recognition of group photos of deceased targets, the mortality count can be automatically calculated using machine vision, avoiding human errors such as omissions and duplicates that may occur during manual counting. The automatic recognition of ear tag optical character recognition eliminates the inefficient method of manual copying of ear tags. Stable recognition capabilities under various shooting conditions and ear tag materials, along with RFID-assisted verification and re-photo guidance mechanisms, ensure reliable ear tag data collection under different farm equipment conditions. The intelligent weighing model adapts to both electronic and mechanical scales, ensuring smooth weighing data collection under different weighing equipment conditions. Through the collaborative work of these three recognition models, farmers only need to follow the guidance to complete the photo-taking operation; all key information is automatically extracted and structured, providing complete and standardized data input for subsequent logical consistency verification and loss assessment analysis.

[0033] Step 103: Based on the on-site evidence data from the user terminal, the policy data, and the key information identification data, perform a logical consistency check on the aquaculture insurance claim request. If the logical consistency check is passed, perform a loss assessment analysis on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model.

[0034] Logical consistency verification refers to a verification process that uses on-site evidence data from the user's end, policy data, and key information identification data as a basis, and cross-compares data from multiple dimensions and multiple data sources to determine whether there are logical contradictions or anomalies among the data.

[0035] Specifically, the user-side on-site evidence data, policy data, and key information identification data are input into the parallel computing channel of the logical consistency verification unit. This channel uses hash index technology to accelerate the data matching process and ensure the consistency of the spatiotemporal attributes of the three. After the verification is passed, the fused feature vector is imported into the loss assessment model for reasoning to determine the loss assessment analysis result.

[0036] In one optional embodiment, the policy data includes a list of insured objects, initial weight data of the insured objects, and geographical location data of the insured objects; the step of performing logical consistency verification on the aquaculture insurance claim based on the user-end on-site evidence data, the policy data, and the key information identification data includes: The ear tag identification result in the key information identification data is compared with the list of insured objects in the policy data to determine whether the ear tag identification result is within the range of the list of insured objects; otherwise, the logical consistency is determined to fail. Based on the initial weight data of the insured object in the policy data and the time interval from the underwriting time to the claim time, the theoretical weight range is calculated according to the preset growth curve model. The weight identification result of the object in the key information identification data is compared with the theoretical weight range to determine whether the weight identification result of the object is within the theoretical weight range. Otherwise, the logical consistency is determined to be unsuccessful. The number of deaths identified in the key information identification data is compared with the number of deaths reported in the livestock insurance claim request. It is determined whether the deviation between the number of deaths identified and the number of deaths reported is within a preset deviation. Otherwise, the logical consistency is deemed to have failed. The pre-trained standard recognition model is invoked to identify the harmless treatment photos in the user terminal's on-site evidence data, and to determine whether the harmless treatment meets the preset standard requirements. If not, the logical consistency is deemed to have failed.

[0037] The insured object list refers to the collection of individual records of insured objects associated with a specific policy, stored in the agricultural insurance policy center of the insurance institution. This list, in the case of aquaculture insurance, must include at least the ear tag number record for each insured object, serving as official registration data proving that the object is included in the insurance coverage. Initial weight data of the insured object refers to the initial weight value of the object recorded and entered into the policy system through weighing or estimation during the insurance underwriting process. Geographical location data refers to the latitude and longitude coordinates or geographical area data of the insured location obtained and entered into the policy system through on-site positioning or map mapping during the insurance underwriting process. The growth curve model is a mathematical function model reflecting the change in the weight of the object over time under normal feeding and management conditions. This model is pre-constructed based on the object's breed, feeding stage, and standard daily weight gain parameters. The standard identification model is an image classification or object detection model pre-trained using a large number of labeled images of harmless treatment sites. This model takes harmless treatment photos as input and outputs a classification label indicating whether the treatment method conforms to standards and the corresponding standard compliance confidence score.

[0038] In this embodiment, after acquiring on-site evidence data, policy data, and key information identification data from the user end, the server performs a logical consistency check on the aquaculture insurance claim request based on the aforementioned data. This check involves cross-comparison of data across multiple independent dimensions; if any dimension fails, the overall logical consistency check is deemed to have failed.

[0039] In the ear tag range verification dimension, the server extracts all identified ear tag number strings from the key information identification data, and simultaneously extracts all ear tag number records contained in the insured object list from the policy data. The server performs an equal-value match between each identified ear tag number and the ear tag number records in the insured object list: if a matching record can be found for a certain ear tag number in the insured object list, the identity verification of the ear tag number is deemed to have passed; if no matching record can be found for a certain ear tag number in the insured object list, the ear tag number is deemed to be outside the coverage scope, the server immediately determines that this logical consistency verification has failed, terminates the execution of subsequent verification dimensions, marks the claim request as an ear tag range anomaly, and transfers it to the manual review queue.

[0040] In terms of weight reasonableness verification, the server extracts the initial weight data of the insured animal from the policy data at the time of underwriting, and extracts the reporting time and underwriting time from the claim request information, calculating the time interval between the underwriting date and the reporting date. The server loads a corresponding preset growth curve model based on the insured animal's breed information. This model includes the daily weight gain parameters and weight gain pattern of the breed under standard feeding conditions. The server uses the initial underwriting weight as the starting point of the growth curve, inputs the time interval as the growth cycle into the growth curve model, and outputs the predicted range of the theoretical weight the insured animal should reach under normal feeding conditions at the current point in time. The server then extracts the target's weight identification result from the key information identification data and compares the weight identification result with the theoretical weight range: if the weight identification result falls within the theoretical weight range (including the endpoint value), the reasonableness verification of the target's weight is deemed to have passed; if the weight identification result falls outside the theoretical weight range, it indicates that the target's current weight deviates significantly from the expected weight under normal feeding conditions, which may mean that the weighing data is inaccurate or that the target is not the same target as the one listed in the policy. The server then determines that the logical consistency verification fails, terminates the execution of subsequent verification dimensions, marks the claim request as having abnormal weighing data, and transfers it to the manual review queue. In cases involving multiple deceased individuals and each individual having an independent initial insured weight record, the server queries the corresponding initial insured weight for each individual, calculates the time interval, calls the growth curve model to calculate the theoretical weight range, and compares the weight identification result of each individual with its respective theoretical weight range. Each of the multiple individuals independently completes the weighing rationality verification. If the verification of any individual fails, the overall verification is deemed to have failed, in order to accurately locate the specific individual with abnormalities.

[0041] In the consistency verification dimension of the number of deaths, the server extracts the death count recognition result output by the intelligent counting model from the key information identification data, and at the same time extracts the number of deaths reported from the claim request information filled in by farmers. The server compares the death count recognition result with the number of deaths reported, and calculates the deviation value between the two: if the deviation value is within the preset allowable deviation range (for example, allowing a small deviation due to the omission of points caused by the partial obscuring of individual targets in the group photo), the consistency verification of the number of deaths is determined to be passed; if the deviation value exceeds the preset allowable deviation range, it means that there is an inexplicable difference between the AI ​​recognition result and the farmer's report, which may mean that the farmer's reported number is inaccurate or the group photo is incomplete. The server immediately determines that this logical consistency verification fails, terminates the execution of subsequent verification dimensions, and marks the claim request as having an abnormal number before transferring it to the manual review queue. As another optional implementation, when the deviation value exceeds the allowable deviation range but the deviation is small, the server does not immediately terminate the verification. Instead, it sends a supplementary photo request to the client, requiring the farmer to retake a group photo after adjusting the placement of the targets and ensuring that all targets are visible. The system then performs intelligent point recognition on the new group photo and compares it with the declared quantity again to eliminate temporary deviations caused by shooting angle or obstruction.

[0042] In the dimension of verifying the harmless treatment standards, the server extracts photos of harmless treatment uploaded by farmers from the on-site evidence data from the user end, and calls a pre-trained standard recognition model to analyze the photos. During the training phase, the standard recognition model has learned and memorized the visual features of the standard operating procedures for harmless treatment stipulated by the local animal husbandry authorities, including the requirements for the selection of treatment sites (such as being carried out in designated areas or treatment pits), the compliance requirements for treatment methods (such as burial depth, disinfectant spray coverage, and the completeness of incineration), and the characteristics of the on-site state after treatment. During the identification process, the standardized identification model extracts image features from the harmless treatment photos and matches the extracted features with the compliant processing patterns learned during the training phase. If the matching degree between the photo features and the compliant processing pattern exceeds a preset compliance threshold, the harmless treatment is deemed to meet the preset standard requirements, and this dimension passes the verification. If the matching degree between the photo features and the compliant processing pattern is lower than the preset compliance threshold, it indicates that the harmless treatment process performed by the farmer is non-compliant. The server immediately determines that this logical consistency verification fails, terminates the execution of subsequent verification dimensions, and marks the claim request as an abnormal harmless treatment process before transferring it to the manual review queue. As an alternative implementation method, when the processing method shown in the harmless treatment photo deviates partially from the standard requirements but the overall harmless treatment objective has been achieved, the server does not directly transfer the case to manual review. Instead, it sends a standard deviation prompt and rectification guidelines to the farmer on the client side, requiring the farmer to complete the supplementary processing according to the guidelines and upload supplementary photos within a specified time. The system then performs standard verification on the supplementary photos again, giving the farmer an opportunity to correct the issue on-site.

[0043] When all the above verification dimensions pass, the server determines that the logical consistency verification has passed and allows the claim request to enter the loss assessment analysis stage.

[0044] In an optional embodiment, the step of performing logical consistency verification on the aquaculture insurance claim based on the user-end on-site evidence data, the policy data, and the key information identification data further includes at least one of the following: Metadata analysis is performed on the photos in the user terminal on-site evidence data to extract the exchangeable image file format information of the photos. The shooting time information, geographical location information and device identification information in the watermark information are compared with the corresponding metadata fields in the exchangeable image file format information. If the watermark information is inconsistent with the exchangeable image file format information, it is determined that the photo has been tampered with and the logical consistency check fails. The geographical location information of the photos in the user terminal on-site evidence data is compared with the insured geographical location data in the policy data to determine whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to have failed. Based on the shooting time information in the watermark information of the user terminal's on-site evidence data and the reporting time information of the aquaculture insurance claim, it is determined whether the shooting time is after the reporting time and whether the time interval between the shooting time and the reporting time is less than a preset watermark time consistency threshold; if not, the logical consistency check is determined to fail; wherein, the camera of the user terminal is a watermark camera; the watermark camera automatically embeds shooting time information into the photo when shooting; Based on the geographical location information in the watermark information and the insured geographical location data in the policy data, it is determined whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to fail; the watermark camera automatically embeds geographical location information into the photo when taking the picture.

[0045] Metadata analysis refers to the technical operation of parsing and extracting metadata fields contained in the underlying storage structure of image files. When an image file is generated, the capturing device automatically writes Exchangeable Image File Format (EFF) information. This information records device-level metadata in standardized fields, including the device's system time at the time of capture, GPS coordinates of the shooting location, the manufacturer and model of the capturing device, image size, and the last modified time of the file. This metadata is automatically written by the hardware system and file system of the capturing device during image generation. It is difficult to tamper with this metadata while maintaining its internal consistency without professional methods. Therefore, the consistency between the fields in the metadata and the consistency between the metadata and the watermark information have evidentiary value in proving the originality and authenticity of the photograph. Watermark information refers to the time information, geographical location information, and device identification information written by the client-side watermark camera into the visible area of ​​the image or the image file's supplementary information area at the moment of shooting, using pixel overlay or digital watermark embedding. This information is obtained and embedded instantly by the client application from the mobile device's system time, positioning module, and hardware identifier at the moment of shooting.

[0046] In this embodiment, in addition to the core verification dimensions such as ear tag range verification, weighing rationality verification, death quantity consistency verification, and harmless treatment standard verification, the server further performs at least one of the following auxiliary verification dimensions to enhance the comprehensiveness and depth of logical consistency verification.

[0047] In terms of verifying the authenticity of photos, the server performs metadata extraction on each photo in the user's on-site evidence data. This involves reading the shooting time, GPS coordinates, device manufacturer and model, and last modified time from the exchangeable image file format (EEFF) information of the photo file. Simultaneously, the server extracts watermark information embedded in the camera from the visible area of ​​the photo or the file's supplementary information area. This watermark information includes the shooting time, geographic coordinates, and device identifier. The server compares the shooting time in the watermark information with the shooting time field in the EEFF information, the geographic coordinates in the watermark information with the GPS coordinate field in the EEFF information, and the device identifier in the watermark information with the device model and unique serial number fields in the EEFF information. If any of the three sets of comparisons above are inconsistent (for example, the shooting time in the watermark and the creation time in the metadata differ from the clock deviation range allowed by the system, or the device model in the watermark does not match the camera model in the metadata), it indicates that the photo may have been modified by image editing software after being taken, resulting in the watermark information and metadata information being out of sync. The server will then determine that the photo has been tampered with, the overall logical consistency check fails, and the claim request is marked as a photo tampering risk and transferred to the manual review queue.

[0048] In the location consistency verification dimension, the server extracts geographic location information from photos in the user's on-site evidence data. This geographic location information can come from the coordinate field of the watermark information embedded in the watermark camera, or from the GPS coordinate field of the photo exchangeable image file format information. If both exist, the watermark information takes priority, and metadata is used for cross-validation. The server calculates the distance between this geographic location coordinate and the insured geographic location coordinate in the policy data. If the calculated distance value is less than the preset location deviation threshold, it means that the accident scene and the insured location are basically consistent in spatial location, and this dimension verification passes. If the distance value is greater than or equal to the preset location deviation threshold, it means that the accident location is too far from the insured location, which may mean that the farmer impersonated the insured location after the accident occurred at a non-insured location, or there are other inaccurate location situations. The server then determines that this logical consistency verification fails, marks the claim request as having an abnormal accident location, and transfers it to the manual review queue.

[0049] In terms of time consistency verification, the server extracts the reporting time from the records of aquaculture insurance claims and the shooting time from the watermark information of photos in the on-site evidence data from the user's end. The shooting time is then compared with the reporting time. First, it is determined whether the shooting time is after the reporting time: if the shooting time is earlier than the reporting time, it means that the photo was taken before the reporting time, which does not conform to the normal process of "collecting on-site evidence after reporting the incident". This may mean that the farmer used historical photos to impersonate evidence of the current incident. The server then determines that the logical consistency verification fails, marks the claim as evidence of timeliness error, and transfers it to the manual review queue. If the shooting time is after the reporting time, the server further calculates the time interval between the shooting time and the reporting time, and compares this time interval with a preset watermark time consistency threshold: if the time interval is less than the threshold, it means that the farmer completed the on-site evidence collection in a timely manner after reporting the case, which meets the requirements of the fast claims settlement for the timeliness of evidence collection, and this dimension of the verification passes; if the time interval is greater than or equal to the threshold, it means that the interval between reporting the case and completing the on-site evidence collection is too long, and the on-site evidence may have been lost or changed due to harmless treatment or other reasons. The server then determines that this logical consistency verification fails, marks the claims request as overdue collection, and transfers it to the manual review queue.

[0050] In the group consistency verification dimension, the server extracts the watermark information of each photo from the user's on-site evidence data and compares the device identification information between the photos for consistency. If the device identification of all photos is completely consistent, it means that all photos were taken by the same device, and this dimension verification passes. If the device identification of one or more photos is different from that of other photos, it means that the photos in this case were not taken by the same device and may have been mixed with image files from other sources. The server then determines that this logical consistency verification fails, marks the claim request as having device abnormality, and transfers it to the manual review queue.

[0051] The aforementioned auxiliary verification dimensions are logically ANDed with the core verification dimensions; that is, if any auxiliary verification dimension fails, the overall logical consistency verification is deemed to have failed. As an alternative implementation, for cases where a single dimension fails the initial verification but there is a reasonable possibility of correction (such as a positional deviation slightly exceeding the threshold or a shooting time slightly later than the time limit requirement), the server does not immediately terminate the verification and transfer it to manual review. Instead, it sends a targeted correction request to the farmer through the client, allowing the farmer to provide reasonable written explanations or supporting evidence (such as transportation records, veterinary visit records, etc.) within a limited time. The server incorporates the supplementary explanations submitted by the farmer into the comprehensive evaluation scope of the logical consistency verification, and after completing the comprehensive evaluation, decides whether to pass the verification or transfer it to manual review.

[0052] Step 104: Based on the loss assessment analysis results, perform rapid claims risk control processing on the aquaculture insurance claim request.

[0053] Specifically, after completing the loss assessment analysis and outputting the results, the server automatically calculates the compensation amount according to the assessed quantity, assessed weight, and underwriting information in the policy data, based on the loss assessment analysis results and the compensation calculation rules stipulated in the corresponding insurance terms. If the total compensation amount exceeds the fast-claim limit, the server determines that the aquaculture insurance claim may be risky and will not execute the fast-claim process, terminating the fast-claim process and transferring the case to the manual review queue. If the total compensation amount does not exceed the fast-claim limit, the server determines that the aquaculture insurance claim is safe, executes the fast-claim process, and automatically initiates the claims decision and payment process.

[0054] In one optional embodiment, the step of performing loss assessment analysis on the user-side on-site evidence data, the policy data, and the key information identification data using a pre-trained loss assessment model includes: Based on the ear tag identification results in the key information identification data and the group photo of the deceased target in the user terminal on-site evidence data, the identity and survival status of the deceased target are confirmed; based on the target weight identification results in the key information identification data, the assessed weight of the deceased target is determined; based on the assessed weight of the deceased target and the assessed weight of the deceased target, the damage assessment analysis result is determined; The rapid claims risk control processing for the aquaculture insurance claims based on the loss assessment results includes: Based on the loss assessment results, the standard weight of the insured, the single insured amount and compensation ratio stipulated in the policy, calculate the single insured amount, and sum up the single insured amounts of all deceased insured persons to obtain the total compensation amount; The total compensation amount is compared with a preset fast compensation amount upper limit threshold. If the total compensation amount is less than or equal to the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a security request and the fast compensation process is executed for automatic payment. If the total compensation amount is greater than the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a risk request, the fast compensation process is terminated, and the request is transferred to the manual review queue.

[0055] In this embodiment, after the logical consistency check passes, the server inputs the user's on-site evidence data, policy data, and key information identification data into the pre-trained loss assessment model to perform loss assessment analysis.

[0056] In the identification and survival status confirmation stage of the deceased insured, the loss assessment model uses the ear tag identification results from the key information identification data to retrieve the insured's file record at the time of underwriting from the policy data. The model then associates the ear tag identification results with a group photo of the deceased insured: by analyzing the image features of each insured in the group photo, the model extracts the position, posture, and identifiable individual features of each deceased insured in the photo. Based on the insured's description information at the time of underwriting in the policy data, a one-to-one mapping relationship is established between each detected deceased in the photo and its corresponding ear tag. After completing this mapping, the model confirms that the insured insured state is present in the photo with the ear tag number, and that its full-body image has been uploaded by the farmer, thus confirming the insured's survival status as "deceased." The model then adds the insured insured to the death list and accumulates the total number of deaths.

[0057] In the loss assessment weight determination stage, the loss assessment model extracts the target weight identification result output by the intelligent weighing model from the key information identification data. The loss assessment model comprehensively compares this weight identification result with the initial weight data entered at the time of underwriting in the policy data and the theoretical weight range calculated based on the growth curve model: if the weight identification result is within a reasonable range of the theoretical weight range, the loss assessment model directly uses the weight identification result as the loss assessment weight for the deceased target; if the weight identification result is within a reasonable range but has a certain degree of deviation, the loss assessment model uses a weighted average method (i.e., assigning appropriate weights to the weight identification result and the theoretical weight value and then calculating a comprehensive value) to determine the final loss assessment weight.

[0058] In the process of generating loss assessment results and comprehensively determining risks, the loss assessment model summarizes the confirmation results and abnormal records from the above-mentioned stages, and generates loss assessment results that include the ear tag number of each deceased object, the corresponding loss assessment weight, and the total number of deaths. In addition, it can also comprehensively determine the overall risk level of the case based on the abnormal situations marked in the logical consistency verification stage and the abnormal signals newly discovered during the loss assessment process.

[0059] In the claims calculation phase, after the loss assessment is completed, the server reads the assessed weight and ear tag number of each deceased insured object from the loss assessment results. For example, the server matches the assessed weight of each deceased insured object with the standard weight stipulated in the insurance policy: when the assessed weight is below a certain percentage lower limit of the standard weight, the corresponding minimum compensation ratio is matched; when the assessed weight is in a certain middle range of the standard weight, the corresponding standard compensation ratio is matched; when the assessed weight is above a certain percentage upper limit of the standard weight, the corresponding maximum compensation ratio is matched. The server multiplies the single-insurance amount stipulated in the policy by the corresponding compensation ratio for that insured object to obtain the single-insurance amount for that deceased insured object. The server then sums up the single-insurance amounts for all deceased insured objects to obtain the total compensation amount for this claim.

[0060] In the fast-payment amount threshold determination stage, the server compares the calculated total compensation amount with the preset fast-payment amount upper limit threshold in the system configuration. If the total compensation amount is less than or equal to the fast-payment amount upper limit threshold, the server determines that the total compensation amount for the case is within the automatic payment range allowed by the fast-payment channel, the case's risk level is identified as low risk, and it is allowed to enter the subsequent automatic payment process. If the total compensation amount is greater than the fast-payment amount upper limit threshold, it means that although the case met the fast-payment conditions in the pre-identification, the actual compensation amount calculated in the final loss assessment has exceeded the set range of the fast-payment channel. The server then terminates the fast-payment process, removes the case from the fast-payment channel, and transfers it to the manual review queue, where a human surveyor or claims reviewer completes the subsequent processing according to the ordinary claims process. By comparing the actual calculated total compensation amount with the preset upper limit in the fast-payment amount threshold determination stage, cases exceeding the fast-payment channel's amount range can be effectively intercepted and transferred to manual review before entering the payment stage. This ensures that all cases processed by the fast-payment channel meet the definition criteria for small-amount cases and avoids the risk of the automated payment process being misused for medium to large-amount cases. The above-mentioned steps together complete the entire loss assessment analysis chain, from death confirmation, weight assessment, compensation calculation to amount verification, making the loss assessment results have the characteristics of sufficient basis, transparent calculation and compliance controllability.

[0061] By applying the technical solution of this embodiment, the system first responds to livestock insurance claims initiated by farmers and obtains claim condition information. Based on this, it identifies whether the request meets preset fast claim conditions. Only after meeting the fast claim conditions is the corresponding on-site evidence data and policy data obtained from the user's end. A pre-trained recognition model automatically extracts key information such as ear tag numbers, number of deaths, and weight of the animal from the on-site evidence. Subsequently, a logical consistency check is performed based on a cross-comparison between the on-site evidence data, policy data, and the identified key information. Only after passing the check is the loss assessment model allowed to complete the loss assessment. Finally, based on the loss assessment analysis results, it is determined whether to execute fast claim processing. This achieves fully automated processing of small-amount cases from reporting to decision-making. It enables eligible small-amount cases to automatically complete loss assessment and determination after effective evidence collection without the need for an investigator to be present, shortening the claim cycle, reducing labor costs, and minimizing the risk of disease transmission from investigators entering and leaving the farm.

[0062] Furthermore, as a refinement and extension of the specific implementation methods of the above embodiments, and to fully illustrate the specific implementation process of this embodiment, another method for claims risk control in aquaculture insurance is provided, such as... Figure 2 As shown, the method includes: Step 201: Based on the damage assessment results and the total compensation amount, automatically generate a case application form; Step 202: Push the case application form to the user terminal to request the user terminal's electronic signature confirmation; Step 203: After receiving the electronic signature confirmation, call the payment gateway to pay the compensation to the policyholder's account.

[0063] In this embodiment, after the total compensation amount is less than or equal to the upper limit threshold of the fast compensation amount and the case risk level is low, the server initiates the automatic payment stage of the fast claims process. The server first reads the ear tag number, assessed weight, and corresponding compensation ratio for each deceased animal from the loss assessment analysis results; it then reads the single-animal compensation amount and total compensation amount for each deceased animal from the compensation calculation results; and it reads the time and location of the accident information filled in by the farmer from the aquaculture insurance claim request. The server dynamically fills in the above data according to the field positions in the preset case application template, generating a complete electronic case application. The main body of this electronic case application includes a detailed loss assessment list presented in tabular form. Each row in the table records the ear tag number, assessed weight, compensation ratio, and single-animal compensation amount for one deceased animal, and the total compensation amount is summarized at the end of the table. The server pushes the generated electronic case application to the farmer's client in encrypted form. After receiving the case application, the client displays the complete application content on the interface for the farmer to review. After reviewing and confirming the information, the farmer clicks the confirmation button on the interface to initiate the electronic signature process. The client generates legally valid electronic signature data and sends this data back to the server along with the electronic file of the case application. Upon receiving a valid electronic signature confirmation instruction, the server begins the payment operation. The server extracts the policyholder's linked bank account information (including the bank name, account holder name, and bank account number) from the policy data, extracts the total compensation amount from the compensation calculation results, and extracts the case number of this claim request from the case record as the payment transaction association number. The server encrypts and assembles the above information according to the requirements of the payment gateway interface specification, generates a payment request message, and sends the payment request message to the payment gateway through a secure network channel. After receiving the request, the payment gateway forwards the transfer instruction to the bank. After completing the fund transfer, the bank returns a transaction result notification to the payment gateway, which forwards the notification to the server. The server parses the transaction result notification returned by the payment gateway and extracts the transaction serial number and payment completion time. If the transaction is successful, the server will associate the payment record with the case record and generate a payment completion notification to be pushed to the client.

[0064] In this embodiment, the case application form is automatically generated based on the damage assessment and compensation calculation results, fully automating the traditional claims document creation process that required manual filling, review, and printing. This eliminates waiting time and data entry errors caused by manual document creation. Furthermore, the standardized template format ensures consistency in the content structure of the case application form across different cases, facilitating farmers' quick location and understanding of key compensation information. Thus, all stages from rapid claims condition identification, multi-source data collection and identification, logical consistency verification, damage assessment analysis to compensation payment are automatically completed without human intervention, forming a complete automated closed loop for rapid claims processing.

[0065] Furthermore, as Figure 1 In terms of specific implementation, this application provides a claims risk control device for aquaculture insurance, such as... Figure 3 As shown, the device includes: The judgment module is used to respond to a livestock insurance claim request, obtain the claim condition information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the fast claim conditions based on the claim condition information; The identification module is used to obtain the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request if the conditions for fast claims are met, and to call the pre-trained identification model to automatically identify the preset key information in the user-side on-site evidence data to obtain key information identification data. The loss assessment module is used to perform logical consistency verification on the aquaculture insurance claim based on the on-site evidence data from the user terminal, the policy data, and the key information identification data. When the logical consistency verification is passed, the module performs loss assessment analysis on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model. The execution module is used to perform rapid claims risk control processing on the aquaculture insurance claim requests based on the loss assessment analysis results.

[0066] In one optional embodiment, the claims condition information includes at least one of the following: the region where the farmer is located, the insured object, the number of claims, and the pre-payment amount; the judgment module is used for: Identify whether the local authorities where the farmers are located have opened fast claims processing services; Identify whether the insured object belongs to the preset quick-claim insurance category catalog; Identify whether the number of claims is less than or equal to a preset threshold for the number of claims in a single quick settlement. Identify whether the estimated compensation amount is less than or equal to a preset upper limit threshold for quick compensation.

[0067] In one optional embodiment, the user-side on-site evidence data includes photos of the deceased targets taken and uploaded by the user-side camera, photos of the ear tags, photos of the weighing, and photos of the harmless treatment; the preset key information includes the quantity information of the deceased targets, the ear tag number information, and the weight information of the targets; the pre-trained recognition model includes a pre-trained ear tag number optical character recognition model, a pre-trained intelligent counting model, and a pre-trained intelligent counting model. The identification module is specifically used for: A pre-trained optical character recognition model for ear tags is invoked to identify the ear tags in the ear tag photo, and the ear tag identification result is obtained. The pre-trained intelligent point counting model is invoked to identify the number of death targets in the group photo of the death targets, and the death count identification result is obtained. The pre-trained intelligent point counting model is invoked to identify the scale readings in the weighing photo and estimate the weight of the target based on the scale readings, thus obtaining the target weight identification result.

[0068] In one optional embodiment, the policy data includes a list of insured objects, initial weight data of the insured objects, and geographical location data of the insured objects; the loss assessment module is specifically used for: The ear tag identification result in the key information identification data is compared with the list of insured objects in the policy data to determine whether the ear tag identification result is within the range of the list of insured objects; otherwise, the logical consistency is determined to fail. Based on the initial weight data of the insured object in the policy data and the time interval from the underwriting time to the claim time, the theoretical weight range is calculated according to the preset growth curve model. The weight identification result of the object in the key information identification data is compared with the theoretical weight range to determine whether the weight identification result of the object is within the theoretical weight range. Otherwise, the logical consistency is determined to be unsuccessful. The number of deaths identified in the key information identification data is compared with the number of deaths reported in the livestock insurance claim request. It is determined whether the deviation between the number of deaths identified and the number of deaths reported is within a preset deviation. Otherwise, the logical consistency is deemed to have failed. The pre-trained standard recognition model is invoked to identify the harmless treatment photos in the user terminal's on-site evidence data, and to determine whether the harmless treatment meets the preset standard requirements. If not, the logical consistency is deemed to have failed.

[0069] In an optional embodiment, the damage assessment module is further configured to: Metadata analysis is performed on the photos in the user terminal on-site evidence data to extract the exchangeable image file format information of the photos. The shooting time information, geographical location information and device identification information in the watermark information are compared with the corresponding metadata fields in the exchangeable image file format information. If the watermark information is inconsistent with the exchangeable image file format information, it is determined that the photo has been tampered with and the logical consistency check fails. The geographical location information of the photos in the user terminal on-site evidence data is compared with the insured geographical location data in the policy data to determine whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to have failed. Based on the shooting time information in the watermark information of the user terminal's on-site evidence data and the reporting time information of the aquaculture insurance claim, it is determined whether the shooting time is after the reporting time and whether the time interval between the shooting time and the reporting time is less than a preset watermark time consistency threshold; if not, the logical consistency check is determined to fail; wherein, the camera of the user terminal is a watermark camera; the watermark camera automatically embeds shooting time information into the photo when shooting; Based on the geographical location information in the watermark information and the insured geographical location data in the policy data, it is determined whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to fail; the watermark camera automatically embeds geographical location information into the photo when taking the picture.

[0070] In an optional embodiment, the damage assessment module is further configured to: Based on the ear tag identification results in the key information identification data and the group photo of the deceased target in the user terminal on-site evidence data, the identity and survival status of the deceased target are confirmed; based on the target weight identification results in the key information identification data, the assessed weight of the deceased target is determined; based on the assessed weight of the deceased target and the assessed weight of the deceased target, the damage assessment analysis result is determined; The execution module is specifically used for: Based on the loss assessment results, the standard weight of the insured, the single insured amount and compensation ratio stipulated in the policy, calculate the single insured amount, and sum up the single insured amounts of all deceased insured persons to obtain the total compensation amount; The total compensation amount is compared with a preset fast compensation amount upper limit threshold. If the total compensation amount is less than or equal to the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a security request and the fast compensation process is executed for automatic payment. If the total compensation amount is greater than the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a risk request, the fast compensation process is terminated, and the request is transferred to the manual review queue.

[0071] In one optional embodiment, the fast claims process includes: Based on the damage assessment results and the total compensation amount, a case application form will be automatically generated; The case application form is pushed to the user terminal to request the user terminal's electronic signature confirmation; After receiving the electronic signature confirmation, the payment gateway is invoked to pay the compensation to the policyholder's account.

[0072] It should be noted that other corresponding descriptions of the functional units involved in the claims risk control device for aquaculture insurance provided in this application embodiment can be found in the following references. Figures 1 to 2 The corresponding descriptions in the method will not be repeated here.

[0073] This application also provides a computer device, which may specifically be a personal computer, a server, a network device, etc. Figure 4 As shown, the computer device includes a bus, a processor, memory, and a communication interface, and may also include an input / output interface and a display device. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores location information. The network interface allows communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the steps in the various method embodiments.

[0074] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0075] In one embodiment, a computer-readable storage medium is provided, which may be non-volatile or volatile, having stored thereon a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0076] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0077] It should be noted that the user personal information involved in the embodiments of this application is all authorized (with the knowledge and consent) by the relevant parties or fully authorized by all parties, and the executing entity can obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with the relevant laws and regulations of the relevant countries and regions, and do not violate public order and good morals. It should be noted that if any software tools or components other than those of this company appear in the embodiments of this application, they are merely illustrative examples and do not represent actual use.

[0078] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, graphics processors, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0079] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0080] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for claims risk control in aquaculture insurance, characterized in that, The method includes: In response to a livestock insurance claim request, obtain the claim conditions information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the conditions for fast claim processing based on the claim conditions information; If the conditions for fast claims settlement are met, the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request are obtained, and the pre-trained recognition model is called to automatically identify the preset key information in the user-side on-site evidence data to obtain key information recognition data. Based on the on-site evidence data from the user terminal, the policy data, and the key information identification data, a logical consistency check is performed on the aquaculture insurance claim. If the logical consistency check is passed, a loss assessment analysis is performed on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model. Based on the loss assessment results, the aquaculture insurance claim requests are processed quickly for risk control.

2. The method according to claim 1, characterized in that, The claim conditions information includes at least one of the following: the region and institution where the farmer is located, the insured object, the number of claims, and the amount of pre-payment; the step of identifying whether the livestock insurance claim request meets the fast claim conditions based on the claim conditions information includes at least one of the following: Identify whether the local authorities where the farmers are located have opened fast claims processing services; Identify whether the insured object belongs to the preset quick-claim insurance category catalog; Identify whether the number of claims is less than or equal to a preset threshold for the number of claims in a single quick settlement. Identify whether the estimated compensation amount is less than or equal to a preset upper limit threshold for quick compensation.

3. The method according to claim 1, characterized in that, The on-site evidence data from the user terminal includes photos of the deceased targets taken and uploaded by the user terminal's camera, photos of the ear tags, photos of the weighing, and photos of the harmless treatment; the preset key information includes the quantity information of the deceased targets, the ear tag number information, and the weight information of the targets; the pre-trained recognition model includes a pre-trained optical character recognition model for ear tag numbers, a pre-trained intelligent counting model, and a pre-trained intelligent counting model. The process involves invoking a pre-trained recognition model to automatically identify preset key information from the user-side on-site evidence data, resulting in key information identification data, including: A pre-trained optical character recognition model for ear tags is invoked to identify the ear tags in the ear tag photo, and the ear tag identification result is obtained. The pre-trained intelligent point counting model is invoked to identify the number of death targets in the group photo of the death targets, and the death count identification result is obtained. The pre-trained intelligent point counting model is invoked to identify the scale readings in the weighing photo and estimate the weight of the target based on the scale readings, thus obtaining the target weight identification result.

4. The method according to claim 3, characterized in that, The policy data includes a list of insured objects, initial weight data of the insured objects, and geographical location data of the insured objects; the logical consistency verification of the aquaculture insurance claim request based on the user-end on-site evidence data, the policy data, and the key information identification data includes: The ear tag identification result in the key information identification data is compared with the list of insured objects in the policy data to determine whether the ear tag identification result is within the range of the list of insured objects; otherwise, the logical consistency is determined to fail. Based on the initial weight data of the insured object in the policy data and the time interval from the underwriting time to the claim time, the theoretical weight range is calculated according to the preset growth curve model. The weight identification result of the object in the key information identification data is compared with the theoretical weight range to determine whether the weight identification result of the object is within the theoretical weight range. Otherwise, the logical consistency is determined to be unsuccessful. The number of deaths identified in the key information identification data is compared with the number of deaths reported in the aquaculture insurance claim request. It is determined whether the deviation between the number of deaths identified and the number of deaths reported is within a preset deviation. Otherwise, the logical consistency is deemed to have failed. The pre-trained standard recognition model is invoked to identify the harmless treatment photos in the user terminal's on-site evidence data, and to determine whether the harmless treatment meets the preset standard requirements. If not, the logical consistency is deemed to have failed.

5. The method according to claim 4, characterized in that, The step of performing logical consistency verification on the aquaculture insurance claim based on the user-end on-site evidence data, the policy data, and the key information identification data also includes at least one of the following: Metadata analysis is performed on the photos in the user terminal on-site evidence data to extract the exchangeable image file format information of the photos. The shooting time information, geographical location information and device identification information in the watermark information are compared with the corresponding metadata fields in the exchangeable image file format information. If the watermark information is inconsistent with the exchangeable image file format information, it is determined that the photo has been tampered with and the logical consistency check fails. The geographical location information of the photos in the user terminal on-site evidence data is compared with the insured geographical location data in the policy data to determine whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to have failed. Based on the shooting time information in the watermark information of the user terminal's on-site evidence data and the reporting time information of the aquaculture insurance claim, it is determined whether the shooting time is after the reporting time and whether the time interval between the shooting time and the reporting time is less than a preset watermark time consistency threshold; if not, the logical consistency check is determined to fail; wherein, the camera of the user terminal is a watermark camera; the watermark camera automatically embeds shooting time information into the photo when shooting; Based on the geographical location information in the watermark information and the insured geographical location data in the policy data, it is determined whether the distance between the accident location and the insured geographical location is less than a preset location deviation threshold; if not, the logical consistency check is determined to fail; the watermark camera automatically embeds geographical location information into the photo when taking the picture.

6. The method according to claim 3, characterized in that, The loss assessment analysis performed using the pre-trained loss assessment model on the user-side on-site evidence data, the policy data, and the key information identification data includes: Based on the ear tag identification results in the key information identification data and the group photo of the deceased target in the user terminal on-site evidence data, the identity and survival status of the deceased target are confirmed; based on the target weight identification results in the key information identification data, the assessed weight of the deceased target is determined; based on the assessed weight of the deceased target and the assessed weight of the deceased target, the damage assessment analysis result is determined; The rapid claims risk control processing for the aquaculture insurance claims based on the loss assessment results includes: Based on the loss assessment results, the standard weight of the insured, the single insured amount and compensation ratio stipulated in the policy, calculate the single insured amount, and sum up the single insured amounts of all deceased insured persons to obtain the total compensation amount; The total compensation amount is compared with a preset fast compensation amount upper limit threshold. If the total compensation amount is less than or equal to the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a security request and the fast compensation process is executed for automatic payment. If the total compensation amount is greater than the fast compensation amount upper limit threshold, the aquaculture insurance claim request is treated as a risk request, the fast compensation process is terminated, and the request is transferred to the manual review queue.

7. The method according to claim 6, characterized in that, The fast claims process includes: Based on the damage assessment results and the total compensation amount, a case application form will be automatically generated; The case application form is pushed to the user terminal to request the user terminal's electronic signature confirmation; After receiving the electronic signature confirmation, the payment gateway is invoked to pay the compensation to the policyholder's account.

8. A claims risk control device for aquaculture insurance, characterized in that, The device includes: The judgment module is used to respond to a livestock insurance claim request, obtain the claim condition information of the livestock insurance claim request, and identify whether the livestock insurance claim request meets the fast claim conditions based on the claim condition information; The identification module is used to obtain the user-side on-site evidence data and policy data corresponding to the aquaculture insurance claim request if the conditions for fast claims are met, and to call the pre-trained identification model to automatically identify the preset key information in the user-side on-site evidence data to obtain key information identification data. The loss assessment module is used to perform logical consistency verification on the aquaculture insurance claim based on the on-site evidence data from the user terminal, the policy data, and the key information identification data. When the logical consistency verification is passed, the module performs loss assessment analysis on the on-site evidence data from the user terminal, the policy data, and the key information identification data using a pre-trained loss assessment model. The execution module is used to perform rapid claims risk control processing on the aquaculture insurance claim requests based on the loss assessment analysis results.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

10. A computer device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

Images