Method, device and equipment for processing multi-source data of agricultural insurance underwriting and medium

Abstract

The present application relates to the field of artificial intelligence technology, which can be applied to the field of financial technology, and discloses a processing method, device and equipment for multi-source data of agricultural insurance underwriting and a medium, comprising: in response to multi-media materials submitted by a user for agricultural insurance underwriting, using multi-modal analysis technology to analyze the multi-media materials to obtain structured text information and crop image visual feature information with unique correlation mapping relationship; taking the correlation mapping relationship as a trigger condition, extracting multi-dimensional feature information through an agricultural insurance scene rule engine; generating a multi-modal prompt field suitable for an agricultural insurance underwriting model according to the multi-dimensional feature information; and through the attention mechanism of the agricultural insurance underwriting model, performing risk correlation reasoning on the multi-dimensional feature information according to the multi-modal prompt field to obtain a risk assessment report for agricultural insurance underwriting. Through the precise adaptation of multi-dimensional feature information and the model, this process realizes multi-dimensional feature cross-linking reasoning, which can effectively mine potential risk correlations behind the data.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology and can be applied to the field of financial technology. In particular, it relates to a method, apparatus, equipment and medium for processing multi-source data in agricultural insurance underwriting. Background Technology

[0002] Agricultural insurance is an important means of mitigating agricultural production risks and protecting farmers' economic interests. The underwriting process is a key step in controlling insurance risks and determining underwriting conditions.

[0003] Current agricultural insurance underwriting scenarios primarily rely on manual review. The processing of underwriting data typically involves underwriters manually verifying and extracting information from multi-source data submitted by farmers. This multi-source data includes various paper materials such as planting certificates, weather reports, historical disaster records, and on-site crop photos. However, due to the complexity and unique nature of agricultural data, agricultural insurance underwriting multi-source data often exhibits strong heterogeneity, and core data is easily affected by external environmental factors. This increases the difficulty of analyzing agricultural insurance underwriting multi-source data, making it difficult to effectively uncover potential risk correlations behind the data and affecting the accuracy of underwriting risk assessment. Summary of the Invention

[0004] This invention provides a method, apparatus, computer equipment, and medium for processing multi-source data in agricultural insurance underwriting, in order to solve the technical problem that related technologies relying on manual review models are unable to effectively uncover potential risk correlations behind the data, thus affecting the accuracy of underwriting risk assessment.

[0005] Firstly, a method for processing multi-source data in agricultural insurance underwriting is provided, including: In response to the multi-media materials submitted by the user for agricultural insurance underwriting, multimodal parsing technology is used to parse the multi-media materials to obtain structured text information and crop image visual feature information with unique association mapping relationship; Using the aforementioned association mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. Based on the multi-dimensional feature information, generate multimodal prompt fields adapted to the agricultural insurance underwriting model; The multimodal cue fields are input into the agricultural insurance underwriting model, and the attention mechanism of the agricultural insurance underwriting model is used to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields to obtain a risk assessment report for agricultural insurance underwriting.

[0006] Secondly, a device for processing multi-source data in agricultural insurance underwriting is provided, including: The parsing module is used to respond to the multi-media materials submitted by the user for agricultural insurance underwriting, and to parse the multi-media materials using multimodal parsing technology to obtain structured text information and crop image visual feature information with unique association mapping relationship; The extraction module is used to extract multi-dimensional feature information through the agricultural insurance scenario rule engine, using the association mapping relationship as a trigger condition. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. The generation module is used to generate multimodal prompt fields adapted to the agricultural insurance underwriting model based on the multidimensional feature information; The reasoning module is used to input the multimodal cue fields into the agricultural insurance underwriting model, so as to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields through the attention mechanism of the agricultural insurance underwriting model, and obtain the risk assessment report for agricultural insurance underwriting.

[0007] Thirdly, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-mentioned method for processing multi-source data for agricultural insurance underwriting.

[0008] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-mentioned method for processing multi-source data in agricultural insurance underwriting.

[0009] In the aforementioned scheme implemented by the processing method, device, computer equipment, and storage medium for multi-source data in agricultural insurance underwriting, the server responds to the multi-media materials submitted by the user for agricultural insurance underwriting, and uses multimodal parsing technology to parse the multi-media materials to obtain structured text information and crop image visual feature information with a unique correlation mapping relationship; using the correlation mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area; multimodal prompt fields adapted to the agricultural insurance underwriting model are generated based on the multi-dimensional feature information; the multimodal prompt fields are input into the agricultural insurance underwriting model, and the attention mechanism of the agricultural insurance underwriting model performs risk correlation reasoning on the multi-dimensional feature information based on the multimodal prompt fields to obtain a risk assessment report for agricultural insurance underwriting, which is then sent to the client. In this invention, multimodal analysis technology is used to process heterogeneous materials of various media, which can differentiate and adapt to the characteristics of heterogeneous data, improve the accuracy of multi-source data in agricultural underwriting, and realize the linkage and traceability of heterogeneous data through a unique association mapping relationship. It also triggers the rule engine to automatically extract features. The rule engine is configured specifically for agricultural insurance scenarios. After being triggered, it only extracts multi-dimensional features that are strongly related to agricultural insurance underwriting risks, so that the multi-dimensional feature information is more in line with the needs of agricultural insurance risk assessment. Furthermore, it combines attention mechanism to guide risk association reasoning. Through the precise adaptation of multi-dimensional feature information and model, it realizes multi-dimensional feature cross-linking reasoning, effectively explores the potential risk associations behind the data, improves the accuracy of underwriting risk judgment, and makes risk judgment more in line with the actual needs of agricultural insurance underwriting. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the application environment of the method for processing multi-source data in agricultural insurance underwriting according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a method for processing multi-source data in agricultural insurance underwriting according to an embodiment of the present invention. Figure 3 yes Figure 2 A schematic diagram of a specific implementation method for step S10; Figure 4 yes Figure 2 A schematic diagram of a specific implementation method for step S20; Figure 5 yes Figure 2A flowchart illustrating another specific implementation of step S20; Figure 6 This is a flowchart illustrating a method for processing multi-source data in agricultural insurance underwriting according to another embodiment of the present invention; Figure 7 yes Figure 6 A schematic diagram of a specific implementation method for step S50; Figure 8 yes Figure 2 A schematic diagram of a specific implementation method for step S30; Figure 9 yes Figure 2 A schematic diagram of a specific implementation of step S40; Figure 10 This is a schematic diagram of the structure of a multi-source data processing device for agricultural insurance underwriting in one embodiment of the present invention; Figure 11 This is a schematic diagram of the structure of a computer device according to an embodiment of the present invention; Figure 12 This is another structural schematic diagram of a computer device according to one embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The method for processing multi-source data in agricultural insurance underwriting provided in this embodiment of the invention can be applied to, for example... Figure 1In this application environment, the server responds to the multi-media materials submitted by the user for agricultural insurance underwriting. It uses multimodal parsing technology to analyze the materials, obtaining structured text information and crop image visual feature information with unique association mapping relationships. Using these association mapping relationships as trigger conditions, the server extracts multi-dimensional feature information through an agricultural insurance scenario rule engine. This multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data for the planting area. Based on this multi-dimensional feature information, a multimodal prompt field adapted to the agricultural insurance underwriting model is generated. This multimodal prompt field is then input into the agricultural insurance underwriting model, which uses its attention mechanism to perform risk association reasoning based on the multimodal prompt field to obtain a risk assessment report for agricultural insurance underwriting. Finally, the risk assessment report is sent to the client. In this invention, multimodal analysis technology is employed to heterogeneously process multi-media materials, enabling differentiated adaptation to heterogeneous data characteristics and improving the accuracy of multi-source data in agricultural underwriting. A unique association mapping relationship is used to achieve linked tracing of heterogeneous data, triggering a rule engine to automatically extract features. This rule engine is specifically configured for agricultural insurance scenarios, extracting only multi-dimensional features strongly correlated with agricultural insurance underwriting risks upon triggering. This ensures that multi-dimensional feature information better aligns with the needs of agricultural insurance risk assessment. Furthermore, an attention mechanism guides risk association reasoning, achieving cross-linked reasoning of multi-dimensional features through precise adaptation of multi-dimensional feature information to the model. This effectively uncovers potential risk associations behind the data, improving the accuracy of underwriting risk judgment and making risk assessment more aligned with the actual needs of agricultural insurance underwriting. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0014] Please see Figure 2 As shown, Figure 2 A flowchart illustrating the method for processing multi-source data in agricultural insurance underwriting provided in this embodiment of the invention includes the following steps: S10: In response to the multi-media materials submitted by the user for agricultural insurance underwriting, multi-modal parsing technology is used to parse the multi-media materials to obtain structured text information and crop image visual feature information with unique association mapping relationship.

[0015] The method for processing multi-source data in agricultural insurance underwriting provided by this invention can be applied to financial property insurance scenarios. Specifically, it is adapted to the entire underwriting process of agricultural insurance in the financial property insurance field. Addressing the industry pain points of complex multi-media underwriting materials and high risk assessment difficulty in agricultural insurance business in financial property insurance scenarios, it improves the underwriting efficiency and risk judgment accuracy of agricultural insurance business in financial property insurance through core technologies such as multimodal parsing, correlation trigger extraction, and multimodal reasoning, and adapts to the business needs of large-scale and compliant financial property insurance.

[0016] Among them, the multi-media materials for agricultural insurance underwriting refer to the various types and heterogeneous data materials submitted by users related to agricultural insurance underwriting. These materials vary in form and cover two core types: text and images. They serve as the core original basis for agricultural insurance underwriting. Specifically, they include text media, such as handwritten insurance applications, printed planting certificates, land ownership certificates, past policy records, and farmer identification documents, in both paper and electronic formats. They also include image media, such as on-site photos of crops, aerial photos of the land, plot boundary diagrams, video frames of crop growth, and photos of damage.

[0017] In this embodiment, the multimodal parsing technology is a dedicated parsing technology fusion scheme adapted to different data modalities of text and images for the aforementioned heterogeneous multi-media materials. Unlike single text parsing or single image parsing, the multimodal parsing technology can output character recognition characteristics adapted to text and visual feature extraction characteristics adapted to images through precise parsing of different modalities and synchronous output linkage, ensuring that data from different media can be efficiently converted into data that is available for certification.

[0018] In real-world applications, isolated and unrelated data from multiple media can lead to underwriting vulnerabilities and compliance risks. Specifically, for example... Figure 3 As shown, step S10, which involves using multimodal analysis technology to analyze the multi-media material and obtain structured text information and crop image visual feature information with unique correlation mapping relationships, includes the following steps: S11: Use multimodal analysis technology to identify text-type materials and image-type materials in the multi-media materials.

[0019] S12: Perform text extraction on the text-type material to obtain structured text information; and perform image extraction on the image-type material to obtain crop image visual feature information.

[0020] S13: Construct a unique association mapping relationship between the structured text information and the crop image visual feature information based on the identifier set in the underwriting business, and obtain structured text information and crop image visual feature information with a unique association mapping relationship.

[0021] In this embodiment, the multimodal parsing technology is an integrated intelligent parsing technology that combines text modality recognition and image modality recognition, targeting the heterogeneous characteristics of multi-media materials in agricultural insurance underwriting. It can simultaneously adapt to the differentiated parsing requirements of character-based text media and visual image media, and realize the automatic classification, accurate parsing, and linkage output of parsing results for multi-media materials.

[0022] Specifically, in the process of extracting text from text-based materials to obtain structured text information, OCR optical character recognition technology can be used to extract valid characters from the text-based materials, filter out invalid information, and then organize them into structured text information containing the planting entity identifier, the insured plot code, and the planting category.

[0023] Specifically, in the process of extracting visual features of crop images from image-based materials, deep learning visual recognition technology can be used to extract the core targets of crops and plots, extract core visual features such as growth stage and plot area, and quantify and convert them into visual feature information of crop images.

[0024] To ensure a one-to-one correspondence between structured text information and crop image visual feature information for the same farmer and the same insured plot, and to avoid data isolation, a unique association identifier can be generated based on the underwriting transaction number as the core of the binding. This association identifier can use the agricultural insurance underwriting-specific code. Then, this unique association identifier is written into the association code field of the structured text information and the feature association label of the crop image visual feature information, respectively. Finally, the bound structured text information and crop image visual feature information are output.

[0025] Understandably, to ensure the validity of the association between structured text information and crop image visual feature information, a consistency verification can be performed on the text information and image features. If the verification passes, an association mapping relationship is established between the structured text information and the crop image visual feature information. For example, verifying whether the planting entity and plot code in the structured text information are consistent with the business filing information associated with the unique business identifier, and verifying whether the crop and plot corresponding to the crop image visual feature information are consistent with the declaration information associated with the unique business identifier.

[0026] S20: Using the aforementioned association mapping relationship as a triggering condition, extract multi-dimensional feature information through the agricultural insurance scenario rule engine.

[0027] In this embodiment, the unique association mapping relationship between structured text information and crop image visual feature information can be used as the trigger threshold for starting the rule engine. When a valid unique association mapping relationship is detected, the agricultural insurance scenario rule engine is automatically triggered. Based on the mapping relationship, text and image data are retrieved in a linked manner, and multi-dimensional feature information that is strongly related to underwriting risk is extracted according to the preset rules of agricultural insurance underwriting.

[0028] In this embodiment, the multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. For the planting area coordinates, the insured plot code in the structured text information is used as an index, linked to the agricultural insurance underwriting GIS geographic database. Simultaneously, the plot boundary features in the crop image visual feature information are combined to calibrate the coordinate accuracy. Specific extraction specifications can use standardized latitude and longitude coordinates, such as XX°XX′XX″N, XX°XX′XXE. For the crop growth stage labels, based on crop growth features in the crop image visual feature information, such as plant height, leaf condition, and planting density, combined with the agricultural insurance crop growth period rule base, standardized classifications can be used, such as for rice: seedling stage / tillering stage / jointing stage / heading stage / maturity stage. For the meteorological factor data of the planting area, the planting area coordinates are used as the geographic reference, and the growth period corresponding to the crop growth stage label is used as the time reference. Targeted extraction is linked to an authoritative meteorological database. Specific extraction specifications can use meteorological indicators, such as temperature, precipitation, sunshine duration, wind speed, and extreme weather warning records.

[0029] In practical applications, the agricultural insurance rule engine has a built-in rule base for agricultural insurance underwriting. This rule base includes a list of associated parameters, feature extraction items, and mapping relationships between data sources. Specifically, for example... Figure 4 As shown, in step S20, that is, using the aforementioned association mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine, including the following steps: S21: Extract the planting entity identifier and the insured land plot code from the association mapping relationship as association parameters.

[0030] S22: Input the associated parameters into the agricultural insurance scenario rule engine, and query the built-in mapping relationship table through the agricultural insurance scenario rule engine to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

[0031] In step S21, the planting entity identifier and the insured plot code are extracted from the structured text information bound by the mapping relationship. The planting entity identifier is a unique identification code for the planting entity, and the insured plot code is a unique geographic code for the insured plot. These two constitute globally unique association parameters, and they correspond to the same planting entity and the same insured plot as the visual feature information of the crop image. It can be understood that the two types of parameters extracted are the core business identifiers of agricultural insurance underwriting, which are consistent with the actual business and ensure that subsequent feature extraction meets the needs of underwriting risk assessment.

[0032] In step S22, the combination of the planting entity identifier extracted from the association mapping relationship and the insured plot code is used as the association parameter and input into the agricultural insurance scenario rule engine according to a preset format. The agricultural insurance scenario rule engine then calls the built-in agricultural insurance mapping relationship table to query the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code. Here, the agricultural insurance mapping relationship table pre-stores a unique correspondence between the two association parameters, feature extraction items, and data sources. For example, entity A + plot A1 corresponds to planting area coordinates, crop growth stage labels, and data sources of structured text + crop images + GIS database. Accordingly, through precise matching of the two parameter combinations, the feature extraction items corresponding to the two parameter combinations and the data sources of each extraction item can be located.

[0033] In practical applications, the association parameters directly determine the subsequent feature extraction items and data sources. Invalid association parameters can lead to the retrieval of incorrect text, images, or third-party database data, rendering the underwriting data unreliable. Verification ensures that the original data corresponding to the association parameters is legitimate and valid. Furthermore, for example... Figure 5 As shown, after step S21, the method further includes the following steps: S23: Perform validity verification on the core associated parameters.

[0034] The validity verification includes at least three verification methods: completeness verification, uniqueness verification, and correlation verification. Completeness verification verifies that the planting entity identifier and the insured plot code are not missing or empty. Uniqueness verification verifies that the planting entity identifier and the insured plot code are not duplicated in the underwriting business. Correlation verification verifies that the planting entity identifier and the insured plot code are bound together.

[0035] Accordingly, S22: Input the verified associated parameters into the agricultural insurance scenario rule engine, and use the agricultural insurance scenario rule engine to query the built-in mapping relationship table to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

[0036] It is understandable that among multi-dimensional feature information, the three types of features—planting area coordinates (spatial dimension), crop growth stage (temporal dimension), and meteorological factors (spatiotemporal correlation dimension)—are naturally strongly correlated, and the sources from which these three types of features are extracted are diverse, naturally leading to a risk of spatiotemporal misalignment. If spatiotemporal misalignment occurs, it will directly lead to incorrect underwriting risk assessment conclusions. For example, matching the coordinates of wheat planting plots in the north with meteorological data from rice-producing areas in the south, or matching wheat seedling stage features with high precipitation meteorological data from the grain-filling stage. To fundamentally avoid this risk, spatiotemporal correction of multi-dimensional feature information is necessary. Furthermore, such as... Figure 6 As shown, after step S20, the method further includes the following steps: S50: Use a spatiotemporal error correction mechanism to perform error correction and verification on the multi-dimensional feature information.

[0037] In this embodiment, the spatiotemporal error correction mechanism is a customized error correction and verification mechanism for multi-dimensional feature information in agricultural insurance underwriting. It ensures that the multi-dimensional feature information is consistent in time and space and fits the actual situation in the field of agricultural insurance, providing accurate and compliant data for subsequent underwriting risk reasoning.

[0038] The specific spatiotemporal error correction mechanism includes a verification module for regional matching, a verification module for reproductive period matching, and a verification module for spatiotemporal linkage adaptability. Through spatial verification, time verification, and spatiotemporal linkage verification, multiple verification logics are formed to identify and correct spatiotemporal misalignment deviations in multi-dimensional feature information.

[0039] Specifically, such as Figure 7 As shown, step S50, which involves using a spatiotemporal error correction mechanism to perform error correction and verification on the multi-dimensional feature information, includes the following steps: S51: Using the coordinates of the planting area as a spatial constraint benchmark, and based on the spatial constraint benchmark, verify the consistency between the associated record of the planting area coordinates and the filing data of the geographical area corresponding to the planting area coordinates.

[0040] S52: Using the crop growth stage label corresponding to the crop growth period as a time constraint benchmark, verify the matching between the collection time of the meteorological factor data of the planting area and the crop growth period.

[0041] In step S51, the coordinates of the planting area in the multi-dimensional features can be anchored as the spatial constraint benchmark. This spatial constraint benchmark is the precise latitude and longitude coordinates of the insured plot, which has a unique geographical attribution attribute. Based on this spatial constraint benchmark, the geographical filing data of the administrative region and cadastral zone corresponding to the coordinates is retrieved from the agricultural insurance underwriting GIS geographic database. Then, the underwriting association record bound to the planting area coordinates is checked to see if it is consistent with the retrieved geographical region filing data, thus verifying the mismatch of the plot's geographical attributes from a spatial dimension.

[0042] Understandably, the spatial constraint is based on the coordinates of the planting area, which is equivalent to the unique geographical identity of the insured plot. It can pinpoint the precise geographical ownership of the plot and constrain all related data to match the geographical attributes corresponding to the coordinates. It is the only benchmark for spatial verification.

[0043] In step S52, the standardized crop growth period corresponding to the crop growth stage label in the multi-dimensional features can be anchored as the time constraint benchmark. This time constraint benchmark is a growth cycle of a fixed duration specific to the corresponding crop category. Based on this time constraint benchmark, the collection time range of the extracted meteorological factor data of the planting area is verified to see if it accurately matches the crop growth period, thus verifying the misalignment between meteorological factors and crop growth stages from the time dimension.

[0044] Understandably, the time constraint benchmark is the standardized crop growth period corresponding to the crop growth stage label, which is mapped from the crop growth stage label. For example, the label wheat seedling stage corresponds to a seedling growth period of 25-30 days, and the label rice grain filling stage corresponds to a grain filling growth period of 18-22 days. These crop growth stage labels are all derived from the crop growth period rule library of agricultural insurance underwriting, and have exclusive, fixed, and standardized attributes, which are the only benchmark for time verification.

[0045] S30: Generate a multimodal prompt field adapted to the agricultural insurance underwriting model based on the multidimensional feature information.

[0046] Understandably, to ensure accurate adaptation of multi-dimensional features to the agricultural insurance underwriting model and to provide high-quality input data for the model's risk inference, multi-dimensional feature information can first undergo feature screening and standardized integration according to the agricultural insurance underwriting model input protocol. Then, the integrated structured features are transformed into semantic and vectorized expressions recognizable by the model, ultimately generating a multi-modal prompt field that integrates text, image, and spatiotemporal modal information. This field possesses both structured format and semantic expression, allowing it to be directly parsed and used for inference by the underwriting model.

[0047] Specifically, such as Figure 8 As shown, step S30, which generates a multimodal prompt field adapted to the agricultural insurance underwriting model based on the multidimensional feature information, includes the following steps: S31: The multi-dimensional feature information is structured and organized to remove redundant and invalid data and supplement the feature semantic description.

[0048] S32: Adjust the representation weight of the multi-dimensional feature information based on the risk association weight rules preset in the agricultural insurance underwriting scenario.

[0049] S33: Based on the multimodal input structure adapted to the agricultural insurance underwriting model, the multidimensional feature information with representation weights is divided into text description modality and structured feature modality.

[0050] S34: Based on the text description modality and structured feature modality, format the data to generate a multimodal prompt field that is adapted to the agricultural insurance underwriting model.

[0051] In step S31, redundant and invalid data removal can be performed according to the agricultural insurance underwriting risk association standard, filtering out redundant features and invalid data that are irrelevant to underwriting risk. Redundant features can include, but are not limited to, irrelevant text in plot notes, duplicate coordinate records, etc., while invalid data can include, but is not limited to, incorrectly formatted meteorological values, semantically conflicting growth stage labels, etc. Supplementing feature semantic descriptions can add precise semantic annotations to structured features to avoid model misunderstandings. For example, coordinate 36°XX′XX can be supplemented to coordinate 36°XX′XX-XX (major wheat producing area), and meteorological factor-temperature 5℃ can be supplemented to meteorological factor-temperature 5℃-suitable temperature range for wheat overwintering.

[0052] In step S32, the core risks of agricultural insurance underwriting are crop growth status and the impact of meteorological disasters. Crop growth stage labels directly reflect the crop's risk resistance capacity, while meteorological factor data directly correlates with the probability of disaster occurrence. These two factors have a greater impact on risk assessment than the planting area coordinates. Generally speaking, the representation weight of crop growth stage labels and meteorological factor data of the planting area is higher than that of the planting area coordinates. In step S33, the text description modality is used for model understanding scenarios, and the structured feature modality is used for model reasoning scenarios. Specifically, based on the dual-modal parallel input structure preset by the agricultural insurance underwriting model, multi-dimensional feature information with representation weights is split according to the feature expression form and model purpose. Semantic and scenario-based features are assigned to the text description modality, and standardized and quantified features are assigned to the structured feature modality. Both modalities retain the difference in representation weights.

[0053] In step S34, using the multimodal input format adapted to the agricultural insurance underwriting model as the standard, the split text description modality and structured feature modality are collaboratively formatted. The dual modalities are accurately bound through a unified business association identifier. The expression format and field sorting of the two modalities are standardized to ensure that high-weight features are presented first. Finally, multimodal prompt fields adapted to the reasoning needs of the agricultural insurance underwriting model are generated.

[0054] S40: Input the multimodal cue field into the agricultural insurance underwriting model, so as to perform risk association reasoning on the multidimensional feature information based on the multimodal cue field through the attention mechanism of the agricultural insurance underwriting model, and obtain the risk assessment report for agricultural insurance underwriting.

[0055] In this embodiment, the multimodal prompt field is input into the agricultural insurance underwriting model. The model identifies the weight identifiers in the field through an attention mechanism, prioritizing high-weight features such as crop growth stage and meteorological factors in the planting area. Then, it calls the agricultural insurance underwriting risk rule base to perform correlation matching between core features and risks. Combined with secondary features such as planting area coordinates and land attributes, it verifies and corrects the data. Finally, it outputs a standardized risk assessment report containing risk level, core risk points, correlation basis, and underwriting recommendations.

[0056] Specifically, such as Figure 9 As shown, step S40 involves inputting the multimodal cue fields into the agricultural insurance underwriting model. The model's attention mechanism then performs risk correlation reasoning on the multidimensional feature information based on the multimodal cue fields to obtain a risk assessment report for agricultural insurance underwriting. This includes the following steps: S41: The multimodal prompt field and the corrected multidimensional feature information are synchronously input into the agricultural insurance underwriting model. Based on the weight annotation in the multimodal prompt field, the associated features of the multidimensional feature information are extracted through the model's attention mechanism.

[0057] S42: Based on the associated features of the multi-dimensional feature information, perform multi-dimensional risk association reasoning in agricultural insurance scenarios to determine crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk.

[0058] S43: Integrate the results of the assessment of crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk to generate a standardized agricultural insurance underwriting risk assessment report.

[0059] In step S41, the associated features include the association features between crop growth stage labels and meteorological factor data of planting area, and the association features between planting area coordinates and past disaster records. Specifically, the multimodal prompt field and the multidimensional feature information that has passed the spatiotemporal error correction verification can be simultaneously input into the agricultural insurance underwriting model. The model first uses an attention mechanism to parse the weight labels in the multimodal prompt field to clarify the priority of feature extraction. Then, guided by the weight labels, it accurately extracts the risk association features between high-weight core features and between core features and secondary features from the multidimensional feature information after error correction. For example, the association between overwintering period and low temperature weather, and the adaptation association between coordinate region and growth stage.

[0060] In step S42, based on the correlation features of multi-dimensional feature information, a pre-built risk reasoning rule library specific to agricultural insurance scenarios is invoked to carry out multi-dimensional collaborative risk correlation reasoning. Through precise matching of correlation features and risk rules, multiple risks are quantitatively determined, including crop growth adaptation risk determination, which is used to determine the degree of adaptation of crop growth with regional climate and current weather; regional meteorological disaster risk determination, which is used to determine whether there are crop-related meteorological disasters in the planting area now or in the future; and planting information matching risk determination, which is used to determine the degree of matching between the declared planting information and the actual situation of the plot. Finally, the crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk are output.

[0061] Accordingly, the risk assessment results are integrated in accordance with the agricultural insurance underwriting standards to generate a standardized risk assessment report that complies with the agricultural insurance underwriting business standards.

[0062] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0063] In one embodiment, a processing device for multi-source agricultural insurance underwriting data is provided, which corresponds one-to-one with the processing method for multi-source agricultural insurance underwriting data in the above embodiments. For example... Figure 10 As shown, the processing device for multi-source agricultural insurance underwriting data includes: a parsing module 101, an extraction module 102, a generation module 103, and an inference module 104. Detailed descriptions of each functional module are as follows: The parsing module 101 is used to respond to the multi-media materials submitted by the user for agricultural insurance underwriting, and to parse the multi-media materials using multimodal parsing technology to obtain structured text information and crop image visual feature information with unique association mapping relationship; Extraction module 102 is used to extract multi-dimensional feature information through agricultural insurance scenario rule engine using the association mapping relationship as the trigger condition. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels and meteorological factor data of the planting area. The generation module 103 is used to generate a multimodal prompt field adapted to the agricultural insurance underwriting model based on the multidimensional feature information; The reasoning module 104 is used to input the multimodal prompt field into the agricultural insurance underwriting model, so as to perform risk association reasoning on the multidimensional feature information based on the multimodal prompt field through the attention mechanism of the agricultural insurance underwriting model, and obtain the risk assessment report for agricultural insurance underwriting.

[0064] In one embodiment, the parsing module 101 is specifically used for: Multimodal analysis technology is used to identify text-based and image-based materials in the multi-media materials. Text extraction is performed on the text-based materials to obtain structured text information; and image extraction is performed on the image-based materials to obtain crop image visual feature information. Based on the identifier set in the underwriting business, a unique association mapping relationship is constructed between the structured text information and the crop image visual feature information, resulting in structured text information and crop image visual feature information with a unique association mapping relationship.

[0065] In one embodiment, the agricultural insurance scenario rule engine has a built-in rule base for agricultural insurance underwriting. The rule base is configured with a list of mapping relationships for associated parameters, feature extraction items, and data sources. The extraction module 102 is specifically used for: The planting entity identifier and the insured plot code are extracted as association parameters from the association mapping relationship; The associated parameters are input into the agricultural insurance scenario rule engine. The built-in mapping relationship table is queried through the agricultural insurance scenario rule engine to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

[0066] In one embodiment, the extraction module 102 is further configured to: After extracting the planting entity identifier and the insured land plot code as association parameters from the association mapping relationship, the validity of the core association parameters is verified. The validity verification includes at least completeness verification, uniqueness verification and association verification. The completeness verification verifies that the planting entity identifier and the insured land plot code are not missing or null. The uniqueness verification verifies that the planting entity identifier and the insured land plot code are not duplicated in the underwriting business. The association verification verifies that the planting entity identifier and the insured land plot code are bound together. Accordingly, the verified associated parameters are input into the agricultural insurance scenario rule engine. The agricultural insurance scenario rule engine queries the built-in mapping relationship table to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

[0067] In one embodiment, the device further includes: The verification module is used to perform error correction and verification on the multi-dimensional feature information after the multi-dimensional feature information is extracted by the agricultural insurance scenario rule engine using the association mapping relationship as the trigger condition; The verification module is specifically used for: Using the coordinates of the planting area as a spatial constraint benchmark, the consistency between the associated records of the planting area coordinates and the filing data of the geographical area corresponding to the planting area coordinates is verified based on the spatial constraint benchmark. Using the crop growth stage label corresponding to the crop growth period as a time constraint benchmark, the matching between the collection time of meteorological factor data in the planting area and the crop growth period is verified.

[0068] In one embodiment, the generation module 103 is further configured to: The multi-dimensional feature information is structured and organized to remove redundant and invalid data and supplement the feature semantic description. Based on the risk association weight rules preset in the agricultural insurance underwriting scenario, the expression weight of the multi-dimensional feature information is adjusted, wherein the expression weight of the crop growth stage label and the meteorological factor data of the planting area is higher than the expression weight of the planting area coordinates. According to the multimodal input structure adapted to the agricultural insurance underwriting model, the multidimensional feature information with representation weights is divided into text description modality and structured feature modality. The text description modality is used for model understanding scenarios, and the structured feature modality is used for model reasoning scenarios. The text description modality and structured feature modality are formatted to generate a multimodal prompt field adapted to the agricultural insurance underwriting model.

[0069] In one embodiment, the inference module 104 is specifically used for: The multimodal prompt field and the corrected multidimensional feature information are synchronously input into the agricultural insurance underwriting model. Based on the weight label in the multimodal prompt field, the model’s attention mechanism is used to extract the association features of the multidimensional feature information. The association features include the association features between crop growth stage labels and meteorological factor data of planting area, and the association features between planting area coordinates and past disaster records. Based on the association features of the multi-dimensional feature information, multi-dimensional risk association reasoning is performed in agricultural insurance scenarios to determine crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk. By integrating the assessment results of crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk, a standardized agricultural insurance underwriting risk assessment report is generated.

[0070] This invention provides a processing device for multi-source agricultural insurance underwriting data. It employs multimodal analysis technology to process heterogeneous materials across multiple media, enabling differentiated adaptation to the characteristics of heterogeneous data and improving the accuracy of multi-source agricultural underwriting data. Through a unique association mapping relationship, it achieves linked tracing of heterogeneous data, triggering a rule engine to automatically extract features. This rule engine is specifically configured for agricultural insurance scenarios, extracting only multi-dimensional features strongly correlated with agricultural insurance underwriting risks after triggering. This makes the multi-dimensional feature information more aligned with the needs of agricultural insurance risk assessment. Furthermore, it combines an attention mechanism to guide risk association reasoning, achieving cross-linked reasoning of multi-dimensional features through precise adaptation of multi-dimensional feature information and the model. This effectively uncovers potential risk associations behind the data, improving the accuracy of underwriting risk judgment and making risk assessment more aligned with the actual needs of agricultural insurance underwriting.

[0071] Specific limitations regarding the processing device for multi-source data in agricultural insurance underwriting can be found in the limitations of the intelligent question-answering method described above, and will not be repeated here. Each module in the aforementioned processing device for multi-source data in agricultural insurance underwriting can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0072] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 11 As shown. The computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side method for processing multi-source data in agricultural insurance underwriting.

[0073] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 12 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of a method for processing multi-source data in agricultural insurance underwriting.

[0074] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: In response to the multi-media materials submitted by the user for agricultural insurance underwriting, multimodal parsing technology is used to parse the multi-media materials to obtain structured text information and crop image visual feature information with unique association mapping relationship; Using the aforementioned association mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. Based on the multi-dimensional feature information, generate multimodal prompt fields adapted to the agricultural insurance underwriting model; The multimodal cue fields are input into the agricultural insurance underwriting model, and the attention mechanism of the agricultural insurance underwriting model is used to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields to obtain a risk assessment report for agricultural insurance underwriting.

[0075] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: In response to the multi-media materials submitted by the user for agricultural insurance underwriting, multimodal parsing technology is used to parse the multi-media materials to obtain structured text information and crop image visual feature information with unique association mapping relationship; Using the aforementioned association mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. Based on the multi-dimensional feature information, generate multimodal prompt fields adapted to the agricultural insurance underwriting model; The multimodal cue fields are input into the agricultural insurance underwriting model, and the attention mechanism of the agricultural insurance underwriting model is used to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields to obtain a risk assessment report for agricultural insurance underwriting.

[0076] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0077] Those skilled in the art will understand that all or part of the processes in the methods of 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 of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0078] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0079] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for processing multi-source data in agricultural insurance underwriting, characterized in that, include: In response to the multi-media materials submitted by the user for agricultural insurance underwriting, multimodal parsing technology is used to parse the multi-media materials to obtain structured text information and crop image visual feature information with unique association mapping relationship; Using the aforementioned association mapping relationship as a trigger condition, multi-dimensional feature information is extracted through the agricultural insurance scenario rule engine. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. Based on the multi-dimensional feature information, generate multimodal prompt fields adapted to the agricultural insurance underwriting model; The multimodal cue fields are input into the agricultural insurance underwriting model, and the attention mechanism of the agricultural insurance underwriting model is used to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields to obtain a risk assessment report for agricultural insurance underwriting.

2. The method for processing multi-source data in agricultural insurance underwriting as described in claim 1, characterized in that, The multimodal analysis technique is used to analyze the multi-media material to obtain structured text information and crop image visual feature information with unique correlation mapping relationships, including: Multimodal analysis technology is used to identify text-based and image-based materials in the multi-media materials. Text extraction is performed on the text-based materials to obtain structured text information; and image extraction is performed on the image-based materials to obtain crop image visual feature information. Based on the identifier set in the underwriting business, a unique association mapping relationship is constructed between the structured text information and the crop image visual feature information, resulting in structured text information and crop image visual feature information with a unique association mapping relationship.

3. The method for processing multi-source data in agricultural insurance underwriting as described in claim 1, characterized in that, The agricultural insurance scenario rule engine has a built-in rule base for agricultural insurance underwriting. This rule base is configured with a list of mapping relationships for related parameters, feature extraction items, and data sources. The process uses these mapping relationships as trigger conditions to extract multi-dimensional feature information through the agricultural insurance scenario rule engine, including: The planting entity identifier and the insured plot code are extracted as association parameters from the association mapping relationship; The associated parameters are input into the agricultural insurance scenario rule engine. The built-in mapping relationship table is queried through the agricultural insurance scenario rule engine to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

4. The method for processing multi-source data in agricultural insurance underwriting as described in claim 3, characterized in that, After extracting the planting entity identifier and the insured plot code as association parameters from the association mapping relationship, the method further includes: The validity of the core associated parameters is verified. The validity verification includes at least completeness verification, uniqueness verification and association verification. The completeness verification verifies that the planting entity identifier and the insured plot code are not missing or empty. The uniqueness verification verifies that the planting entity identifier and the insured plot code are not duplicated in the underwriting business. The association verification verifies that the planting entity identifier and the insured plot code are bound together. Accordingly, the verified associated parameters are input into the agricultural insurance scenario rule engine. The agricultural insurance scenario rule engine queries the built-in mapping relationship table to determine the feature extraction items and data sources corresponding to the planting entity identifier and the insured plot code.

5. The method for processing multi-source data for agricultural insurance underwriting as described in claim 1, characterized in that, After using the aforementioned association mapping relationship as a trigger condition to extract multi-dimensional feature information through the agricultural insurance scenario rule engine, the method further includes: The multi-dimensional feature information is corrected and verified using a spatiotemporal error correction mechanism. The step of using a spatiotemporal error correction mechanism to correct and verify the multi-dimensional feature information includes: Using the coordinates of the planting area as a spatial constraint benchmark, the consistency between the associated records of the planting area coordinates and the filing data of the geographical area corresponding to the planting area coordinates is verified based on the spatial constraint benchmark. Using the crop growth stage label corresponding to the crop growth period as a time constraint benchmark, the matching between the collection time of meteorological factor data in the planting area and the crop growth period is verified.

6. The method for processing multi-source data for agricultural insurance underwriting as described in any one of claims 1-5, characterized in that, The process of generating multimodal prompt fields adapted to the agricultural insurance underwriting model based on the multidimensional feature information includes: The multi-dimensional feature information is structured and organized to remove redundant and invalid data and supplement the feature semantic description. Based on the risk association weight rules preset in the agricultural insurance underwriting scenario, the expression weight of the multi-dimensional feature information is adjusted, wherein the expression weight of the crop growth stage label and the meteorological factor data of the planting area is higher than the expression weight of the planting area coordinates. According to the multimodal input structure adapted to the agricultural insurance underwriting model, the multidimensional feature information with representation weights is divided into text description modality and structured feature modality. The text description modality is used for model understanding scenarios, and the structured feature modality is used for model reasoning scenarios. The text description modality and structured feature modality are formatted to generate a multimodal prompt field adapted to the agricultural insurance underwriting model.

7. The method for processing multi-source data for agricultural insurance underwriting as described in any one of claims 1-6, characterized in that, The step of inputting the multimodal cue fields into the agricultural insurance underwriting model, and using the attention mechanism of the agricultural insurance underwriting model to perform risk correlation reasoning on the multidimensional feature information based on the multimodal cue fields, to obtain a risk assessment report for agricultural insurance underwriting, includes: The multimodal prompt field and the corrected multidimensional feature information are synchronously input into the agricultural insurance underwriting model. Based on the weight label in the multimodal prompt field, the model’s attention mechanism is used to extract the association features of the multidimensional feature information. The association features include the association features between crop growth stage labels and meteorological factor data of planting area, and the association features between planting area coordinates and past disaster records. Based on the association features of the multi-dimensional feature information, multi-dimensional risk association reasoning is performed in agricultural insurance scenarios to determine crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk. By integrating the assessment results of crop growth adaptation risk, regional meteorological disaster risk, and planting information matching risk, a standardized agricultural insurance underwriting risk assessment report is generated.

8. A processing device for multi-source data in agricultural insurance underwriting, characterized in that, include: The parsing module is used to respond to the multi-media materials submitted by the user for agricultural insurance underwriting, and to parse the multi-media materials using multimodal parsing technology to obtain structured text information and crop image visual feature information with unique association mapping relationship; The extraction module is used to extract multi-dimensional feature information through the agricultural insurance scenario rule engine, using the association mapping relationship as a trigger condition. The multi-dimensional feature information includes at least the coordinates of the planting area, crop growth stage labels, and meteorological factor data of the planting area. The generation module is used to generate multimodal prompt fields adapted to the agricultural insurance underwriting model based on the multidimensional feature information; The reasoning module is used to input the multimodal cue fields into the agricultural insurance underwriting model, so as to perform risk association reasoning on the multidimensional feature information based on the multimodal cue fields through the attention mechanism of the agricultural insurance underwriting model, and obtain the risk assessment report for agricultural insurance underwriting.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for processing multi-source data for agricultural insurance underwriting as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for processing multi-source data for agricultural insurance underwriting as described in any one of claims 1 to 7.

Images