Agricultural insurance loss assessment method and device, computer equipment and readable storage medium

Abstract

The invention provides an agricultural insurance loss assessment method and device, computer equipment and a readable storage medium, and relates to the field of data processing and financial science and technology. The method comprises the following steps: acquiring agricultural data of different sources of an insured farmland, and extracting feature information of the agricultural data of different sources; determining a feature fusion weight of each piece of agricultural data based on target data, and performing feature fusion on the plurality of pieces of feature information according to the feature fusion weight corresponding to each piece of agricultural data to obtain a first comprehensive feature vector, the target data comprises at least one of disaster scene information and a credibility score of the agricultural data; and inputting the first comprehensive feature vector into a trained loss assessment prediction model, and outputting loss assessment data of the insured farmland, the loss assessment data including a loss assessment amount and disaster-inducing factors.

Description

Technical Field

[0001] This application relates to the fields of data processing and financial technology, and in particular to an agricultural insurance loss assessment method, apparatus, computer equipment, and computer-readable storage medium. Background Technology

[0002] Agricultural insurance, as an important means of mitigating risks in agricultural production and operation, plays a vital role in promoting modern agricultural development and rural industrial revitalization. However, the current agricultural insurance loss assessment process faces numerous challenges. Traditional loss assessment methods mainly rely on manual on-site inspections, which not only consumes a significant amount of manpower, resources, and time, but also makes the assessment results susceptible to subjective factors, compromising accuracy and impartiality. Furthermore, with the continuous expansion of agricultural production scale and the increasing complexity of business models, a single data source can no longer meet the demand for accurate loss assessment. Summary of the Invention

[0003] In view of this, this application provides an agricultural insurance loss assessment method, apparatus, computer equipment, and computer-readable storage medium, which solves the problem of poor loss assessment accuracy in related technologies.

[0004] In a first aspect, embodiments of this application provide a method for assessing losses in agricultural insurance, including: Acquire agricultural data from different sources for insured farmland, and extract feature information from the agricultural data from different sources; Based on the target data, the feature fusion weights of each of the agricultural data are determined, and the feature information of multiple data is fused according to the feature fusion weights corresponding to each of the agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data. The first comprehensive feature vector is input into the trained loss assessment prediction model, and the loss assessment data for the insured farmland is output. The loss assessment data includes the loss amount and the disaster-causing factors.

[0005] The method described in the embodiments of this application may also have the following additional technical features: Optionally, in the above technical solution, the method further includes: On a geographic information system map, insured farmland with different degrees of disaster damage is represented by different visual variables; In response to a request to view the target insured farmland, the loss assessment data of the target insured farmland is displayed.

[0006] In any of the above technical solutions, optionally, agricultural data from different sources of the insured farmland is obtained, including: The satellite remote sensing image data of the insured farmland is obtained, the land ownership confirmation data of the insured farmland is obtained from the land ownership confirmation database, and the pixel units of the obtained satellite remote sensing image data are semantically tagged and bound to the land ownership confirmation data to generate the association relationship between the pixel units of the satellite remote sensing image data and the land ownership confirmation data. The system acquires IoT sensor data of the insured farmland. When the IoT sensor data exceeds the disaster sensitivity threshold, it pushes a data supplementation prompt to the terminal device of the farmer corresponding to the insured farmland, acquires the supplementation data provided by the terminal device, and acquires historical satellite remote sensing image data and historical meteorological data of the insured farmland for the same period. Obtain meteorological data and farmer declaration data for the insured farmland; The meteorological data, the farmer reporting data, the satellite remote sensing image data, the correlation data, the IoT sensor data, the historical satellite remote sensing image data, and the historical meteorological data are used as the agricultural data.

[0007] In any of the above technical solutions, optionally, after obtaining agricultural data from different sources of insured farmland, the method further includes: The agricultural data from different sources are preprocessed separately, and the data preprocessing includes at least one of the following: data cleaning, data format conversion, data semantic normalization, and data spatiotemporal alignment. Specifically, when performing spatiotemporal alignment of the data, the precision of the spatiotemporal grid is adjusted, increasing the spatiotemporal grid during non-disaster periods and decreasing it during disaster periods; when performing semantic normalization of the data, for environmental data in the agricultural data, linear normalization is used to map the values ​​to the [0,1] interval, and semantic weights are set according to the disaster type; for disaster data in the agricultural data, interval mapping is used to map the values ​​to the [-1,1] interval, and semantic weights are set according to the crop growth period.

[0008] In any of the above technical solutions, optionally, determining the feature fusion weights of each of the agricultural data based on the target data includes: The agricultural data is evaluated for credibility, and feature fusion weights are determined based on the credibility scores, wherein the credibility scores are directly proportional to the feature fusion weights; or, The determination of feature fusion weights for each of the agricultural data based on the target data includes: The disaster scenario is determined based on the disaster scenario information, and the feature fusion weight of the agricultural data is determined based on the disaster scenario, wherein the feature fusion weight of the agricultural data is different under different disaster scenarios.

[0009] Optionally, in any of the above technical solutions, the method further includes: Obtain agricultural sample data from different sources for insured farmland; Feature information is extracted from agricultural sample data from different sources, and multiple feature information is fused to obtain a second comprehensive feature vector; Loss assessment is performed separately using different agricultural sample data to obtain multiple sub-loss assessment conclusions, and these multiple sub-loss assessment conclusions are then merged to obtain the loss assessment conclusion data. The deep learning model is trained based on the second comprehensive feature vector and the damage assessment conclusion data to generate the damage assessment prediction model. During training, the deep learning model is trained by freezing the bottom feature extraction layer and fine-tuning the top decision layer. A damage assessment causal analysis submodule is set in the deep learning model to identify and output the disaster-causing factors.

[0010] In any of the above technical solutions, optionally, loss assessment is performed separately using different agricultural sample data to obtain multiple sub-loss assessment conclusions, and the multiple sub-loss assessment conclusions are then fused to obtain loss assessment conclusion data, including: Damage assessment is conducted using satellite remote sensing imagery data to evaluate the affected area and extent of damage to farmland. Damage assessment is conducted using meteorological data to predict the impact of weather on crop yields; Damage assessment is conducted using IoT sensor data to determine the constraints of the farmland environment on crop growth. The data reported by farmers is used to assess the damage, determine the affected area, the extent of the damage, and the estimated losses of farmland; The data for loss assessment are obtained by integrating information on the affected area and severity of farmland, the impact of weather on crop yield, the constraints of farmland environment on crop growth, estimated losses, and insurance parameters.

[0011] Secondly, embodiments of this application provide an agricultural insurance loss assessment device, comprising: The first data processing module is used to acquire agricultural data from different sources of insured farmland and extract feature information from the agricultural data from different sources. The second data processing module is used to determine the feature fusion weight of each of the agricultural data based on the target data, and to perform feature fusion on multiple feature information according to the feature fusion weight corresponding to each of the agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data. The loss assessment prediction module is used to input the first comprehensive feature vector into the trained loss assessment prediction model and output loss assessment data for the insured farmland. The loss assessment data includes the loss amount and the disaster-causing factors.

[0012] Thirdly, embodiments of this application provide a computer device including a processor and a memory, the memory storing programs or instructions that can run on the processor, the programs or instructions implementing the steps of the method as described in the first aspect when executed by the processor.

[0013] Fourthly, embodiments of this application provide a computer-readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method as described in the first aspect.

[0014] This application's agricultural insurance loss assessment method, apparatus, computer equipment, and computer-readable storage medium acquire agricultural data from different sources of insured farmland and extract feature information from the agricultural data from different sources; determine feature fusion weights for each of the agricultural data based on target data, and perform feature fusion on multiple feature information according to the feature fusion weights corresponding to each of the agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data; input the first comprehensive feature vector into a trained loss assessment prediction model, and output loss assessment data for the insured farmland, including the loss amount and disaster-causing factors. By fusing multi-source heterogeneous data, the advantages of different data sources are fully utilized to comprehensively and accurately reflect the disaster situation of crops, reduce errors caused by a single data source, and improve the accuracy and reliability of loss assessment results. In addition, during fusion, the fusion weights can be dynamically adapted according to data quality and disaster scenario to improve fusion accuracy. Furthermore, this application employs automated data collection and intelligent damage assessment models, reducing the workload of manual on-site investigation and calculation, shortening the damage assessment cycle, improving claims efficiency, and reducing the costs of manual investigation and claims disputes caused by inaccurate damage assessment through accurate risk assessment and pricing.

[0015] 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, specific embodiments of this application are given below. Attached Figure Description

[0016] 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 schematic diagram illustrating the application environment of the agricultural insurance loss assessment method according to an embodiment of this application is shown; Figure 2 A flowchart illustrating the agricultural insurance loss assessment method according to an embodiment of this application is shown; Figure 3 A structural block diagram of an agricultural insurance loss assessment device according to an embodiment of this application is shown; Figure 4 One of the structural block diagrams of a computer device according to an embodiment of this application is shown; Figure 5 A second structural block diagram of a computer device according to an embodiment of this application is shown. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0018] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0019] The agricultural insurance loss assessment method, apparatus, computer equipment, and computer-readable storage medium provided in this application will be described in detail below with reference to the accompanying drawings and through specific embodiments and application scenarios. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0020] The agricultural insurance loss assessment method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The server acquires agricultural data from different sources of insured farmland and extracts feature information from the agricultural data from different sources; based on the target data, it determines the feature fusion weights of each agricultural data, and performs feature fusion on multiple feature information according to the feature fusion weights corresponding to each agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data; the first comprehensive feature vector is input into a trained loss assessment prediction model, and the model outputs loss assessment data for the insured farmland, including the loss assessment amount and the disaster-causing factor.

[0021] By integrating multi-source heterogeneous data and fully leveraging the advantages of different data sources, this application comprehensively and accurately reflects the damage situation of crops, reduces errors caused by a single data source, and improves the accuracy and reliability of damage assessment results. Furthermore, this application employs automated data collection and an intelligent damage assessment model, reducing the workload of manual on-site investigation and calculation, shortening the damage assessment cycle, improving claims efficiency, and reducing the costs of manual investigation and claims disputes caused by inaccurate damage assessment through precise risk assessment and pricing.

[0022] 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 now be described in detail through specific embodiments.

[0023] This application provides a method for assessing losses in agricultural insurance, such as... Figure 2 As shown, the method includes: Step 201: Obtain agricultural data from different sources for the insured farmland, and extract feature information from the agricultural data from different sources.

[0024] In this step, agricultural data from different sources is obtained for the insured farmland. In one embodiment of this application, the agricultural data from different sources includes at least two of the following: satellite remote sensing image data, meteorological data, Internet of Things sensor data, and farmer declaration data.

[0025] Satellite remote sensing imagery data refers to periodically acquired remote sensing images of insured farmland using high-resolution satellite remote sensing technology. By analyzing spectral information across different bands, key information such as crop planting area, growth status, and pest and disease conditions is extracted. For example, near-infrared bands are used to monitor crop health, and the Normalized Difference Vegetation Index (NDVI) is used to assess crop growth vitality. A data interface is established with meteorological departments to obtain real-time meteorological data for the insured area, including temperature, precipitation, sunlight, wind speed, and humidity. Historical meteorological data is also collected to analyze meteorological trends and their long-term impact on crop growth. Various IoT sensors are deployed in the farmland, such as soil moisture sensors, soil fertility sensors, and pest and disease monitoring sensors, to collect real-time data on farmland environmental parameters and crop growth status. Sensor data is transmitted to a data processing center via wireless communication technology. A mobile application is developed for farmers, allowing them to report damage to their crops by taking photos or providing text descriptions, including the time, location, type of damage, and estimated losses. Meanwhile, farmers can upload relevant supporting documents, such as photos and videos of the disaster site. This application uses multiple data sources as the basis for damage assessment, improving the accuracy of the assessment.

[0026] In one embodiment of this application, obtaining agricultural data from different sources for insured farmland includes: The system acquires satellite remote sensing image data of insured farmland, retrieves land ownership confirmation data of insured farmland from the land ownership confirmation database, and binds the pixel units of the acquired satellite remote sensing image data with the land ownership confirmation data using semantic tags to generate the association relationship between the pixel units of the satellite remote sensing image data and the land ownership confirmation data. The system acquires IoT sensor data of insured farmland. When the IoT sensor data exceeds the disaster sensitivity threshold, it pushes a data supplementation prompt to the terminal device of the farmer corresponding to the insured farmland, and acquires the supplementation data provided by the terminal device, as well as historical satellite remote sensing image data and historical meteorological data of the insured farmland for the same period. Obtain meteorological data for insured farmland and data submitted by farmers; Meteorological data, farmer-reported data, satellite remote sensing image data, correlation data, IoT sensor data, historical satellite remote sensing image data, and historical meteorological data are used as agricultural data.

[0027] In this embodiment, considering that the multi-source data is collected in isolation, satellite remote sensing image data can only obtain macroscopic information about crops and cannot be associated with the ownership of specific farmers' plots; IoT sensors can only passively upload environmental parameters, which are disconnected from farmers' on-site disaster feedback, resulting in data that is "data without association and information without context". Subsequent fusion requires a lot of manual verification, which is extremely inefficient.

[0028] Therefore, this application employs semantic tag embedded acquisition and exception-triggered collaborative acquisition, specifically as follows: (1) Semantic Tag Embedded Acquisition: In the satellite remote sensing image data acquisition stage, by connecting with the "land parcel ownership database" of the agricultural and rural departments, the pixel units of each satellite remote sensing image data are automatically bound to the semantic tags of the land parcel ownership data. The land parcel ownership data is "land parcel code-farmer's name-crop variety-growth period-insurance information". After binding, a one-to-one correspondence is formed between remote sensing pixels, physical land parcels, and ownership information, eliminating the need for manual secondary matching and solving the problem of disconnection between remote sensing data and land parcel ownership. For example, if a farmer insures 10 mu of paddy fields, the pixel area of ​​the 10 mu land parcel can be directly identified during satellite remote sensing acquisition, and the tag "farmer Wang Mou-rice-booting period-policy number" can be automatically marked.

[0029] (2) Anomaly-triggered collaborative data collection: When deploying IoT sensors, disaster sensitivity thresholds are pre-set. For example, if soil moisture is <15%, drought warnings are triggered, and if pest sensors capture spectral signals of specific pests, pest warnings are triggered. When sensor monitoring data exceeds the disaster sensitivity threshold, a "data supplementation task" prompt is automatically pushed to the terminal device APP of the corresponding farmers. The data supplementation task includes taking close-up photos of the affected parts of the crop (which must include the plot sign), filling in the field management records of the past few days (e.g., 3 days) (e.g., whether fertilization or irrigation was carried out), and estimating the affected area, etc., with structured options. At the same time, historical satellite remote sensing image data and historical meteorological data of the same period in history are automatically retrieved for the plot, forming a collaborative data collection closed loop of "sensor warning - farmer on-site verification - historical data corroboration", which solves the defects of passive data collection and lack of scene details.

[0030] Finally, meteorological data, farmer reporting data, satellite remote sensing image data, correlation data, IoT sensor data, and if historical satellite remote sensing image data and historical meteorological data are available, will be added to the satellite remote sensing image data and meteorological data as agricultural data to achieve data correlation.

[0031] After acquiring agricultural data from different sources of insured farmland, the process further involves feature extraction from this data to obtain feature information for each piece of agricultural data. In one embodiment of this application, extracting feature information from agricultural data from different sources includes: Extracting texture and shape features of crops from satellite remote sensing image data; Extracting the changing characteristics of meteorological elements from meteorological data; Extracting fluctuation characteristics of farmland environmental parameters from IoT sensor data; Extract disaster-related characteristics from the data submitted by farmers.

[0032] In this embodiment, feature information from multiple sources is extracted, such as extracting texture and shape features of crops from satellite remote sensing image data, extracting change features of meteorological elements from meteorological data, extracting fluctuation features of farmland environmental parameters from IoT sensor data, and extracting disaster reporting features from farmer reporting data.

[0033] In one embodiment of this application, after obtaining agricultural data from different sources on the insured farmland, the method further includes: Agricultural data from different sources are preprocessed separately. Data preprocessing includes at least one of the following: data cleaning, data format conversion, data semantic normalization, and data spatiotemporal alignment. Specifically, when performing spatiotemporal alignment of data, the precision of the spatiotemporal grid is adjusted. The spatiotemporal grid is increased during non-disaster periods and decreased during disaster periods. During non-disaster periods, the spatiotemporal grid updates data according to the first cycle, and during disaster periods, the spatiotemporal grid updates data according to the second cycle, with the first cycle being longer than the second cycle. When performing semantic normalization of data, for environmental data in agricultural data, linear normalization is used to map the values ​​to the [0,1] interval, and semantic weights are set according to the disaster type. For disaster data in agricultural data, interval mapping is used to map the values ​​to the [-1,1] interval, and semantic weights are set according to the crop growth period.

[0034] In this embodiment, the agricultural data from different sources are sequentially preprocessed, including data cleaning, data format conversion, data normalization, and data spatiotemporal alignment.

[0035] For data cleaning, this involves removing noisy, duplicate, and erroneous data from the collected multi-source data. For example, cloud noise in satellite remote sensing image data is removed using threshold segmentation and morphological processing methods; outliers in meteorological data are identified and corrected using statistical methods.

[0036] For data format conversion: Multi-source data in different formats are uniformly converted into a standard format that can be recognized and processed. For example, satellite remote sensing image data is converted from the original TIFF format to HDF5 format for easier data storage and retrieval; meteorological data is converted from text format to structured database table format.

[0037] For data normalization: Normalization is performed on multi-source data with different dimensions and value ranges to make the data comparable. For example, for soil moisture data and temperature data, the value range is unified to the [0,1] interval by using linear normalization methods.

[0038] For spatiotemporal alignment of data: Based on the temporal and spatial attributes of the data, multi-source data are spatiotemporally aligned. For example, satellite remote sensing image data and meteorological data are aligned at certain time intervals (e.g., daily, weekly) and spatial resolutions (e.g., 100m × 100m grids); for IoT sensor data and farmer reporting data, they are matched with the corresponding spatiotemporal grids using geolocation information.

[0039] In this embodiment of the application, the acquired data is preprocessed to obtain more standardized data, and then the characteristic information of agricultural data is extracted to ensure the accuracy of the final loss assessment data of the insured farmland.

[0040] It is worth noting that some data preprocessing can only achieve "basic data cleaning + format conversion," resulting in low spatiotemporal alignment accuracy (mostly at the township level spatial granularity and monthly temporal granularity). Furthermore, different types of data (such as soil moisture percentage, air temperature in degrees Celsius, and remote sensing NDVI index) have large differences in dimensions and lack semantic correlation, leading to spatiotemporal misalignment and semantic conflicts during subsequent fusion, resulting in distorted fusion results. This application proposes a deep alignment mechanism of "dynamic spatiotemporal grid + semantic normalization": (1) Dynamic spatiotemporal grid construction: Breaking the limitations of the existing "fixed spatiotemporal granularity", the spatiotemporal grid accuracy is dynamically adjusted according to the disaster type and data characteristics: 1) Spatial Dimension: During non-disaster periods, a 100m x 100m grid is used, while during disaster periods (such as after heavy rain or typhoons), it automatically switches to a 10m x 100m grid to ensure accurate location of affected plots. At the same time, GIS technology is used to accurately map the installation locations of IoT sensors and the latitude and longitude coordinates of the plots reported by farmers to the corresponding grid cells, achieving complete spatial alignment of data from "satellite remote sensing - meteorology - sensors - farmer reports".

[0041] 2) Time dimension: Data is updated weekly during non-disaster periods, daily during disaster warning periods, and every 6 hours after a disaster occurs, ensuring that the data time granularity matches the loss assessment requirements and solving the problem of information lag or redundancy caused by fixed spatiotemporal granularity in existing technologies.

[0042] (2) Semantic normalization processing: Breaking through the limitations of the existing "single numerical normalization", a three-dimensional normalization model of "data type-semantic weight-numerical interval" is established: 1) For environmental data, such as soil moisture and temperature: linear normalization is used to map the values ​​to the [0,1] interval. At the same time, semantic weights are assigned according to the disaster type. For example, during drought, the semantic weight of soil moisture is set to 0.4 and the semantic weight of temperature is set to 0.2; during pests and diseases, the semantic weight of soil moisture is set to 0.2 and the semantic weight of temperature is set to 0.1.

[0043] 2) For disaster-related data, such as affected area and NDVI index: use interval mapping to map the values ​​to the interval [-1,1], where -1 represents complete crop failure and 1 represents no disaster. Also, set semantic weights by associating the crop growth period coefficients. For example, the disaster correction coefficient for rice during the booting stage is 1.2 and that during the grain-filling stage is 1.1, which solves the problem of loss assessment bias caused by the difference in disaster sensitivity of different crop growth stages.

[0044] Step 202: Determine the feature fusion weights of each of the agricultural data based on the target data, and perform feature fusion on multiple feature information according to the feature fusion weights corresponding to each of the agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data.

[0045] In this step, based on at least one of the disaster scenario information and the credibility score of agricultural data, the feature fusion weight of each agricultural data point is determined. The feature fusion weight refers to the weight calculated for the corresponding data during subsequent data fusion. Further, according to the feature fusion weight, the extracted feature information is fused to construct a first comprehensive feature vector.

[0046] In one embodiment, after determining the feature fusion weights of the satellite remote sensing image data, the features obtained based on the satellite remote sensing image data are calculated according to the feature fusion weights of the satellite remote sensing image data during feature fusion.

[0047] In one embodiment of this application, determining the feature fusion weights of various agricultural data based on target data includes: The credibility score of agricultural data is evaluated, and the feature fusion weights are determined based on the credibility score, where the credibility score is directly proportional to the feature fusion weights; or, Based on the target data, the feature fusion weights of each agricultural data point are determined, including: The disaster scenario is determined based on disaster scenario information, and the feature fusion weight of agricultural data is determined based on the disaster scenario. The feature fusion weight of agricultural data is different under different disaster scenarios.

[0048] In this embodiment, considering that using "single feature-level fusion" (such as performing PCA dimensionality reduction and stitching only on remote sensing and meteorological data) or "fixed-weight decision-level fusion" (such as fusing data with a weight of 40% for remote sensing, 30% for meteorological data, and 30% for sensor data regardless of disaster type), the fusion strategy cannot be dynamically adapted according to data quality and disaster scenario, resulting in a sharp drop in the accuracy of the fusion results under complex scenarios such as cloud cover and sensor failure in satellite remote sensing. Specifically, single feature-level fusion, such as performing PCA (Principal Component Analysis) dimensionality reduction and stitching only on remote sensing and meteorological data, and fixed-weight decision-level fusion, such as fusing data with a weight of 30% for satellite remote sensing image data, 30% for meteorological data, 30% for IoT sensor data, and 10% for farmer reporting data regardless of disaster type, are problematic.

[0049] Therefore, this application proposes to determine the fusion weights based on the credibility score of agricultural data and disaster scenario information, thereby enabling subsequent dynamic feature fusion: (1) Feature-level dynamic fusion, that is, determining the fusion weight based on the credibility score of agricultural data: extracting the "scenario-based features" of multi-source data and adjusting the feature contribution in real time according to the data quality.

[0050] First, a data quality assessment is conducted: a data source credibility scoring model is constructed, using a three-dimensional index of data integrity × 0.3 + data timeliness × 0.4 + data consistency × 0.3 to obtain the credibility score. Each item can be set to a maximum score of 100 points, with data integrity having a weight of 0.3, data timeliness having a weight of 0.4, and data consistency having a weight of 0.3. For example, when satellite remote sensing is obstructed by clouds, the integrity score drops from 90 to 40, and the credibility score decreases accordingly.

[0051] Then, feature contribution adjustments are made: feature contribution is dynamically allocated based on the credibility score, which is also the feature fusion weight. For example, in a drought scenario: if the credibility score of meteorological data is 90 (complete precipitation data), the credibility score of IoT sensor data is 85 (continuous soil moisture record), the credibility score of satellite remote sensing image data is 60 (partial cloud obscuration), and the credibility score of farmer-reported data is 80, then the feature contribution of meteorological data is set to 0.4, the feature contribution of IoT sensor data is set to 0.3, the feature contribution of satellite remote sensing image data is set to 0.2, and the feature contribution of farmer-reported data is set to 0.1.

[0052] In this application scenario, when performing feature fusion, an "improved PCA dimensionality reduction algorithm" can be used to retain detailed information of high-contribution features during the dimensionality reduction process, avoiding feature loss caused by excessive dimensionality reduction in traditional PCA.

[0053] (2) Decision-level dynamic fusion, that is, determining the fusion weight based on disaster scenario information: generating “independent loss assessment decision” based on different data sources and using “Dempster-Shafer evidence theory + scenario adaptation rules” for fusion.

[0054] First, independent damage assessment decisions are generated: the "affected area and degree" are assessed using satellite remote sensing image data, the "yield loss rate" is predicted using meteorological data, the "impact of environmental disaster factors" is determined using IoT sensor data, and the "on-site disaster details are corrected" using data reported by farmers, thus forming four independent damage assessment decisions.

[0055] Then, dynamic evidence fusion is performed: different feature fusion weights are set according to the type of disaster, for example: 1) Drought scenario: The feature fusion weight of meteorological data is set to 0.4, the feature fusion weight of IoT sensor data is set to 0.3, the feature fusion weight of satellite remote sensing image data is set to 0.2, and the feature fusion weight of farmer reporting data is set to 0.1.

[0056] 2) Pest and disease scenario: The feature fusion weight of IoT sensor data is set to 0.45, the feature fusion weight of farmer reporting data is set to 0.3, the feature fusion weight of satellite remote sensing image data is set to 0.2, and the feature fusion weight of meteorological data is set to 0.05.

[0057] In this application scenario, when performing feature fusion, the Dempster-Shafer evidence theory is used to handle data conflicts. For example, if the remote sensing assessment indicates that the affected area is 10 mu, and the farmer reports 8 mu, the final loss assessment decision is obtained through "conflict coefficient calculation - evidence correction - synthesis rules", which solves the problem that the fixed weight in the existing technology cannot handle data conflicts.

[0058] Step 203: Input the first comprehensive feature vector into the trained loss assessment prediction model and output the loss assessment data for the insured farmland. The loss assessment data includes the loss amount and the disaster-causing factors.

[0059] In this step, the first comprehensive feature vector is used as an input parameter and input into the pre-trained loss assessment prediction model to obtain the loss assessment data for the insured farmland. The loss assessment data includes at least the loss amount and the disaster-causing factor. In other embodiments, the loss assessment data may also include at least one of the following: loss rate, disaster level, and insurance parameters, including the corresponding disaster liability, deductible rules, and other clauses.

[0060] This application embodiment, by fusing multi-source heterogeneous data, fully leverages the advantages of different data sources to comprehensively and accurately reflect the disaster situation of crops, reducing errors caused by a single data source and improving the accuracy and reliability of damage assessment results. Furthermore, during fusion, fusion weights can be dynamically adapted based on data quality and disaster scenario to improve fusion accuracy. Moreover, this application employs automated data collection and an intelligent damage assessment model, reducing the workload of manual on-site investigation and calculation, shortening the damage assessment cycle, improving claims efficiency, and reducing manual investigation costs and claims dispute costs due to inaccurate damage assessment through precise risk assessment and pricing. When outputting damage assessment data, both the disaster-causing factors and the damage assessment amount are output together. The disaster-causing factors are used for damage assessment interpretation, solving the problem of uninterpreted damage assessment results in related technologies.

[0061] Furthermore, this application utilizes an intelligent loss assessment model to monitor and analyze multi-source data in real time. The output of the disaster level and the estimated potential loss rate can help identify potential risks in agricultural production in advance, such as weather disaster warnings and pest and disease outbreak warnings. This provides risk prevention advice to insurance institutions and farmers, thereby reducing disaster losses.

[0062] In one embodiment of this application, the method further includes: On a geographic information system map, insured farmland with different degrees of disaster damage is represented by different visual variables; In response to a request to view the target insured farmland, the loss assessment data for the target insured farmland is displayed.

[0063] In this embodiment, a damage assessment data visualization function is provided. Geographic Information System (GIS) technology is used to display the damage assessment data on a map in a visual manner.

[0064] In one implementation, different visual variables, such as different colors and symbols, are used on the GIS map to represent insured farmland areas with different degrees of damage. For example, different colors are used to mark affected plots: red indicates severe damage (yield reduction > 50%), yellow indicates moderate damage (yield reduction 20%-50%), and green indicates minor damage (yield reduction < 20%). In one embodiment, the GIS map supports zooming and panning.

[0065] Furthermore, if a user clicks on a target insured farmland area on the map, in response to a request to view the target insured farmland, detailed information about the target insured farmland is displayed, including at least loss assessment data.

[0066] This application provides a data visualization function for damage assessment, allowing users to intuitively view the disaster situation and damage assessment results in different areas, which facilitates claims management and risk assessment by insurance institutions.

[0067] In one embodiment, the detailed information of the insured farmland may also include: multi-source data comparison charts, causal analysis charts, loss assessment calculation processes, etc., so that farmers and insurance institutions can intuitively understand the basis for loss assessment. Among them, multi-source data comparison charts include comparisons of remote sensing images before and after the disaster, time-series curves of sensor data, photos reported by farmers, etc.; causal analysis charts include pie charts of the proportion of disaster-causing factors, etc.; and loss assessment calculation processes include feature fusion weights, decision synthesis results, etc.

[0068] In one embodiment of this application, the method further includes: in response to an objection request, displaying dispute retrospective data, the dispute retrospective data including raw data from multi-source agricultural data, credibility scores, weight adjustment records, and model calculation logs, to ensure that the focus of the dispute is traceable.

[0069] In one embodiment of this application, the method further includes: responding to an objection request, obtaining preliminary results of manual investigation uploaded by an insurance specialist, performing loss assessment calculation based on manually set parameters corresponding to the preliminary results of manual investigation, obtaining loss assessment results, comparing the loss assessment results with loss assessment data output by a loss assessment prediction model, and generating a comparison report.

[0070] For example, the preliminary results of manual investigation are: the affected area is 9 mu (approximately 0.67 hectares) and the yield reduction rate is 35%. Based on the same multi-source data, two loss assessments are performed using "manually set parameters" (such as the manually determined affected area and yield reduction rate) and "system intelligent parameters" respectively, generating a comparison report. For example, the loss assessment using manual parameters is 4,500 yuan, and the loss assessment using system parameters is 5,000 yuan, with the difference being a deviation of 1 mu (approximately 0.67 hectares) in the determination of the affected area. This helps the disputing parties reach a consensus quickly and shortens the dispute resolution cycle.

[0071] In one embodiment of this application, the method further includes: constructing a loss assessment prediction model.

[0072] In one embodiment, the method of constructing a loss assessment prediction model includes: Obtain agricultural sample data from different sources for insured farmland; Feature information is extracted from agricultural sample data from different sources, and multiple feature information is fused to obtain a second comprehensive feature vector; Loss assessment is performed separately using different agricultural sample data to obtain multiple sub-loss assessment conclusions, and these multiple sub-loss assessment conclusions are then merged to obtain the loss assessment conclusion data. The deep learning model is trained based on the second comprehensive feature vector and the damage assessment conclusion data to generate the damage assessment prediction model. During training, the deep learning model is trained by freezing the bottom feature extraction layer and fine-tuning the top decision layer. A damage assessment causal analysis submodule is set in the deep learning model to identify and output the disaster-causing factors.

[0073] In this embodiment, agricultural sample data from different sources of insured farmland is acquired. The agricultural sample data includes at least two of the following: satellite remote sensing image data, meteorological data, IoT sensor data, and farmer declaration data. The acquired agricultural sample data undergoes data preprocessing, including data cleaning, data format conversion, data normalization, and data spatiotemporal alignment.

[0074] Feature information is extracted from multi-source agricultural sample data. For example, texture and shape features of crops are extracted from satellite remote sensing imagery data, variation features of meteorological elements are extracted from meteorological data, fluctuation features of farmland environmental parameters are extracted from IoT sensor data, and disaster reporting features are extracted from farmer reporting data. These feature information are then fused to construct a second comprehensive feature vector.

[0075] Furthermore, loss assessment decisions are made separately using different data sources, and then these decision results are fused. In one embodiment of this application, loss assessment is performed separately using different agricultural sample data to obtain multiple sub-loss assessment conclusions, and these multiple sub-loss assessment conclusions are fused to obtain loss assessment conclusion data, including: loss assessment using satellite remote sensing image data to assess the affected area and degree of damage to farmland; loss assessment using meteorological data to predict the impact of meteorology on crop yield, such as yield loss rate; loss assessment using IoT sensor data to determine the constraints of farmland environment on crop growth; loss assessment using farmer-reported data to determine the affected area, degree of damage, and estimated loss of farmland; and fusion of the affected area and degree of damage to farmland, the impact of meteorology on crop yield, the constraints of farmland environment on crop growth, estimated loss, and insurance parameters to obtain loss assessment conclusion data. In this embodiment, damage assessment is performed using satellite remote sensing image data, resulting in sub-assessment conclusions regarding the affected area and severity of damage to farmland; damage assessment is performed using meteorological data, resulting in sub-assessment conclusions regarding the impact of meteorology on crop yield; damage assessment is performed using IoT sensor data, resulting in sub-assessment conclusions regarding the constraints of the farmland environment on crop growth; and damage assessment is performed using data reported by farmers, resulting in sub-assessment conclusions regarding the affected area, severity of damage, and estimated losses to farmland.

[0076] Finally, these sub-assessment conclusions are integrated using PCA, weighted voting, or Dempster-Shafer evidence theory to obtain the final assessment conclusion data. The assessment conclusion data includes the loss rate, disaster level, insurance parameters, and the assessment amount derived from these parameters.

[0077] Deep learning models, such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and their variants, Long Short-Term Memory Networks (LSTMs), are constructed. These models are trained using the second comprehensive feature vector and damage assessment data, enabling them to learn the complex mapping relationship between multi-source data and crop damage losses. For model optimization and evaluation, methods such as cross-validation and regularization are employed to optimize the deep learning model, preventing overfitting and improving its generalization ability. Simultaneously, metrics such as accuracy, recall, and mean squared error (MSE) are used to evaluate model performance. By continuously adjusting model parameters and structure, the model is optimized to achieve the best damage assessment results.

[0078] It is worth noting that the loss assessment models in related technologies are mostly "static deep learning models" that rely on a large amount of local historical data. Newly insured areas have poor generalization ability due to the lack of historical data. Alternatively, the models are "black box models" that only output the loss amount and cannot explain the cause of the loss, leading to frequent disputes between farmers and insurance institutions.

[0079] This application proposes a model based on transfer learning and causal inference: (1) Cross-regional transfer learning training: to solve the problem of data scarcity in new regions.

[0080] 1) Model initialization: Use historical model parameters from regions with "same crop type + similar climate zone" as initial parameters. For example, use the model parameters from the rice planting area in the middle and lower reaches of the Yangtze River as initial parameters to reduce the "cold start" cost of training new regional models.

[0081] 2) Incremental fine-tuning: Incremental training is performed using a small amount of data from the new region. By freezing the bottom feature extraction layer and fine-tuning the top decision layer, the model can quickly adapt to the soil and climate characteristics of the new region while retaining the generalization ability of the original model. The accuracy of the new region model is improved by more than 35% compared with the existing technology.

[0082] (2) Causal inference and explanatory function: embed the loss assessment causal analysis sub-module into the model to achieve dual output of "result + cause".

[0083] 1) Causal factor identification: By analyzing the causal relationship between multi-source data and losses through Bayesian network analysis, disaster-causing factors such as insufficient rainfall, pests and diseases, and soil fertility can be identified.

[0084] 2) Loss Attribution Output: The output includes both the assessed loss amount and the causative factors. In one embodiment, the assessed loss amount and causative factors are used together to generate a causal analysis report. For example, the causal analysis report might read: "Assessment loss for a certain plot of land: 5000 yuan (40% yield reduction). Insufficient rainfall contributes 55% (rainfall in the past 30 days was 60% lower than the historical average for the same period), pests and diseases contribute 30% (sensors detected a rice planthopper density of 15 per plant), and insufficient soil fertility contributes 15% (soil nitrogen content was 25 mg / kg, lower than the normal level of 35 mg / kg)." Furthermore, the original agricultural data can be linked, thus solving the problem of unexplained loss assessment results in related technologies.

[0085] As a specific implementation of the above-mentioned agricultural insurance loss assessment method, this application provides an agricultural insurance loss assessment device. For example... Figure 3 As shown, the agricultural insurance loss assessment device 300 includes: a first data processing module 301, a second data processing module 302, and a loss assessment prediction module 303.

[0086] The first data processing module 301 is used to acquire agricultural data from different sources of insured farmland and extract feature information from the agricultural data from different sources. The second data processing module 302 is used to determine the feature fusion weight of each of the agricultural data based on the target data, and to perform feature fusion on multiple feature information according to the feature fusion weight corresponding to each of the agricultural data to obtain a first comprehensive feature vector. The target data includes at least one of disaster scenario information and the credibility score of the agricultural data. The loss assessment prediction module 303 is used to input the first comprehensive feature vector into the trained loss assessment prediction model and output loss assessment data for the insured farmland, wherein the loss assessment data includes the loss amount and the disaster-causing factors.

[0087] Furthermore, the device also includes: a display module, used for: On a geographic information system map, insured farmland with different degrees of disaster damage is represented by different visual variables; In response to a request to view the target insured farmland, the loss assessment data of the target insured farmland is displayed.

[0088] Furthermore, the first data processing module 301 is specifically used for: The satellite remote sensing image data of the insured farmland is obtained, the land ownership confirmation data of the insured farmland is obtained from the land ownership confirmation database, and the pixel units of the obtained satellite remote sensing image data are semantically tagged and bound to the land ownership confirmation data to generate the association relationship between the pixel units of the satellite remote sensing image data and the land ownership confirmation data. The system acquires IoT sensor data of the insured farmland. When the IoT sensor data exceeds the disaster sensitivity threshold, it pushes a data supplementation prompt to the terminal device of the farmer corresponding to the insured farmland, acquires the supplementation data provided by the terminal device, and acquires historical satellite remote sensing image data and historical meteorological data of the insured farmland for the same period. Obtain meteorological data and farmer declaration data for the insured farmland; The meteorological data, the farmer reporting data, the satellite remote sensing image data, the correlation data, the IoT sensor data, the historical satellite remote sensing image data, and the historical meteorological data are used as the agricultural data.

[0089] Furthermore, the first data processing module 301 is also used to: perform data preprocessing on the agricultural data from different sources, wherein the data preprocessing includes at least one of the following: data cleaning, data format conversion, data semantic normalization, and data spatiotemporal alignment; Specifically, during the spatiotemporal alignment of the data, the precision of the spatiotemporal grid is adjusted. The spatiotemporal grid is increased during non-disaster periods and decreased during disaster periods. During non-disaster periods, the spatiotemporal grid updates data according to a first cycle, and during disaster periods, it updates data according to a second cycle, where the first cycle is longer than the second cycle. During data semantic normalization, for environmental data in the agricultural data, linear normalization is used to map the values ​​to the [0,1] interval, and semantic weights are set according to the disaster type. For disaster-related data in the agricultural data, interval mapping is used to map the values ​​to the [-1,1] interval, and semantic weights are set according to the crop growth period.

[0090] Furthermore, the second data processing module 302 is specifically used for: The agricultural data is evaluated for credibility, and feature fusion weights are determined based on the credibility scores, wherein the credibility scores are directly proportional to the feature fusion weights; or, The disaster scenario is determined based on the disaster scenario information, and the feature fusion weight of the agricultural data is determined based on the disaster scenario, wherein the feature fusion weight of the agricultural data is different under different disaster scenarios.

[0091] Furthermore, the device also includes: a model building module, used for: Obtain agricultural sample data from different sources for insured farmland; Feature information is extracted from agricultural sample data from different sources, and multiple feature information is fused to obtain a second comprehensive feature vector; Loss assessment is performed separately using different agricultural sample data to obtain multiple sub-loss assessment conclusions, and these multiple sub-loss assessment conclusions are then merged to obtain the loss assessment conclusion data. The deep learning model is trained based on the second comprehensive feature vector and the damage assessment conclusion data to generate the damage assessment prediction model. During training, the deep learning model is trained by freezing the bottom feature extraction layer and fine-tuning the top decision layer. A damage assessment causal analysis submodule is set in the deep learning model to identify and output the disaster-causing factors.

[0092] Furthermore, the model building module is specifically used for: Damage assessment is conducted using satellite remote sensing imagery data to evaluate the affected area and extent of damage to farmland. Damage assessment is conducted using meteorological data to predict the impact of weather on crop yields; Damage assessment is conducted using IoT sensor data to determine the constraints of the farmland environment on crop growth. The data reported by farmers is used to assess the damage, determine the affected area, the extent of the damage, and the estimated losses of farmland; The data for loss assessment are obtained by integrating information on the affected area and severity of farmland, the impact of weather on crop yield, the constraints of farmland environment on crop growth, estimated losses, and insurance parameters.

[0093] The agricultural insurance loss assessment device 300 in this application embodiment can be a computer device or a component of a computer device, such as an integrated circuit or a chip. The computer device can be a terminal or other devices besides a terminal. For example, the computer device can be a mobile phone, tablet computer, laptop computer, PDA, Ultra-Mobile Personal Computer (UMPC), netbook, or Personal Digital Assistant (PDA), etc., and can also be a server, Network Attached Storage (NAS), Personal Computer (PC), etc. This application embodiment does not specifically limit the specific device.

[0094] The agricultural insurance loss assessment device 300 provided in this application embodiment can achieve... Figure 2 The various processes implemented in the agricultural insurance loss assessment method implementation example will not be repeated here to avoid duplication.

[0095] This application also provides a computer device, such as... Figure 4As shown, the computer device 400 includes a processor 401 and a memory 402. The memory 402 stores programs or instructions that can run on the processor 401. When the program or instructions are executed by the processor 401, they implement the various steps of the above-described agricultural insurance loss assessment method embodiment and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0096] The memory 402 can be used to store software programs and various data. The memory 402 may primarily include a first storage area for storing programs or instructions and a second storage area for storing data. The first storage area may store the operating system, application programs or instructions required for at least one function (such as sound playback, image playback, etc.). Furthermore, the memory 402 may include volatile memory or non-volatile memory, or both. The non-volatile memory may be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory can be random access memory (RAM), static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DRRAM). The memory 402 in this embodiment includes, but is not limited to, these and any other suitable types of memory.

[0097] Processor 401 may include one or more processing units; optionally, processor 401 integrates an application processor and a modem processor, wherein the application processor mainly handles operations involving the operating system, user interface, and applications, and the modem processor mainly handles wireless communication signals, such as a baseband processor. It is understood that the aforementioned modem processor may also not be integrated into processor 401.

[0098] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 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 the agricultural insurance loss assessment method server side.

[0099] This application also provides a computer-readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the above-described agricultural insurance loss assessment method embodiments and achieve the same technical effects. To avoid repetition, they will not be described again here.

[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0101] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. An agricultural insurance loss determination method characterized by comprising: The method comprises: acquiring agricultural data of different sources of the insured farmland, and extracting feature information of the agricultural data of different sources; determining feature fusion weights of each of the agricultural data based on target data, and performing feature fusion on a plurality of the feature information according to the feature fusion weights corresponding to each of the agricultural data to obtain a first comprehensive feature vector, the target data including at least one of disaster scenario information and a credibility score of the agricultural data; inputting the first comprehensive feature vector into a trained loss prediction model to output loss data of the insured farmland, the loss data including a loss amount and a disaster-causing factor.

2. The method of claim 1, wherein, The method further comprises: on a geographic information system map, representing the insured farmlands with different disaster levels according to different visual variables; in response to a viewing request for a target insured farmland, displaying loss data of the target insured farmland.

3. The method of claim 1, wherein, The acquisition of the agricultural data of different sources of the insured farmland comprises: acquiring satellite remote sensing image data of the insured farmland, acquiring plot right data of the insured farmland in a plot right database, and binding pixel units of the acquired satellite remote sensing image data and the plot right data with semantic labels to generate an association relationship between the pixel units of the satellite remote sensing image data and the plot right data; acquiring Internet of Things sensor data of the insured farmland, pushing a data supplement prompt to a terminal device of a farmer corresponding to the insured farmland when the Internet of Things sensor data exceeds a disaster sensitivity threshold, acquiring supplement data provided by the terminal device, and acquiring historical satellite remote sensing image data and historical meteorological data of the insured farmland in the same period; acquiring meteorological data and farmer declaration data of the insured farmland; using the meteorological data, the farmer declaration data, the satellite remote sensing image data, the data of the association relationship, the Internet of Things sensor data, the historical satellite remote sensing image data, and the historical meteorological data as the agricultural data.

4. The method of claim 1, wherein, After the acquisition of the agricultural data of different sources of the insured farmland, the method further comprises: respectively performing data preprocessing on the agricultural data of different sources, the data preprocessing including at least one of data cleaning, data format conversion, data semantic normalization, and data space-time alignment; wherein, when performing the data space-time alignment, the accuracy of a space-time grid is adjusted, the space-time grid is increased in a non-disaster period and is reduced in a disaster period, in the non-disaster period, the space-time grid updates data according to a first period, and in the disaster period, the space-time grid updates data according to a second period, the first period being greater than the second period; when performing data semantic normalization, for environmental data in the agricultural data, a linear normalization is used to map numerical values to the interval [0, 1], and a semantic weight is set according to a disaster type, and for disaster data in the agricultural data, an interval mapping is used to map numerical values to the interval [-1, 1], and a semantic weight is set according to a crop growth period.

5. The method of claim 1, wherein, The determination of the feature fusion weights of each of the agricultural data based on the target data comprises: The agricultural data is evaluated for a credibility score, and a feature fusion weight of the agricultural data is determined according to the credibility score of the agricultural data, wherein the credibility score is directly proportional to the feature fusion weight; or The feature fusion weight of each agricultural data is determined based on target data, including: A disaster scene is determined according to the disaster scene information, and a feature fusion weight of the agricultural data is determined based on the disaster scene, wherein the feature fusion weight of the agricultural data is different under different disaster scenes.

6. The method according to any one of claims 1 to 5, characterized in that, The method further includes: Obtaining agricultural sample data of different sources of the insured farmland; Extracting feature information of the agricultural sample data of different sources, and fusing a plurality of the feature information to obtain a second comprehensive feature vector; Performing loss assessment on different agricultural sample data respectively to obtain a plurality of sub-loss assessment conclusions, and fusing a plurality of the sub-loss assessment conclusions to obtain loss assessment conclusion data; Training a deep learning model based on the second comprehensive feature vector and the loss assessment conclusion data to generate the loss assessment prediction model; wherein, during the training, the deep learning model is trained in a manner of freezing a bottom feature extraction layer of the deep learning model and fine-tuning a top decision layer of the deep learning model, and a loss assessment causal analysis sub-module is set in the deep learning model, and the loss assessment causal analysis sub-module is used to identify and output the disaster-causing factor.

7. The method of claim 6, wherein, Performing loss assessment on different agricultural sample data respectively to obtain a plurality of sub-loss assessment conclusions, and fusing a plurality of the sub-loss assessment conclusions to obtain loss assessment conclusion data, including: Performing loss assessment through satellite remote sensing image data to evaluate the disaster area and the disaster degree of the farmland; Performing loss assessment through meteorological data to predict the influence of the meteorological condition on the crop yield; Performing loss assessment through Internet of Things sensor data to determine the restriction of the farmland environment on the crop growth; Performing loss assessment through farmer declaration data to determine the disaster area, the disaster degree and the estimated loss of the farmland; Fusing the disaster area and the disaster degree of the farmland, the influence of the meteorological condition on the crop yield, the restriction of the farmland environment on the crop growth, the estimated loss and the insurance parameters to obtain the loss assessment conclusion data.

8. An agricultural insurance loss determining apparatus characterized by comprising: Including: A first data processing module is configured to obtain agricultural data of different sources of the insured farmland, and extract feature information of the agricultural data of different sources; A second data processing module is configured to determine a feature fusion weight of each agricultural data based on target data, and fuse a plurality of the feature information according to the feature fusion weight corresponding to each agricultural data to obtain a first comprehensive feature vector, wherein the target data includes at least one of disaster scene information and a credibility score of the agricultural data; A loss assessment prediction module is configured to input the first comprehensive feature vector into a trained loss assessment prediction model, and output loss assessment data of the insured farmland, wherein the loss assessment data includes a loss assessment amount and a disaster-causing factor.

9. A computer device, comprising: A computer program product comprising a computer readable medium, the computer readable medium having stored thereon the program or instructions executable by a computer processor to cause the computer processor to perform the steps of the agricultural insurance loss assessment method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a program or instructions, characterized in that, The program or instructions executable by a computer processor to cause the computer processor to perform the steps of the agricultural insurance loss assessment method according to any one of claims 1 to 7.

Images