Multi-scale remote sensing-based regional risk classification assessment system for agricultural insurance

Abstract

This invention relates to the field of regional risk grading technology for agricultural insurance, and discloses a multi-scale remote sensing-based regional risk grading assessment system for agricultural insurance. The system includes a multi-scale remote sensing data acquisition and access module, and a satellite remote sensing data access sub-module for automatically accessing multi-source satellite remote sensing data, including but not limited to medium- and low-resolution optical satellite data. This multi-scale remote sensing-based regional risk grading assessment system for agricultural insurance integrates multi-scale observation data from satellite remote sensing, UAV aerial photography, and IoT sensing. The system can simultaneously acquire macro-level regional conditions, meso-level details of plots, and micro-level conditions of plants, forming an integrated sky-ground three-dimensional monitoring network. This greatly enriches the data dimensions and accuracy of risk assessment. Furthermore, by utilizing machine learning and deep learning algorithms, the system automatically extracts features and performs correlation analysis on the fused multi-scale data, enabling real-time identification and quantitative inversion of crop growth status and environmental stress factors.

Description

Technical Field

[0001] This invention relates to the field of regional risk classification technology for agricultural insurance, and in particular to a multi-scale remote sensing-based regional risk classification and assessment system for agricultural insurance. Background Technology

[0002] Agricultural insurance, as an important financial tool for mitigating natural agricultural risks and stabilizing farmers' income, has always had scientific pricing and accurate risk assessment as core issues in the industry. In existing technologies, agricultural insurance risk assessment methods mainly rely on historical meteorological data, yield statistics, and some ground monitoring information, and attempt to construct a statistical relationship between meteorological indices and yield reduction rates, thereby enabling the design of insurance products and the determination of premium rates. For example, the Chinese patent with announcement number CN113674099B discloses a comprehensive risk assessment method for apple drought and flood disasters. By analyzing the curve relationship between the water surplus / deficit rate and yield reduction rate during the apple fruit expansion period, it determines the weather index threshold for drought and flood disasters, and combines disaster-prone environmental and disaster-bearing factors such as topography, vegetation, and planting area to construct a comprehensive risk index model for regional differentiation correction of pure insurance premium rates.

[0003] Alternatively, research has also attempted to integrate multi-source data such as soil, crops, and heavy metals into a paddy field heavy metal risk assessment system, as indicated in the announcement number CN120561503A. This system aims to establish an integrated database and a multi-level risk assessment model (such as single-factor index, Nemerow index, potential ecological risk index, etc.), and use GIS technology to achieve spatial visualization and dynamic early warning of risks.

[0004] However, existing technologies still have the following shortcomings: 1) Limited data sources and spatiotemporal resolution: Existing methods mostly rely on meteorological data from stations or periodic sampling data, which makes it difficult to achieve large-scale, high-frequency, and full-coverage dynamic monitoring, especially with a lag in response to sudden and local disasters; 2) Risk assessment models lack multi-scale information fusion: Most models do not fully integrate multi-scale observation data from macro satellite remote sensing, meso drone aerial photography to near-ground sensor networks, making it difficult to balance spatial granularity and timeliness in risk assessment, and failing to accurately characterize the spatial heterogeneity of risks within the region.

[0005] 3) The assessment results are static and lack dynamic grading and early warning mechanisms: Existing systems are often based on historical data or periodic assessments, which makes it difficult to achieve real-time tracking, dynamic grading and early warning of risks throughout the entire crop growth period. The timeliness and adaptability of insurance pricing and risk management are not strong. 4) Low system scalability and automation: Existing solutions are mostly designed for specific crops or disaster types, with weak model generalization ability and insufficient automation and intelligence in data processing and evaluation processes, making it difficult to support the rapid deployment and operation of large-scale, multi-type agricultural insurance business. Summary of the Invention

[0006] In view of the problems mentioned above, such as the single source of existing data and limited spatiotemporal resolution, insufficient fusion of multi-scale information in risk assessment models, static assessment results, lack of dynamic classification and early warning mechanisms, and low system scalability and automation, this invention is proposed.

[0007] Therefore, the purpose of this invention is to provide a multi-scale remote sensing agricultural insurance regional risk classification and assessment system. The purpose is to provide an agricultural insurance regional risk assessment system that can integrate multi-scale remote sensing data, realize dynamic risk classification and assessment, and has high automation and scalability, so as to improve the accuracy of agricultural insurance risk identification, the scientific nature of rate determination, and the level of intelligent business management.

[0008] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a multi-scale remote sensing agricultural insurance regional risk classification and assessment system, comprising: The multi-scale remote sensing data acquisition and access module includes: Satellite remote sensing data access submodule: used to automatically access multi-source satellite remote sensing data, including but not limited to medium and low resolution optical satellite data, high resolution commercial satellite data and synthetic aperture radar satellite data, supporting standard format parsing and metadata extraction; UAV aerial photography data acquisition submodule: Used to control the UAV to acquire centimeter-resolution images according to preset routes or emergency commands, and has the functions of rapid image stitching and orthorectification. Ground IoT data aggregation submodule: used to aggregate real-time environmental data collected by weather stations, soil sensors, and crop growth sensors deployed in the field, including precipitation, temperature, humidity, soil moisture, and leaf temperature;

[0009] The data fusion and intelligent feature extraction module includes: The data preprocessing submodule is used to perform spatiotemporal registration, radiometric correction, atmospheric correction, and format standardization on multi-source remote sensing data. Multi-source data spatiotemporal fusion submodule: used to fuse remote sensing data with different spatiotemporal resolutions to generate a data sequence with high spatiotemporal consistency; Deep learning feature extraction submodule: It adopts a multi-scale feature fusion network based on attention mechanism to adaptively and weightedly fuse remote sensing information with different spatial resolutions and spectral features, and outputs crop type classification map, leaf area index map, vegetation health index map, soil moisture distribution map, disease and pest infection suspicion distribution map and flood inundation range map. The dynamic risk grading and assessment model library includes: Model Management Submodule: Used to manage various machine learning and deep learning risk assessment models, supporting model version control, training, and deployment; Risk assessment calculation submodule: Based on the input crop characteristics, environmental stress characteristics and static background data, it calls the corresponding model to calculate the indices of four dimensions: the risk of disaster-causing factors, the sensitivity of the disaster-prone environment, the vulnerability of the disaster-bearing body and the disaster prevention and mitigation capacity. Model update submodule: Based on newly added disaster samples and loss data, the model is incrementally trained and its performance is optimized on a regular basis. The risk assessment results visualization and output module includes: Spatial Visualization Submodule: Used to visualize the assessment results such as risk level, disaster distribution, and vulnerable areas in the form of interactive heat maps, hierarchical thematic maps, and 3D scenes; Report generation submodule: Used to automatically generate structured assessment reports that include risk spatial distribution, level statistics, main disaster-causing factor analysis, and trend prediction, and supports export in multiple formats; Insurance business integration interfaces include: Rate determination auxiliary submodule: used to perform spatial overlay analysis of risk classification map and insured land vector boundary, automatically calculate the risk level and recommended benchmark rate for each insured unit; Claims assistance submodule: After a claim is triggered, it is used to quickly extract information on the scope and extent of damage by comparing the features of pre- and post-disaster remote sensing images, thus assisting in loss assessment.

[0010] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention, the satellite remote sensing data access submodule in the multi-scale remote sensing data acquisition and access module supports access to multiple satellite data sources, including Sentinel series, Landsat series, Gaofen series, PlanetScope, and WorldView, and has an automatic data subscription and push mechanism; the UAV aerial photography data acquisition submodule supports multi-UAV collaborative operation and real-time video stream transmission; and the ground IoT data aggregation submodule adopts low-power wide area network technology to achieve stable data transmission of large-area field sensor networks.

[0011] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention, the multi-source data spatiotemporal fusion submodule in the data fusion and intelligent feature extraction module adopts a method based on a spatiotemporal adaptive reflectivity fusion model to fuse low temporal resolution high spatial resolution data with high temporal resolution low spatial resolution data to generate a data sequence with high spatiotemporal consistency; the deep learning feature extraction submodule adopts an encoder-decoder architecture and combines an attention mechanism to achieve precise identification of key crop growth stages and environmental stresses.

[0012] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification and assessment system of the present invention, the risk assessment calculation submodule in the dynamic risk classification and assessment model library includes built-in risk assessment algorithm models such as random forest, gradient boosting tree, convolutional neural network and recurrent neural network models; the model input features include dynamic features inverted from remote sensing data, frequency and intensity of disasters in the same historical period, topographic factors derived from digital elevation model, soil type and texture data, and crop planting area and output value data from statistical data; the model output includes the probability of disaster occurrence, expected yield loss rate and comprehensive risk index.

[0013] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention, the system further includes a risk dynamic simulation and early warning module. This module accesses short-term weather forecast data and climate prediction data, combines them with the dynamic risk classification assessment model library, performs scenario simulation and deduction of agricultural disaster risks in the next 7-30 days, and issues early warning information for areas or time periods where the risk level may be upgraded.

[0014] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention, the system is deployed based on a cloud platform architecture, including a data storage and computing sublayer, a model service sublayer and an application display sublayer, and supports multi-user concurrent access, task queue management and customized configuration of the assessment process.

[0015] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification and assessment system of the present invention, wherein: the comprehensive agricultural risk index defined in the risk assessment calculation submodule The calculation formula is as follows: =

[0016] in, As a hazard index for disaster-causing factors, This is an index of environmental sensitivity to disasters. As a vulnerability index for disaster-bearing bodies, For disaster prevention and mitigation capability index; , These are the weights of each index, and they satisfy... =1;; All indices have been normalized and their values ​​range from [0, 1]; The risk index Disaster intensity and historical occurrence probability calculation based on real-time inversion; sensitivity index Vulnerability index is calculated based on topographic relief, soil permeability, drainage conditions, and vegetation cover. The disaster prevention and mitigation capability index is calculated based on crop economic value density, insurance concentration, and farmers' risk resistance capacity. Calculated based on regional irrigation facility density, disaster prevention investment, and insurance coverage.

[0017] As a preferred embodiment of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention, wherein: the disaster-causing factor hazard index The calculation formula is: =

[0018] in, This is a normalized value of disaster intensity retrieved based on current remote sensing monitoring data. The frequency of disasters during the same period in history, The adjustment factor ranges from 0.6 to 0.8.

[0019] To achieve the above objectives, the present invention provides the following technical solution: an assessment method for a multi-scale remote sensing-based regional risk classification assessment system for agricultural insurance, comprising the following steps: S1: Automatically collect and access multi-source remote sensing data of the target area according to a preset cycle or triggering conditions through the multi-scale remote sensing data acquisition and access module; S2: The data fusion and intelligent feature extraction module is used to preprocess, perform spatiotemporal fusion and feature extraction on the raw data to obtain the crop planting distribution, growth parameters and environmental stress factor distribution; S3: Call the risk assessment calculation submodule in the dynamic risk grading assessment model library, combine it with static background data, calculate the disaster-causing factor hazard, disaster-prone environment sensitivity, disaster-bearing body vulnerability and disaster prevention and mitigation capacity index, and then weight and fuse them to obtain the comprehensive agricultural risk index and classify the risk level. S4: Through the risk assessment results visualization and output module, the risk level results are displayed in the form of interactive heat maps and hierarchical thematic maps, and a structured assessment report is automatically generated. S5: Through the insurance business integration interface, the assessment results are pushed to the core insurance business system and applied to rate setting, risk zoning, policy underwriting, claims assistance and reinsurance decisions.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects: 1. This invention integrates multi-scale observation data from satellite remote sensing, UAV aerial photography, and IoT sensing. The system can simultaneously acquire the macro-level situation of the region, the meso-level details of the plot, and the micro-level condition of the plant, forming a three-dimensional monitoring network that integrates the sky and the ground. This greatly enriches the data dimensions and accuracy of risk assessment. At the same time, by using machine learning and deep learning algorithms, the system automatically extracts features and performs correlation analysis on the fused multi-scale data, realizing real-time identification and quantitative inversion of crop growth status and environmental stress factors. Therefore, a dynamic comprehensive risk index model is constructed that considers the hazard of disaster-causing factors, the sensitivity of the disaster-prone environment, the vulnerability of the disaster-bearing body, and the disaster prevention and mitigation capabilities, and realizes the automatic classification and spatial mapping of risk levels.

[0021] 2. The dynamic risk grading map and quantitative assessment report output by the system of this invention can be directly applied to differentiated premium rate determination, policy risk review, claims settlement assistance, and reinsurance arrangements in agricultural insurance. By accurately linking risk with spatial location and time series, personalized insurance product design can be achieved, basis risk can be reduced, and insurance operational efficiency can be improved.

[0022] 3. The system of this invention adopts a modular design and has built-in functions such as automatic data preprocessing, model self-training optimization, and automatic report generation, which greatly reduces manual intervention. At the same time, the system architecture supports flexible access to new data sources, risk assessment algorithms, and insurance business rules, and can quickly adapt to the agricultural insurance risk assessment needs of different crops, different types of disasters, and different regions. It has good platform attributes and industry promotion value. Based on real-time monitoring and dynamic assessment results, the system can provide early warnings for high-risk areas and key growth periods, prompting insurance institutions and insured parties to take preventive measures, realizing the transformation of risk management paradigm from post-disaster compensation to pre-disaster prevention, and improving the overall resilience of the agricultural system. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the organizational framework of the multi-scale remote sensing agricultural insurance regional risk classification and assessment system of the present invention. Figure 2 This is a schematic diagram illustrating the steps of the assessment method of the multi-scale remote sensing agricultural insurance regional risk classification assessment system of the present invention. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Example 1 System hardware and software environment deployment Hardware deployment: Cloud server cluster: Employs high-performance cloud servers configured with GPU computing cards for deep learning model training and inference. Deploys Geographic Information System (GIS) servers, database servers, and application servers.

[0026] Data acquisition terminal: Satellite data receiving station, used to receive or acquire satellite data such as Sentinel-2 (optical), Sentinel-1 (radar), Landsat 8 / 9, etc.

[0027] Unmanned aerial vehicle (UAV) system: Multiple rotary-wing UAVs equipped with multispectral cameras and high-definition visible light cameras.

[0028] Ground-based Internet of Things (IoT): Deploy wireless soil temperature and humidity sensors and field microclimate stations in typical fields.

[0029] System module collaborative workflow Step 1: Multi-scale remote sensing data acquisition and access The satellite data access and scheduling submodule is activated: The system automatically sets a data acquisition plan based on the winter wheat growth calendar. During the critical greening-jointing period, it automatically retrieves and downloads Sentinel-2 Level-1C data (for growth and drought monitoring) and Sentinel-1 GRD data (for soil moisture retrieval) covering the target area daily. The module automatically parses metadata, filters cloud cover, and pushes qualified data slices to the preprocessing queue.

[0030] The UAV collaborative data acquisition and control submodule responds as follows: When satellite monitoring detects suspected drought or low-temperature anomalies in a certain area, or when a blue frost warning is received from the meteorological department, the system automatically generates an emergency monitoring task. A swarm of UAVs is planned to conduct intensive flights over key areas, acquiring multispectral images with centimeter-level resolution, and the data is transmitted back to the ground station in real time.

[0031] The ground-based sensor network data fusion submodule operates as follows: Field sensors collect soil moisture content (0-20cm, 20-40cm), air temperature and humidity, and surface temperature data every 10 minutes. The module aggregates this data in real time, removes outliers, and interpolates missing data caused by communication interruptions, forming standardized environmental variable time-series curves, which are then provided to the risk assessment model via an API interface.

[0032] Step 2: Data Fusion and Intelligent Feature Extraction

[0033] The multi-scale data preprocessing and registration submodule performs the following: atmospheric correction (Sen2Cor processor) on the received Sentinel-2 imagery to generate surface reflectance products; radiometric calibration, speckle filtering, and terrain correction on the Sentinel-1 imagery. All satellite and UAV imagery is uniformly registered to the WGS84 UTM coordinate system and spatially correlated with ground sensor locations.

[0034] Work of the multimodal remote sensing information deep fusion submodule: By employing a water cloud model and a change detection algorithm, and by synergistically utilizing the Sentinel-1 backscattering coefficient and Sentinel-2 optical information, the volumetric water content distribution map of the 0-5cm surface soil was obtained.

[0035] Canopy height was estimated by combining high-precision DSM and Sentinel-2 data from UAVs.

[0036] The final result is a fused data cube containing "reflectivity (multi-band), soil moisture content, canopy height, and surface temperature (derived from Landsat or UAV thermal infrared)".

[0037] Work on the risk-oriented intelligent interpretation submodule for agricultural characteristics: Load the pre-trained U-Net++ model (trained on a large number of labeled crop remote sensing images) to perform pixel-by-pixel interpretation of the fused data cube.

[0038] Step 3: Dynamic Risk Classification and Assessment

[0039] Multi-hazard risk assessment model integration sub-module selection model: For this assessment, the "drought disaster risk assessment model" and the "spring low temperature freezing damage risk assessment model" are selected, and the "multi-hazard integrated risk aggregation model" is activated.

[0040] The risk assessment dynamic calculation engine submodule performs the calculations: Input: LAI map, MSI map, soil moisture map, low temperature stress map generated in the previous step, as well as historical drought frequency map, historical frost damage occurrence map, topographic slope map, soil quality map, cultivated land plot vector and insurance rate data.

[0041] calculate: Drought risk: Using = formula.

[0042] Frost damage risk: Similarly, calculations are based on the current level of low temperature stress and the frequency of historical frost damage.

[0043] Environmental sensitivity: Calculate the sensitivity to drought and frost damage based on topography (slope, elevation) and soil water retention capacity (texture).

[0044] Vulnerability of disaster-bearing bodies: calculated based on the proportion of winter wheat planting area, average yield per mu, and policy-based insurance coverage in the region.

[0045] Disaster prevention and mitigation capabilities: calculated based on irrigation facility density (obtained from data from the water resources department) and the distribution of artificial hail suppression operation sites.

[0046] application = The formula generates a raster chart of the comprehensive agricultural risk index.

[0047] Based on preset thresholds (e.g., low risk: RI<0.3; medium risk: 0.3≤RI<0.6; high risk: RI≥0.6), the risk is divided into three levels, generating a risk zoning map.

[0048] Model Adaptive Optimization and Knowledge Iteration Submodule: After this evaluation, the system will store the input features, the calculated risk index, and subsequent actual disaster loss reports (if any) as new sample pairs in the sample library for use in the next round of model iteration training.

[0049] Step 4: Visualization of Risk Assessment Results and Decision Support

[0050] The multi-dimensional dynamic risk map visualization submodule works by loading a risk zoning map onto the WebGIS platform, overlaying it with satellite imagery, township boundaries, and major rivers. Users can click on any location to query its risk level and the contribution of major disaster-causing factors (e.g., drought contributes 70%, frost damage contributes 30%). A time slider is provided to compare risk changes over different dates (e.g., before and after a frost damage warning).

[0051] The structured report and early warning product generation submodule works as follows: The system automatically generates an assessment report, which includes: a risk spatial distribution map, area statistics for each risk level, a list of high-risk townships (e.g., Township A, Township B), analysis of the main risk sources, and countermeasures such as "recommendations to strengthen irrigation and frost warnings in the high-risk area of ​​Township A". Simultaneously, for areas with an RI > 0.7, it automatically generates "high-risk early warning" information and pushes it to relevant agricultural management departments and insurance institutions via an interface.

[0052] Step 5: Intelligent adaptation of the entire insurance business process

[0053] The risk-differentiated premium rate determination submodule works as follows: Insurance institutions upload the vector boundaries of the farmland plots they intend to insure this year through the system interface. The system automatically performs spatial overlay analysis on these plots and the risk zoning map. For plots located in high-risk areas, the system, based on the benchmark pure premium rate (e.g., 2%), increases it by a certain percentage according to the risk level (e.g., 30% for high-risk plots), and combines this with operating costs to provide a suggested differentiated premium rate (e.g., 2.6%). For low-risk areas, a lower premium rate can be suggested.

[0054] The intelligent verification auxiliary module for underwriting and claims processing works as follows: Underwriting verification: Underwriters use the system to retrieve crop type interpretation maps and historical images of the insured plots of farmers to quickly verify the accuracy of the planted crops and area.

[0055] Claims Assessment: Suppose a severe cold snap occurred in early April. After the disaster, the system automatically dispatches remote sensing images and compares the changes in LAI, NDVI, and other feature layers between before (March) and after (April). Combined with the low-temperature stress identification map, it automatically delineates the affected area and preliminarily classifies the losses into mild (10-30% production reduction), moderate (30-50%), and severe (>50%) levels based on the degree of feature attenuation. This generates an auxiliary claims assessment report, greatly improving the objectivity and efficiency of the assessment.

[0056] Through the implementation of this embodiment, the system has successfully achieved integrated air-ground dynamic monitoring of drought and frost disasters during the critical growth period of winter wheat, completed the automated and intelligent production from raw data to risk level products, and accurately applied the assessment results to differentiated calculation of insurance premiums and rapid auxiliary loss assessment for claims, thus verifying the effectiveness, advancement and practical value of the system of the present invention.

[0057] It should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-scale remote sensing-based regional risk classification and assessment system for agricultural insurance, characterized in that, include: The multi-scale remote sensing data acquisition and access module includes: Satellite remote sensing data access submodule: used for automatic access to multi-source satellite remote sensing data; UAV aerial photography data acquisition submodule: used to control the UAV to acquire centimeter-resolution images according to preset routes or emergency commands; Ground IoT data aggregation submodule: used to aggregate real-time environmental data collected by weather stations, soil sensors, and crop growth sensors deployed in the field; The data fusion and intelligent feature extraction module includes: The data preprocessing submodule is used to perform spatiotemporal registration, radiometric correction, atmospheric correction, and format standardization on multi-source remote sensing data. Multi-source data spatiotemporal fusion submodule: used to fuse remote sensing data with different spatiotemporal resolutions to generate a data sequence with high spatiotemporal consistency; Deep learning feature extraction submodule: It adopts a multi-scale feature fusion network based on attention mechanism to adaptively and weightedly fuse remote sensing information with different spatial resolutions and spectral features; The dynamic risk grading and assessment model library includes: Model Management Submodule: Used to manage various machine learning and deep learning risk assessment models, supporting model version control, training, and deployment; Risk assessment calculation submodule: used to calculate based on input crop characteristics, environmental stress characteristics, and static background data; Model update submodule: Based on newly added disaster samples and loss data, the model is incrementally trained and its performance is optimized on a regular basis. The risk assessment results visualization and output module includes: Spatial Visualization Submodule: Used to visualize the assessment results such as risk level, disaster distribution, and vulnerable areas in the form of interactive heat maps, hierarchical thematic maps, and 3D scenes; The report generation submodule is used to automatically generate structured assessment reports that include risk spatial distribution, level statistics, main disaster-causing factor analysis, and trend prediction. Insurance business integration interfaces include: Rate determination auxiliary submodule: used to perform spatial overlay analysis of risk classification map and insured land parcel vector boundary; Claims assistance submodule: After a claim is triggered, it is used to quickly extract information on the scope and extent of damage by comparing the features of pre- and post-disaster remote sensing images, thus assisting in loss assessment.

2. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The satellite remote sensing data access submodule in the multi-scale remote sensing data acquisition and access module supports access to multiple satellite data sources, including Sentinel series, Landsat series, Gaofen series, PlanetScope, and WorldView, and has an automatic data subscription and push mechanism; the UAV aerial photography data acquisition submodule supports multi-UAV collaborative operation and real-time video stream transmission; the ground IoT data aggregation submodule uses low-power wide area network technology to achieve stable data transmission of large-area field sensor networks.

3. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The multi-source data spatiotemporal fusion submodule in the data fusion and intelligent feature extraction module adopts a method based on the spatiotemporal adaptive reflectivity fusion model to fuse low temporal resolution high spatial resolution data with high temporal resolution low spatial resolution data to generate a data sequence with high spatiotemporal consistency. The deep learning feature extraction submodule adopts an encoder-decoder architecture and combines an attention mechanism to achieve precise identification of key crop growth stages and environmental stresses.

4. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The risk assessment calculation submodule in the dynamic risk grading assessment model library includes built-in risk assessment algorithm models such as random forest, gradient boosting tree, convolutional neural network, and recurrent neural network models. The model input features include dynamic features inverted from remote sensing data, the frequency and intensity of disasters in the same historical period, topographic factors derived from digital elevation models, soil type and texture data, and crop planting area and output value data from statistical data. The model output includes the probability of disaster occurrence, expected yield loss rate, and comprehensive risk index.

5. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The system also includes a risk dynamic simulation and early warning module. This module accesses short-term weather forecast data and climate prediction data, and combines them with the dynamic risk classification assessment model library to conduct scenario simulation and extrapolation of agricultural disaster risks in the next 7-30 days, and issues early warning information for areas or periods where the risk level may be upgraded.

6. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The system is deployed based on a cloud platform architecture, including a data storage and computing sublayer, a model service sublayer, and an application display sublayer. It supports multi-user concurrent access, task queue management, and customized configuration of the evaluation process.

7. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 1, characterized in that: The comprehensive agricultural risk index defined in the risk assessment calculation submodule The calculation formula is as follows: = in, As a hazard index for disaster-causing factors, This is an index of environmental sensitivity to disasters. As a vulnerability index for disaster-bearing bodies, For disaster prevention and mitigation capability index; , These are the weights of each index, and they satisfy... =1;; All indices have been normalized and their values ​​range from [0, 1]; The risk index Disaster intensity and historical occurrence probability calculation based on real-time inversion; sensitivity index Vulnerability index is calculated based on topographic relief, soil permeability, drainage conditions, and vegetation cover. The disaster prevention and mitigation capability index is calculated based on crop economic value density, insurance concentration, and farmers' risk resistance capacity. Calculated based on regional irrigation facility density, disaster prevention investment, and insurance coverage.

8. The multi-scale remote sensing agricultural insurance regional risk classification and assessment system according to claim 7, characterized in that: The hazard index of the disaster-causing factor The calculation formula is: = in, This is a normalized value of disaster intensity retrieved based on current remote sensing monitoring data. The frequency of disasters during the same period in history, The adjustment factor ranges from 0.6 to 0.

8.

9. A method for regional risk classification and assessment of agricultural insurance using the system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Automatically collect and access multi-source remote sensing data of the target area according to a preset cycle or triggering conditions through the multi-scale remote sensing data acquisition and access module; S2: The data fusion and intelligent feature extraction module is used to preprocess, perform spatiotemporal fusion and feature extraction on the raw data to obtain the crop planting distribution, growth parameters and environmental stress factor distribution; S3: Call the risk assessment calculation submodule in the dynamic risk grading assessment model library, combine it with static background data, calculate the disaster-causing factor hazard, disaster-prone environment sensitivity, disaster-bearing body vulnerability and disaster prevention and mitigation capacity index, and then weight and fuse them to obtain the comprehensive agricultural risk index and classify the risk level. S4: Through the risk assessment results visualization and output module, the risk level results are displayed in the form of interactive heat maps and hierarchical thematic maps, and a structured assessment report is automatically generated. S5: Through the insurance business integration interface, the assessment results are pushed to the core insurance business system and applied to rate setting, risk zoning, policy underwriting, claims assistance and reinsurance decisions.

Images