A farmland digital twin model construction method

Abstract

The application relates to a farmland digital twin model construction method. The method comprises the following steps: acquiring multi-source farmland monitoring data of a target farmland area and extracting a farmland monitoring feature set; constructing a first farmland twin model by adopting machine learning joint modeling based on the farmland monitoring feature set; performing regional clustering division on the first farmland twin model by using a farmland index evaluation system to obtain a farmland index regional field; constructing a parameter updating mechanism of the farmland index regional field and embedding the first farmland twin model to obtain a second farmland twin model. By using the method, the real-time performance and accuracy of the farmland digital twin model can be improved through dynamic parameter updating and regional clustering, and the decision management in intelligent agriculture can be effectively supported.

Description

Technical Field

[0001] This invention belongs to the field of smart agriculture, and in particular relates to a method for constructing a digital twin model of farmland. Background Technology

[0002] With the development of smart agriculture and digital twin technology, farmland digital twin modeling technology has emerged. This technology enables the mapping and interaction between physical farmland and virtual models, providing data support and decision-making basis for precise farmland management, and has become an important technological direction for the intelligent development of agriculture. Currently, farmland digital twin model construction often uses a single remote sensing data source or simple multi-source data integration. Traditional technologies rely solely on satellite remote sensing data for full-area modeling or only on UAV remote sensing data for small-scale, detailed modeling. While some methods integrate multiple types of farmland monitoring data, they fail to achieve in-depth feature mining and spatiotemporal correlation analysis, and lack targeted regional division and dynamic parameter update mechanisms. Existing methods suffer from difficulties in balancing model accuracy and coverage, insufficient correlation analysis of farmland elements, and lagging model parameter updates, failing to dynamically match the real-time state of farmland. This results in low fidelity and practicality of the twin models, making it difficult to meet the refined and dynamic management needs of smart agriculture. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for constructing a digital twin model of farmland that can solve the above problems.

[0004] Firstly, this application provides a method for constructing a digital twin model of farmland, including:

[0005] Acquire multi-source farmland monitoring data for the target farmland area, and extract farmland monitoring feature sets based on the multi-source farmland monitoring data;

[0006] Based on the farmland monitoring feature set, a joint modeling method based on machine learning models is used to construct the first farmland twin model;

[0007] Based on the first farmland twin model, a farmland index evaluation system is used to perform regional clustering to obtain the farmland index regional field.

[0008] A parameter update mechanism for the regional field of farmland indicators is constructed, and the parameter update mechanism is embedded into the first farmland twin model to obtain the second farmland twin model.

[0009] In one embodiment, the multi-source farmland monitoring data includes multi-source remote sensing data, meteorological observation data, soil sensor monitoring data, and basic soil data;

[0010] Based on multi-source farmland monitoring data, a farmland monitoring feature set is extracted, including:

[0011] Geometric radiometric correction is performed on multi-source remote sensing data to obtain calibrated remote sensing data. Based on the calibrated remote sensing data, remote sensing features are extracted to obtain a remote sensing feature set, which includes spectral features and texture features.

[0012] Based on meteorological observation data, meteorological time-series features are extracted to obtain a meteorological feature set; the meteorological time-series features include temperature time-series features, humidity time-series features, precipitation time-series features, and sunshine time-series features;

[0013] Based on soil sensor monitoring data, soil monitoring time-series features are extracted to obtain a soil monitoring feature set, which includes soil physical property time-series features and soil electrochemical property time-series features.

[0014] Based on basic soil data, principal component analysis was used to extract a set of basic soil properties.

[0015] By integrating remote sensing feature sets, meteorological feature sets, soil monitoring feature sets, and soil basic property sets, a farmland monitoring feature set is obtained.

[0016] In one embodiment, based on a farmland monitoring feature set, a first farmland twin model is constructed using a joint modeling method based on a machine learning model, including:

[0017] Calculate vegetation indices based on remote sensing feature sets;

[0018] Based on the meteorological feature set, a long short-term memory network is used to construct the meteorological factor correlation matrix;

[0019] Based on the soil monitoring feature set, the empirical mode decomposition method was used to extract the soil physicochemical change characteristics;

[0020] By aligning vegetation indices, meteorological factor correlation matrices, soil physicochemical change characteristics, and soil basic property sets in multiple dimensions, basic farmland elements are obtained.

[0021] Based on basic farmland elements, a graph neural network is used to mine spatiotemporal association rules to obtain the topological relationships between basic farmland elements.

[0022] Based on topological relationships and combined with basic farmland elements, an initial twin feature matrix is ​​generated;

[0023] Historical farmland basic elements are obtained, and based on these elements, the weight parameters of the initial twin feature matrix are optimized using the backpropagation algorithm to obtain the optimized twin feature matrix.

[0024] Based on the optimized twin feature matrix, a twin model of the first farmland was constructed.

[0025] In one embodiment, the farmland index evaluation system includes crop growth index, meteorological disaster index, pest and disease stress index, farmland pollution index, and nutrient surplus and deficit index;

[0026] Based on the first farmland twin model, a farmland index evaluation system is used for regional clustering to obtain a farmland index regional field, including:

[0027] Based on crop growth indicators and vegetation indices, crop growth grading zones are divided according to the vegetation index grading thresholds.

[0028] Based on crop growth grading regions and meteorological disaster indicators, the disaster-causing meteorological grading thresholds for disasters caused by meteorological disasters in corresponding crop growth grading regions are determined, and meteorological disaster grading regions are divided based on the disaster-causing meteorological grading thresholds and the meteorological factor correlation matrix.

[0029] Based on crop growth grading regions and nutrient surplus / deficient indicators, crop nutrient grading requirements are determined, and based on crop nutrient grading requirements, soil physicochemical change characteristics, and soil basic property set, nutrient surplus / deficient grading regions are divided.

[0030] Based on crop growth grading regions and pest and disease stress indicators, pest and disease suitable environment grading parameters are determined, and pest and disease stress grading regions are divided based on pest and disease suitable environment grading parameters, meteorological factor correlation matrix, soil physicochemical change characteristics and soil basic property set.

[0031] Pollution classification thresholds are determined based on farmland pollution indicators, and farmland pollution classification areas are divided based on pollution classification thresholds, soil physicochemical change characteristics, and soil basic property sets.

[0032] Cross-regional clustering was performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus / deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain farmland indicator regional fields.

[0033] In one embodiment, the parameter update mechanism includes a periodic update mechanism and an event-triggered update mechanism;

[0034] A parameter update mechanism for the regional field of farmland indicators is constructed, and this mechanism is embedded into the first farmland twin model to obtain the second farmland twin model, including:

[0035] Extract the dynamic parameter set of each farmland indicator area;

[0036] Based on the dynamic parameter set, parameter change features are extracted, including the parameter change rate and the parameter abrupt change magnitude.

[0037] Based on the rate of parameter change, the update frequency of each farmland indicator area field is determined;

[0038] Based on the magnitude of parameter mutations, the immediate update threshold for each farmland indicator area field is determined.

[0039] Based on the topological relationship of the first farmland twin model, the update priority of each farmland indicator area field is determined;

[0040] A periodic update mechanism is generated based on update priority and update frequency.

[0041] An event-triggered update mechanism is generated based on update priority and immediate update threshold.

[0042] By embedding the periodic update mechanism and the event-triggered update mechanism into the first farmland twin model, the second farmland twin model is obtained.

[0043] In one embodiment, the formula for calculating the update frequency of each farmland index area field based on the parameter change rate is as follows:

[0044]

[0045] in, Let be the update frequency of the i-th farmland index area field. Let be the velocity sensitivity coefficient of the i-th farmland index area field. For the statistical time window of farmland dynamic parameters, Let be the rate of change of parameters in the i-th farmland index area field. The mean rate of change of parameters of all farmland indicators within the target farmland area. Let the standard deviation of the rate of change of parameters in the regional field of all farmland indicators within the target farmland area be denoted as . It is a minimal positive real number.

[0046] In one embodiment, cross-regional clustering is performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus / deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain a farmland index regional field, including:

[0047] Based on crop growth classification regions, meteorological disaster classification regions, nutrient surplus and deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions, spatial geographic coordinates of each classification region are extracted, and a spatial adjacency matrix is ​​constructed based on the spatial geographic coordinates.

[0048] Based on the hierarchical level corresponding to each hierarchical region, the spatial adjacency relation matrix is ​​weighted by the region to obtain the weighted spatial adjacency matrix.

[0049] A spatial clustering algorithm is used to divide the weighted spatial adjacency matrix into regional clusters to obtain the initial farmland index cluster regions;

[0050] By combining the grading thresholds corresponding to each grading region, the boundaries of the initial farmland index cluster regions are corrected to obtain the optimized cluster regions.

[0051] Spatial topological fusion of optimized cluster regions yields farmland index regional fields.

[0052] Secondly, this application also provides a device for constructing a digital twin model of farmland, comprising:

[0053] The multi-source data acquisition module is used to acquire multi-source farmland monitoring data of the target farmland area and extract farmland monitoring feature sets based on the multi-source farmland monitoring data;

[0054] The model joint construction module is used to construct the first farmland twin model based on the farmland monitoring feature set and the joint modeling method based on machine learning model.

[0055] The regional clustering module is used to perform regional clustering based on the first farmland twin model and the farmland index evaluation system to obtain the farmland index regional field.

[0056] The model mechanism embedding module is used to construct the parameter update mechanism of the farmland indicator regional field, and embed the parameter update mechanism into the first farmland twin model to obtain the second farmland twin model.

[0057] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described method for constructing a digital twin model of farmland.

[0058] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for constructing a digital twin model of farmland.

[0059] The aforementioned method for constructing a digital twin model of farmland overcomes the limitations of traditional single data sources by acquiring multi-source farmland monitoring data of the target farmland area and extracting farmland monitoring feature sets. This achieves deep integration of multi-source data such as remote sensing, meteorology, and soil, expanding the model's coverage and enhancing feature diversity. Based on this feature set, a first farmland twin model is constructed using machine learning joint modeling. Intelligent algorithms are used to deeply mine the spatiotemporal correlation rules between farmland elements, improving the model's representation accuracy of complex farmland environments and overcoming the shortcomings of insufficient correlation analysis in existing technologies. A farmland indicator evaluation system is used to perform regional clustering on the first farmland twin model to obtain farmland indicator regional fields. This achieves hierarchical regional clustering based on indicators such as crop growth and meteorological disasters, balancing overall coverage with local refinement and solving the problem of balancing model coverage and accuracy. A parameter update mechanism is constructed and embedded into the first farmland twin model to obtain a second farmland twin model. The dynamic parameter update mechanism enables real-time adjustment of model parameters, improving the twin model's response speed and fidelity to changes in farmland conditions, supporting refined decision-making and management in smart agriculture. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of a method for constructing a digital twin model of farmland according to the present invention;

[0062] Figure 2 This is a structural diagram of a farmland digital twin model construction device according to the present invention. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0064] In one embodiment, such as Figure 1As shown, a method for constructing a digital twin model of farmland is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and to a system including both a terminal and a server, and is implemented through the interaction between the terminal and the server. In the implementation environment of this application, the hardware architecture mainly consists of terminal devices deployed in the farmland (such as multispectral remote sensing sensors, meteorological observation stations, soil sensing probes, and UAV platforms). These terminals are responsible for collecting multi-source farmland monitoring data in real time, including satellite remote sensing images, meteorological time-series information, and soil physicochemical parameters. The collected data can be transmitted to a cloud server through an IoT gateway or 5G network. The server runs machine learning algorithms (such as long short-term memory networks and graph neural networks) to perform feature fusion and model construction, forming a hierarchical processing system. In application scenarios, when a smart agriculture management platform needs to dynamically respond to changes in farmland status (such as crop growth monitoring or disaster early warning), terminal devices continuously collect and preprocess data. After receiving the data, the server performs feature extraction, regional clustering, and parameter update mechanisms. Model parameters are adjusted through event triggering or periodic update mechanisms to generate a high-fidelity digital twin model. The results are then fed back to the terminal interface to guide decisions such as variable fertilization and precision irrigation, achieving real-time mapping and closed-loop optimization between the physical farmland and the virtual model. In this embodiment, the method includes the following steps:

[0065] S01, acquire multi-source farmland monitoring data of the target farmland area, and extract farmland monitoring feature set based on the multi-source farmland monitoring data.

[0066] The multi-source farmland monitoring data includes remote sensing data (such as satellite multispectral imagery and UAV hyperspectral imagery), meteorological observation data (such as temperature, humidity, and precipitation time series), soil sensing data (such as physical properties and electrochemical parameters), and basic soil attribute data. This data can be collected in real time via IoT terminals or network transmission devices. During the feature extraction stage, the raw data can be corrected (e.g., geometric radiometric calibration and atmospheric correction for remote sensing data to eliminate environmental interference). Based on the corrected data, data processing algorithms (such as principal component analysis for dimensionality reduction and gray-level co-occurrence matrix for texture feature calculation) are used to extract spectral, texture, and temporal features, which are then integrated into a high-dimensional farmland monitoring feature set to support subsequent model construction. Multi-source complementarity expands data coverage, enhances feature diversity, and overcomes the limitations of a single data source.

[0067] S02, based on the farmland monitoring feature set, uses a joint modeling method based on machine learning models to construct the first farmland twin model.

[0068] Based on farmland monitoring feature sets, multi-dimensional alignment techniques (such as spatiotemporal registration and feature standardization) can be used to integrate basic farmland elements such as vegetation indices, meteorological factor correlation matrices, and soil change characteristics, ensuring data consistency and modelability. The joint modeling method includes various machine learning models, such as using Long Short-Term Memory (LSTM) networks to mine temporal dependencies in meteorological data, using Graph Neural Networks (GNNs) to capture spatial topology to reveal nonlinear relationships between elements, and combining signal processing techniques such as Empirical Mode Decomposition (EMD) to extract dynamic features. An initial twin feature matrix is ​​generated based on these features, and weight parameters can be optimized using a backpropagation algorithm driven by historical data to improve the model's accuracy and robustness. Through the synergistic effect of machine learning models, the limitations of traditional single modeling are overcome, enabling the rapid construction of high-fidelity digital twins and providing a foundation for farmland condition simulation and decision support.

[0069] S03, based on the first farmland twin model, uses the farmland index evaluation system to perform regional clustering and obtain the farmland index regional field.

[0070] The farmland indicator evaluation system can include quantitative indicators from multiple agricultural fields, such as crop growth, meteorological disasters, pest and disease stress, farmland pollution, and nutrient surplus and deficiency. In implementation, based on elements generated from the first farmland twin model (such as the correlation matrix between vegetation indices and meteorological factors), a grading threshold can be set (e.g., dividing crop growth areas by vegetation indices or defining meteorological disaster areas by combining disaster-causing thresholds) to achieve preliminary regional division. Clustering algorithms (such as weighted clustering based on adjacency matrices or topological fusion methods) can be used to integrate these grading regions across indicators. By extracting geographic coordinates to construct spatial relationships and weighting and optimizing cluster boundaries, a unified farmland indicator regional field is formed. This method dynamically adapts to the overlap and conflict of different indicators, making regional division both globally consistent and locally specific, thereby improving the accuracy and practicality of the regional field and providing spatial support for precise decision-making in smart agriculture.

[0071] S04. Construct a parameter update mechanism for the regional field of farmland indicators, and embed the parameter update mechanism into the first farmland twin model to obtain the second farmland twin model.

[0072] The parameter update mechanism can include periodic update mechanisms and event-triggered update mechanisms. In implementation, it can extract parameter change characteristics (such as the rate of parameter change and the magnitude of abrupt changes) based on the dynamic parameter set of the farmland indicator regional field (such as changes in crop growth and soil physicochemical indicators). Statistical methods (such as calculating update frequency using statistical time windows to quantify update rules, and combining this with the topological relationship of the first farmland twin model to determine the regional field update priority) are used to generate periodic batch updates and event-driven instant update logic. These mechanisms are then embedded into the first farmland twin model to obtain the second farmland twin model. Alternatively, machine learning-assisted threshold adaptation or distributed computing optimization can be used to enable the model to dynamically respond to changes in farmland status. This method improves the real-time performance and accuracy of the twin model through the intelligent embedding of the parameter update mechanism.

[0073] In one embodiment, the multi-source farmland monitoring data includes multi-source remote sensing data, meteorological observation data, soil sensor monitoring data, and basic soil data;

[0074] Based on multi-source farmland monitoring data, a farmland monitoring feature set is extracted, including:

[0075] S01.1 Perform geometric radiometric correction on multi-source remote sensing data to obtain calibrated remote sensing data, and extract remote sensing features based on the calibrated remote sensing data to obtain a remote sensing feature set, which includes spectral features and texture features.

[0076] S01.2, Based on meteorological observation data, extract meteorological time-series features to obtain a meteorological feature set; the meteorological time-series features include temperature time-series features, humidity time-series features, precipitation time-series features, and sunshine time-series features;

[0077] S01.3 Based on soil sensor monitoring data, extract soil monitoring time-series features to obtain a soil monitoring feature set. The soil monitoring time-series features include soil physical property time-series features and soil electrochemical property time-series features.

[0078] S01.4, Based on basic soil data, extract the basic soil property set using principal component analysis;

[0079] S01.5 integrates the remote sensing feature set, meteorological feature set, soil monitoring feature set, and soil basic property set to obtain the farmland monitoring feature set.

[0080] Specifically, multi-source farmland monitoring data can include multi-source remote sensing data, meteorological observation data, soil sensor monitoring data, and basic soil data. In implementation, geometric radiometric correction can be performed on multi-source remote sensing data, such as using the Sen2Cor algorithm for atmospheric correction and geometric calibration to eliminate environmental interference and obtain calibrated remote sensing data. Based on the calibrated data, spectral and textural features can be extracted using methods such as gray-level co-occurrence matrix to form a remote sensing feature set. Based on meteorological observation data, time-series analysis techniques can be used to extract the temporal characteristics of temperature, humidity, precipitation, and illumination to construct a meteorological feature set. For soil sensor monitoring data, empirical mode decomposition can be used to process the time-series sequences of soil physical and electrochemical properties, extracting soil monitoring time-series features to form a soil monitoring feature set. Based on basic soil data, principal component analysis can be used to reduce dimensionality and extract key basic soil properties, such as organic matter content or texture classification, to simplify data dimensionality. Remote sensing feature sets, meteorological feature sets, soil monitoring feature sets, and soil basic property sets can be integrated through multi-dimensional alignment techniques. For example, spatiotemporal registration and feature standardization can be used to ensure data consistency and generate high-dimensional farmland monitoring feature sets, providing a comprehensive and fusionable data foundation for subsequent model construction.

[0081] In one embodiment, based on a farmland monitoring feature set, a first farmland twin model is constructed using a joint modeling method based on a machine learning model, including:

[0082] S02.1 Calculate vegetation index based on remote sensing feature set;

[0083] S02.2, Based on the meteorological feature set, a long short-term memory network is used to construct the meteorological factor correlation matrix;

[0084] S02.3, Based on the soil monitoring feature set, the empirical mode decomposition method is used to extract the soil physicochemical change characteristics;

[0085] S02.4 aligns vegetation indices, meteorological factor correlation matrices, soil physicochemical change characteristics, and soil basic property sets in multiple dimensions to obtain basic farmland elements;

[0086] S02.5 Based on basic farmland elements, a graph neural network is used to mine spatiotemporal association rules to obtain the topological relationships between basic farmland elements;

[0087] S02.6, Based on topological relationships and combined with basic farmland elements, an initial twin feature matrix is ​​generated;

[0088] S02.7, Obtain historical farmland basic elements, and based on historical farmland basic elements, use the backpropagation algorithm to optimize the weight parameters of the initial twin feature matrix to obtain the optimized twin feature matrix;

[0089] S02.8, Based on the optimized twin feature matrix, the first farmland twin model is constructed.

[0090] For example, vegetation indices can be calculated based on remote sensing feature sets. Spectral data can be processed using formulas for the Normalized Difference Vegetation Index (NDVI) or the Enhanced Vegetation Index (EVI) to quantify crop growth status. Based on meteorological feature sets, a long short-term memory network (LSTM) can be used to construct a meteorological factor correlation matrix. The time-series modeling capability of LSTM can be utilized to analyze the long-term dependencies of data such as temperature and humidity, generating a matrix structure reflecting the interaction of meteorological elements. Based on soil monitoring feature sets, empirical mode decomposition (EMD) can be used to extract soil physicochemical change characteristics. Intrinsic mode functions of soil physical properties (such as water content) and electrochemical properties (such as conductivity) can be separated using signal decomposition techniques to capture dynamic change patterns. By aligning vegetation indices, meteorological factor correlation matrices, soil physicochemical change characteristics, and soil basic property sets in multiple dimensions, and using spatiotemporal registration and feature standardization techniques, data scale consistency can be ensured, integrating them into basic farmland elements. Building upon this foundation, Graph Neural Networks (GNNs) can be employed for spatiotemporal association rule mining. By constructing a graph structure where nodes represent farmland elements and edges represent spatial adjacency relationships, message passing algorithms are applied to mine topological relationships among elements, revealing nonlinear interactions between regions. Based on these topological relationships and fundamental farmland elements, an initial twin feature matrix is ​​generated, such as by fusing the weights of each element through matrix multiplication. After obtaining historical farmland elements, the weight parameters of the initial twin feature matrix can be optimized using the backpropagation algorithm. Gradient descent is then used to minimize the prediction error, and network parameters are adjusted to improve the model's generalization ability, resulting in an optimized twin feature matrix. Based on the optimized twin feature matrix, a first farmland twin model is constructed. An encoder-decoder structure is used to map the physical farmland to the virtual model, and the model can be deployed using software tools (such as Python's PyTorch library), enabling the model to simulate farmland conditions in real time and providing a high-fidelity foundation for smart agriculture decision-making.

[0091] In one embodiment, the farmland index evaluation system includes crop growth index, meteorological disaster index, pest and disease stress index, farmland pollution index, and nutrient surplus and deficit index;

[0092] Based on the first farmland twin model, a farmland index evaluation system is used for regional clustering to obtain a farmland index regional field, including:

[0093] S03.1, Based on crop growth indicators and vegetation indices, crop growth grading zones are divided according to the vegetation index grading thresholds.

[0094] S03.2, Based on crop growth grading regions and meteorological disaster indicators, determine the disaster-causing meteorological grading threshold for disasters caused by meteorological disasters in corresponding crop growth grading regions, and divide meteorological disaster grading regions based on the disaster-causing meteorological grading threshold and meteorological factor correlation matrix;

[0095] S03.3, based on crop growth grading areas and nutrient surplus / deficient indicators, determine crop nutrient grading requirements, and based on crop nutrient grading requirements, soil physicochemical change characteristics and soil basic property set, divide nutrient surplus / deficient grading areas.

[0096] S03.4 Based on crop growth grading regions and pest and disease stress indicators, determine pest and disease suitable environment grading parameters, and divide pest and disease stress grading regions based on pest and disease suitable environment grading parameters, meteorological factor correlation matrix, soil physicochemical change characteristics and soil basic property set.

[0097] S03.5, determine the pollution classification threshold based on farmland pollution indicators, and divide farmland pollution classification areas based on pollution classification thresholds, soil physicochemical change characteristics and soil basic property set;

[0098] S03.6, cross-regional clustering is performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus and deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain farmland indicator regional fields.

[0099] Specifically, based on crop growth indicators and vegetation indices, spatial division can be performed using geographic information system (GIS) tools according to preset vegetation index grading thresholds (e.g., NDVI values ​​less than 0.3 are classified as poor growth areas, 0.3-0.6 as moderate areas, and greater than 0.6 as good areas), generating crop growth grading regions. Based on crop growth grading regions and meteorological disaster indicators, disaster-causing meteorological grading thresholds can be determined by analyzing historical disaster data (e.g., continuous high temperatures exceeding 35 degrees Celsius or precipitation less than 5 mm as disaster thresholds), and spatial interpolation algorithms can be applied to divide meteorological disaster grading regions in conjunction with the meteorological factor correlation matrix (time-series correlation results output by the LSTM network). Based on crop growth grading regions and nutrient surplus / deficient indicators, nutrient grading requirements can be determined according to crop growth models (e.g., NPK demand curves), and soil physicochemical change characteristics (e.g., nutrient fluctuations after EMD decomposition) and soil basic property sets (key parameters after dimensionality reduction by principal component analysis) can be integrated to divide nutrient surplus / deficient grading regions using a threshold segmentation method. Similarly, based on crop growth grading regions and pest and disease stress indicators, environmental suitability models (such as humidity greater than 80% and temperature 20-30 degrees Celsius) are used to determine pest and disease stress grading parameters. These parameters are then combined with meteorological factor correlation matrices, soil characteristics, and basic property sets. Spatial overlay analysis is used to delineate pest and disease stress grading regions. Simultaneously, pollution grading thresholds can be set based on farmland pollution indicators (such as heavy metal content limits), and buffer zone analysis is used to delineate farmland pollution grading regions based on soil change characteristics and basic property sets. All these grading regions are then clustered across regions: spatial geographic coordinates of each region are extracted to construct an adjacency matrix. Weights are assigned according to the grading level (e.g., regions with good growth have higher weights), and spatial clustering algorithms (such as DBSCAN or K-means) are used for cluster division. The initial clustering results are boundary-corrected using grading thresholds (e.g., morphological operations to smooth edges), and optimized cluster regions are topologically fused (e.g., Voron graph integration) to generate a unified farmland indicator regional field.

[0100] In one embodiment, the parameter update mechanism includes a periodic update mechanism and an event-triggered update mechanism;

[0101] A parameter update mechanism for the regional field of farmland indicators is constructed, and this mechanism is embedded into the first farmland twin model to obtain the second farmland twin model, including:

[0102] S04.1 Extract the dynamic parameter set of each farmland indicator area field;

[0103] S04.2, Based on the dynamic parameter set, extract parameter change features, including parameter change rate and parameter abrupt change magnitude;

[0104] S04.3, Based on the rate of parameter change, determine the update frequency of each farmland indicator area field;

[0105] S04.4, Based on the parameter mutation amplitude, determine the immediate update threshold of each farmland indicator area field;

[0106] S04.5, Based on the topological relationship of the first farmland twin model, determine the update priority of each farmland indicator area field;

[0107] S04.6, Generate a periodic update mechanism based on update priority and update frequency;

[0108] S04.7, Based on update priority and immediate update threshold, generate an event-triggered update mechanism;

[0109] S04.8, embed the periodic update mechanism and the event-triggered update mechanism into the first farmland twin model to obtain the second farmland twin model.

[0110] For example, the parameter update mechanism includes a periodic update mechanism and an event-triggered update mechanism, used to improve the real-time performance and accuracy of the model through dynamic adjustments. In implementation, parameter change characteristics, such as the rate of parameter change and the magnitude of parameter mutations, can be extracted based on the dynamic parameter set. The rate of parameter change is quantified by calculating the rate of change using time-series data (such as the difference value within a sliding window); the magnitude of parameter mutations is used to identify abnormal fluctuations by monitoring the standard deviation or range of the data. The update frequency of each regional field can be determined based on the rate of parameter change. A formula is used to calculate the update frequency, which integrates the local velocity sensitivity coefficient, statistical time window, global velocity mean, and standard deviation, and is adaptively adjusted through an exponential term to prioritize updates for highly dynamic regions. An immediate update threshold is set based on the magnitude of parameter mutations, such as triggering an immediate update when the mutation magnitude exceeds a specific multiple of the historical benchmark. The update priority of each regional field can be determined based on the topological relationships of the first farmland twin model (such as spatial connectivity weights mined through graph neural networks), prioritizing critical or highly interconnected regions. A periodic update mechanism is generated by combining update priority and update frequency, such as a batch parameter adjustment update mechanism executed at a set frequency through scheduled tasks; an event-triggered update mechanism is generated by combining priority and immediate update threshold, enabling driven updates when the threshold is exceeded. The periodic update and event-triggered update mechanisms are embedded into the first farmland twin model, which can be achieved through software middleware or API integration, allowing the model to dynamically respond to changes in farmland status, resulting in a second farmland twin model. This process, through the quantification of parameter change characteristics, priority allocation, and mechanism fusion, makes model updates intelligent and effective, supporting refined decision-making in smart agriculture.

[0111] In one embodiment, S04.3.1, the formula for calculating the update frequency of each farmland index area field based on the parameter change rate is as follows:

[0112]

[0113] in, Let be the update frequency of the i-th farmland index area field. Let be the velocity sensitivity coefficient of the i-th farmland index area field. For the statistical time window of farmland dynamic parameters, Let be the rate of change of parameters in the i-th farmland index area field. The mean rate of change of parameters of all farmland indicators within the target farmland area. Let the standard deviation of the rate of change of parameters in the regional field of all farmland indicators within the target farmland area be denoted as . It is a minimal positive real number.

[0114] Specifically, the calculation formula for determining the update frequency of each farmland index area field based on the parameter change rate is as follows: This method enables adaptive parameter updates, automatically adjusting the model update rhythm by quantifying dynamic changes in farmland, thereby improving the real-time performance and accuracy of the digital twin model. The overall formula maps changes in farmland state to an operable update frequency through mathematical relationships, where there is a synergistic effect between variables: As a velocity-sensitive coefficient, it is calibrated based on historical data to assign field weights to different regions, reflecting local specificity; Provide a time scale benchmark for statistical time windows (such as 7 days); It is the rate of change of parameters in the i-th region field, which is obtained by calculating the sliding window difference of time series data (such as crop growth indicators) and can directly reflect the intensity of dynamic changes in farmland status. and These are the mean and standard deviation of the global rate of change, respectively, used for normalization to make the formula adaptable to the overall farmland variability; The numerator should be a very small positive real number (e.g., 1e-5) to prevent the denominator from being zero and ensure numerical stability. Structurally, the numerator... Emphasizing the product of the rate of change and the time window, the denominator... Perform global scaling, exponential term A nonlinear adjustment is introduced to make the update frequency increase exponentially as the rate of change deviates from the mean, prioritizing responses to highly dynamic areas. The formula as a whole is based on the dynamic parameter set in the parameter update mechanism (such as real-time monitoring data extracted from farmland indicator regional fields), and generates a periodic update mechanism by combining it with update priority. For example, in the software implementation, the output frequency of this formula is used to drive a timed task, which is embedded in the event loop of the first farmland twin model. It works in conjunction with the event-triggered update mechanism (based on an immediate update threshold) to enable the second farmland twin model to dynamically adapt to changes in farmland status, supporting the refined decision-making of smart agriculture.

[0115] In one embodiment, cross-regional clustering is performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus / deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain a farmland index regional field, including:

[0116] S03.6.1 Based on crop growth classification regions, meteorological disaster classification regions, nutrient surplus and deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions, the spatial geographic coordinates of each classification region are extracted, and a spatial adjacency matrix is ​​constructed based on the spatial geographic coordinates.

[0117] S03.6.2, Based on the hierarchical level corresponding to each hierarchical region, assign regional weights to the spatial adjacency matrix to obtain the weighted spatial adjacency matrix;

[0118] S03.6.3, a spatial clustering algorithm is used to divide the weighted spatial adjacency matrix into regional clusters to obtain the initial farmland index cluster regions;

[0119] S03.6.4, Combining the grading thresholds corresponding to each grading region, the boundaries of the initial farmland index cluster regions are corrected to obtain the optimized cluster regions;

[0120] S03.6.5, Spatial topological fusion is performed on the optimized cluster regions to obtain the farmland index regional field.

[0121] Specifically, cross-regional clustering can be performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus / deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain a farmland indicator regional field. Based on the spatial geographic coordinates of each classification region, the boundary point coordinates of each region are extracted using Geographic Information System (GIS) tools, and a spatial adjacency matrix is ​​constructed. For example, Euclidean distance can be used to calculate the proximity between regions, forming a binary or distance-weighted adjacency matrix to quantify spatial relationships. According to the classification level corresponding to each classification region (e.g., crop growth is divided into excellent, medium, and poor levels, with weights of 3, 2, and 1 respectively), regional weights are assigned to the spatial adjacency matrix. A linear weighting method (e.g., matrix elements multiplied by level weights) can be used to obtain a weighted spatial adjacency matrix, highlighting the influence of high-priority regions. Spatial clustering algorithms (such as density-based DBSCAN or partitioning K-means) can be used to divide the weighted spatial adjacency matrix into regional clusters. By setting clustering parameters (such as the eps distance threshold and minimum sample size for DBSCAN), spatially dense regions are identified, resulting in initial farmland indicator cluster regions. Combining the corresponding grading thresholds for each grading region (such as vegetation index thresholds or pollution concentration limits), the boundaries of the initial cluster regions are corrected. Morphological operations (such as opening or closing operations) can be used to smooth the boundaries, or threshold overlap analysis can be used to adjust the cluster outlines to ensure regional consistency, resulting in optimized cluster regions. Spatial topological fusion is then performed on the optimized cluster regions. Voronoi diagrams or Delaunay triangulation techniques can be applied to integrate adjacent regions, eliminating regional overlaps and gaps. Tools such as QGIS and PostGIS are used to complete spatial merging, generating a farmland indicator regional field. By improving the accuracy and practicality of regional division, dynamic spatial support is provided for smart agriculture decision-making.

[0122] The aforementioned method for constructing a digital twin model of farmland acquires multi-source farmland monitoring data of the target farmland area and extracts farmland monitoring feature sets. It employs techniques such as geometric radiometric correction, temporal feature extraction, and principal component analysis to overcome the limitations of traditional single remote sensing data sources. This achieves deep fusion of multi-source data, including remote sensing, meteorological, and soil data, expanding the model's coverage and enhancing feature diversity. It also addresses the problem of insufficient model representation capabilities caused by single data sources in existing technologies. Based on this farmland monitoring feature set, a first farmland twin model is constructed using a machine learning joint modeling method. This involves calculating vegetation indices, constructing meteorological factor correlation matrices using long short-term memory networks, extracting soil physicochemical change characteristics through empirical mode decomposition, and utilizing graph neural networks for spatiotemporal correlation rule mining to reveal nonlinear topological relationships between farmland elements. This improves the model's representation accuracy of complex farmland environments and overcomes the shortcomings of superficial correlation analysis in traditional methods. This paper utilizes a farmland indicator evaluation system to perform regional clustering on the first farmland twin model, resulting in a farmland indicator regional field. Based on indicators such as crop growth, meteorological disasters, nutrient surplus / deficiency, pest and disease stress, and farmland pollution, grading thresholds are set. Cross-regional integration is achieved through spatial adjacency matrix weighting and clustering algorithms, balancing overall coverage with localized fine-grained segmentation, thus addressing the challenge of balancing model accuracy and coverage. A parameter update mechanism for the farmland indicator regional field is constructed and embedded into the first farmland twin model to obtain a second farmland twin model. This mechanism extracts the rate of change and abrupt change of dynamic parameter sets, determines update priorities based on topological relationships, and uses an adaptive formula including periodic and event-triggered updates to dynamically adjust parameters, improving the twin model's response speed and fidelity to changes in farmland status, supporting refined decision-making and management in smart agriculture. This method, through the orderly integration of multi-source data fusion, intelligent algorithm mining, regional clustering, and dynamic update mechanisms, significantly improves the real-time performance, accuracy, and practicality of digital twin models, providing a reliable technical foundation for the dynamic needs of farmland management.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] Based on the same inventive concept, this application also provides a farmland digital twin model construction apparatus for implementing the above-mentioned farmland digital twin model construction method. The solution provided by this apparatus is similar to the implementation scheme described in the above-described method. Therefore, the specific limitations of one or more embodiments of the farmland digital twin model construction apparatus provided below can be found in the limitations of the farmland digital twin model construction method described above, and will not be repeated here.

[0125] In one exemplary embodiment, such as Figure 2 As shown, a device for constructing a digital twin model of farmland is provided, comprising:

[0126] The multi-source data acquisition module 101 is used to acquire multi-source farmland monitoring data of the target farmland area and extract farmland monitoring feature sets based on the multi-source farmland monitoring data;

[0127] The model joint construction module 102 is used to construct the first farmland twin model based on the farmland monitoring feature set and the joint modeling method based on the machine learning model.

[0128] The regional clustering module 103 is used to perform regional clustering based on the first farmland twin model and the farmland index evaluation system to obtain the farmland index regional field.

[0129] The model mechanism embedding module 104 is used to construct the parameter update mechanism of the farmland indicator regional field and embed the parameter update mechanism into the first farmland twin model to obtain the second farmland twin model.

[0130] In one embodiment, the multi-source data acquisition module 101 includes multi-source farmland monitoring data, meteorological observation data, soil sensor monitoring data, and basic soil data.

[0131] The multi-source data acquisition module 101 is also used for:

[0132] Geometric radiometric correction is performed on multi-source remote sensing data to obtain calibrated remote sensing data. Based on the calibrated remote sensing data, remote sensing features are extracted to obtain a remote sensing feature set, which includes spectral features and texture features.

[0133] Based on meteorological observation data, meteorological time-series features are extracted to obtain a meteorological feature set; the meteorological time-series features include temperature time-series features, humidity time-series features, precipitation time-series features, and sunshine time-series features;

[0134] Based on soil sensor monitoring data, soil monitoring time-series features are extracted to obtain a soil monitoring feature set, which includes soil physical property time-series features and soil electrochemical property time-series features.

[0135] Based on basic soil data, principal component analysis was used to extract a set of basic soil properties.

[0136] By integrating remote sensing feature sets, meteorological feature sets, soil monitoring feature sets, and soil basic property sets, a farmland monitoring feature set is obtained.

[0137] In one embodiment, the model federation construction module 102 is further configured to:

[0138] Calculate vegetation indices based on remote sensing feature sets;

[0139] Based on the meteorological feature set, a long short-term memory network is used to construct the meteorological factor correlation matrix;

[0140] Based on the soil monitoring feature set, the empirical mode decomposition method was used to extract the soil physicochemical change characteristics;

[0141] By aligning vegetation indices, meteorological factor correlation matrices, soil physicochemical change characteristics, and soil basic property sets in multiple dimensions, basic farmland elements are obtained.

[0142] Based on basic farmland elements, a graph neural network is used to mine spatiotemporal association rules to obtain the topological relationships between basic farmland elements.

[0143] Based on topological relationships and combined with basic farmland elements, an initial twin feature matrix is ​​generated;

[0144] Historical farmland basic elements are obtained, and based on these elements, the weight parameters of the initial twin feature matrix are optimized using the backpropagation algorithm to obtain the optimized twin feature matrix.

[0145] Based on the optimized twin feature matrix, a twin model of the first farmland was constructed.

[0146] In one embodiment, the farmland index evaluation system in the regional clustering module 103 includes crop growth index, meteorological disaster index, pest and disease stress index, farmland pollution index and nutrient surplus and deficit index.

[0147] The region clustering module 103 is also used for:

[0148] Based on crop growth indicators and vegetation indices, crop growth grading zones are divided according to the vegetation index grading thresholds.

[0149] Based on crop growth grading regions and meteorological disaster indicators, the disaster-causing meteorological grading thresholds for disasters caused by meteorological disasters in corresponding crop growth grading regions are determined, and meteorological disaster grading regions are divided based on the disaster-causing meteorological grading thresholds and the meteorological factor correlation matrix.

[0150] Based on crop growth grading regions and nutrient surplus / deficient indicators, crop nutrient grading requirements are determined, and based on crop nutrient grading requirements, soil physicochemical change characteristics, and soil basic property set, nutrient surplus / deficient grading regions are divided.

[0151] Based on crop growth grading regions and pest and disease stress indicators, pest and disease suitable environment grading parameters are determined, and pest and disease stress grading regions are divided based on pest and disease suitable environment grading parameters, meteorological factor correlation matrix, soil physicochemical change characteristics and soil basic property set.

[0152] Pollution classification thresholds are determined based on farmland pollution indicators, and farmland pollution classification areas are divided based on pollution classification thresholds, soil physicochemical change characteristics, and soil basic property sets.

[0153] Cross-regional clustering was performed on crop growth classification regions, meteorological disaster classification regions, nutrient surplus / deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions to obtain farmland indicator regional fields.

[0154] In one embodiment, the parameter update mechanism in the model mechanism embedding module 104 includes a periodic update mechanism and an event-triggered update mechanism.

[0155] The model mechanism embedding module 104 is also used for:

[0156] Extract the dynamic parameter set of each farmland indicator area;

[0157] Based on the dynamic parameter set, parameter change features are extracted, including the parameter change rate and the parameter abrupt change magnitude.

[0158] Based on the rate of parameter change, the update frequency of each farmland indicator area field is determined;

[0159] Based on the magnitude of parameter mutations, the immediate update threshold for each farmland indicator area field is determined.

[0160] Based on the topological relationship of the first farmland twin model, the update priority of each farmland indicator area field is determined;

[0161] A periodic update mechanism is generated based on update priority and update frequency.

[0162] An event-triggered update mechanism is generated based on update priority and immediate update threshold.

[0163] By embedding the periodic update mechanism and the event-triggered update mechanism into the first farmland twin model, the second farmland twin model is obtained.

[0164] In one embodiment, the calculation formula for determining the update frequency of each farmland indicator area field based on the parameter change rate in the model mechanism embedding module 104 is as follows:

[0165]

[0166] in, Let be the update frequency of the i-th farmland index area field. Let be the velocity sensitivity coefficient of the i-th farmland index area field. For the statistical time window of farmland dynamic parameters, Let be the rate of change of parameters in the i-th farmland index area field. The mean rate of change of parameters of all farmland indicators within the target farmland area. Let the standard deviation of the rate of change of parameters in the regional field of all farmland indicators within the target farmland area be denoted as . It is a minimal positive real number.

[0167] In one embodiment, the region clustering partitioning module 103 is further configured to:

[0168] Based on crop growth classification regions, meteorological disaster classification regions, nutrient surplus and deficiency classification regions, pest and disease stress classification regions, and farmland pollution classification regions, spatial geographic coordinates of each classification region are extracted, and a spatial adjacency matrix is ​​constructed based on the spatial geographic coordinates.

[0169] Based on the hierarchical level corresponding to each hierarchical region, the spatial adjacency relation matrix is ​​weighted by the region to obtain the weighted spatial adjacency matrix.

[0170] A spatial clustering algorithm is used to divide the weighted spatial adjacency matrix into regional clusters to obtain the initial farmland index cluster regions;

[0171] By combining the grading thresholds corresponding to each grading region, the boundaries of the initial farmland index cluster regions are corrected to obtain the optimized cluster regions.

[0172] Spatial topological fusion of optimized cluster regions yields farmland index regional fields.

[0173] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method for constructing a digital twin model of farmland as described above.

[0174] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0175] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0176] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for constructing a digital twin model of farmland, characterized in that, The method includes: Acquire multi-source farmland monitoring data for the target farmland area, and extract farmland monitoring feature sets based on the multi-source farmland monitoring data; Based on the farmland monitoring feature set, a first farmland twin model is constructed using a joint modeling method based on machine learning models. Based on the first farmland twin model, a farmland index evaluation system is used to perform regional clustering to obtain the farmland index regional field. A parameter update mechanism for the farmland index regional field is constructed, and the parameter update mechanism is embedded into the first farmland twin model to obtain the second farmland twin model.

2. The method according to claim 1, characterized in that, The multi-source farmland monitoring data includes multi-source remote sensing data, meteorological observation data, soil sensor monitoring data, and basic soil data; The extraction of farmland monitoring feature sets based on the multi-source farmland monitoring data includes: Geometric radiometric correction is performed on the multi-source remote sensing data to obtain calibrated remote sensing data. Based on the calibrated remote sensing data, remote sensing features are extracted to obtain a remote sensing feature set, which includes spectral features and texture features. Based on the meteorological observation data, meteorological time-series features are extracted to obtain a meteorological feature set; the meteorological time-series features include temperature time-series features, humidity time-series features, precipitation time-series features, and sunshine time-series features; Based on the soil sensing and monitoring data, soil monitoring time-series features are extracted to obtain a soil monitoring feature set, which includes soil physical property time-series features and soil electrochemical property time-series features. Based on the aforementioned basic soil data, a set of basic soil properties was extracted using principal component analysis. The farmland monitoring feature set is obtained by integrating the remote sensing feature set, the meteorological feature set, the soil monitoring feature set, and the soil basic property set.

3. The method according to claim 2, characterized in that, The construction of a first farmland twin model based on the farmland monitoring feature set and employing a joint modeling method based on machine learning models includes: Based on the remote sensing feature set, calculate the vegetation index; Based on the meteorological feature set, a meteorological factor correlation matrix is ​​constructed using a long short-term memory network; Based on the aforementioned soil monitoring feature set, the empirical mode decomposition method was used to extract soil physicochemical change characteristics; By aligning the vegetation index, the meteorological factor correlation matrix, the soil physicochemical change characteristics, and the soil basic property set in multiple dimensions, the basic elements of farmland are obtained. Based on the basic farmland elements, a graph neural network is used to mine spatiotemporal association rules to obtain the topological relationships between the basic farmland elements. Based on the aforementioned topological relationships and combined with the basic farmland elements, an initial twin feature matrix is ​​generated; Historical farmland basic elements are obtained, and based on the historical farmland basic elements, the weight parameters of the initial twin feature matrix are optimized using the backpropagation algorithm to obtain the optimized twin feature matrix; Based on the optimized twin feature matrix, the first farmland twin model is constructed.

4. The method according to claim 3, characterized in that, The farmland index evaluation system includes crop growth index, meteorological disaster index, pest and disease stress index, farmland pollution index, and nutrient surplus and deficit index. Based on the first farmland twin model, a farmland index evaluation system is used for regional clustering to obtain a farmland index regional field, including: Based on the crop growth index and the vegetation index, the crop growth grading areas are divided according to the vegetation index grading threshold. Based on the crop growth grading region and the meteorological disaster index, the disaster-causing meteorological grading threshold for the disaster caused by the corresponding crop growth grading region is determined, and the meteorological disaster grading region is divided based on the disaster-causing meteorological grading threshold and the meteorological factor correlation matrix. Based on the crop growth grading regions and the nutrient surplus / deficient index, the crop nutrient grading requirements are determined, and based on the crop nutrient grading requirements, the soil physicochemical change characteristics, and the soil basic property set, nutrient surplus / deficient grading regions are divided. Based on the crop growth grading regions and the pest and disease stress indicators, pest and disease suitable environment grading parameters are determined, and pest and disease stress grading regions are divided based on the pest and disease suitable environment grading parameters, the meteorological factor correlation matrix, the soil physicochemical change characteristics, and the soil basic property set. Based on the farmland pollution indicators, pollution classification thresholds are determined, and farmland pollution classification areas are divided based on the pollution classification thresholds, the soil physicochemical change characteristics, and the soil basic property set. The crop growth classification region, the meteorological disaster classification region, the nutrient surplus / deficiency classification region, the pest and disease stress classification region, and the farmland pollution classification region are clustered across regions to obtain the farmland index regional field.

5. The method according to claim 3, characterized in that, The parameter update mechanism includes a periodic update mechanism and an event-triggered update mechanism; The construction of the parameter update mechanism for the farmland index regional field, and the embedding of the parameter update mechanism into the first farmland twin model to obtain the second farmland twin model, includes: Extract the dynamic parameter set of each of the farmland index regional fields; Based on the dynamic parameter set, parameter change features are extracted, including parameter change rate and parameter abrupt change magnitude; Based on the rate of change of the parameters, the update frequency of each of the farmland index areas is determined; Based on the magnitude of the parameter mutation, the immediate update threshold for each of the farmland index regional fields is determined; Based on the topological relationship of the first farmland twin model, the update priority of each farmland indicator regional field is determined; The periodic update mechanism is generated based on the update priority and the update frequency; The event-triggered update mechanism is generated based on the update priority and the immediate update threshold. The periodic update mechanism and the event-triggered update mechanism are embedded into the first farmland twin model to obtain the second farmland twin model.

6. The method according to claim 5, characterized in that, The formula for determining the update frequency of each farmland index area field based on the rate of change of the parameters is as follows: in, Let be the update frequency of the i-th farmland index area field. Let be the velocity sensitivity coefficient of the i-th farmland index area field. For the statistical time window of farmland dynamic parameters, Let be the rate of change of parameters in the i-th farmland index area field. The mean rate of change of parameters of all farmland indicators within the target farmland area. Let the standard deviation of the rate of change of parameters in the regional field of all farmland indicators within the target farmland area be denoted as . It is a minimal positive real number.

7. The method according to claim 4, characterized in that, The process of cross-regional clustering of the crop growth classification region, the meteorological disaster classification region, the nutrient surplus / deficiency classification region, the pest and disease stress classification region, and the farmland pollution classification region to obtain the farmland indicator regional field includes: Based on the crop growth classification area, the meteorological disaster classification area, the nutrient surplus / deficiency classification area, the pest and disease stress classification area, and the farmland pollution classification area, the spatial geographic coordinates of each classification area are extracted, and a spatial adjacency matrix is ​​constructed based on the spatial geographic coordinates. Based on the hierarchical level corresponding to each hierarchical region, the spatial adjacency relation matrix is ​​weighted to obtain a weighted spatial adjacency matrix. A spatial clustering algorithm is used to divide the weighted spatial adjacency matrix into regional clusters to obtain the initial farmland index cluster regions; By combining the grading thresholds corresponding to each grading region, the boundaries of the initial farmland index cluster regions are corrected to obtain optimized cluster regions. Spatial topological fusion is performed on the optimized cluster region to obtain the farmland index region field.

8. A device for constructing a digital twin model of farmland, characterized in that, The device includes: The multi-source data acquisition module is used to acquire multi-source farmland monitoring data of the target farmland area, and extract farmland monitoring feature sets based on the multi-source farmland monitoring data; The model joint construction module is used to construct a first farmland twin model based on the farmland monitoring feature set and a joint modeling method based on machine learning models. The regional clustering module is used to perform regional clustering based on the first farmland twin model and the farmland index evaluation system to obtain the farmland index regional field. The model mechanism embedding module is used to construct the parameter update mechanism of the farmland index regional field, and embed the parameter update mechanism into the first farmland twin model to obtain the second farmland twin model.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

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

Images