A farmland condition monitoring method based on a digital twin model

Abstract

This invention discloses a crop monitoring method based on a digital twin model, relating to the fields of smart agriculture and digital twin technology. The method includes: acquiring multi-source crop data; fusing crop data to obtain target data; obtaining source point clouds based on the target data; registering the source point clouds to obtain crop point clouds; constructing a physical crop growth model based on the crop point clouds; constructing a target mapping model based on growth space data; dividing the crop's growth cycle into several physiological stages based on the crop's morphological characteristics, with each physiological stage corresponding to an accumulated temperature range, a growth rate coefficient, and a maximum growth amount; obtaining an initial crop growth rate based on the accumulated temperature range, growth rate coefficient, and maximum growth amount; correcting the initial crop growth rate to obtain the target crop growth rate; constructing a virtual crop growth model based on the physical crop growth model, the target mapping model, and the target crop growth rate; and obtaining crop monitoring results based on the virtual crop growth model. This method addresses the shortcomings of existing crop growth models that ignore the specificity of physiological stages, leading to distortion in virtual crop growth simulation and insufficient prediction accuracy.

Description

Technical Field

[0001] This invention relates to the fields of smart agriculture and digital twin technology, specifically to a method for monitoring agricultural conditions based on a digital twin model. Background Technology

[0002] With the development of agricultural modernization and intelligentization, digital twin technology provides a new paradigm for building a virtual-physical integrated agricultural management platform. By establishing a digital twin model of farmland environment and crop growth, it is theoretically possible to achieve real-time mapping, process simulation, and decision optimization of physical systems, thereby promoting the implementation of precision agriculture.

[0003] However, the core simulation and prediction module of current agricultural digital twin systems, namely crop growth models, faces fundamental challenges in accuracy, severely restricting the system's practicality and reliability. Specifically, this manifests in the following aspects:

[0004] Traditional crop growth models (such as accumulated temperature models) typically rely solely on accumulated temperature as a single variable to drive crop development, assuming a constant response rate to accumulated temperature throughout the entire growth period. However, in reality, crops exhibit significant differences in their sensitivity to and utilization efficiency of accumulated temperature at different physiological stages (such as the transition from tillering to jointing). This leads to systematic biases in growth rate predictions during critical growth stage transitions, failing to accurately depict the dynamic characteristics of crop growth stages. Consequently, the developmental process in the virtual twin becomes asynchronous with the physical entity, resulting in distorted virtual crop growth simulations and insufficient prediction accuracy. Summary of the Invention

[0005] To address the shortcomings of existing crop growth models that ignore the specificity of physiological stages, leading to distortion in virtual crop growth simulations and insufficient prediction accuracy, this invention provides a crop condition monitoring method based on a digital twin model. The method includes:

[0006] Crop data from different data sources is acquired and fused to obtain target data. The crop data includes growth environment data, growth space data, and growth morphology data.

[0007] Based on the target data, several source point clouds are obtained, and several crop point clouds are obtained by registration of the source point clouds. Based on the crop point clouds, an entity crop growth model is constructed.

[0008] A target mapping model is constructed based on the aforementioned growth space data;

[0009] Based on the morphological characteristics of crops, the growth cycle of crops is divided into several physiological stages. Each physiological stage corresponds to an accumulated temperature range, a growth rate coefficient, and a maximum growth amount. Based on the accumulated temperature range, the growth rate coefficient, and the maximum growth amount, the initial crop growth rate is obtained.

[0010] The initial crop growth rate is corrected to obtain the target crop growth rate;

[0011] Based on the physical crop growth model, the target mapping model, and the target crop growth rate, a virtual crop growth model is constructed.

[0012] The monitoring results of crops are obtained based on the virtual crop growth model.

[0013] This method introduces crop physiological stage division and stage-specific parameters to construct an initial crop growth rate coupled with accumulated temperature drive and physiological stage regulation. The growth rate coefficient is dynamically adjusted according to the current physiological stage of the crop, making the growth rate prediction more consistent with the actual growth law of the crop, and the virtual crop growth simulation more realistic, thus improving the prediction accuracy.

[0014] Furthermore, the specific steps for fusing the crop data to obtain the target data include:

[0015] Based on the crop data, the mean square error and data consistency index of different data sources are obtained, and the basic weights of different data sources are obtained based on the mean square error and the data consistency index.

[0016] The base weights are enhanced based on the critical reproductive period and attention coefficient to obtain the target weights;

[0017] The target data is obtained by fusing crop data from different data sources based on the target weight;

[0018] The first calculation formula for obtaining the basic weights is:

[0019] ;

[0020] ;

[0021] in, Indicates data source The basic weights, Indicates data source The mean square error, Indicates data source Data consistency metrics Indicates data source The mean square error, Indicates data source Data consistency metrics and Both represent integers greater than or equal to 1. Indicates data source Measured values ​​of single-crop data, This represents the mean of all data sources;

[0022] The second calculation formula for obtaining the target weight is:

[0023] ;

[0024] in, Indicates data source The target weight, This indicates the numerical value corresponding to the critical reproductive period. Indicates data source Attention coefficient Indicates data source The basic weights, Indicates data source Attention coefficient.

[0025] Traditional weighted fusion algorithms assign fixed weights based solely on the mean squared error (MSE) of each data source, failing to consider the dynamic correlation of data under different agricultural scenarios (such as different crop growth stages and different environmental complexities). Furthermore, they cannot highlight the contribution of highly reliable data during critical periods (such as the critical growth stages of crops), resulting in insufficient robustness of the fused data in complex environments. This method introduces an adaptive weight adjustment mechanism and an attention mechanism. On the one hand, it dynamically updates the weights based on the real-time quality of the data. On the other hand, it assigns higher attention weights to data during critical growth stages of crops (such as the grain-filling stage and the tillering stage), realizing a fusion logic that adapts to the scenario and strengthens key information, thereby improving the accuracy of the fused data in supporting agricultural production decisions.

[0026] Furthermore, the specific steps for obtaining the crop point cloud include:

[0027] The local curvature features and texture features of the source point cloud are obtained, and a feature descriptor is obtained based on the local curvature features and texture features;

[0028] The source point cloud and the target point cloud are matched to obtain an initial transformation matrix, and the source point cloud is aligned to the target coordinate system of the target point cloud based on the initial transformation matrix;

[0029] Based on the feature weight factor, the feature descriptor, and the objective function, the source point cloud and the target point cloud are aligned to obtain the optimal transformation matrix, and the position of the source point cloud is updated based on the optimal transformation matrix to obtain the crop point cloud;

[0030] The third formula for calculating the objective function is:

[0031] ;

[0032] in, Indicates the number of source point clouds. Represents source point cloud coordinates Represents the target point cloud coordinates Indicates the feature weight factor. Represents source point cloud Feature descriptors, Represents the target point cloud Feature descriptors, Represents an integer greater than or equal to 1.

[0033] Traditional ICP algorithms rely solely on the spatial location information of point clouds for iterative registration, neglecting the unique characteristics of agricultural point cloud data (such as crop leaf edges and straw textures). This leads to registration drift in densely cropped areas (such as rice paddies and wheat fields) and slow iterative convergence (especially when the amount of point cloud data is large). This method introduces local feature constraints of crop point clouds (such as curvature and texture features), combining feature matching with spatial location matching to construct a two-stage registration logic of feature pre-registration and spatial fine registration. This improves registration accuracy and accelerates convergence.

[0034] Furthermore, the specific steps for constructing an entity crop growth model based on the crop point cloud include:

[0035] The crop point cloud is divided into several coarse-scale point clouds and several fine-scale point clouds based on a preset scale.

[0036] The Poisson equation is solved on the coarse-scale point cloud and the fine-scale point cloud based on a low-resolution voxel mesh to obtain the coarse-scale model.

[0037] The Poisson equation is solved for the fine-scale point cloud in the coarse-scale model based on a high-resolution voxel mesh to obtain the fine-scale model.

[0038] The coarse-scale model and the fine-scale model are combined to obtain the actual crop growth model.

[0039] Traditional Poisson reconstruction algorithms are prone to detail loss or excessive smoothing when dealing with small components of crops (such as young leaves and delicate roots), and are sensitive to noise, resulting in the reconstruction model failing to accurately reproduce the microscopic growth state of crops and making it difficult to further reduce the fitting error. This method introduces a multi-scale reconstruction strategy and detail enhancement mechanism. First, the overall morphology of the crop is constructed through coarse-scale reconstruction, and then the microscopic details are restored through fine-scale reconstruction. At the same time, point cloud noise hierarchical suppression technology is combined to reduce noise interference while preserving details, thereby achieving ultra-high fidelity reconstruction of crop morphology.

[0040] Furthermore, the specific steps for constructing the target mapping model include:

[0041] Based on the growth space data, the spatial locations of the actual planting area and the virtual planting area are registered to obtain an initial mapping model;

[0042] The nonlinear region of the initial mapping model is obtained based on the growth space data, and the nonlinear region is compensated based on the nonlinear compensation operator to obtain the target mapping model.

[0043] The fourth calculation formula of the target mapping model is:

[0044] ;

[0045] in, This represents the original 3D input coordinates of the target mapping model. It is a target mapping model that represents the original 3D input coordinates. The virtual space mapping coordinates obtained after linear affine transformation and nonlinear compensation It is the initial mapping model, which directly acts on the original 3D input coordinates to achieve linear matching of the overall space. This indicates the number of characteristic control points within the nonlinear region. The compensation weights can be obtained by minimizing the control point mapping error using the least squares method. Represents radial basis functions. Represents the first term within the nonlinear region. The original coordinates of the feature control points This represents the index of the feature control points within the nonlinear region.

[0046] Traditional affine transformations can only handle linear spatial mapping relationships, while agricultural planting areas may have nonlinear spatial characteristics such as topographic relief (e.g., mountain terraces) and irregular planting boundaries. This leads to a significant increase in the mapping error between the virtual model and the real location under complex terrain, which cannot meet the spatial matching requirements of precision agriculture. This method adopts a hybrid mapping strategy of linear affine transformation and nonlinear compensation. First, linear matching of the overall space is achieved through affine transformation, and then local nonlinear compensation is performed for nonlinear areas such as topographic relief and irregular boundaries to ensure the spatial mapping accuracy under complex scenarios.

[0047] Furthermore, the fifth calculation formula for obtaining the initial crop growth rate is:

[0048] ;

[0049] ;

[0050] in, Indicates the initial crop growth rate. Indicates physiological stage Maximum growth Indicates physiological stage The growth rate coefficient, Indicates physiological stage The cumulative temperature, Indicates physiological stage The basic temperature for crop growth Indicates physiological stage The end of the accumulated temperature, Indicates physiological stage Inner Time temperature, Indicates physiological stage The growth cycle, and Both represent integers greater than or equal to 1.

[0051] Furthermore, the specific steps for obtaining the growth rate of the target crop include:

[0052] Based on environmental factors, key factors are obtained, and interaction terms between any two of the key factors are acquired.

[0053] A growth status index is obtained based on the real-time growth of the crop and the maximum growth, and a growth status feedback coefficient is obtained based on the growth status index.

[0054] The initial crop growth rate is corrected based on the interaction term, the growth state feedback coefficient, and the correction coefficient to obtain the target crop growth rate;

[0055] The sixth calculation formula for obtaining the aforementioned growth status index is:

[0056] ;

[0057] in, Indicators representing growth status Indicates real-time growth;

[0058] The seventh calculation formula for obtaining the growth state feedback coefficient is:

[0059] ;

[0060] in, Indicates the feedback coefficient;

[0061] The eighth calculation formula for obtaining the growth rate of the target crop is:

[0062] ;

[0063] in, Indicates the growth rate of the target crop. , , , , and These represent correction factors for temperature, humidity, fertility, light intensity, wind speed, and groundwater level, respectively. Indicates environmental factors With environmental factors Interactive items, and , This represents the sum of all interactive items.

[0064] Traditional environmental factor models use fixed linear correction coefficients, fail to consider the interactions between environmental factors (such as the synergistic inhibitory effect of high temperature and drought), and lack a feedback mechanism for growth status response to the environment (such as higher nutrient requirements and greater sensitivity to fertility deficiency during the vigorous growth period of crops), resulting in virtual models that do not realistically respond to complex environments. This method introduces environmental factor interaction terms and growth status feedback coefficients to construct a target crop growth rate that is dynamically coupled with growth status feedback regulation, achieving dual optimization of the synergistic effect of environmental factors and the dynamic adaptation of crop growth.

[0065] Furthermore, the method also includes:

[0066] Obtain soil data and agricultural data to be operated in the farmland area of ​​the crop, construct a spatial heterogeneous distribution field based on the soil data, and obtain a soil property model based on the spatial heterogeneous distribution field;

[0067] Construct an agricultural operation logic library and an environmental response knowledge graph. The agricultural operation logic library includes several logical rules, each of which corresponds to a standardized event. The standardized event includes operation type, operation resources, agricultural machinery to be executed, and expected time consumption.

[0068] The pre-trained agricultural mechanism model, the soil property model, the agricultural operation logic library, and the environmental response knowledge graph are embedded into the virtual crop growth model;

[0069] The agricultural data to be operated is input into the virtual crop growth model to obtain growth prediction results.

[0070] Considering that the aforementioned virtual crop growth model lacks mechanistic models to support key life and environmental processes such as crop growth and soil water and fertilizer transport, and lacks a core process model with predictive capabilities that can drive growth, it cannot make realistic growth predictions for operations such as fertilization and irrigation, nor can it effectively extrapolate and predict future growth status, yield, and stress risks; it cannot predict the future impact of different management measures (such as irrigation, fertilization, and pesticide application), and the system is unable to provide optimization decision suggestions based on simulation, so users still rely on experience to make judgments.

[0071] This method constructs a soil property model to improve the accuracy of subsequent water and fertilizer transport simulations, demonstrating a detailed characterization of farmland spatial heterogeneity. Through an agricultural mechanism model, it enables the understanding and deduction of the intrinsic processes of crop growth, providing a deeper understanding of agricultural conditions and overcoming the limitations of existing systems that can only view data. The introduction of an agricultural operation logic library and environmental response rules allows the twin to not only see and calculate, but also to make preliminary judgments based on the rules, achieving a closed loop of perception, cognition, and decision support. This provides a virtual test field with zero cost and zero risk, allowing users to assess the effects of different management measures in advance, realizing a shift from experience-based decision-making to simulation-based optimization. It can quantitatively predict crop status, yield, and stress risks for the next few hours to the entire growing season, providing a scientific basis for forward-looking management.

[0072] Furthermore, the specific steps for constructing a spatially heterogeneous distribution field based on the soil data include:

[0073] The farmland area is divided into several geographic grids in the horizontal direction and several soil profile layers in the vertical direction. A three-dimensional spatial grid system is obtained based on the geographic grids and the soil profile layers.

[0074] Historical data and sampling and testing data of the farmland area are obtained, and prior spatial distribution information and statistical characteristic values ​​of soil properties are obtained based on the historical data. The soil properties include soil moisture, soil temperature, soil texture, organic matter content and pH value.

[0075] Inverting the soil-related data in the target data to obtain isometric data;

[0076] Based on the sampled test data, the prior spatial distribution information, the statistical feature values, and the areal data, the stable attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained;

[0077] Based on the soil data, the dynamic attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained;

[0078] Based on the stable attribute values, the dynamic attribute values, and the three-dimensional spatial grid system, the spatial heterogeneity distribution field is obtained.

[0079] This study employs a multi-source data fusion method, combining real-time soil data from laboratory benchmarks, IoT sensor points, remote sensing inversion surfaces, and historical prior knowledge, to overcome the limitations of single data sources and achieve low-cost, high-precision, and comprehensive soil attribute field construction. Soil attributes are categorized into relatively stable baseline attributes (such as texture) and highly dynamic process attributes (such as moisture and nutrients), with static and dynamic attributes processed separately to construct a dynamic digital soil attribute field capable of finely characterizing its internal spatial variability. This attribute field serves as the spatial data foundation for all subsequent water and fertilizer transport simulations and precision management decisions, better aligns with agricultural science principles, and optimizes computational resource allocation.

[0080] Furthermore, the specific steps for constructing the environmental response knowledge graph include:

[0081] The core entities and core relationships that constitute the graph;

[0082] Obtain simulation results of crops under different stress scenarios, and obtain several triples based on the simulation results;

[0083] Obtain historical disaster records and corresponding multi-dimensional environmental data, and obtain statistical relationships based on the historical disaster records and the multi-dimensional environmental data;

[0084] Based on expert experience, several target entities and several target relationships are obtained;

[0085] The environmental response knowledge graph is constructed based on the core entities, core relationships, triples, statistical relationships, target entities, and target relationships.

[0086] The Environmental Response Knowledge Graph organizes fragmented agricultural knowledge (concepts, entities, and relationships) into a semantic network. Through graph computing and reasoning technologies, it enables correlation analysis, root cause tracing, and intelligent recommendation of complex agricultural conditions. This achieves a leap from isolated judgment to associative cognition of knowledge and can handle complex, multi-cause, and multi-effect agricultural scenarios.

[0087] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0088] This method introduces crop physiological stage division and stage-specific parameters to construct an initial crop growth rate coupled with accumulated temperature drive and physiological stage regulation. The growth rate coefficient is dynamically adjusted according to the current physiological stage of the crop, making the growth rate prediction more consistent with the actual growth law of the crop. By introducing environmental factor interaction terms and growth status feedback coefficients, a target crop growth rate is constructed by combining multi-factor dynamic coupling with growth status feedback regulation. This achieves dual optimization of environmental factor synergy effect and crop growth dynamic adaptation, making the virtual crop growth simulation more realistic and improving prediction accuracy. Attached Figure Description

[0089] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0090] Figure 1 This is a flowchart illustrating an agricultural monitoring method based on a digital twin model, as described in this invention.

[0091] Figure 2 This is a system architecture diagram of an agricultural monitoring method based on a digital twin model, as described in this invention. Detailed Implementation

[0092] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0093] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0094] Example 1

[0095] refer to Figures 1-2 This embodiment provides a method for agricultural monitoring based on a digital twin model, the method comprising:

[0096] Crop data from different data sources is acquired and fused to obtain target data. The crop data includes growth environment data, growth space data, and growth morphology data. For example, multiple types of sensors (such as temperature sensors and humidity sensors) are deployed in farmland to collect crop growth environment data in real time. At the same time, spatial information and morphological data of crop growth status are obtained by combining drone aerial photography and satellite remote sensing images.

[0097] Based on the target data, several source point clouds are obtained, and several crop point clouds are obtained by registration of the source point clouds. Based on the crop point clouds, an entity crop growth model is constructed.

[0098] A target mapping model is constructed based on the aforementioned growth space data;

[0099] Physiological stage division: Based on the morphological characteristics of crops, the growth cycle of crops is divided into several physiological stages. For example, rice is divided into sowing period, tillering period, jointing period, grain filling period and maturity period.

[0100] Stage-specific parameters: Each physiological stage corresponds to an accumulated temperature range (including the starting temperature and the ending temperature), a growth rate coefficient (which can be obtained by fitting field monitoring data, such as rice tillering stage = 0.003, grain-filling stage = 0.0015), and a maximum growth (such as tillering stage = 0.5m, maturity stage = 1.2m). Based on the accumulated temperature range, the growth rate coefficient, and the maximum growth, the initial crop growth rate is obtained.

[0101] The initial crop growth rate is corrected to obtain the target crop growth rate;

[0102] Based on the physical crop growth model, the target mapping model, and the target crop growth rate, a virtual crop growth model is constructed.

[0103] The system obtains crop monitoring results based on the virtual crop growth model. Early warning values ​​are set for key indicators such as soil moisture, temperature, and pest and disease indices. When monitored data exceeds or falls below preset warning thresholds, the system automatically issues warning messages to remind agricultural producers to take appropriate measures. For example, when soil moisture falls below a set lower limit, the system reminds users to irrigate through a visual interface using flashing warning lights and pop-up windows; when abnormalities in pest and disease-related indicators are detected, timely warning information is pushed out and prevention and control suggestions are provided. Simultaneously, historical data and machine learning algorithms are combined to analyze data trends and predict potential anomalies in advance.

[0104] Threshold determination algorithm: Based on historical data, the threshold can be calculated using statistical methods, with the following formula:

[0105] (Equation 1)

[0106] in The mean of the data. The standard deviation is used to determine an appropriate threshold range based on the 95% confidence interval, ensuring the accuracy of early warnings.

[0107] Trend warning algorithm: It uses LSTM (Long Short-Term Memory Network) to predict the trend of time series data. The core is to capture the long-term dependence of data. The model is trained by historical environmental data and growth data to predict the trend of indicator changes in the next 24-72 hours and give early warning of potential risks.

[0108] This system enables multi-dimensional visualization of crop growing environment data (both above and below ground). It integrates and analyzes environmental data such as soil, meteorology, and water quality, displaying the distribution and changing patterns of the data in various forms, including charts, maps, and 3D models. The system also presents the evolution of the growth process through animation within the visualization interface. For example, it uses underground soil profile maps to show the spatial distribution of underground environmental data such as soil fertility and groundwater level; and utilizes dynamic meteorological maps to display real-time changes in above-ground environmental data such as sunlight, wind speed, and rainfall, helping users comprehensively understand the development and changing trends of the crop growing environment.

[0109] The specific steps for obtaining the target data by fusing the crop data include:

[0110] Based on the crop data, the mean square error and data consistency index (numerical deviation of different data sources at the same spatiotemporal point) of different data sources are obtained, and the basic weights of different data sources are obtained based on the mean square error and the data consistency index.

[0111] The base weights are enhanced based on the critical reproductive period and attention coefficient to obtain the target weights;

[0112] The target data is obtained by fusing crop data from different data sources based on the target weight;

[0113] The first calculation formula for obtaining the basic weights is:

[0114] (Equation 2)

[0115] (Equation 3)

[0116] in, Indicates data source The basic weights, Indicates data source The mean square error, Indicates data source Data consistency metrics Indicates data source The mean square error, Indicates data source Data consistency metrics and Both represent integers greater than or equal to 1. Indicates data source Measured values ​​of single-crop data, This represents the mean of all data sources;

[0117] Attention enhancement: Determining the current critical growth stage through a crop growth stage identification model ( or non-critical reproductive period ( ), Introducing attention coefficient ( Key data sources A value of 1.2-1.5 is acceptable for non-critical data sources. The second calculation formula for obtaining the target weight is:

[0118] (Equation 4)

[0119] in, Indicates data source The target weight, This indicates the numerical value corresponding to the critical reproductive period. Indicates data source Attention coefficient Indicates data source The basic weights, Indicates data source Attention coefficient.

[0120] The specific steps for obtaining the crop point cloud include:

[0121] Feature pre-registration: Obtaining the local curvature features of the source point cloud. (Neighborhood surface curvature of the source point cloud) and texture features (e.g., constructing feature descriptors based on grayscale gradient features from UAV imagery) ;

[0122] The source point cloud and the target point cloud are matched using the K-nearest neighbor algorithm to obtain an initial transformation matrix. Based on the initial transformation matrix, the source point cloud is aligned to the target coordinate system of the target point cloud.

[0123] Spatial fine registration: Based on pre-registration, an improved ICP iterative strategy is adopted. Based on the feature weight factor, the feature descriptor and the objective function, the source point cloud and the target point cloud are aligned to obtain the optimal transformation matrix. The position of the source point cloud is updated based on the optimal transformation matrix to obtain the crop point cloud.

[0124] The third formula for calculating the objective function is:

[0125] (Equation 5)

[0126] in, Indicates the number of source point clouds. Represents source point cloud coordinates Represents the target point cloud coordinates Indicates the feature weight factor. Represents source point cloud Feature descriptors, Represents the target point cloud Feature descriptors, Represents an integer greater than or equal to 1.

[0127] Iteration termination condition: Stop when the registration error is <0.008m or the number of iterations reaches 50.

[0128] The specific steps for constructing a physical crop growth model based on the crop point cloud include:

[0129] Multi-scale point cloud preprocessing: Based on a preset scale, the crop point cloud is divided into several coarse-scale point clouds (such as retaining the core structure of the crop stem and main root, with a downsampling rate of 50%) and several fine-scale point clouds (such as retaining the detailed structure of the leaf edge and lateral root, with a downsampling rate of 10%). Statistical filtering (coarse scale, wavelength = 1.5) and bilateral filtering (fine scale, balancing noise reduction and detail preservation) are then applied to these points respectively.

[0130] Multiscale Poisson reconstruction:

[0131] Coarse-scale reconstruction: Based on a low-resolution voxel mesh (e.g., voxel size 0.1m), the Poisson equation is solved on the coarse-scale point cloud and the fine-scale point cloud to obtain a coarse-scale model, and a three-dimensional mesh model of the crop is constructed to ensure the accuracy of the overall shape.

[0132] Fine-scale reconstruction: Based on the coarse-scale model, the Poisson equation is solved on the fine-scale point cloud in the coarse-scale model using a high-resolution voxel grid (e.g., voxel size 0.02m) to obtain the fine-scale model. The detailed isosurfaces are extracted, and the coarse-scale model and the fine-scale model are fused to obtain the actual crop growth model.

[0133] This embodiment may also include:

[0134] Detail enhancement processing: Calculate the normal vector deviation of the fine-scale model, refine the local mesh in detailed areas such as leaf edges and root branches, optimize mesh smoothness through Laplacian smoothing, and preserve detailed features.

[0135] The specific steps for constructing the target mapping model include:

[0136] Global Affine Transformation: The overall spatial matching is completed using traditional affine transformation formulas. Based on the growth space data, the spatial positions of the actual planting area and the virtual planting area are registered to obtain an initial mapping model. The calculation formula for the initial mapping model can be:

[0137] (Equation 6)

[0138] in, Represents the initial mapping model. This represents a 3×3 linear transformation matrix (including rotation, scaling, and shearing). Represents the translation vector. Represents the original 3D input coordinates. The column vector form representing the original 3D input coordinates is the original 3D input coordinates Represented in the standard column vector format of matrix operations, it is used to adapt to the multiplication rules of the 3×3 linear transformation matrix A (the matrix needs to be multiplied with the column vector), and is the standardized coordinate input form of matrix operations in this affine transformation formula.

[0139] The linear transformation matrix is ​​specifically decomposed as follows:

[0140] Rotation transformation, the rotation matrix about the X-axis is The rotation matrix around the Y-axis is The rotation matrix around the Z-axis is , , and All are rotation angles corresponding to the axes, and the total rotation matrix is... ;

[0141] Scaling transformation ( (All are three-axis scaling factors);

[0142] Linear transformation matrix ;

[0143] Translation vector This can be obtained by calibrating the four corners of the real planting area with the origin of the virtual scene. The core advantage of affine transformation is that it maintains the proportions of the crop morphology, ensuring a one-to-one correspondence between the virtual model and the spatial position of the real planting area, thus reducing mapping errors.

[0144] Nonlinear region identification: Based on the growth space data, nonlinear regions of the initial mapping model are obtained, such as by using UAV terrain elevation data and planting area boundary vector data to identify nonlinear regions (such as regions with terrain slope > 5° or boundary curvature > 0.1).

[0145] Local nonlinear compensation: Radial basis function (RBF) is introduced as a nonlinear compensation operator. The nonlinear region is compensated based on the nonlinear compensation operator to obtain the target mapping model.

[0146] The fourth calculation formula of the target mapping model is:

[0147] (Equation 7)

[0148] in, This represents the original 3D input coordinates of the target mapping model. It is a target mapping model that represents the original 3D input coordinates. The virtual space mapping coordinates obtained after linear affine transformation and nonlinear compensation It is the initial mapping model, which directly acts on the original 3D input coordinates to achieve linear matching of the overall space. This indicates the number of characteristic control points within the nonlinear region. The compensation weights can be obtained by minimizing the control point mapping error using the least squares method. Representing radial basis functions, using Gaussian functions. , =0.8, This represents the shape parameter of the Gaussian radial basis function, used to control the decay rate of the Gaussian function and determine the range of influence of the radial basis function on the neighborhood of the characteristic control points. Indicates the coordinates of the current point to be mapped and the first... Euclidean distance between the coordinates of the feature control points , Represents the first term within the nonlinear region. The original coordinates of the feature control points This represents the index of the feature control points within the nonlinear region.

[0149] In this embodiment, the feature control points can be terrain inflection points, boundary feature points, etc.

[0150] The fifth calculation formula for obtaining the initial crop growth rate is as follows:

[0151] (Equation 8)

[0152] (Equation 9)

[0153] in, Indicates the initial crop growth rate. Indicates physiological stage Maximum growth Indicates physiological stage The growth rate coefficient, Indicates physiological stage The cumulative temperature, Indicates physiological stage The basic temperature for crop growth Indicates physiological stage The end of the accumulated temperature, Indicates physiological stage Inner Time temperature, Indicates physiological stage The growth cycle, and Both represent integers greater than or equal to 1.

[0154] The specific steps for obtaining the growth rate of the target crop include:

[0155] Environmental factor collection: Key factors are obtained based on environmental factors, such as aboveground environmental factors (temperature, humidity, light intensity, etc.) which are collected in real time by sensors (temperature accuracy ±0.1℃, humidity accuracy ±2% RH); underground environmental factors (soil fertility, pH value, groundwater level), with soil fertility collected by soil nutrient sensors (detecting N / P / K content), pH value collected by electrochemical sensors, and groundwater level collected by liquid level sensors; extended environmental factors (wind speed, rainfall, etc.), with wind speed collected by ultrasonic anemometers and rainfall collected by tipping bucket rain gauges; and pest and disease related factors, which are retrieved through UAV multispectral image inversion (NDVI index, pest and disease index).

[0156] Construction of environmental factor interaction terms: Obtain interaction terms between any two of the key factors; for example, select key environmental factors affecting crop growth (e.g., temperature T, humidity W, fertility F, light L, etc.) and construct pairwise interaction terms (e.g., T×W, F×L) to quantify the synergistic or antagonistic effects between factors. For example, high temperature (T> +3℃) and drought (W< The co-inhibition term of ×0.6) is , Indicates real-time temperature. Indicates the suitable temperature. Indicates suitable humidity. This indicates the real-time humidity.

[0157] Growth status feedback coefficient: The growth status index is obtained based on the real-time growth of the crop and the maximum growth, and the growth status feedback coefficient is obtained based on the growth status index.

[0158] The initial crop growth rate is corrected based on the interaction term, the growth state feedback coefficient, and the correction coefficient to obtain the target crop growth rate;

[0159] The sixth calculation formula for obtaining the aforementioned growth status index is:

[0160] (Equation 10)

[0161] in, Indicators representing growth status Indicates real-time growth;

[0162] The more vigorous the growth ( The larger the value of the feedback coefficient, the higher the crop's sensitivity to environmental factors. The seventh formula for calculating the growth state feedback coefficient is as follows:

[0163] (Equation 11)

[0164] in, Indicates the feedback coefficient;

[0165] The eighth calculation formula for obtaining the growth rate of the target crop is:

[0166] (Equation 12)

[0167] in, Indicates the growth rate of the target crop. , , , , and These represent correction factors for temperature, humidity, fertility, light intensity, wind speed, and groundwater level, respectively. Indicates environmental factors With environmental factors Interactive items, and , This represents the sum of all interactive items.

[0168] The correction coefficients are optimized into a dynamic form, such as:

[0169] Temperature correction factor: , This is the actual temperature. (for suitable temperature)

[0170] Humidity correction factor: ( This represents the actual soil moisture. (for suitable humidity)

[0171] Correction factor for fertility: ( For actual soil fertility, (for suitable fertility)

[0172] Lighting correction factor: ( For actual photosynthetically effective radiation, To provide suitable light intensity;

[0173] Wind speed correction factor: ( This refers to the actual wind speed. (for suitable wind speed)

[0174] Correction factor for groundwater level: ( This represents the actual groundwater level. (to suit the groundwater level)

[0175] Incorporating growth status feedback, The sum of all environmental factor interaction terms ( (To avoid excessive suppression).

[0176] In this embodiment, the method further includes:

[0177] A visualization and interactive platform is constructed using 3D graphics visualization technology, featuring a brand-new visualization interface that provides an immersive, real-time 3D display of agricultural data. Users can view various states of the crop growth environment from all angles by wearing virtual reality (VR) devices or on the 3D visualization platform, gaining real-time access to information about the surrounding environment and intuitively comparing environmental parameters. For example, the visualization interface uses different colors and graphics to visually display the distribution of soil moisture and dynamic curves to present temperature change trends, enabling users to quickly obtain key information.

[0178] Example 2

[0179] Based on Embodiment 1, this embodiment illustrates the process of the method with an example:

[0180] In a rice-growing area, a ground-based sensor network was deployed, including temperature and humidity sensors, light sensors, soil pH sensors, soil fertility sensors, and groundwater level sensors, collecting environmental data every 10 minutes. Simultaneously, drones conducted a weekly full-area aerial survey to acquire images and spatial information of the rice plants, and satellite remote sensing imagery was updated monthly. The collected sensor data, drone imagery, and satellite remote sensing imagery were transmitted via data transmission and interface layer to a data processing center for data cleaning, calibration, and fusion. Using 3D modeling software, a 3D model of the rice and its growing environment was constructed based on point cloud data and image information. During the modeling process, the visualization of environmental parameters in the model was updated in real time based on sensor data. For example, different colored grids represented the distribution of soil moisture (green = suitable, yellow = slightly dry, red = arid), and dynamic arrows showed wind direction changes, achieving a real-time 3D presentation of agricultural information.

[0181] Data processing:

[0182] Data cleaning: Outliers can be removed using the 3σ criterion, the formula is as follows:

[0183] (Equation 13)

[0184] in , The mean and standard deviation of the sensor data for the past 30 days are given, and outliers are filled using linear interpolation.

[0185] Data calibration: Calibrate the sensor using standard equipment and establish the existing calibration equations:

[0186] (Equation 14)

[0187] in , These are all calibration coefficients, which can be obtained through linear regression of standard values ​​and raw data, such as those used in temperature and humidity sensors. , , This represents the data before calibration. This indicates the data after calibration.

[0188] Multi-source data fusion: An adaptive weighted attention fusion algorithm is used to calculate the real-time MSE and data consistency index of each data source, and the fused data is obtained by substituting into Equations 2-4.

[0189] 3D model construction:

[0190] Point cloud preprocessing: Multi-scale preprocessing is performed on UAV point cloud data. Statistical filtering is used for coarse-scale point clouds, and bilateral filtering is used for fine-scale point clouds to preserve detailed features.

[0191] Point cloud registration: The feature-constrained ICP fusion algorithm is applied. First, the initial transformation matrix is ​​calculated through feature matching, and then iterative fine registration is performed until the registration error is <0.008m.

[0192] 3D Reconstruction: A multi-scale Poisson-detail enhancement reconstruction algorithm is adopted. The voxel size of the coarse-scale reconstruction is 0.1m, and the voxel size of the fine-scale reconstruction is 0.02m. The coarse and fine-scale models are fused and detailed processing is performed to generate a 3D mesh model of rice and environment.

[0193] Example of virtual crop growth modeling:

[0194] Based on the biological characteristics and growth patterns of rice (e.g., the growth cycle of indica rice is 120 days, the suitable temperature during the tillering stage is 25-30℃, the suitable soil moisture content during the grain-filling stage is 70%-90%, and the basic temperature...), Each physiological stage and They are respectively: tillering stage , Jointing period , Grouting period , Maturity , Using environmental data, a virtual rice growth model was constructed using the Unity3D development platform. The model sets parameters to reflect the effects of different environmental factors on rice growth, such as: for every 1°C drop in temperature below the suitable range, the growth rate decreases by 5%. ( (This is the growth status feedback coefficient); when soil moisture content is below 60%, root growth decreases by 12% × (1 + ... When light intensity is insufficient, the chlorophyll content of leaves decreases, and the leaf color in the model changes from dark green to light green. Furthermore, through spatial mapping and transformation technology, the virtual rice model is precisely matched with the real-world planting area, enabling the virtual rice's growth status to reflect the real-world rice's growth in real time.

[0195] Coordinate system calibration: Obtain the GPS coordinates of the four corners of the actual planting area (e.g., A(118.5°E, 30.2°N), B(118.6°E, 30.2°N) etc.), establish a virtual coordinate system in Unity3D, and first calculate the coordinates through a global affine transformation. (1:1 modeling) Then, nonlinear regions with terrain slope > 5° are identified, feature control points are set, compensation weights are calculated through radial basis functions, and an adaptive affine nonlinear hybrid mapping model is constructed to achieve accurate matching of virtual and real coordinates.

[0196] Growth model construction:

[0197] Crop growth rate model: Substitute biological parameters into the initial crop growth rate, such as the grain-filling stage (initial temperature). , Substitution formula 8:

[0198] (Equation 15)

[0199] (Equation 16)

[0200] Environmental correction: Substituting biological parameters into the target crop growth rate, such as the temperature correction factor.

[0201] ( ); (Equation 17)

[0202] Soil moisture correction factor:

[0203] ( ); (Equation 18)

[0204] Fertility correction factor:

[0205] ( ); (Equation 19)

[0206] Light intensity correction factor:

[0207] ( μmol / (m²・s);(Equation 20);

[0208] The corrected crop growth rate is obtained (Equation 12).

[0209] in, The sum of all environmental factor interaction terms ( (to avoid excessive suppression), such as (when and hour);

[0210] Root growth model: Corresponding to soil fertility and root growth, the formula is as follows:

[0211] (Equation 21)

[0212] in This indicates the amount of root growth.

[0213] Virtual-Real Mapping Update: Receive sensor fusion data every 10 minutes, calculate correction coefficients, interaction terms, and feedback coefficients, and update... , This drives the virtual model to change dynamically.

[0214] Examples of visualization and interactive presentation:

[0215] A WebGL-based 3D visualization and interactive platform was developed, accessible to users via a browser. The platform interface displays the overall environment of the rice-growing area as a 3D map, allowing users to view agricultural data for different regions through mouse dragging, zooming, and other operations.

[0216] In the digital twin layer, based on historical and environmental data, early warning values ​​are set for data such as soil moisture, temperature, and the probability of pests and diseases. For example, an early warning is triggered when soil moisture falls below 60%. When the monitored data reaches the early warning conditions, the corresponding icon in the platform interface will flash red, and an early warning prompt box will pop up, displaying specific early warning information and suggested measures. For the visualization of planting environment data, bar charts are used to compare soil pH values ​​in different areas, heat maps are used to present the spatial distribution of temperature, and 3D profile maps are used to show the distribution of underground soil fertility and groundwater. For the visualization of crop growth status models, animations simulate the entire growth process of rice from sowing to maturity (tillering stage, jointing stage, and grain-filling stage), and simulated pest and disease transmission paths and control measures are displayed during periods of high incidence of pests and diseases (such as rice planthoppers). Users can also click on virtual rice plants to view their detailed growth parameters and health status, achieving comprehensive dynamic interaction.

[0217] Example 3

[0218] Based on the above embodiments, in this embodiment, the method further includes:

[0219] Acquire soil data and agricultural data to be performed in the crop's farmland area (such as preset irrigation, fertilization, and spraying data), construct a spatial heterogeneous distribution field based on the soil data, and obtain a soil property model based on the spatial heterogeneous distribution field.

[0220] In this embodiment, IoT nodes are deployed in farmland, and mobile or fixed soil multi-parameter sensors are used to periodically or in real time acquire measured soil data at representative points within the grid. The soil data includes at least: volumetric water content at different depths, soil temperature, electrical conductivity, and nitrogen, phosphorus, and potassium content.

[0221] Construct an agricultural operation logic library and an environmental response knowledge graph. The agricultural operation logic library includes several logical rules, each of which corresponds to a standardized event. The standardized event includes operation type, operation resources, agricultural machinery to be executed, and expected time consumption.

[0222] In this embodiment, each agricultural operation is defined as an executable standardized event, including metadata such as operation type, required operation resources (e.g., water, fertilizer, pesticide dosage), executing agricultural machinery, and expected time consumption. An IF-THEN rule base is established, such as:

[0223] Conditions (IF section), state judgment: IF [crop growth stage == jointing stage] AND [soil moisture field mean (depth = 0-20cm) < field capacity * 0.65] AND [probability of rainfall in the next 24 hours < 30%];

[0224] Action (THEN section): This refers to the corresponding agricultural operation suggestion or automatic trigger command.

[0225] THEN [Execution Event: Jointing Stage Irrigation] AND [Recommended Parameters: Irrigation Volume = 35mm, Irrigation Method = Drip Irrigation].

[0226] The pre-trained agricultural mechanism model, the soil property model, the agricultural operation logic library, and the environmental response knowledge graph are embedded into the virtual crop growth model;

[0227] The agricultural data to be operated on is input into the virtual crop growth model to obtain growth prediction results. A predetermined agricultural operation plan is input, and the mechanistic model performs forward simulation to predict crop growth indicators (such as biomass, leaf area, soil moisture, and nutrient concentration), the arrival time of critical growth periods, and the final yield potential at specific future time points.

[0228] In this embodiment, the agricultural mechanism model includes, but is not limited to:

[0229] Crop growth models (such as existing DSSAT, AquaCrop, WOFOST or their simplified / localized versions) are used to simulate crop photosynthesis, respiration, dry matter accumulation and distribution, and growth cycle processes;

[0230] Soil water and nutrient transport models (such as existing Hydrus, SWAP or their core algorithms) are used to simulate the movement of water and nutrient transformation and migration in the soil-plant-atmosphere continuum;

[0231] The pest and disease occurrence prediction model predicts the probability and severity of pest and disease occurrence based on the microenvironment and crop growth status.

[0232] The specific steps for constructing a spatially heterogeneous distribution field based on the soil data include:

[0233] The farmland area is divided into several geographic grids in the horizontal direction and several soil profile layers in the vertical direction. A three-dimensional spatial grid system is obtained based on the geographic grids and the soil profile layers. Each grid unit is a container or record card that carries all soil properties.

[0234] Historical data and sampling and testing data of the farmland area are obtained. Based on the historical data, the prior spatial distribution information and statistical characteristic values ​​of soil properties are obtained. The soil properties include soil moisture, soil temperature, soil texture (such as sand / silt / clay ratio), organic matter content and pH value, and may also include bulk density, permanent wilting point and nutrients such as nitrogen, phosphorus and potassium.

[0235] In this embodiment, historical data includes historical soil survey reports and farmland soil type maps of farmland areas; sampling and testing data: soil samples are collected before and after the planting season according to a preset sampling strategy (such as grid method or zonal random method) and sent to the laboratory for testing and analysis to obtain high-precision benchmark values. The data includes: mechanical composition (percentage of sand, silt, and clay particles), total nitrogen, available phosphorus, available potassium, cation exchange capacity, etc.

[0236] Areal data is obtained by inverting the soil-related data in the target data; for example, remote sensing monitoring can be conducted during key growth periods using UAV / satellite platforms equipped with multispectral, hyperspectral, or thermal infrared sensors. Areal indicator factors related to soil properties are extracted from remote sensing images through physical model inversion or machine learning algorithms, for example:

[0237] Spatial distribution of soil moisture can be retrieved by utilizing the diurnal variation of surface temperature.

[0238] Spatial distribution of soil organic carbon content was retrieved using visible-near-infrared spectral characteristics.

[0239] Spatial differences in crop growth can be used to indirectly infer the spatial heterogeneity of subsoil fertility.

[0240] Based on the sampled test data, the prior spatial distribution information, the statistical feature values, and the areal data, stable attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained; for example, using laboratory sampled test data as high-precision reference points, prior spatial distribution information as trend surface constraints, and combining areal data retrieved from remote sensing as spatial correlation auxiliary information, co-kriging interpolation or a spatial prediction model based on machine learning is used to calculate the attribute values ​​of each grid cell in the three-dimensional spatial grid.

[0241] Based on the soil data, the dynamic attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained; for example, by using a fusion interpolation method, the initial spatial distribution is generated using measured data.

[0242] Based on the stable attribute values, the dynamic attribute values, and the three-dimensional spatial grid system, the spatial heterogeneity distribution field is obtained.

[0243] The specific steps for constructing the environmental response knowledge graph include:

[0244] The core entities and core relationships that constitute the graph;

[0245] Obtain simulation results of crops under different stress scenarios, and obtain several triples (entity-relationship-entity or entity-attribute-value) based on the simulation results; for example, using existing simulation models, driven by historical meteorological data, conduct a large number of simulations covering normal and various stress scenarios (water, nutrients, salinity, diseases, etc.), and extract the following from the conclusions: (water stress leads to a decrease in stomatal conductance), (a decrease in stomatal conductance manifests as an increase in canopy temperature).

[0246] Obtain historical disaster records and corresponding multi-dimensional environmental data. Based on the historical disaster records and the multi-dimensional environmental data, obtain statistical relationships. For example, use machine learning algorithms such as decision trees and association rule mining to perform association analysis on historical agricultural data and disaster records, and extract statistical relationships: (continuous rain during the heading period, induces Fusarium head blight, confidence level: 0.8).

[0247] Based on expert experience, several target entities and several target relationships can be obtained; for example, natural language processing technology can be used to automatically extract entities and relationships from agricultural manuals, research papers, and expert reports.

[0248] The environmental response knowledge graph is constructed based on the core entities, core relationships, triples, statistical relationships, target entities, and target relationships.

[0249] Real-time sensed abnormal signals (such as sensor X reporting an abnormal increase in canopy temperature of 3°C) are used as query seeds to search for matching patterns in the graph.

[0250] For example, the following subgraph patterns are automatically matched:

[0251] The current rise in canopy temperature manifests as potential stress, leading to a decrease in stomatal conductance.

[0252] The current low soil moisture leads to water stress, which manifests as an increase in canopy temperature.

[0253] By using a graph traversal algorithm, all potential stressing entities and paths associated with the current abnormal signal can be quickly identified, thereby obtaining a risk transmission warning after the current preset agricultural operation data.

[0254] In this embodiment, the core entity may include:

[0255] Environmental factors: temperature, humidity, light, rainfall, soil moisture, soil EC value, etc.

[0256] Crop status: growth stage, leaf area index, canopy temperature, water stress index, nitrogen nutrition index, etc.

[0257] Types of stress: water stress, salt stress, nitrogen deficiency, diseases (such as powdery mildew), pests (such as aphids), etc.

[0258] Agricultural operations: irrigation, fertilization, pesticide application, and tilling.

[0259] Management recommendations: watering recommendations, fertilization recommendations, and pest and disease control recommendations.

[0260] Core relationships may include:

[0261] Causes / Induces: For example, sustained high temperatures can lead to increased transpiration, and high humidity can induce downy mildew.

[0262] The manifestations / symptoms include: for example, water stress manifests as leaf wilting and elevated canopy temperature.

[0263] Mitigation / Exacerbation: Timely irrigation can alleviate water stress, while excessive nitrogen application can exacerbate the risk of disease.

[0264] Occurs during: For example, stripe rust occurs from the jointing to heading stage.

[0265] It has a threshold: such as the temperature having a threshold >35℃ (high temperature critical value).

[0266] Virtual crop growth model: A holographic mapping and unique source of truth for physical farmland in the digital world. It hosts and uniformly manages the input, output, and state of all other models and data. Specific functions include:

[0267] Geometry and attribute containers: store 3D models, spatial heterogeneous distribution fields of soil attributes, real-time crop status, etc.

[0268] Data Fusion Center: Receives and synchronizes all real-time sensing data;

[0269] Status presentation interface: This is the final display platform for all visualizations, interactions, and decision-making suggestions;

[0270] Connecting bridges: providing a unified spatiotemporal framework and data interface for other modules and data.

[0271] Agricultural Mechanism Model: A mathematical simulator based on the laws of physics, chemistry, and biology, providing in-depth extrapolation and scientific prediction capabilities. Specific functions include:

[0272] Process simulation: Accurately calculates intrinsic processes that cannot be directly observed, such as crop growth and soil water and fertilizer transport;

[0273] State prediction: Predicting the future state of crops and soil under given boundary conditions (weather, management);

[0274] Provides deep state variables: outputs key physiological indicators such as water stress index and nitrogen uptake rate for use in environmental response knowledge graph diagnosis.

[0275] Agricultural Operation Logic Library: This library formalizes and structures agricultural management knowledge, and is responsible for generating, managing, and evaluating specific agricultural operation plans. Specific functions include:

[0276] Strategy generation: Based on preset rules or optimization algorithms, generate specific operation instructions on when, where, what to do, and how much to do (e.g., irrigate area A with 20 cubic meters of water at 10:00 AM tomorrow).

[0277] Policy management: Stores standard agronomic plans and historical operation records;

[0278] Virtual actuators: These execute operational instructions in a virtual test field, driving the mechanism model to simulate the process.

[0279] Environmental Response Knowledge Graph: An interconnected knowledge network responsible for rapid diagnosis, root cause analysis, risk chain reasoning, and providing interpretable decision-making support. Specific functions include:

[0280] Real-time diagnostics: Based on the current state of the virtual crop growth model, quickly identify the type and degree of stress.

[0281] Cause-and-effect tracing: answering why this happened and locating the root cause of the problem;

[0282] Related recommendations: Intelligently associate diagnosed problems with solution strategies in the logic library.

[0283] Overall process:

[0284] 1) The agricultural data to be operated is input into the virtual crop growth model. The agricultural operation logic library parses, verifies and transforms it into a standardized event sequence that can be recognized within the system, including precise time, spatial range, operation type and intensity parameters.

[0285] 2) Using the standardized event sequence of future operations, future weather forecasts, and soil models as driving conditions, the agricultural mechanism model is started to perform forward time simulation. The model begins to calculate the changes in the state of crops and soil every day and even every hour.

[0286] 3) As the agricultural mechanism model progresses, its intermediate output states (such as soil moisture reaching X and leaf area index reaching Y on day 3 of the simulation) are updated in real time to the environmental response knowledge graph. The knowledge graph immediately performs graph matching and reasoning on these intermediate states. For example, when it detects that the canopy humidity is consistently >90% and the temperature is between 15-25℃, it will immediately diagnose a high risk of downy mildew and record the time and location of the risk.

[0287] 4) Derivative risks diagnosed by the knowledge graph (such as the disease risks mentioned above) will be immediately fed back to the agricultural operation logic library. The logic library may automatically generate a secondary response strategy based on the rules (such as recommending preventive spraying on the 5th day of the risk period).

[0288] 5) After the simulation is complete, the virtual crop growth model summarizes all results, which may include:

[0289] Final indicators: forecasted output, quality, total resource consumption, and economic benefits;

[0290] Process indicators: Water stress index curve throughout the entire growth period, nitrogen absorption dynamics;

[0291] Risk Report: A list of all warning events and their severity, generated from a knowledge graph and arranged chronologically.

[0292] Causal chain: Root cause analysis of key issues (such as production constraints) generated from knowledge graphs.

[0293] In this embodiment, statistical relationships refer to: using association rule mining, time-series causal analysis, hypothesis testing, or machine learning models to discover statistically significant association patterns, causal relationships, or risk threshold ranges between environmental factors, agricultural operations, and disaster events, and assigning quantified confidence, support, or risk intensity values ​​to these patterns as computable relationship edges added to the environmental response knowledge graph. For example:

[0294] Co-occurrence relationships and association rules: These identify events or state conditions that frequently occur simultaneously or sequentially. For example, analyzing historical data using association rule mining algorithms can uncover a strong association rule: {Heading stage, consecutive rainy days ≥ 5 days, average relative humidity > 85%} — {Fusarium head blight occurs}. This rule has a support of 8% (representing the frequency of this scenario in history) and a confidence level of 82% (representing the probability of Fusarium head blight occurring when this scenario occurs). This rule can be transformed into a relation edge in a knowledge graph: (Heading stage, consecutive rainy days and high humidity) + [Inducement, confidence level = 0.82] — (Fusarium head blight).

[0295] Causal and predictive relationships in time series: Discovering changes in certain early-stage indicators can significantly predict the occurrence of specific disasters later. For example, causal analysis revealed that the average minimum temperature in early May (T1) is the Granger causative factor for the aphid outbreak index (Y1) in June. A statistical model was established: Y1 = f(T1,...), with an explained variance R² of 0.75. When T1 > 15°C, the risk of aphid outbreaks increases significantly, forming the relationship: (early May high temperature) + [predictive, intensity = 0.75] - (aphid risk). The threshold of 15°C can be added as an attribute to this relationship.

[0296] Statistically significant risk interval: Through statistical testing, the specific numerical range of key environmental factors that significantly increase the probability of disaster occurrence is determined. For example, by using hypothesis testing or conditional probability distribution analysis, comparing samples that experienced frost damage with those that did not, a significant difference was found in the distribution of the lowest nighttime ground temperature during the flowering period. In samples that experienced frost damage, this temperature fell within the range of -2°C to 2°C for more than 90% of the time. This interval is the statistically significant risk threshold interval. The relationship is: (flowering period, nighttime ground temperature ∈ [-2, 2]°C) + [leading to, risk ratio = XX] - (frost damage).

[0297] Nonlinear patterns of multi-factor interactions: Disaster occurrence is found to be a complex interaction of multiple factors. For example, using a random forest model to analyze lodging events, the interaction between rainfall during the jointing stage and total nitrogen application was found to be the most important. Partial dependency plots show that when total nitrogen application > 20 kg / mu and rainfall during the jointing stage > 100 mm, the model-predicted lodging probability rises sharply to over 40%. This forms a complex relationship: (high nitrogen fertilizer && heavy rainfall during the jointing stage) + [synergistic effect] — (surge in lodging risk).

[0298] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0299] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for agricultural condition monitoring based on a digital twin model, characterized in that, The method includes: Crop data from different data sources is acquired and fused to obtain target data. The crop data includes growth environment data, growth space data, and growth morphology data. Based on the target data, several source point clouds are obtained, and several crop point clouds are obtained by registration of the source point clouds. Based on the crop point clouds, an entity crop growth model is constructed. A target mapping model is constructed based on the aforementioned growth space data; Based on the morphological characteristics of crops, the growth cycle of crops is divided into several physiological stages. Each physiological stage corresponds to an accumulated temperature range, a growth rate coefficient, and a maximum growth amount. Based on the accumulated temperature range, the growth rate coefficient, and the maximum growth amount, the initial crop growth rate is obtained. The initial crop growth rate is corrected to obtain the target crop growth rate; Based on the physical crop growth model, the target mapping model, and the target crop growth rate, a virtual crop growth model is constructed. The monitoring results of crops are obtained based on the virtual crop growth model.

2. The agricultural monitoring method based on a digital twin model according to claim 1, characterized in that, The specific steps for fusing the crop data to obtain the target data include: Based on the crop data, the mean square error and data consistency index of different data sources are obtained, and the basic weights of different data sources are obtained based on the mean square error and the data consistency index. The base weights are enhanced based on the critical reproductive period and attention coefficient to obtain the target weights; The target data is obtained by fusing crop data from different data sources based on the target weight; The first calculation formula for obtaining the basic weights is: ； ； in, Indicates data source The basic weights, Indicates data source The mean square error, Indicates data source Data consistency metrics Indicates data source The mean square error, Indicates data source Data consistency metrics and Both represent integers greater than or equal to 1. Indicates data source The numerical values ​​of crop data, This represents the mean of all data sources; The second calculation formula for obtaining the target weight is: ； in, Indicates data source The target weight, This indicates the numerical value corresponding to the critical reproductive period. Indicates data source Attention coefficient Indicates data source The basic weights, Indicates data source Attention coefficient.

3. The agricultural monitoring method based on a digital twin model according to claim 1, characterized in that, The specific steps for obtaining the crop point cloud include: The local curvature features and texture features of the source point cloud are obtained, and a feature descriptor is obtained based on the local curvature features and texture features; The source point cloud and the target point cloud are matched to obtain an initial transformation matrix, and the source point cloud is aligned to the target coordinate system of the target point cloud based on the initial transformation matrix; Based on the feature weight factor, the feature descriptor, and the objective function, the source point cloud and the target point cloud are aligned to obtain the optimal transformation matrix, and the position of the source point cloud is updated based on the optimal transformation matrix to obtain the crop point cloud; The third formula for calculating the objective function is: ； in, Indicates the number of source point clouds, Represents source point cloud coordinates Represents the target point cloud coordinates Indicates the feature weight factor. Represents source point cloud Feature descriptors, Represents the target point cloud Feature descriptors, Represents an integer greater than or equal to 1.

4. The agricultural monitoring method based on a digital twin model according to claim 1, characterized in that, The specific steps for constructing a physical crop growth model based on the crop point cloud include: The crop point cloud is divided into several coarse-scale point clouds and several fine-scale point clouds based on a preset scale. The Poisson equation is solved on the coarse-scale point cloud and the fine-scale point cloud based on a low-resolution voxel mesh to obtain the coarse-scale model. The Poisson equation is solved for the fine-scale point cloud in the coarse-scale model based on a high-resolution voxel mesh to obtain the fine-scale model. The coarse-scale model and the fine-scale model are combined to obtain the actual crop growth model.

5. The agricultural monitoring method based on a digital twin model according to claim 1, characterized in that, The specific steps for constructing a target mapping model include: Based on the growth space data, the spatial locations of the actual planting area and the virtual planting area are registered to obtain an initial mapping model; The nonlinear region of the initial mapping model is obtained based on the growth space data, and the nonlinear region is compensated based on the nonlinear compensation operator to obtain the target mapping model. The fourth calculation formula of the target mapping model is: ； in, Represents the target mapping model. Represents the initial mapping model. This indicates the number of characteristic control points within the nonlinear region. Indicates the compensation weight. Represents radial basis functions. This represents the original 3D input coordinates of the target mapping model. Represents the first term within the nonlinear region. The coordinates of the feature control points This represents the index of the feature control points within the nonlinear region.

6. The agricultural monitoring method based on a digital twin model according to claim 5, characterized in that, The fifth calculation formula for obtaining the initial crop growth rate is: ； ； in, Indicates the initial crop growth rate. Indicates physiological stage Maximum growth Indicates physiological stage The growth rate coefficient, Indicates physiological stage The cumulative temperature, Indicates physiological stage The basic temperature for crop growth Indicates physiological stage The end of the accumulated temperature, Indicates physiological stage Inner Time temperature, Indicates physiological stage The growth cycle, and Both represent integers greater than or equal to 1.

7. The agricultural monitoring method based on a digital twin model according to claim 6, characterized in that, The specific steps for obtaining the growth rate of the target crop include: Based on environmental factors, key factors are obtained, and interaction terms between any two of the key factors are acquired. A growth status index is obtained based on the real-time growth of the crop and the maximum growth, and a growth status feedback coefficient is obtained based on the growth status index. The initial crop growth rate is corrected based on the interaction term, the growth state feedback coefficient, and the correction coefficient to obtain the target crop growth rate; The sixth calculation formula for obtaining the aforementioned growth status index is: ； in, Indicators representing growth status Indicates real-time growth; The seventh calculation formula for obtaining the growth state feedback coefficient is: ； in, Indicates the feedback coefficient; The eighth calculation formula for obtaining the growth rate of the target crop is: ； in, Indicates the growth rate of the target crop. , , , , and These represent correction factors for temperature, humidity, fertility, light intensity, wind speed, and groundwater level, respectively. Indicates environmental factors With environmental factors Interactive items, and , This represents the sum of all interactive items.

8. The agricultural monitoring method based on a digital twin model according to claim 1, characterized in that, The method further includes: Obtain soil data and agricultural data to be operated in the farmland area of ​​the crop, construct a spatial heterogeneous distribution field based on the soil data, and obtain a soil property model based on the spatial heterogeneous distribution field; Construct an agricultural operation logic library and an environmental response knowledge graph. The agricultural operation logic library includes several logical rules, each of which corresponds to a standardized event. The standardized event includes operation type, operation resources, agricultural machinery to be executed, and expected time consumption. The pre-trained agricultural mechanism model, the soil property model, the agricultural operation logic library, and the environmental response knowledge graph are embedded into the virtual crop growth model; The agricultural data to be operated is input into the virtual crop growth model to obtain growth prediction results.

9. The agricultural monitoring method based on a digital twin model according to claim 8, characterized in that, The specific steps for constructing a spatially heterogeneous distribution field based on the aforementioned soil data include: The farmland area is divided into several geographic grids in the horizontal direction and several soil profile layers in the vertical direction. A three-dimensional spatial grid system is obtained based on the geographic grids and the soil profile layers. Historical data and sampling and testing data of the farmland area are obtained, and prior spatial distribution information and statistical characteristic values ​​of soil properties are obtained based on the historical data. The soil properties include soil moisture, soil temperature, soil texture, organic matter content and pH value. Inverting the soil-related data in the target data to obtain isometric data; Based on the sampled test data, the prior spatial distribution information, the statistical feature values, and the areal data, the stable attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained; Based on the soil data, the dynamic attribute values ​​of each grid cell in the three-dimensional spatial grid system are obtained; Based on the stable attribute values, the dynamic attribute values, and the three-dimensional spatial grid system, the spatial heterogeneity distribution field is obtained.

10. A method for agricultural monitoring based on a digital twin model according to claim 9, characterized in that, The specific steps for constructing the environmental response knowledge graph include: The core entities and core relationships that constitute the graph; Obtain simulation results of crops under different stress scenarios, and obtain several triples based on the simulation results; Obtain historical disaster records and corresponding multi-dimensional environmental data, and obtain statistical relationships based on the historical disaster records and the multi-dimensional environmental data; Based on expert experience, several target entities and several target relationships are obtained; The environmental response knowledge graph is constructed based on the core entities, core relationships, triples, statistical relationships, target entities, and target relationships.

Images