A multi-source agricultural data fusion method and system based on semantic Gaussian field

Abstract

The application relates to the field of intelligent agricultural information, and provides a multi-source agricultural data fusion method and system based on a semantic Gaussian field. The multi-source agricultural data fusion method based on the semantic Gaussian field comprises the following steps: acquiring multi-source agricultural data, extracting a plurality of characteristic values, and associating the characteristic values with corresponding pixel geographical coordinates to obtain all spatial feature points; performing space-time alignment on all the spatial feature points to obtain an observation data set; creating a semantic Gaussian atom for each observation point, initializing a covariance matrix of the multi-source agricultural data, setting an initial weight value, and constructing a semantic Gaussian atom set; introducing an ecological radius constraint, a terrain constraint and a space-time attenuation constraint to construct a semantic Gaussian field; based on the semantic Gaussian field, a prediction model is used to obtain a spatial semantic heat map; based on the spatial semantic heat map, a space-time field is constructed, and time-series dynamic fusion is performed on the space-time field to identify a growth abnormal area. Full-chain automation from data acquisition to decision support is realized.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture information technology, and in particular to a method and system for multi-source agricultural data fusion based on semantic Gaussian fields. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] In precision agriculture, the accurate perception and modeling of plant growth status relies on the fusion of multi-source data, including point data obtained by various environmental sensors (such as temperature and humidity sensors and soil sensors) deployed in the field, as well as area image data (from which vegetation indices, leaf area indices, etc. can be extracted) obtained through drones or satellite remote sensing technology.

[0004] Currently common fusion methods are mostly based on simple spatial interpolation (such as Kriging interpolation) or traditional statistical models. These methods have the following significant drawbacks:

[0005] 1. Insufficient representation of spatial heterogeneity: Point data (sensors) and area data (remote sensing) differ significantly in spatial distribution characteristics (anisotropy, diffusion range). Traditional methods rely on idealized assumptions (such as uniform distribution) and cannot accurately depict the spatial variation characteristics of farmland micro-environment.

[0006] 2. Lack of physical mechanism constraints: The fusion process relies solely on mathematical statistics and does not incorporate prior knowledge of agricultural environmental dynamics (such as the influence of wind speed on humidity diffusion and the effect of terrain slope on water transport), which may lead to fusion results that violate actual physical laws and are out of touch with the real environment of crop growth. Summary of the Invention

[0007] To address the technical problems mentioned above, this invention provides a multi-source agricultural data fusion method and system based on semantic Gaussian fields. The data fusion method provided by this invention can uniformly represent multi-source agricultural data, incorporate domain knowledge, and provide a transparent and interpretable fusion process, achieving full-chain automation from data acquisition to decision support.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] The first aspect of the present invention provides a method for fusing multi-source agricultural data based on semantic Gaussian fields.

[0010] A multi-source agricultural data fusion method based on semantic Gaussian fields includes:

[0011] Multi-source agricultural data is acquired, preprocessed, and multiple feature values ​​are extracted and associated with the geographic coordinates of the corresponding pixels to obtain all spatial feature points. All spatial feature points are spatiotemporally aligned to obtain the observation dataset. The multi-source agricultural data includes: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images.

[0012] For each observation point in the observation dataset, a semantic Gaussian atom is created, the covariance matrix of the multi-source agricultural data is initialized, and initial weight values ​​are set to construct a semantic Gaussian atom set; based on the semantic Gaussian atom set, ecological radius constraints, terrain constraints, and spatiotemporal decay constraints are introduced to construct a semantic Gaussian field;

[0013] Based on the semantic Gaussian field, a prediction model is used to obtain a spatial semantic heatmap; a spatiotemporal field is constructed based on the spatial semantic heatmap, and the spatiotemporal field is dynamically fused in time series to identify abnormal growth regions.

[0014] Furthermore, the preprocessing includes: performing analog-to-digital conversion and calibration on air temperature, air humidity, soil volumetric water content, soil electrical conductivity, and light intensity, and removing outliers; and performing radiometric calibration, atmospheric correction, and geometric correction on farmland remote sensing images.

[0015] Furthermore, the method for initializing the covariance matrix of multi-source agricultural data includes: isotropic initialization of physical quantities with diffusion isotropicity, anisotropic initialization of humidity characteristics affected by wind direction, setting an initial diffusion range for farmland remote sensing image features according to image resolution, and obtaining the covariance matrix of the initialized multi-source agricultural data.

[0016] Furthermore, the ecological radius constraint is as follows:

[0017]

[0018] The terrain constraints are as follows:

[0019]

[0020] The spatiotemporal decay constraint is:

[0021]

[0022] in, Represents the eigenvalues ​​of the covariance matrix; Indicates the radius of maximum ecological impact; This represents the covariance matrix after terrain constraints. Represents the rotation matrix; Indicates slope aspect; Indicates slope; The variance represents the principal direction. This represents the variance of the secondary diffusion direction; This represents the effective weight of the i-th observation point at time t; Indicates the initial weights; This represents the observation timestamp, the time when the data was collected at the i-th observation point; This represents the time decay coefficient.

[0023] Furthermore, during the training of the prediction model, a multi-task loss function is constructed to optimize the hyperparameters of the prediction model. The multi-task loss function is as follows:

[0024]

[0025] in, Indicates multitasking loss; Indicates the predicted loss; Indicates the regularization loss; This represents the loss of physical consistency. The weight hyperparameters representing the regularization loss; The weight hyperparameters represent the loss of physical consistency.

[0026] Furthermore, the method involves constructing a spatiotemporal field based on a spatial semantic heatmap and performing temporal dynamic fusion of the spatiotemporal field to identify abnormal growth regions; the method includes:

[0027] A spatiotemporal field is constructed based on a spatial semantic heatmap; the spatiotemporal field is:

[0028]

[0029] in, Represents the spacetime field; (t) represents the atomic weights; This represents the standardized semantic feature vector; Represents the Gaussian kernel function in two-dimensional space; Represents the coordinates of the atom's center; Represent the covariance matrix;

[0030] An LSTM model is used to perform temporal dynamic fusion of the spatiotemporal field, identify abnormal growth areas, and generate agricultural suggestions.

[0031] A second aspect of the present invention provides a multi-source agricultural data fusion system based on semantic Gaussian fields.

[0032] A multi-source agricultural data fusion system based on semantic Gaussian fields includes:

[0033] The data acquisition and processing module is configured to: acquire multi-source agricultural data, perform preprocessing, extract multiple feature values, and associate them with the geographic coordinates of the corresponding pixels to obtain all spatial feature points; perform spatiotemporal alignment on all spatial feature points to obtain the observation dataset; wherein, the multi-source agricultural data includes: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images.

[0034] The semantic Gaussian field construction module is configured to: create semantic Gaussian atoms for each observation point in the observation dataset, initialize the covariance matrix of multi-source agricultural data, set initial weight values, and construct a set of semantic Gaussian atoms; based on the set of semantic Gaussian atoms, introduce ecological radius constraints, terrain constraints, and spatiotemporal decay constraints to construct a semantic Gaussian field.

[0035] The prediction module is configured to: obtain a spatial semantic heatmap based on a semantic Gaussian field and a prediction model; construct a spatiotemporal field based on the spatial semantic heatmap and perform temporal dynamic fusion of the spatiotemporal field to identify abnormal growth regions.

[0036] A third aspect of the present invention provides a computer device comprising:

[0037] A processor, adapted to execute computer programs;

[0038] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the multi-source agricultural data fusion method based on semantic Gaussian fields as described in the first aspect above.

[0039] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program adapted to be loaded by a processor and to execute steps in the multi-source agricultural data fusion method based on semantic Gaussian fields as described in the first aspect above.

[0040] The fifth aspect of the present invention provides a computer program product or computer program.

[0041] This invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps in the semantic Gaussian field-based multi-source agricultural data fusion method described in the first aspect above.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] This invention proposes a parameterized representation method for semantic Gaussian atoms, unifying multi-source agricultural data such as air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing imagery into parameterized atoms with clear physical meaning. This approach provides a unified "communication basis" for previously fragmented data, effectively overcoming the problem of inconsistent data representation in traditional methods. Through this unified representation, the inherent connections between different types of data are clearly revealed, laying a solid foundation for subsequent data fusion and analysis.

[0044] This invention develops an end-to-end optimization framework with physical constraints, cleverly integrating prior knowledge of agricultural environmental dynamics into the loss function in the form of differentiable constraints. In this way, the model not only learns the statistical regularities in the data during training but also follows the physical principles of agriculture. When dealing with the relationship between air temperature, air humidity, and soil volumetric water content, constraints are constructed based on the physical equations of water evaporation and energy exchange, ensuring that the model's predicted soil moisture changes conform to the actual physical processes and avoiding results that contradict reality. This optimization framework, which integrates physical wisdom, ensures that the fused results not only highly conform to the regularities presented by the data but also strictly meet the requirements of physical consistency, laying a solid scientific foundation for precise analysis and decision-making in agricultural production. Attached Figure Description

[0045] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0046] Figure 1 This is a flowchart illustrating a multi-source agricultural data fusion method based on semantic Gaussian fields, as shown in an embodiment of the present invention.

[0047] Figure 2 This is a structural diagram of a multi-source agricultural data fusion system based on semantic Gaussian fields, as shown in an embodiment of the present invention.

[0048] Figure 3 This is a structural diagram of a computer device shown in an embodiment of the present invention. Detailed Implementation

[0049] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0050] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0051] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0052] To facilitate understanding of the technical solutions of this invention, some technical terms involved in this invention will be introduced below.

[0053] Semantic Gaussian Field (SGaussian Field) is a scene understanding technique that combines semantic information with 3D Gaussian representation. It aims to achieve efficient and accurate semantic scene reconstruction and understanding through explicit 3D Gaussian models. Its core idea is to transfer the semantic knowledge of pre-trained 2D visual models (such as CLIP and SAM) to 3D Gaussian representations, thereby endowing the scene with semantic understanding capabilities for open-ended vocabulary.

[0054] As described in the background section, existing multi-source agricultural data fusion methods suffer from insufficient expression of spatial heterogeneity and a lack of physical mechanism constraints. Furthermore, these methods also suffer from poor model interpretability. Because traditional statistical models are often "black box" structures, they cannot quantify the contribution weights of different data sources to the fusion results, making it difficult for users to trace the basis of their decisions. This reduces the credibility of agricultural operation recommendations and hinders the practical application of smart agriculture. To address at least one of the above technical problems, this invention provides a multi-source agricultural data fusion method and system based on semantic Gaussian fields, offering a reliable data fusion and decision support tool for smart agriculture. The technical solution of this invention is described in detail below through several embodiments.

[0055] Figure 1 This is a flowchart illustrating a multi-source agricultural data fusion method based on semantic Gaussian fields, as shown in an embodiment of the present invention; see reference. Figure 1 The method includes:

[0056] Multi-source agricultural data is acquired, preprocessed, and multiple feature values ​​are extracted and associated with the geographic coordinates of the corresponding pixels to obtain all spatial feature points. All spatial feature points are spatiotemporally aligned to obtain the observation dataset. The multi-source agricultural data includes: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images.

[0057] For each observation point in the observation dataset, a semantic Gaussian atom is created, the covariance matrix of the multi-source agricultural data is initialized, and initial weight values ​​are set to construct a semantic Gaussian atom set; based on the semantic Gaussian atom set, ecological radius constraints, terrain constraints, and spatiotemporal decay constraints are introduced to construct a semantic Gaussian field;

[0058] Based on the semantic Gaussian field, a prediction model is used to obtain a spatial semantic heatmap; a spatiotemporal field is constructed based on the spatial semantic heatmap, and the spatiotemporal field is dynamically fused in time series to identify abnormal growth regions.

[0059] The core of this invention lies in constructing a multi-source agricultural data fusion framework based on learnable Gaussian fields. Through the parameterized representation of semantic Gaussian atoms and optimization of physical constraints, it achieves spatially continuous feature fusion that conforms to the agricultural environment mechanism.

[0060] The technical solution of this embodiment will be described in detail below:

[0061] Step 1: Collection and Standardization Preprocessing of Multi-Source Heterogeneous Agricultural Data

[0062] Step 1.1 Data Acquisition

[0063] High-frequency environmental sensor data, including air temperature, air humidity, soil volumetric water content, soil electrical conductivity (EC value), and light intensity, are collected through IoT nodes. The sampling frequency range is 1-60 minutes, and all data are accompanied by high-precision GNSS coordinates (accuracy better than 0.1 meters).

[0064] Using a drone platform equipped with a multispectral camera, farmland remote sensing images with a spatial resolution better than 5 cm are collected weekly; satellite remote sensing data (such as Sentinel-2 multispectral data) are also used as a supplement, with a spatial resolution of 10 meters and a revisit cycle of 5 days.

[0065] Step 1.2 Data Preprocessing and Feature Extraction

[0066] Sensor data: The raw current / voltage signals are converted and calibrated, converted into physical quantity values ​​using the following formula, and outliers are removed (using the 3σ principle).

[0067]

[0068] in, Represents a physical quantity. Indicates the reading. , Indicates calibration parameters,

[0069] Image data: Radiometric calibration, atmospheric correction, and geometric correction were performed on the original images; the following vegetation indices were calculated.

[0070]

[0071]

[0072] Canopy coverage

[0073] in, Indicates the normalized vegetation index; Indicates near-infrared reflectivity; Indicates the reflectivity in the red light band; Indicates leaf area index; Indicates a specific coefficient for crop canopy; NDVI values ​​representing soil background; This represents the NDVI value for pure vegetation.

[0074] All extracted feature values ​​are associated with the geographic coordinates of the corresponding pixels (using the UTM projection coordinate system) to form spatial feature points.

[0075] Step 1.3 Spatiotemporal Alignment and Standardization

[0076] Unify all data to the same time base (UTC timestamp) and spatial base (WGS84 coordinate system);

[0077] For eigenvectors Perform Z-score standardization: ,in and The first The mean and variance of each feature dimension;

[0078] Output a spatiotemporally aligned, standardized observation dataset:

[0079]

[0080] in, This represents the mean of the i-th data point; This represents the timestamp of the i-th data item. This represents the data source identifier for the i-th data.

[0081] Step 2: Construction and parameter initialization of semantic Gaussian atoms

[0082] Step 2.1 Definition of atomic structure

[0083] Each semantic Gaussian atom is defined as a six-tuple:

[0084]

[0085] in, Represents semantic Gaussian atoms; Indicates the coordinates of the atom center (obtained directly from observation data); Represents the covariance matrix, which controls the diffusion pattern of atoms in space; This represents the standardized semantic feature vector; Indicates atomic weights, reflecting data reliability; Indicates the time decay coefficient (used for time series fusion); Represents a data type identifier.

[0086] Step 2.2 Covariance Matrix Initialization Strategy

[0087] Employing a combination of data-driven approaches and prior knowledge:

[0088] Step 2.2.1 Isotropic Initialization: For physical quantities with isotropic diffusion such as temperature, let... The variance , Set according to the effective radius of the sensor (default 20 meters).

[0089] Step 2.2.2 Anisotropy Initialization: For humidity data affected by wind direction, initialize the principal axis direction based on meteorological station data:

[0090]

[0091] in, Represents the rotation matrix. Indicates the current wind direction and angle. , The variance represents the principal direction. This represents the variance of the secondary diffusion direction. Indicates wind speed level.

[0092] Image feature initialization: For features extracted by remote sensing, the initial diffusion range is set according to the image resolution. , This indicates the spatial resolution of the remote sensing image.

[0093] Step 2.2.3 Weight initialization rules:

[0094] Different initial weight values ​​are set according to the type and quality of the data source: high-precision sensor data is weighted at 0.9, commercial-grade sensor data at 0.7, UAV image features at 0.8, and satellite image features at 0.6.

[0095] Step 2.2.4 Output the initialized semantic Gaussian atom set .

[0096] Step 3: Construction of Semantic Gaussian Fields for Physical Constraints

[0097] Step 3.1 Basic Field Construction Formula

[0098] any point in space The fusion feature calculation at the location uses an adaptive Gaussian mixture model:

[0099]

[0100]

[0101] in, This represents the Gaussian kernel function in two-dimensional space.

[0102] Step 3.2 Physical constraint mechanism

[0103] Ecological radius constraint: by modifying the eigenvalues ​​of the covariance matrix accomplish:

[0104]

[0105] in, This indicates the maximum ecological impact radius, which is typically set to 2-5 meters for the crop root zone.

[0106] Terrain constraints: Calculate slope using a digital elevation model (DEM). and slope Adjust the diffusion direction of water-related atoms:

[0107]

[0108] in, This represents the rotation matrix.

[0109] Spatiotemporal decay constraint: For time series data, an exponential decay term is introduced:

[0110]

[0111] in, This represents the effective weight of the i-th observation point at time t; This indicates the initial weights, reflecting the reliability of the data (such as sensor accuracy and signal-to-noise ratio). This represents the observation timestamp, the time when the data was collected at the i-th observation point.

[0112] Step 3.3 Output the continuous semantic field of physical constraints .

[0113] Step 4: End-to-end joint optimization framework

[0114] Step 4.1 Differentiable Prediction Network

[0115] Employing a multilayer perceptron and residual connection structure:

[0116]

[0117]

[0118]

[0119]

[0120] in, Represents input features; This represents the weight matrix of the first layer; Indicates the bias term of the first layer; Indicates the activation function; This represents the output of the first hidden layer; This represents the weight matrix of the second layer; Indicates the bias term of the second layer; This represents the output of the second hidden layer; This represents the weight matrix of the third layer; This represents the bias term of the third layer; This represents the output of the third hidden layer; This represents the weight matrix of the output layer; Indicates the bias term of the output layer; This represents the model's predicted output, which is the growth state prediction vector (biomass, stress index, etc.).

[0121] Step 4.2 Multi-task loss function

[0122] .

[0123] in, Indicates multitasking loss; Indicates the predicted loss; Indicates the regularization loss; This represents the loss of physical consistency. The weight hyperparameters representing the regularization loss; The weight hyperparameters representing the physical consistency loss; Indicates the number of samples; Indicates the sample index; This represents the model's predicted output value for the j-th sample; This represents the true label value of the j-th sample; Indicates the number of parameters or features in the regularization term; Let i represent the covariance matrix to be regularized; Represents the covariance matrix The square of the Frobenius norm; Representation matrix The largest eigenvalue; This indicates the set threshold.

[0124] Step 4.3 Optimization Strategy:

[0125] Using the Adam optimizer, the learning rate decays exponentially: ,in, This represents the learning rate used in the t-th training iteration; t represents the initial learning rate; t represents the number of training iterations.

[0126] The covariance matrix is ​​parameterized using Cholesky decomposition: To ensure positive definiteness; among them, Let i represent the i-th covariance matrix; Represents the covariance matrix Cholesky decomposition factor.

[0127] Training cycle: 200-500 rounds, with a tolerance of 20 rounds for the early stop strategy.

[0128] Output optimized atomic parameters and predict network parameters .

[0129] Step 5: Dynamic Temporal Extension and Decision Support

[0130] Step 5.1 Spatiotemporal Field Construction

[0131] Extend the semantic field to:

[0132]

[0133] Step 5.2 LSTM Dynamic Prediction:

[0134]

[0135]

[0136] in, This represents the hidden state at the current moment, used to store the context information of the current time step; Representing cell states, responsible for conveying long-term dependency information in time series, ensuring the model can remember key historical states; the state at the previous time step. and The data is recursively passed as input to the current time step, thus enabling effective modeling of time series data. The final prediction output... By linear transformation, Represents the weight matrix. These two parameters, representing the bias term, together determine how the hidden states of the LSTM are processed. This is mapped to specific crop growth status prediction results, such as key agricultural indicators like biomass and stress index.

[0137] This invention achieves a spatiotemporally unified continuous field representation by introducing a time decay factor and LSTM dynamic modeling. The introduction of the time decay factor automatically adjusts the weights of data based on their temporal distance. Recently collected data, such as air temperature and soil moisture, are given higher weights because they more directly and accurately reflect the current crop growth status; while earlier data, whose influence on the current state gradually weakens over time, have lower weights. This dynamic weight adjustment mechanism allows the model to focus more on real-time data changes, effectively reducing the interference of noise from historical data on the current analysis.

[0138] Step 5.3 Anomaly Detection and Decision Rules

[0139] Based on the aforementioned spatiotemporal semantic field Key agricultural environmental parameters are extracted from various spatial locations, including soil salinity, normalized differentiating vegetation index (NDVI), soil moisture content, air temperature, and air humidity. Multi-dimensional joint criteria are established for typical stress patterns during crop growth to achieve intelligent identification and tiered early warning of abnormal areas.

[0140] Specifically, the system executes the following decision logic: when the soil salinity of a certain area exceeds the preset salinity stress threshold, its NDVI value is lower than the lower limit threshold for healthy vegetation growth, and the soil moisture content is lower than the minimum irrigation guarantee threshold, the area is determined to be under severe stress, the risk level is marked as "high risk", and corresponding agricultural management suggestions are generated: "Prioritize irrigation and apply soil conditioners to alleviate salinity damage".

[0141] If the above conditions are not met, but the temperature in the area is detected to be higher than the high-temperature stress threshold and the air humidity is lower than the drought warning threshold, then it is determined that there is a risk of combined heat and drought stress, the risk level is marked as "moderate risk", and the management recommendation is generated: "Increase the irrigation frequency to reduce water deficit caused by evapotranspiration".

[0142] For areas that do not meet either the high-risk or medium-risk criteria, the growth status is considered normal, the risk level is marked as "normal", and it is recommended to maintain the current agricultural operation plan without intervention or adjustment.

[0143] This invention effectively solves the three key technical challenges of spatial heterogeneity, physical consistency, and poor resolvability in multi-source agricultural data fusion through the above technical solution, providing a reliable data fusion and decision support tool for smart agriculture.

[0144] The above combination Figure 1 The multi-source agricultural data fusion method based on semantic Gaussian field provided in the embodiments of the present invention has been described in detail. Next, the multi-source agricultural data fusion system based on semantic Gaussian field provided in the embodiments of the present invention will be described in conjunction with the accompanying drawings.

[0145] Figure 2 This is a schematic diagram of the structure of a multi-source agricultural data fusion system based on semantic Gaussian fields, as shown in an embodiment of the present invention. Figure 2 The system described in this invention includes:

[0146] The data acquisition and processing module is configured to: acquire multi-source agricultural data, perform preprocessing, extract multiple feature values, and associate them with the geographic coordinates of the corresponding pixels to obtain all spatial feature points; perform spatiotemporal alignment on all spatial feature points to obtain the observation dataset; wherein, the multi-source agricultural data includes: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images.

[0147] The semantic Gaussian field construction module is configured to: create semantic Gaussian atoms for each observation point in the observation dataset, initialize the covariance matrix of multi-source agricultural data, set initial weight values, and construct a set of semantic Gaussian atoms; based on the set of semantic Gaussian atoms, introduce ecological radius constraints, terrain constraints, and spatiotemporal decay constraints to construct a semantic Gaussian field.

[0148] The prediction module is configured to: obtain a spatial semantic heatmap based on a semantic Gaussian field and a prediction model; construct a spatiotemporal field based on the spatial semantic heatmap and perform temporal dynamic fusion of the spatiotemporal field to identify abnormal growth regions.

[0149] In some embodiments, the preprocessing includes: performing analog-to-digital conversion and calibration on air temperature, air humidity, soil volumetric water content, soil electrical conductivity, and light intensity, and removing outliers; and performing radiometric calibration, atmospheric correction, and geometric correction on farmland remote sensing images.

[0150] In some embodiments, the initialization of the covariance matrix of multi-source agricultural data includes: isotropic initialization of physical quantities with diffusion isotropicity, anisotropic initialization of humidity characteristics affected by wind direction, setting an initial diffusion range for farmland remote sensing image features according to image resolution, and obtaining the initialized covariance matrix of multi-source agricultural data.

[0151] In some embodiments, the ecological radius constraint is:

[0152]

[0153] The terrain constraints are as follows:

[0154]

[0155] The spatiotemporal decay constraint is:

[0156]

[0157] in, Represents the eigenvalues ​​of the covariance matrix; Indicates the radius of maximum ecological impact; This represents the covariance matrix after terrain constraints. Represents the rotation matrix; Indicates slope aspect; Indicates slope; The variance represents the principal direction. This represents the variance of the secondary diffusion direction; This represents the effective weight of the i-th observation point at time t; Indicates the initial weights; This represents the observation timestamp, the time when the data was collected at the i-th observation point; This represents the time decay coefficient.

[0158] In some embodiments, during the training of the prediction model, a multi-task loss function is constructed to optimize the hyperparameters of the prediction model. The multi-task loss function is:

[0159]

[0160] in, Indicates multitasking loss; Indicates the predicted loss; Indicates the regularization loss; This represents the loss of physical consistency. The weight hyperparameters representing the regularization loss; The weight hyperparameters represent the loss of physical consistency.

[0161] In some embodiments, the method of constructing a spatiotemporal field based on a spatial semantic heatmap and performing temporal dynamic fusion of the spatiotemporal field to identify abnormal growth regions includes:

[0162] A spatiotemporal field is constructed based on a spatial semantic heatmap; the spatiotemporal field is:

[0163]

[0164] in, Represents the spacetime field; Indicates atomic weights; This represents the standardized semantic feature vector; Represents the Gaussian kernel function in two-dimensional space; Represents the coordinates of the atom's center; Represent the covariance matrix;

[0165] An LSTM model is used to perform temporal dynamic fusion of the spatiotemporal field, identify abnormal growth areas, and generate agricultural suggestions.

[0166] The system deployment architecture of this embodiment includes: (1) Edge layer: NVIDIA Jetson AGX Orin module is responsible for real-time data preprocessing and atomization; (2) Cloud layer: AWS EC2 P3 instance is responsible for large-scale field computing and model training; (3) Application layer: WebGIS platform provides heat map visualization and agricultural advice push.

[0167] This embodiment achieves automation and intelligence across the entire chain from data acquisition to decision support through an edge-cloud collaborative computing architecture. Data flows efficiently and collaboratively between edge devices and the cloud, providing strong technical support for precision agriculture. This architecture not only improves the efficiency and accuracy of agricultural data processing but also provides a feasible technical implementation plan for refined management, optimal resource allocation, and sustainable development in agricultural production.

[0168] According to embodiments of the present invention, the multi-source agricultural data fusion system based on semantic Gaussian fields can correspond to the execution of the methods described in the embodiments of the present invention, and the above and other operations and / or functions of each module of the multi-source agricultural data fusion system based on semantic Gaussian fields are respectively for the purpose of implementing... Figure 1 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0169] See Figure 3 The diagram shows the structure of a computer device, which includes a processor, a communication interface, and a computer-readable storage medium. The processor, communication interface, and computer-readable storage medium are connected via a bus or other means. The communication interface is used to receive and send data. The computer-readable storage medium can be stored in the computer device's memory. The computer-readable storage medium stores computer programs, including program instructions, and the processor executes the program instructions stored in the computer-readable storage medium. The processor (or CPU, Central Processing Unit) is the computing and control core of the computer device, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to implement the corresponding steps in the embodiment of the multi-source agricultural data fusion method based on semantic Gaussian fields.

[0170] This embodiment provides a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the processing system of the computer device.

[0171] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or non-volatile memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0172] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes one or more instructions stored in the computer-readable storage medium to implement the corresponding steps in the above-described embodiment of the multi-source agricultural data fusion method based on semantic Gaussian fields.

[0173] This embodiment provides a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the corresponding steps in the above-described embodiment of the multi-source agricultural data fusion method based on semantic Gaussian fields.

[0174] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0175] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0176] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0177] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0178] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0179] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-source agricultural data fusion method based on semantic Gaussian fields, characterized in that, include: Acquire multi-source agricultural data, preprocess it, extract multiple feature values, and associate them with the geographic coordinates of the corresponding pixels to obtain all spatial feature points; All spatial feature points were spatiotemporally aligned to obtain the observation dataset; among them, multi-source agricultural data included: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images. For each observation point in the observation dataset, a semantic Gaussian atom is created, the covariance matrix of the multi-source agricultural data is initialized, and initial weight values ​​are set to construct a semantic Gaussian atom set; based on the semantic Gaussian atom set, ecological radius constraints, terrain constraints, and spatiotemporal decay constraints are introduced to construct a semantic Gaussian field; The method for initializing the covariance matrix of multi-source agricultural data includes: isotropic initialization of physical quantities with diffusion isotropicity, anisotropic initialization of humidity characteristics affected by wind direction, setting an initial diffusion range for farmland remote sensing image features according to image resolution, and obtaining the covariance matrix of the initialized multi-source agricultural data. Based on the semantic Gaussian field, a prediction model is used to obtain a spatial semantic heatmap; A spatiotemporal field is constructed based on a spatial semantic heatmap, and the spatiotemporal field is dynamically fused over time to identify regions of abnormal growth; the spatiotemporal field is as follows: in, Represents the spacetime field; Indicates atomic weights; This represents the standardized semantic feature vector; Represents the Gaussian kernel function in two-dimensional space; Represents the coordinates of the atom's center; Let represent the covariance matrix.

2. The multi-source agricultural data fusion method based on semantic Gaussian fields according to claim 1, characterized in that, The preprocessing includes: performing analog-to-digital conversion and calibration on air temperature, air humidity, soil volumetric water content, soil electrical conductivity, and light intensity, and removing outliers; and performing radiometric calibration, atmospheric correction, and geometric correction on farmland remote sensing images.

3. The multi-source agricultural data fusion method based on semantic Gaussian fields according to claim 1, characterized in that, The ecological radius constraint is: The terrain constraints are as follows: The spatiotemporal decay constraint is: in, Represents the eigenvalues ​​of the covariance matrix; Indicates the radius of maximum ecological impact; This represents the covariance matrix after terrain constraints. Represents the rotation matrix; Indicates slope aspect; Indicates slope; The variance represents the principal direction. This represents the variance of the secondary diffusion direction; This represents the effective weight of the i-th observation point at time t; Indicates the initial weights; This represents the observation timestamp, the time when the data was collected at the i-th observation point; This represents the time decay coefficient.

4. The multi-source agricultural data fusion method based on semantic Gaussian fields according to claim 1, characterized in that, During the training of the prediction model, a multi-task loss function is constructed to optimize the hyperparameters of the prediction model. The multi-task loss function is as follows: in, Indicates multitasking loss; Indicates the predicted loss; Indicates the regularization loss; Indicates the loss of physical consistency; The weight hyperparameters representing the regularization loss; The weight hyperparameters represent the loss of physical consistency.

5. The multi-source agricultural data fusion method based on semantic Gaussian fields according to claim 1, characterized in that, An LSTM model is used to perform temporal dynamic fusion of the spatiotemporal field, identify abnormal growth areas, and generate agricultural suggestions.

6. A multi-source agricultural data fusion system based on semantic Gaussian fields, characterized in that, include: The data acquisition and processing module is configured to: acquire multi-source agricultural data, perform preprocessing, extract multiple feature values, and associate them with the geographic coordinates of the corresponding pixels to obtain all spatial feature points; All spatial feature points were spatiotemporally aligned to obtain the observation dataset; among them, multi-source agricultural data included: air temperature, air humidity, soil volumetric water content, soil electrical conductivity, light intensity, and farmland remote sensing images. The semantic Gaussian field construction module is configured to: create semantic Gaussian atoms for each observation point in the observation dataset, initialize the covariance matrix of multi-source agricultural data, set initial weight values, and construct a set of semantic Gaussian atoms; based on the set of semantic Gaussian atoms, introduce ecological radius constraints, terrain constraints, and spatiotemporal decay constraints to construct a semantic Gaussian field. The method for initializing the covariance matrix of multi-source agricultural data includes: isotropic initialization of physical quantities with diffusion isotropicity, anisotropic initialization of humidity characteristics affected by wind direction, setting an initial diffusion range for farmland remote sensing image features according to image resolution, and obtaining the covariance matrix of the initialized multi-source agricultural data. The prediction module is configured to: obtain a spatial semantic heatmap based on a semantic Gaussian field and a prediction model; construct a spatiotemporal field based on the spatial semantic heatmap; and perform temporal dynamic fusion of the spatiotemporal field to identify abnormal growth regions. The spatiotemporal field is as follows: in, Represents the spacetime field; Indicates atomic weights; This represents the standardized semantic feature vector; Represents the Gaussian kernel function in two-dimensional space; Represents the coordinates of the atom's center; Let represent the covariance matrix.

7. A computer device, characterized in that, A processor, adapted to execute computer programs; A computer-readable storage medium storing a computer program, which, when executed by the processor, implements the steps of the multi-source agricultural data fusion method based on semantic Gaussian fields as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program adapted to be loaded by a processor and to execute the steps of the multi-source agricultural data fusion method based on semantic Gaussian fields as described in any one of claims 1-5.

9. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps in the multi-source agricultural data fusion method based on semantic Gaussian fields as described in any one of claims 1-5.

Images