An intelligent evaluation method and application system for arable land quality based on space graph learning

Abstract

The application discloses a kind of based on space diagram learning's cultivated land quality intelligent evaluation method and application system, it is related to cultivated land quality evaluation technical field, including the acquisition area size cultivated land soil sample obtains cultivated land soil data;Design cultivated land soil quality diagram network evaluation model;Cultivated land soil quality diagram network evaluation model is trained and optimized;The optimal cultivated land soil quality diagram network evaluation model of training completion is applied test, carries out final evaluation.The application fully excavates the spatial structure relationship between sampling points by graph neural network, the graph neural network can model the spatial relationship between sampling points and their neighborhood as graph nodes and edges by constructing graph structure, and then capture the complex dependency relationship between different sampling points. This kind of spatial structured modeling not only retains the spatial correlation of data, but also can automatically learn the feature mode of different regional sample data, improve the adaptability of the model to the regional change of cultivated land quality.

Description

Technical Field

[0001] This invention relates to the field of farmland quality assessment technology, and in particular to an intelligent farmland quality assessment method and application system based on spatial graph learning. Background Technology

[0002] With the transformation of agricultural production models and the continued impact of human activities on land, the decline in arable land quality has become a major challenge to the sustainable development of global agriculture. Arable land quality is directly related to crop yield and quality, which in turn affects food security and the stability of the ecological environment. Therefore, timely and accurate assessment of arable land quality, and understanding key factors such as soil fertility, structure, and pollution status, has become a necessary means to ensure the sustainable and healthy development of agricultural production.

[0003] Traditional methods for assessing arable land quality largely rely on the analysis of sampling point data, such as soil fertility testing and heavy metal content testing. While these methods can provide information on arable land quality in local areas to some extent, they often fail to reveal the spatial heterogeneity and complexity of soil quality. Furthermore, traditional assessment methods typically depend on manual sampling, which is time-consuming and costly, and unsuitable for the needs of large-scale, dynamic agricultural environmental monitoring. With the development of remote sensing technology, sensor technology, and big data analytics, research on arable land quality assessment is gradually shifting towards intelligent assessment methods based on spatial data. In recent years, graph neural networks, as a deep learning technique capable of handling complex spatial data structures, have begun to demonstrate unique advantages in environmental monitoring and resource assessment. Graph neural networks can flexibly capture and learn the inherent dependencies of spatial data by constructing complex relationships between nodes (such as soil sample points and monitoring equipment) and edges (such as spatial relationships between adjacent sample points), making them particularly suitable for processing arable land quality data with non-Euclidean structures.

[0004] Graph neural network-based farmland quality assessment methods can effectively overcome spatial heterogeneity and nonlinearity problems, and can also capture the complex relationships between historical data and different regions through deep learning models, thereby improving the accuracy and timeliness of the assessment. However, the application of graph neural networks in the field of intelligent farmland quality assessment still faces a series of challenges: how to customize models for soil characteristics in different regions, how to combine multi-source data to achieve cross-scale farmland quality assessment, and how to improve computational efficiency in large-scale agricultural data environments are issues that urgently need to be addressed. Summary of the Invention

[0005] Traditional methods for assessing arable land quality assume that the data is stationary and linear. However, the spatial distribution of arable land quality is typically non-linear, influenced by factors such as soil structure, climate, and fertilization, making it difficult for traditional methods to capture complex variations. Furthermore, arable land quality data is high-dimensional and complex, encompassing multiple factors including soil physicochemical properties and microbial communities. Traditional methods struggle to effectively process this multidimensional data, often resulting in inaccurate predictions. Existing methods suffer from low efficiency in large-scale data processing, particularly in long-term, continuous monitoring, exhibiting poor real-time performance and failing to meet the demands of modern agriculture. This invention addresses these issues.

[0006] Therefore, the problem to be solved by this invention is how to provide an intelligent assessment method for arable land quality based on spatial graph learning.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] In a first aspect, embodiments of the present invention provide an intelligent assessment method for arable land quality based on spatial graph learning, comprising: collecting and analyzing arable land soil samples of regional size, obtaining arable land soil data through data processing; designing an arable land soil quality graph network assessment model based on the arable land soil data; training and optimizing the arable land soil quality graph network assessment model, selecting the optimal arable land soil quality graph network assessment model; applying and testing the trained optimal arable land soil quality graph network assessment model, and conducting a final assessment.

[0009] As a preferred embodiment of the intelligent assessment method for arable land quality based on spatial graph learning described in this invention, the collection and data processing of arable land soil samples at regional dimensions includes the following steps:

[0010] Regional division and sampling point layout: Obtain the latitude and longitude information of the sampling points, divide the area to be sampled into several sub-regions based on geographical environment, soil type, and agricultural management measures, and reasonably deploy multiple soil sampling points in each sub-region;

[0011] Soil sample collection and analysis: Soil samples are collected using professional tools, and standardized analytical methods are used to determine the physicochemical properties of the soil, analyze the microbial community, and quantitatively analyze the heavy metal content.

[0012] Data processing and cleaning: The collected data is preprocessed and cleaned to obtain complete and accurate soil data.

[0013] As a preferred embodiment of the intelligent assessment method for arable land quality based on spatial graph learning described in this invention, the design of the arable land soil quality map network assessment model includes the following steps:

[0014] Geospatial map network construction: Based on the latitude and longitude information of the sampling points, the geospatial distance between the sampling points is calculated, and a geospatial map network is constructed according to the set threshold.

[0015] Graph network node embedding: Fourteen representative indicators were selected from four key aspects: topography, soil conditions, socio-economic factors, and ecological security; a 14-dimensional feature vector was embedded for each sampling point to represent its soil quality characteristics.

[0016] As a preferred embodiment of the intelligent farmland quality assessment method based on spatial graph learning described in this invention, the calculation formula for constructing the geospatial graph network is as follows:

[0017]

[0018]

[0019] Where lat1, lat2, lon1, and lon2 are the longitude and latitude of sampling points P1 and P2, respectively; a is an intermediate variable used to calculate the spherical distance between the two points; R is the Earth's radius; and d is the geographic spatial distance between the sampling point pairs.

[0020] Define the geospatial threshold as d thr The specific details of the geographic spatial distance between the sampling point pairs are as follows:

[0021] If the geographic spatial distance d between sampling point pairs is less than d thr This indicates a potential interaction between the heavy metal contents of the two sampling points;

[0022] If the geographic spatial distance d ≥ d thr This indicates that the interaction between the heavy metal contents of the two sampling points is extremely weak and negligible.

[0023] As a preferred embodiment of the intelligent farmland quality assessment method based on spatial graph learning described in this invention, the method involves: based on the constructed graph network model and embedding a 14-dimensional feature vector into each node, dividing all nodes into three parts: a training set, a validation set, and a test set.

[0024] Training set: 70% of the nodes are randomly selected from all nodes and used for training the farmland soil quality map network evaluation model;

[0025] Validation set: 15% of the nodes are selected as the validation set, which is used to evaluate the model performance and adjust hyperparameters during the training of the farmland soil quality map network evaluation model; the validation set also includes expert scores as real labels.

[0026] Test set: 15% of the nodes are selected as the test set, and the farmland quality level of this part of the data is unknown.

[0027] As a preferred embodiment of the intelligent assessment method for arable land quality based on spatial graph learning described in this invention, the training and optimization of the arable land soil quality map network assessment model based on the aforementioned training set, validation set, and test set includes the following steps:

[0028] Model Training: Using labeled data from the training set, a graph neural network is used to train the nodes. The parameters of the graph neural network are continuously optimized using the backpropagation algorithm. During training, the cross-entropy loss function is used as the objective function to measure the gap between the predicted results of the farmland soil quality graph network assessment model and the true labels.

[0029] Model validation and tuning: During each round of training, the validation set is used for evaluation to monitor the performance of the farmland soil quality map network and avoid overfitting.

[0030] Training Termination and Optimal Model Selection: The training process can be stopped when the performance on the validation set no longer improves, or when the loss function of the farmland soil quality map network evaluation model on the validation set tends to stabilize. The model that performs best during training is selected as the final farmland soil quality map network evaluation model to ensure its predictive ability on unknown data.

[0031] Test set evaluation: The trained optimal farmland soil quality map network evaluation model is applied to the test set for final evaluation.

[0032] As a preferred embodiment of the intelligent assessment method for arable land quality based on spatial graph learning described in this invention, the core formula for training and optimizing the arable land soil quality graph network assessment model is:

[0033]

[0034]

[0035] in, It is the feature representation of node i at the k-th level, N r (i) is the set of neighboring nodes of node i under the r-th relation. It is the normalized coefficient between node i and its neighbor node j under the r-th relation. It is the weight matrix related to the r-th relationship at the k-th level. It is the bias term related to the r-th relation in the k-th layer, σ is the Sigmoid activation function, and W out It is the weight matrix of the output layer, b out This is the output layer bias term, used to adjust the output results. M is the number of relation types, and y i It is the final output of node i;

[0036] The loss function used to measure the difference between the predictions of the arable land soil quality map network assessment model and the true labels is as follows:

[0037]

[0038] Where L is the loss function, representing the difference between the model's predicted result and the true label, and N is the summation points, representing the number of samples involved in the calculation. y is the true label of node i, representing the farmland quality grade of that node. i It is the predicted value of node i, representing the model's assessment of the farmland quality at that node.

[0039] Secondly, to further address the problems existing in farmland quality assessment, this invention provides an application system for an intelligent farmland quality assessment method based on spatial graph learning. The system includes: a preprocessing module for collecting farmland soil samples of varying regional sizes to obtain farmland soil data; a model design module for training and optimizing a farmland soil quality graph network assessment model; a training and optimization module for training and optimizing the farmland soil quality graph network assessment model and selecting the optimal model; and a testing module for applying the trained optimal model to a test and conducting a final evaluation.

[0040] Thirdly, embodiments of the present invention provide a computer device, including a memory and a processor, wherein the memory stores a computer program, and the computer program, when executed by the processor, implements any step of the intelligent assessment method for arable land quality based on spatial graph learning as described in the first aspect of the present invention.

[0041] Fourthly, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the intelligent assessment method for arable land quality based on spatial graph learning as described in the first aspect of the present invention.

[0042] The beneficial effects of this invention are as follows:

[0043] 1. Compared with traditional methods, the key improvement of this invention lies in the introduction of graph neural network technology, which overcomes the limitations of traditional methods in processing arable land quality data. Traditional methods typically assume that the data is stationary and linear, making it difficult to capture the nonlinear characteristics and spatial dependencies of arable land quality. Through graph neural networks, this invention can simultaneously consider multi-dimensional information (such as soil physicochemical properties, climate, fertilization, etc.), effectively capturing complex spatial variation patterns and improving prediction accuracy. Furthermore, this invention also possesses powerful high-dimensional data processing capabilities, enabling efficient and real-time assessments in large-scale data and long-term monitoring, meeting the demands of modern agriculture for precision and rapid response.

[0044] 2. Compared with existing artificial intelligence methods (such as random forests and perceptron networks), this invention has a stronger spatial structure mining capability. Traditional machine learning methods, when assessing farmland quality, typically treat sampled point data as independent features, making it difficult to fully utilize the spatial structural relationships between sampled points. Although random forests and perceptron networks can model through feature selection, they often ignore the spatial autocorrelation and neighborhood relationships in farmland quality data, thus reducing the spatial accuracy of predictions.

[0045] 3. The core innovation of this invention lies in fully exploring the spatial structural relationships between sampling points through graph neural networks. Graph neural networks can construct graph structures to model the spatial relationships between sampling points and their neighborhoods as graph nodes and edges, thereby capturing the complex dependencies between different sampling points. This spatially structured modeling not only preserves the spatial correlation of the data but also automatically learns the feature patterns of sample data from different regions, improving the model's adaptability to regional changes in farmland quality. Attached Figure Description

[0046] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0047] Figure 1 This is a flowchart illustrating the implementation of the present invention in Example 1.

[0048] Figure 2 This is a schematic diagram of the geospatial map network of the present invention in Example 1. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] 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 those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0052] Example 1

[0053] Reference Figures 1 to 2 This is the first embodiment of the present invention, which provides an intelligent assessment method for arable land quality based on spatial graph learning, including the following steps:

[0054] S1: Collect and analyze farmland soil samples of regional size, and obtain farmland soil data through data processing;

[0055] Specifically:

[0056] S11. To effectively assess farmland quality at the regional scale, the first step is to conduct scientific and reasonable regional division and sampling point layout to ensure that the sampling data is representative and scientifically sound.

[0057] Regional division: Obtain the latitude and longitude information of the sampling points, and divide the area to be sampled into several sub-regions based on geographical environment, soil type, and agricultural management practices. Each sub-region should have relatively uniform environmental characteristics and be able to represent the changing trend of arable land quality within the region;

[0058] Sampling point layout: Based on the area of ​​the sub-region and the actual monitoring needs, 20-50 soil sampling points should be reasonably laid out to ensure the representativeness and reliability of the sample data. The distribution of sampling points should cover different terrains, climate conditions and agricultural activity areas.

[0059] S12. The collection and analysis of farmland soil samples is the core of assessing soil quality. Ensuring the representativeness of the samples and the scientific rigor of the analytical methods are key to obtaining high-quality data.

[0060] Soil sample collection: Use professional soil sampling tools (such as borehole samplers and shovel samplers) to collect soil samples, ensuring the uniformity and representativeness of the samples. The standard sampling depth is 0-20 cm and 20-40 cm to cover most crop root systems. For special soil problems, deeper sampling (e.g., 40-60 cm) is required. Soil samples from each sampling point should be thoroughly mixed, aliquoted, and immediately sealed in clean containers (such as plastic bags or glass bottles) and stored at a low temperature (4°C) to prevent sample deterioration.

[0061] Heavy metal content analysis: Commonly used standardized analytical methods are employed, such as soil physicochemical property determination (pH, nutrient content, organic matter), microbial community analysis, and heavy metal content analysis. Inductively coupled plasma mass spectrometry and atomic absorption spectrometry are used for quantitative analysis of heavy metals (such as lead (Pb), cadmium (Cd), and arsenic (As)) to ensure the accuracy and reliability of the analytical data.

[0062] S13, Data Processing and Cleaning: The collected farmland quality data needs to be preprocessed and cleaned to ensure the integrity and accuracy of the data, and to provide a high-quality foundation for subsequent spatial analysis and intelligent assessment.

[0063] It should be noted that the data processing and cleaning employed methods commonly used in the field, including missing value imputation, outlier detection and handling, and data standardization. Since these processing steps are existing and mature technologies, they will not be elaborated upon here. These processing steps ensured the reliability of the soil data and provided effective support for the subsequent design and analysis of the arable land soil quality map network assessment model.

[0064] S2: Design a map network assessment model for arable land soil quality.

[0065] Specifically:

[0066] S21, Geospatial Map Network Construction: Based on the latitude and longitude information of the sampling points, the geospatial distance between the sampling points is calculated, and a geospatial map network is constructed according to a set threshold. The calculation formula is as follows:

[0067]

[0068] Where lat1, lat2, lon1, and lon2 are the longitude and latitude of sampling points P1 and P2, respectively, in radians; a is an intermediate variable used to calculate the spherical distance between the two points; R is the Earth's radius, usually taken as 6371 kilometers; and d is the geographic distance between the sampling point pairs.

[0069] It should be noted that this formula is based on Haversine's formula in spherical trigonometry and is used to calculate the geographic distance between two points on Earth. The formula first calculates the difference in latitude and longitude between the two points, then calculates the intermediate variable 'a' using trigonometric functions, and finally calculates the actual distance 'd' between the two points using inverse trigonometric functions and the Earth's radius R.

[0070] Define the geospatial threshold as d thr The specific details of the geographic spatial distance between the sampling point pairs are as follows:

[0071] If the geographic spatial distance d between sampling point pairs is less than d thr This indicates a potential interaction between the heavy metal contents of the two sampling points;

[0072] If the geographic spatial distance d ≥ d thr This indicates that the interaction between the heavy metal contents of the two sampling points is extremely weak and negligible.

[0073] It should be noted that when constructing the graph network G = (V, E), the sampling points are used as nodes V. Nodes with potential interactions are connected to each other via edges in E, while nodes without potential interactions are not connected. This ultimately yields a geospatial graph network, such as... Figure 2 As shown.

[0074] S22: Graph Network Node Embedding: From four key aspects—topography, soil conditions, socio-economic factors, and ecological security—14 representative indicators were selected to construct a comprehensive soil quality evaluation index system applicable to the study area. A 14-dimensional feature vector was embedded for each sampling point to represent its soil quality characteristics.

[0075] It should be noted that the 14 representative indicators cover the natural and social factors affecting soil quality, and can comprehensively reflect the physical, chemical and ecological characteristics of soil, as well as its importance in socio-economic activities, including topography (slope, topographic location, topsoil texture, effective soil thickness), soil conditions (soil organic matter, soil pH, available phosphorus, available potassium), socio-economic factors (drainage capacity, irrigation capacity, ease of cultivation, accessibility), and ecological security (biodiversity, cleanliness).

[0076] Among the 14 representative indicators mentioned above, slope, topographic location, topographic texture, drainage capacity, irrigation capacity, biodiversity, and cleanliness are all calculated using qualitative indicators. The values ​​of the qualitative indicators are obtained using the Delphi method, as shown in Table 1 below.

[0077] Among the 14 representative indicators mentioned above, effective soil layer thickness, soil organic matter, soil pH, soil available phosphorus, soil available potassium, ease of cultivation, and accessibility are calculated using quantitative indicators. The values ​​of the quantitative indicators are calculated using membership functions, as shown in Table 2 below.

[0078] Table 1 Qualitative Indicator Levels and Their Membership Degrees

[0079]

[0080] Table 2 Membership Functions of Quantitative Indicators

[0081]

[0082] In the formula, y represents the membership degree, a represents the coefficient, b represents the intercept, c represents the standard index, and u represents the measured value. When the function type is upper bound type, y = 0 when u is less than or equal to the lower limit; y = 1 when u is greater than or equal to the upper limit. When the function type is peak type, y = 0 when u is less than or equal to the lower limit or greater than or equal to the upper limit. When the function type is negative linear type, y = 1 when u is less than or equal to the lower limit; y = 0 when u is greater than or equal to the upper limit.

[0083] Based on the above metrics, a 14-dimensional feature vector X = {x1, x2, ..., xn} is embedded into the nodes of the graph network. 14 This feature vector can uniquely represent the soil quality characteristics of each sampling point, ensuring the accuracy of soil quality assessment.

[0084] S3: Train and optimize the farmland soil quality map network evaluation model, and select the optimal farmland soil quality map network evaluation model.

[0085] Specifically:

[0086] S31: Based on the existing graph network model and the embedding of a 14-dimensional feature vector for each node, divide all nodes into three parts: training set, validation set, and test set.

[0087] Training set: 70% of the nodes were randomly selected from all nodes for training the farmland soil quality map network assessment model. Each node in the training set has an expert score as a true label for the farmland quality level. The main goal of the training set is to optimize the parameters of the farmland soil quality map network assessment model by learning from the labeled data, so that it can capture the patterns and spatial dependencies of farmland quality.

[0088] Validation set: 15% of the nodes are selected as the validation set. This set is used to evaluate model performance and adjust hyperparameters during the training process of the farmland soil quality map network evaluation model. The validation set also includes expert scores as real labels to provide real-time feedback during training, avoid overfitting, and improve the model's generalization ability.

[0089] Test set: 15% of the nodes are selected as the test set, and the farmland quality level of this data is unknown. The test set is not used for model training, but only for evaluating the final effect of the farmland soil quality map network evaluation model. The label of the test set will be evaluated by the farmland soil quality map network evaluation model.

[0090] S32: Train and optimize the farmland soil quality map network assessment model based on the above training set, validation set, and test set, including the following steps:

[0091] Model Training: Using labeled data from the training set, a graph neural network is used to train the nodes. The parameters of the graph neural network are continuously optimized through backpropagation to ensure that the feature vector of each node accurately reflects its farmland quality level. During training, the cross-entropy loss function is used as the objective function to measure the gap between the predicted results of the farmland soil quality graph network assessment model and the true labels.

[0092] It should be noted that, unlike ordinary graph convolution operations, this method captures complex dependencies between nodes through multiple relationships, rather than relying on just one. Each relationship has an independent weight matrix and normalization coefficients. This allows the network to more flexibly handle complex graph structures and uncover multi-dimensional relationships, thereby improving the model's expressive power and prediction accuracy, as detailed below:

[0093]

[0094]

[0095] in, It is the feature representation of node i at the k-th level, N r (i) is the set of neighboring nodes of node i under the r-th relation. It is the normalized coefficient between node i and its neighbor node j under the r-th relationship, used to balance the contributions of different nodes. This is the weight matrix related to the r-th relationship at the k-th layer, used to perform a linear transformation on the features of neighboring nodes. It is the bias term related to the r-th relation in the k-th layer, used to adjust the result of the linear transformation. σ is the Sigmoid activation function, used to introduce a nonlinear transformation. W out This is the weight matrix of the output layer, used to map node features to the output space, b out This is the output layer bias term, used to adjust the output result. M is the number of relation types, indicating that there may be multiple different relations between nodes. i It is the final output of node i, representing the farmland quality assessment result of that node;

[0096] It should be noted that this formula describes the update process of node features in a graph neural network. For each node i, its features at layer k are represented as follows: It is obtained by aggregating the features of its neighboring nodes; specifically, for each relation r, node i collects feature information from its neighboring node j and uses a weight matrix. and bias terms A linear transformation is performed; the feature information of all relations is weighted and summed, then a nonlinear transformation is performed through the activation function σ to obtain a new feature representation of node i. Finally, the weight matrix W of the output layer is used to...out and bias term b out The node features are mapped to the output space to obtain the farmland quality assessment result y for node i. i .

[0097] Preferably, the loss function used to measure the difference between the predicted results of the farmland soil quality map network assessment model and the true labels is as follows:

[0098]

[0099] Where l is the loss function, representing the difference between the model's predicted result and the true label, and N is the summation points, representing the number of samples involved in the calculation. y is the true label of node i, representing the farmland quality grade of that node. i It is the predicted value of node i, representing the model's assessment of the farmland quality at that node;

[0100] It should be noted that the loss function is mean squared error, used to measure the difference between the model's predictions and the true labels. The formula calculates the squared errors between the predicted values ​​and the true values ​​for all nodes, and takes the average as the final loss value. Mean squared error is a commonly used loss function in regression tasks, effectively reflecting the model's prediction accuracy. By minimizing this loss function, the model can progressively optimize its parameters and improve prediction accuracy.

[0101] Preferably, the graph neural network parameters are continuously optimized using the backpropagation algorithm, as follows:

[0102] First, calculate the gradient of the output layer, which is the gradient of the loss function with respect to the final node features:

[0103]

[0104] Then, the gradient of the output layer parameters is calculated using the chain rule:

[0105]

[0106]

[0107] For each hidden layer k, we need to consider the error of the output layer. The error is propagated back to each layer using the chain rule. First, we calculate the error at each layer:

[0108]

[0109] Next, calculate the gradient of the weights and biases for each layer:

[0110]

[0111]

[0112] Finally, the weights and biases are updated using gradient descent:

[0113]

[0114]

[0115] in, It is the error of the output layer, representing the output y of the loss function at node i. i The gradient; It is the error of the k-th layer, representing the loss function applied to the k-th layer features of node i. The gradient of the activation function; δ′ is the derivative of the activation function, representing the gradient of the activation function in that layer; Θ is the weight matrix related to the r-th relation in the (k+1)-th layer; Θ is the element-wise multiplication (Hadamard product); η is the learning rate;

[0116] It should be noted that the above formula describes the backpropagation process in a graph neural network. First, the error of the output layer is calculated. The loss function output y at node i i The gradient is calculated; then, using the chain rule, the error is backpropagated from the output layer to each subsequent layer. For each layer k, the error... By using the weight matrix of this layer Error with the next layer The error is obtained by multiplying the two layers and then multiplying by the derivative δ′ of the activation function. This process can calculate the error of each layer and use it to update the parameters of the model. Then, the gradient of the weights and biases of each layer is calculated. Finally, the weights and biases of the farmland soil quality map network evaluation model are updated by gradient descent.

[0117] Model Validation and Tuning: During each training round, a validation set is used for evaluation to monitor the performance of the farmland soil quality map network and avoid overfitting. Model performance can be optimized by adjusting hyperparameters such as the learning rate, number of hidden layers, and number of nodes. The validation set provides real-time feedback to allow for timely adjustments to hyperparameters, ensuring model stability and performance improvement during training.

[0118] Training Termination and Optimal Model Selection: The training process can be stopped when the performance on the validation set no longer improves, or when the loss function of the farmland soil quality map network evaluation model on the validation set tends to stabilize. The model that performs best during training is selected as the final farmland soil quality map network evaluation model to ensure its predictive ability on unknown data.

[0119] S4: Apply the trained optimal farmland soil quality map network evaluation model to the test set for final evaluation;

[0120] Specifically, since the farmland quality level of the test set nodes is unknown, the model will output the predicted farmland quality level based on the feature vector of each node. Evaluation metrics include accuracy, precision, recall, and F1 score, used to comprehensively evaluate the model's performance.

[0121] This embodiment also provides an application system for an intelligent assessment method of arable land quality based on spatial graph learning, including: collecting and analyzing arable land soil samples of regional size, obtaining arable land soil data through data processing; designing an arable land soil quality map network assessment model based on the arable land soil data; training and optimizing the arable land soil quality map network assessment model, selecting the optimal arable land soil quality map network assessment model; and applying the trained optimal arable land soil quality map network assessment model to a test set for final evaluation.

[0122] This embodiment also provides a computer device applicable to a method for intelligent assessment of arable land quality based on spatial graph learning, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the method for intelligent assessment of arable land quality based on spatial graph learning as proposed in the above embodiment.

[0123] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0124] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements a method for intelligent assessment of arable land quality based on spatial graph learning as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0125] In summary, this invention fully exploits the spatial structural relationships between sampling points using graph neural networks. Graph neural networks can construct graph structures to model the spatial relationships between sampling points and their neighborhoods as graph nodes and edges, thereby capturing the complex dependencies between different sampling points. This spatially structured modeling not only preserves the spatial correlation of the data but also automatically learns the feature patterns of sample data from different regions, improving the model's adaptability to regional changes in farmland quality.

[0126] Example 2

[0127] This invention proposes a method for assessing arable land quality based on spatial graph learning by combining graph neural networks and spatial data characteristics. This method not only alleviates the shortcomings of traditional methods and improves the accuracy and efficiency of arable land quality prediction, but also enhances the intelligence of arable land quality monitoring and reduces monitoring costs.

[0128] Example 1: Given the physicochemical properties of 1500 sampling points of farmland soil in a certain area, and the expert assessment results of the farmland quality at these sampling points, 200 new sampling points of farmland are now added to the area, requiring an assessment of their soil quality. Using the graph model proposed in this patent, it is unnecessary to invite experts to assess the farmland quality of the new sampling points; a graph model containing 1700 nodes can be directly established to assess the farmland quality of the new nodes. This operation can save assessment costs.

[0129] Example 2: A certain region needs to assess its arable land soil quality every three years. However, the expert team members differ each time, leading to significant subjectivity and inconsistent quality grades, failing to accurately reflect the actual changes in the region's arable land soil. Using the method proposed in this patent, historical assessment results can be incorporated into the model, encompassing the opinions of all experts as comprehensively as possible, and providing a holistic assessment result.

[0130] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for intelligent assessment of arable land quality based on spatial graph learning, characterized in that: include: S1. Collect and analyze farmland soil samples of regional size, and obtain farmland soil data through data processing. S2. Based on arable land soil data, design a map network assessment model for arable land soil quality. S3, train and optimize the farmland soil quality map network evaluation model, and select the optimal farmland soil quality map network evaluation model. S4. The trained optimal farmland soil quality map network evaluation model will be tested and evaluated for the final result. The design of a network assessment model for arable land soil quality includes the following steps: Geospatial map network construction: Based on the latitude and longitude information of the sampling points, the geospatial distance between the sampling points is calculated, and a geospatial map network is constructed according to the set threshold. Graph network node embedding: Fourteen representative indicators were selected from four key aspects: topography, soil conditions, socio-economic factors, and ecological security; a 14-dimensional feature vector was embedded for each sampling point to represent its soil quality characteristics. The calculation formula for constructing a geospatial map network is as follows: ； in, , , , These are the longitude and latitude of the collection points P1 and P2, respectively; R is an intermediate variable used to calculate the spherical distance between two points; R is the Earth's radius. It is the geographic spatial distance between sampling point pairs; Define the geospatial threshold as The specific details of the geographic spatial distance between the sampling point pairs are as follows: If the geographic distance of the sampling point pair < This indicates that there is a potential interaction between the heavy metal contents of the two sampling points; If the geographic distance of the sampling point pair ≥ , indicating that the interaction between the heavy metal contents of the two sampling points is extremely weak and negligible.

2. The intelligent assessment method for cultivated land quality based on spatial graph learning as described in claim 1, characterized in that: Regional-scale farmland soil sample collection and data processing includes the following steps: Regional division and sampling point layout: Obtain the latitude and longitude information of the sampling points, divide the area to be sampled into several sub-regions based on geographical environment, soil type, and agricultural management measures, and reasonably deploy multiple soil sampling points in each sub-region; Soil sample collection and analysis: Soil samples are collected using professional tools, and standardized analytical methods are used to determine the physicochemical properties of the soil, analyze the microbial community, and quantitatively analyze the heavy metal content. Data processing and cleaning: The collected data is preprocessed and cleaned to obtain complete and accurate soil data.

3. The intelligent assessment method for cultivated land quality based on spatial graph learning as described in claim 1, characterized in that: Based on the existing graph network model and the embedding of a 14-dimensional feature vector for each node, all nodes are divided into three parts: training set, validation set, and test set. Training set: 70% of the nodes are randomly selected from all nodes for training the farmland soil quality map network evaluation model; each node in the training set has an expert score as the true label of farmland quality level. Validation set: 15% of the nodes are selected as the validation set to evaluate the model performance and adjust hyperparameters during the training process of the farmland soil quality map network evaluation model; The validation set also includes expert scores as real labels; Test set: 15% of the nodes are selected as the test set, and the farmland quality level of this part of the data is unknown.

4. The intelligent assessment method for cultivated land quality based on spatial graph learning as described in claim 3, characterized in that: The farmland soil quality map network assessment model is trained and optimized based on the above training set, validation set, and test set, including the following steps: Model training: Using the labeled data in the training set, a graph neural network is used to train the nodes. The parameters of the graph neural network are continuously optimized through the backpropagation algorithm. During the training process, the cross-entropy loss function is used as the objective function to measure the gap between the prediction results of the arable land soil quality graph network assessment model and the true labels. Model validation and tuning: During each round of training, the validation set is used for evaluation to monitor the performance of the farmland soil quality map network and avoid overfitting. Training Termination and Optimal Model Selection: When the performance on the validation set no longer improves, or the loss function of the farmland soil quality map network evaluation model on the validation set tends to stabilize, the training process can be stopped. The model that performs best during training is selected as the final farmland soil quality map network evaluation model to ensure its predictive ability on unknown data. Test set evaluation: The trained optimal farmland soil quality map network evaluation model is applied to the test set for final evaluation.

5. The intelligent assessment method for cultivated land quality based on spatial graph learning as described in claim 4, characterized in that: The core formula for training and optimizing the arable land soil quality map network assessment model is: ； in, yes Layer Node Feature representation, yes Node at the The set of neighbor nodes under this relationship It is the first Nodes under this relationship and neighboring nodes The normalization coefficients between them It is the first Layer The weight matrix related to the relationship, It is the first Layer Bias terms related to the relationship, It is the Sigmoid activation function. It is the weight matrix of the output layer. These are the output layer bias terms, used to adjust the output results. It is the number of types of relations. It is a node The final output; The loss function used to measure the difference between the predictions of the arable land soil quality map network assessment model and the true labels is as follows: ； in, It is the loss function, representing the difference between the model's prediction and the true label. This is the summary point count, representing the number of samples involved in the calculation. It is a node The actual label indicates the farmland quality grade of that node. It is a node The predicted value represents the model's assessment of the farmland quality at that node.

6. An application system for an intelligent assessment method of arable land quality based on spatial graph learning, based on any one of claims 1 to 5, characterized in that: include: The preprocessing module is used to collect regional-sized farmland soil samples to obtain farmland soil data; The model design module trains and optimizes the farmland soil quality map network assessment model. The training and optimization module trains and optimizes the farmland soil quality map network evaluation model and selects the optimal farmland soil quality map network evaluation model. The testing module applies the trained optimal farmland soil quality map network evaluation model to the test set for final evaluation.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the intelligent assessment method for cultivated land quality based on spatial graph learning as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the intelligent assessment method for cultivated land quality based on spatial graph learning as described in any one of claims 1 to 6.

Images