Regional soil health assessment method and device based on machine learning

Abstract

The invention is suitable for the field of artificial intelligence, and provides a regional soil health assessment method and device based on machine learning. In the embodiment, the linear features and the nonlinear features of the soil are determined through the historical feature data of the soil, and then the training data of the soil are determined through the boundary conditions of the linear features and the nonlinear features, so that the linear and nonlinear reduced-order models of the soil are determined through the training data. When the soil in the to-be-detected area is detected subsequently, the corresponding reduced-order model is determined according to the detection data of the soil, and health assessment is carried out. Therefore, the complex characteristics of the soil can be accurately identified and analyzed, and a more targeted evaluation result can be provided according to different soil conditions.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method and device for regional soil health assessment based on machine learning. Background Art

[0002] In the field of soil assessment, it is crucial to accurately assess the quality and characteristics of soil. Traditional soil assessment methods mainly rely on physical and chemical analysis of soil samples. Although these methods can provide basic soil characteristics to a certain extent, they often lack an in-depth understanding of the complex characteristics in the soil. The limitations of these traditional methods are particularly obvious when dealing with soils with significant linear and nonlinear characteristics.

[0003] Existing soil assessment models often fail to effectively distinguish between linear and nonlinear characteristics in soil. Linear characteristics usually refer to those characteristics in soil that can be predicted or described by simple linear relationships, such as soil pH, organic matter content, etc. Nonlinear characteristics involve more complex interactions and changes, such as soil water retention capacity, microbial activity, etc. Due to the complexity of these characteristics, traditional assessment models often cannot provide accurate predictions when dealing with these characteristics.

[0004] Due to the lack of in-depth analysis of soil characteristics, existing assessment models are often too generalized and lack pertinence when outputting assessment results, which not only affects the accuracy of the assessment results, but also limits its effectiveness in practical applications.

[0005] Soil properties may vary significantly under different soil types and environmental conditions. Existing assessment models often adopt a "one-size-fits-all" approach and fail to adjust to different soil conditions, which limits their applicability in different application scenarios. Summary of the invention

[0006] In view of this, the present application provides a regional soil health assessment method and device based on machine learning to more accurately identify and analyze the complex characteristics of the soil, and to provide more targeted assessment results based on different soil conditions.

[0007] The first aspect of the present application provides a regional soil health assessment method based on machine learning, characterized in that the method comprises:

[0008] Collecting historical characteristic data of target soil under various climatic conditions, wherein the historical characteristic data includes linear characteristics and nonlinear characteristics;

[0009] Determine the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determine the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions;

[0010] The pre-constructed linear initial reduced order model is trained by the linear training data to generate the linear reduced order model of the target soil, and the pre-constructed nonlinear initial reduced order model is trained by the non-linear training data to generate the non-linear reduced order model of the target soil;

[0011] The detection data of the soil in the area to be detected is obtained, a target reduced-order model for health assessment is determined according to the detection data, a target boundary condition of the soil in the area to be detected is determined by using the detection data and the historical characteristic data, and the target boundary condition is input into the target reduced-order model to obtain health estimation information of the target soil.

[0012] Optionally, the determining the target boundary condition of the soil in the area to be detected by using the detection data and the historical characteristic data includes:

[0013] Constructing a first model and a second model of the target soil by using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic;

[0014] Optimizing the linear features and the nonlinear features in the target feature data respectively by using the first model and the second model to obtain the processed target feature data, wherein the optimization processing includes at least one of correction, deduplication and deletion of invalid data;

[0015] The target boundary conditions of the detection data are then determined by the preset method.

[0016] Optionally, the preset method includes at least one of numerical simulation method, machine learning method, statistical analysis method, and mathematical modeling method.

[0017] Optionally, the determining the linear training data and the nonlinear training data of the target soil according to the linear boundary condition and the nonlinear boundary condition includes:

[0018] Constructing an evaluation model based on a machine learning algorithm according to the historical feature data;

[0019] Determining a first corresponding relationship between the linear feature and a preset soil health state according to the linear boundary condition, and determining the first corresponding relationship as the linear training data;

[0020] A second corresponding relationship between the nonlinear feature and a preset soil health state is determined according to the nonlinear boundary condition, and the second corresponding relationship is determined as the nonlinear training data.

[0021] Optionally, determining a target reduced-order model for health assessment according to the detection data includes:

[0022] When the detection data only contains linear features, determining that the target reduced-order model for health assessment using the detection data is a linear reduced-order model; or,

[0023] When the detection data only contains nonlinear features, determining that the target reduced-order model for health assessment using the detection data is a nonlinear reduced-order model; or,

[0024] When the detection data contains both linear features and nonlinear features, it is determined that the target reduced-order model of the detection data used for health assessment is a linear reduced-order model and a nonlinear reduced-order model.

[0025] A second aspect of the present application provides a regional soil health assessment device based on machine learning, the device comprising:

[0026] A feature collection unit, used to collect historical feature data of target soil under various climatic conditions, wherein the historical feature data includes linear features and nonlinear features;

[0027] A data determination unit, used to determine the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determine the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions;

[0028] A model generation unit, configured to train a pre-constructed linear initial reduced order model using the linear training data to generate a linear reduced order model of the target soil, and to train a pre-constructed nonlinear initial reduced order model using the non-linear training data to generate a non-linear reduced order model of the target soil;

[0029] A soil assessment unit is used to obtain detection data of the soil in the area to be detected, determine a target reduced-order model for health assessment according to the detection data, determine a target boundary condition of the soil in the area to be detected through the detection data and the historical characteristic data, and input the target boundary condition into the target reduced-order model to obtain health estimation information of the target soil.

[0030] Optionally, the data determination unit determines the target boundary condition of the soil in the to-be-detected area by using the detection data and the historical characteristic data, including:

[0031] Constructing a first model and a second model of the target soil by using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic;

[0032] Optimizing the linear features and the nonlinear features in the target feature data respectively by using the first model and the second model to obtain the processed target feature data, wherein the optimization processing includes at least one of correction, deduplication and deletion of invalid data;

[0033] The target boundary conditions of the detection data are then determined by the preset method.

[0034] Optionally, the preset method in the data determination unit includes at least one of a numerical simulation method, a machine learning method, a statistical analysis method, and a mathematical modeling method.

[0035] Optionally, the step of determining the linear training data and the nonlinear training data of the target soil according to the linear boundary condition and the nonlinear boundary condition in the data determination unit includes:

[0036] Constructing an evaluation model based on a machine learning algorithm according to the historical feature data;

[0037] Determining a first corresponding relationship between the linear feature and a preset soil health state according to the linear boundary condition, and determining the first corresponding relationship as the linear training data;

[0038] A second corresponding relationship between the nonlinear feature and a preset soil health state is determined according to the nonlinear boundary condition, and the second corresponding relationship is determined as the nonlinear training data.

[0039] Optionally, determining a target reduced-order model for health assessment according to the detection data in the soil assessment unit includes:

[0040] When the detection data only contains linear features, determining that the target reduced-order model for health assessment using the detection data is a linear reduced-order model; or,

[0041] When the detection data only contains nonlinear features, determining that the target reduced-order model for health assessment using the detection data is a nonlinear reduced-order model; or,

[0042] When the detection data contains both linear features and nonlinear features, it is determined that the target reduced-order model of the detection data used for health assessment is a linear reduced-order model and a nonlinear reduced-order model.

[0043] In the embodiment provided by the present application, the linear and nonlinear characteristics of the soil are first determined through the historical characteristic data of the soil, and then the training data of the soil is determined through the boundary conditions of the linear and nonlinear characteristics, so as to determine the linear and nonlinear reduced-order models of the soil through the training data. When the soil in the area to be tested is subsequently tested, the corresponding reduced-order model is first determined through the test data of the soil and a health assessment is performed. This allows the present application to accurately identify and analyze the complex characteristics of the soil, and can provide more targeted evaluation results according to different soil conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of the method provided in the embodiment of the present application;

[0045] Figure 2 A structural diagram of a device provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0048] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0049] It should be understood that although the terms first, second, third, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0050] The present application provides a regional soil health assessment method and device based on machine learning to solve the problem that the assessment results output by the existing soil assessment model lack pertinence, resulting in inaccurate assessment results.

[0051] The technical solution of the present application is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0052] like Figure 1 As shown, it is a flowchart of a regional soil health assessment method based on machine learning provided by this application, and the process may include the following steps:

[0053] Step S101, collecting historical characteristic data of target soil under various climatic conditions, wherein the historical characteristic data includes linear characteristics and nonlinear characteristics.

[0054] In this embodiment, the target soil may be soil in a specific area or soil in any area. For soil in a specific area, its historical characteristic data may be obtained through early collection. For soil in any area, its historical characteristic data may be obtained through some public databases, such as the China Soil Database, the International Soil Data Network, etc.

[0055] After obtaining the historical characteristic data, it can be divided into linear characteristics or nonlinear characteristics according to the various soil properties in the data. For example, the historical characteristic data contains characteristic values ​​of pH value, organic matter content, total nitrogen content, available phosphorus content, available potassium content, salt, water content, microbial diversity, and crop yield under each characteristic value, etc. For every 1 unit increase in pH value, the crop yield increases by a certain amount, and the pH value can be determined as a linear characteristic. Total nitrogen content, available phosphorus content, and available potassium content can also be determined as linear characteristics. The effects of soil moisture, salt, and microbial diversity on yield are discontinuous and can be determined as nonlinear characteristics.

[0056] Step S102, determining the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determining the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions.

[0057] In this embodiment, the linear feature takes pH as an example, and the pH range when there is a linear relationship between pH and yield is determined through historical feature data, and this linear relationship is valid within this range. For example, if the analysis shows that crop yield and pH are linearly related between pH 4.0 and 7.0, then this range is the boundary condition of the linear feature pH.

[0058] Taking organic matter content as an example of nonlinear characteristics, a scatter plot of organic matter content and crop yield was drawn through the above historical characteristic data, the turning point or saturation point of the relationship between organic matter content and crop yield was determined, and these points were determined as the boundary conditions of the nonlinear characteristic organic matter content.

[0059] After determining the boundary conditions of linear and nonlinear features, the boundary conditions are input into the machine learning algorithm evaluation model constructed by historical feature data to determine the corresponding relationship between linear features and soil health status, and determine it as linear training data. Then, the corresponding relationship between nonlinear features and soil health status is determined and determined as nonlinear training data.

[0060] For example, assume that a set of historical characteristic data A contains five parameters: pH, organic matter, total nitrogen, available phosphorus and salt. First, determine the weight vector (w_1, w_2, w_3, w_4, w_5) of the historical characteristic data, which represents the weights of pH, organic matter, total nitrogen, available phosphorus and salt, respectively, and (w_1+w_2+w_3+w_4+w_5=1). There are many specific methods for determining weights, such as through expert opinions, statistical analysis or machine learning, which will not be repeated here. And because the measurement units and magnitudes of different soil properties may be different, the collected data needs to be standardized. This usually involves converting the raw data into dimensionless fractions or percentages for comparison and weighting.

[0061] Therefore, after determining the weights and standardizing each parameter, the specific value of the soil health index SHI can be determined through the formula SHI = w_1*{standardized pH value}+w_2*{standardized organic matter}+w_3*{standardized total nitrogen}+w_4*{standardized available phosphorus}+w_5*{standardized salt}.

[0062] As shown in the following table:

[0063]

[0064]

[0065] The corresponding relationship between pH value, organic matter, total nitrogen, available phosphorus and soil health index SHI in the above table is the linear training data of historical characteristic data A, and the corresponding relationship between salt and soil health index SHI is the nonlinear training data of historical characteristic data A.

[0066] In another embodiment, the boundary conditions of the above-mentioned linear characteristics and nonlinear characteristics can be determined by at least one of numerical simulation method, machine learning method, statistical analysis method, and mathematical modeling method.

[0067] Step S103, training the pre-constructed linear initial reduced order model with the linear training data to generate the linear reduced order model of the target soil, and training the pre-constructed nonlinear initial reduced order model with the non-linear training data to generate the non-linear reduced order model of the target soil.

[0068] In this embodiment, a dimensionality reduction technique will be used to construct a low-dimensional model. This can be achieved through a variety of methods, such as principal component analysis (PCA), singular value decomposition (SVD), etc. Assuming that PCA is selected as the dimensionality reduction technique, we project the original data into a low-dimensional space composed of principal components to obtain a low-dimensional representation.

[0069] For example, assume that the collected linear features include three parameters: pH value (PH), total nitrogen content (N), and available phosphorus content (P), and each parameter has N data points. First, organize these data into a matrix X = [PH1, N1, P1; PH2, N2, P2; ... PH N ,N N ,P N ]

[0070] Each row represents a data point and each column represents a parameter. Before performing PCA, it is usually necessary to standardize the data to eliminate the dimensional effects between different parameters. Each column of the standardized data matrix Xs will have zero mean and unit variance.

[0071] Then the covariance matrix C = (1 / (N-1))*Xs.T*Xs of the standardized data matrix Xs is calculated, and the covariance matrix C is eigendecomposed by the function eig to obtain the eigenvalue λ and the corresponding eigenvector V. The eig function is used to calculate the eigenvalue and eigenvector of the square matrix.

[0072] Then, according to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues, and these eigenvectors constitute the projection matrix W. Then, the original data matrix Xs is projected to the low-dimensional space formed by the projection matrix W through the formula Xp=Xs*W to obtain the reduced-dimensional data matrix Xp. Xp is the data representation after dimensionality reduction, which is the above-mentioned linear initial reduction model.

[0073] After the linear initial reduced-order model is determined, it is trained through linear training data to achieve iterative optimization, thereby obtaining a linear reduced-order model of soil.

[0074] Similarly, the nonlinear reduced-order model of soil is obtained by the above method.

[0075] Step S104, obtaining detection data of the soil in the area to be detected, determining a target reduced-order model for health assessment based on the detection data, determining target boundary conditions of the soil in the area to be detected through the detection data and the historical characteristic data, and inputting the target boundary conditions into the target reduced-order model to obtain health estimation information of the target soil.

[0076] In this embodiment, since the detection data may be missing, abnormal, etc., the detection data can be optimized. For example, by combining and comparing the detection data with historical feature data, invalid data in the detection data can be removed and missing data can be corrected.

[0077] After the optimization of the test data is completed, it is determined whether each feature in the test data is a linear feature or a nonlinear feature. After the target boundary conditions of the soil in the area to be tested are determined by the above-mentioned preset method, such as the statistical analysis method, a corresponding reduced order model is assigned to it. When the test data only contains linear features, it is determined that the target reduced order model of the test data for health assessment is a linear reduced order model. When the test data only contains nonlinear features, it is determined that the target reduced order model of the test data for health assessment is a nonlinear reduced order model. When the test data contains both linear features and nonlinear features, it is determined that it needs to be evaluated simultaneously by a linear reduced order model and a nonlinear reduced order model.

[0078] In another embodiment, a first model and a second model of the target soil may also be constructed using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic;

[0079] Optimizing the linear features and the nonlinear features in the target feature data respectively by using the first model and the second model to obtain the processed target feature data, wherein the optimization processing includes at least one of correction, deduplication and deletion of invalid data;

[0080] The target boundary conditions of the detection data are then determined by the preset method.

[0081] In this embodiment, the detection data of the soil in the area to be detected is optimized by using a model, so that the optimized detection data can be more accurate, thereby improving the effectiveness of the detection data.

[0082] So far, completed Figure 1 The process shown.

[0083] In an embodiment of the present application, the linear and nonlinear characteristics of the soil are first determined through the historical characteristic data of the soil, and then the training data of the soil is determined through the boundary conditions of the linear and nonlinear characteristics, so as to determine the linear and nonlinear reduced-order models of the soil through the training data. When the soil in the area to be tested is subsequently tested, the corresponding reduced-order model is first determined through the test data of the soil and a health assessment is performed. This allows the present application to accurately identify and analyze the complex characteristics of the soil, and can provide more targeted evaluation results based on different soil conditions.

[0084] like Figure 2 As shown, the present application also provides a regional soil health assessment device based on machine learning, the device comprising:

[0085] A feature collection unit 201 is used to collect historical feature data of target soil under various climatic conditions, wherein the historical feature data includes linear features and nonlinear features;

[0086] A data determination unit 202 is used to determine the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determine the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions;

[0087] A model generation unit 203 is used to train a pre-constructed linear initial reduced order model through the linear training data to generate a linear reduced order model of the target soil, and to train a pre-constructed nonlinear initial reduced order model through the non-linear training data to generate a non-linear reduced order model of the target soil;

[0088] The soil assessment unit 204 is used to obtain the test data of the soil in the area to be tested, determine the target reduced-order model for health assessment according to the test data, determine the target boundary conditions of the soil in the area to be tested through the test data and the historical characteristic data, and input the target boundary conditions into the target reduced-order model to obtain health estimation information of the target soil.

[0089] In another embodiment, the data determination unit determines the target boundary condition of the soil in the to-be-detected area by using the detection data and the historical characteristic data, including:

[0090] Constructing a first model and a second model of the target soil by using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic;

[0091] By using the first model and the second model, the linear features and the nonlinear features in the target feature data are optimized respectively to obtain the processed target feature data;

[0092] The target boundary conditions of the detection data are then determined by the preset method.

[0093] In another embodiment, the preset method in the data determination unit includes at least one of a numerical simulation method, a machine learning method, a statistical analysis method, and a mathematical modeling method.

[0094] In another embodiment, the data determination unit determines the linear training data and the nonlinear training data of the target soil according to the linear boundary condition and the nonlinear boundary condition, including:

[0095] Constructing an evaluation model based on a machine learning algorithm according to the historical feature data;

[0096] Determining a first corresponding relationship between the linear feature and a preset soil health state according to the linear boundary condition, and determining the first corresponding relationship as the linear training data;

[0097] A second corresponding relationship between the nonlinear feature and a preset soil health state is determined according to the nonlinear boundary condition, and the second corresponding relationship is determined as the nonlinear training data.

[0098] In another embodiment, determining the target reduced-order model for health assessment according to the detection data in the soil assessment unit includes:

[0099] When the detection data only contains linear features, determining that the target reduced-order model for health assessment using the detection data is a linear reduced-order model; or,

[0100] When the detection data only contains nonlinear features, determining that the target reduced-order model for health assessment using the detection data is a nonlinear reduced-order model; or,

[0101] When the detection data contains both linear features and nonlinear features, it is determined that the target reduced-order model of the detection data used for health assessment is a linear reduced-order model and a nonlinear reduced-order model.

[0102] The above embodiment of the present invention provides a method for regional soil health assessment based on machine learning, and based on this method, provides a method device for regional soil health assessment based on machine learning. Through the above method and device, the present application can solve the problem of accurately identifying and analyzing the complex characteristics of soil, and can provide more targeted assessment results according to different soil conditions.

[0103] This embodiment also discloses a computer device, such as Figure 3As shown, the computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement any of the above-mentioned regional soil health assessment methods based on machine learning.

[0104] In addition, in the implementation of the regional soil health assessment device based on machine learning, the logical division of each program module is only an example. In actual applications, the above functions can be assigned to different program modules as needed, for example, for the convenience of corresponding hardware configuration requirements or software implementation. That is, the internal structure of the regional soil health assessment device based on machine learning is divided into different program modules to complete all or part of the functions described above.

[0105] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the scope of protection of the present application.

Claims

1. A regional soil health assessment method based on machine learning, characterized in that: The method comprises: Collecting historical characteristic data of target soil under various climatic conditions, wherein the historical characteristic data includes linear characteristics and nonlinear characteristics; Determine the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determine the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions; The pre-constructed linear initial reduced order model is trained by the linear training data to generate the linear reduced order model of the target soil, and the pre-constructed nonlinear initial reduced order model is trained by the non-linear training data to generate the non-linear reduced order model of the target soil; The detection data of the soil in the area to be detected is obtained, a target reduced-order model for health assessment is determined according to the detection data, a target boundary condition of the soil in the area to be detected is determined by using the detection data and the historical characteristic data, and the target boundary condition is input into the target reduced-order model to obtain health estimation information of the target soil.

2. The method according to claim 1, characterized in that Determining the target boundary condition of the soil in the area to be detected by using the detection data and the historical characteristic data includes: Constructing a first model and a second model of the target soil by using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic; Optimizing the linear features and the nonlinear features in the target feature data respectively by using the first model and the second model to obtain the processed target feature data, wherein the optimization processing includes at least one of correction, deduplication and deletion of invalid data; The target boundary conditions of the detection data are then determined by the preset method.

3. The method according to claim 1, characterized in that The preset method includes at least one of a numerical simulation method, a machine learning method, a statistical analysis method, and a mathematical modeling method.

4. The method according to claim 1, characterized in that: The step of determining the linear training data and the nonlinear training data of the target soil according to the linear boundary condition and the nonlinear boundary condition comprises: Constructing an evaluation model based on a machine learning algorithm according to the historical feature data; Determining a first corresponding relationship between the linear feature and a preset soil health state according to the linear boundary condition, and determining the first corresponding relationship as the linear training data; A second corresponding relationship between the nonlinear feature and a preset soil health state is determined according to the nonlinear boundary condition, and the second corresponding relationship is determined as the nonlinear training data.

5. The method according to claim 1, characterized in that Determining a target reduced-order model for health assessment according to the detection data comprises: When the detection data only contains linear features, determining that the target reduced-order model for health assessment using the detection data is a linear reduced-order model; or, When the detection data only contains nonlinear features, determining that the target reduced-order model for health assessment using the detection data is a nonlinear reduced-order model; or, When the detection data contains both linear features and nonlinear features, it is determined that the target reduced-order model for health assessment using the detection data is a linear reduced-order model and a nonlinear reduced-order model.

6. A regional soil health assessment device based on machine learning, characterized in that: The device comprises: A feature collection unit, used to collect historical feature data of target soil under various climatic conditions, wherein the historical feature data includes linear features and nonlinear features; A data determination unit, used to determine the linear boundary conditions of the linear feature and the nonlinear boundary conditions of the nonlinear feature by a preset method, and then determine the linear training data and the nonlinear training data of the target soil according to the linear boundary conditions and the nonlinear boundary conditions; A model generation unit, configured to train a pre-constructed linear initial reduced order model using the linear training data to generate a linear reduced order model of the target soil, and to train a pre-constructed nonlinear initial reduced order model using the non-linear training data to generate a non-linear reduced order model of the target soil; A soil assessment unit is used to obtain detection data of the soil in the area to be detected, determine a target reduced-order model for health assessment according to the detection data, determine a target boundary condition of the soil in the area to be detected through the detection data and the historical characteristic data, and input the target boundary condition into the target reduced-order model to obtain health estimation information of the target soil.

7. The device according to claim 6, characterized in that The data determination unit determines the target boundary condition of the soil in the area to be detected by using the detection data and the historical characteristic data, including: Constructing a first model and a second model of the target soil by using the historical characteristic data, wherein the first model corresponds to the linear characteristic and the second model corresponds to the nonlinear characteristic; Optimizing the linear features and the nonlinear features in the target feature data respectively by using the first model and the second model to obtain the processed target feature data, wherein the optimization processing includes at least one of correction, deduplication and deletion of invalid data; The target boundary conditions of the detection data are then determined by the preset method.

8. The device according to claim 6, characterized in that The preset method in the data determination unit includes at least one of a numerical simulation method, a machine learning method, a statistical analysis method, and a mathematical modeling method.

9. The device according to claim 6, characterized in that The data determination unit determines the linear training data and the nonlinear training data of the target soil according to the linear boundary condition and the nonlinear boundary condition, including: Constructing an evaluation model based on a machine learning algorithm according to the historical feature data; Determining a first corresponding relationship between the linear feature and a preset soil health state according to the linear boundary condition, and determining the first corresponding relationship as the linear training data; A second corresponding relationship between the nonlinear feature and a preset soil health state is determined according to the nonlinear boundary condition, and the second corresponding relationship is determined as the nonlinear training data.

10. The device according to claim 6, characterized in that Determining the target reduced-order model for health assessment according to the detection data in the soil assessment unit includes: When the detection data only contains linear features, determining that the target reduced-order model for health assessment using the detection data is a linear reduced-order model; or, When the detection data only contains nonlinear features, determining that the target reduced-order model for health assessment using the detection data is a nonlinear reduced-order model; or, When the detection data contains both linear features and nonlinear features, it is determined that the target reduced-order model for health assessment using the detection data is a linear reduced-order model and a nonlinear reduced-order model.

Images