Straw returning fertilization optimization method fused with soil carbon nitrogen ratio

Abstract

The invention relates to the field of agricultural fertilization, in particular to a soil carbon nitrogen ratio fused straw returning fertilization optimization method, which comprises the following steps of: performing standardization processing on farmland static characteristic data to obtain a farmland standardization characteristic vector and a characteristic standardization threshold value; performing coupling analysis on the straw returning amount and the soil carbon-nitrogen ratio in the farmland standardized feature vector set to obtain risk factors; performing path gradient integration on the risk potential energy field function to obtain a symmetric path cost gain factor; performing path cost gain correction on the Euclidean distance to obtain adaptive distance measurement; and fuzzy C-means clustering is performed on the farmland standardized feature vectors through adaptive distance measurement to obtain differentiated fertilization management partitions, so that the problem of inaccurate fertilization decision caused by incapability of identifying key feature nonlinear coupling risks in existing management partitions based on Euclidean distance is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of agricultural fertilization, and particularly relates to a straw returning to field fertilization optimization method fusing soil carbon-nitrogen ratio. BACKGROUND

[0002] In the practice of modern precision agriculture, in order to realize the differentiation and fine management of large-area farmland, it is usually necessary to preform management zoning. The purpose of management zoning is to divide the land blocks with similar soil properties and production potential into the same management unit, so as to facilitate the application of uniform field operation standards. At present, the fuzzy C-means (FCM) clustering algorithm is a common technical means to realize the management zoning of farmland. This method collects and analyzes the soil physical and chemical property data of multiple sampling points in the farmland, such as soil organic matter content, total nitrogen content, available phosphorus content, available potassium content and pH value, and represents each sampling point as a feature vector composed of multiple static attributes. Subsequently, the FCM algorithm calculates the similarity according to the Euclidean distance of each feature vector in the multi-dimensional space, and through an iterative optimization process, the land blocks with high similarity are divided into the same category, thereby forming the management zoning. This kind of zoning method based on static soil property data provides a basis for guiding fertilization, irrigation and other agricultural activities under normal conditions.

[0003] However, with the wide application of straw returning to field agricultural mode, the management zoning method based on the standard FCM algorithm gradually shows limitations. After straw returning to field, the nitrogen balance and nitrogen availability of the land block not only depend on its inherent static soil properties, but also are significantly affected by the dynamic biochemical process dominated by straw characteristics, especially the nitrogen stealing effect caused by microorganisms in the decomposition of straw. The core similarity measurement of the existing FCM algorithm still depends on the Euclidean distance, which is an isotropic linear space measurement method. It can only independently measure the numerical difference of each feature dimension and perform linear accumulation, and it is difficult to capture the possible nonlinear coupling effect between different feature dimensions. In the actual scene of straw returning to field, there is a significant interaction between soil carbon-nitrogen ratio and straw returning amount. When high carbon-nitrogen ratio and high straw returning amount appear in the same land block at the same time, the risk of microbial nitrogen stealing they cause presents a sharply increasing nonlinear trend, rather than a simple addition of two independent risks. The traditional FCM algorithm ignores this kind of nonlinear coupling effect, which may divide the land block with multiple high-risk characteristics into the same zoning as the land block with only single high-risk characteristic or no risk characteristic, resulting in deviation of the zoning result from the actual needs of agronomy. The limitation of distance measurement makes the management zoning result of the existing method inaccurate in the scene of straw returning to field, and it is difficult to provide accurate and reliable decision basis for subsequent fertilization optimization. SUMMARY

[0004] In view of this, the present invention aims to propose a straw return fertilization optimization method integrating soil carbon-nitrogen ratio to solve the problem that the existing management zoning based on Euclidean distance cannot identify the nonlinear coupling risk of key features and leads to inaccurate fertilization decisions.

[0005] To achieve the above object, the technical solution of the present invention is achieved as follows:

[0006] A straw return fertilization optimization method integrating soil carbon-nitrogen ratio comprises the following steps:

[0007] Step S1: obtaining the farmland standardized feature vector and feature standardization threshold by standardizing the farmland static feature data;

[0008] Step S2: Obtain risk factors by coupling analysis of the amount of straw returned to the field and the soil carbon-nitrogen ratio in the standardized characteristic vector of the farmland;

[0009] Step S3: Obtain a symmetric path cost gain factor by performing path gradient integration on the risk factor;

[0010] Step S4: Obtaining an adaptive distance metric between grid sampling points in the farmland by performing path cost gain correction on the Euclidean distance between grid sampling points in the farmland;

[0011] Step S5: Perform fuzzy C-means clustering on the standardized feature vectors of the farmland through the adaptive distance metric between the grid sampling points in the farmland to obtain differentiated fertilization management zones.

[0012] Furthermore, the method of obtaining the farmland standardized feature vector and feature standardized threshold by standardizing the farmland static feature data includes:

[0013] Gridding was performed within the target farmland area, and sampling points were set up in each grid. Soil samples were collected at each sampling point, and static soil characteristic data corresponding to the sampling location were recorded, including soil organic matter content, soil total nitrogen content, amount of straw returned to the field, and straw type.

[0014] The soil total carbon content is obtained by approximate analysis of the soil organic matter content; the soil carbon-nitrogen ratio is obtained by evaluating the ratio of the soil total carbon content to the soil total nitrogen content;

[0015] For any target grid sampling point in the target farmland area, the straw return amount, soil carbon-nitrogen ratio, soil organic matter content, soil total nitrogen content, and soil total carbon content of the target grid sampling point are Z-score standardized in all grid sampling points, and the corresponding multidimensional feature vector is used as the multidimensional feature vector of the target grid sampling point;

[0016] The straw return amount threshold and the soil carbon-nitrogen ratio threshold were set, and the straw return amount threshold and the soil carbon-nitrogen ratio threshold were Z-score standardized in the straw return amount feature dimension and the soil carbon-nitrogen ratio feature dimension, respectively, to obtain the standardized straw return amount threshold and the soil carbon-nitrogen ratio threshold.

[0017] Furthermore, the risk factors are obtained by coupling analysis of the amount of straw returned to the field and the soil carbon-nitrogen ratio in the standardized characteristic vector sink of farmland, including:

[0018] By performing threshold integration processing on the standardized characteristic data of straw return amount and soil basic carbon-nitrogen ratio, a nonlinear risk response assessment is obtained; and by performing characteristic coupling analysis on the nonlinear risk response assessment, the risk factor is obtained.

[0019] Furthermore, the threshold integration processing is performed on the standardized characteristic data of the amount of straw returned to the field and the basic carbon-nitrogen ratio of the soil to obtain a nonlinear risk response assessment, including:

[0020] For any target grid sampling point in the target farmland area, the standardized threshold value of straw return amount is divided by the square of the calculated result of the straw return amount at the target grid sampling point as the threshold difference assessment of the straw return amount at the target grid sampling point; the reciprocal of the sum of the constant 1 and the threshold difference assessment of the straw return amount at the target grid sampling point is used as the state conversion function of the straw return amount at the target grid sampling point; the state conversion function of the straw return amount at the target grid sampling point is used as the integral of the integrand in the numerical range from the constant 0 to the straw return amount at the target grid sampling point as the risk response assessment of the straw return amount at the target grid sampling point;

[0021] The standardized threshold of soil carbon-nitrogen ratio is divided by the square of the calculated result of soil carbon-nitrogen ratio at the target grid sampling point as the soil carbon-nitrogen ratio threshold difference assessment at the target grid sampling point; the reciprocal of the sum of constant 1 and the soil carbon-nitrogen ratio threshold difference assessment at the target grid sampling point is used as the state conversion function of soil carbon-nitrogen ratio at the target grid sampling point; the state conversion function of soil carbon-nitrogen ratio at the target grid sampling point is used as the integral function in the range from constant 0 to the soil carbon-nitrogen ratio value of the target grid sampling point as the soil carbon-nitrogen ratio risk response assessment at the target grid sampling point;

[0022] The result of multiplying the risk response assessment of the straw return amount at the target grid sampling point by the risk response assessment of the soil carbon-nitrogen ratio at the target grid sampling point is used as the nonlinear risk response assessment of the target grid sampling point.

[0023] Furthermore, the risk factors are obtained by performing characteristic coupling analysis on the nonlinear risk response assessment, including:

[0024] Set a risk gain coefficient; multiply the risk gain coefficient with the nonlinear risk response evaluation of the target grid sampling point and add the calculation result of constant 1 as the risk factor of the target grid sampling point.

[0025] Further, the symmetric path cost gain factor is obtained by path gradient integration of the risk potential field function, including:

[0026] The risk gradient vector of each sampling point on the path is obtained by gradient calculation processing of the risk factor; the cumulative risk gradient value of the path is obtained by integrating the modulus of the risk gradient vector on the path; the symmetric path cost gain factor is obtained by path scale normalization processing of the cumulative risk gradient value.

[0027] Further, the risk gradient vector of each sampling point on the path is obtained by gradient calculation processing of the risk factor, including:

[0028] For any target grid sampling point in the target farmland area, the partial derivative of the risk factor to the characteristic dimension is calculated as the gradient component on the characteristic dimension in the straw returning amount characteristic dimension and the soil carbon-nitrogen ratio characteristic dimension respectively; the gradient component is set to 0 in the characteristic dimensions of soil organic matter content, soil total nitrogen content and soil total carbon content; the gradient components of each characteristic dimension are combined according to the characteristic dimensions to form a vector as the risk gradient vector of the target grid sampling point.

[0029] Further, the cumulative risk gradient value of the path is obtained by integrating the modulus of the risk gradient vector on the path, including:

[0030] Set the integral step; in the process of clustering the grid sampling points, all cluster center points in the clustering iteration process are obtained; the path between the target grid sampling point and any target cluster center is divided into multiple continuous sampling positions according to the set integral step; the modulus value of the risk gradient vector is calculated at each sampling position, and the calculation result of the cumulative sum of the risk gradient vector modulus values of each sampling position is taken as the cumulative risk gradient value of the path.

[0031] Further, the symmetric path cost gain factor is obtained by path scale normalization processing of the cumulative risk gradient value, including:

[0032] Set the path cost gain coefficient; set the benchmark distance scale parameter through the Euclidean distance between all non-repeated point pairs in all grid sampling points; add the square of the Euclidean distance between the target grid sampling point and the target cluster center to the benchmark distance scale parameter and use it as the path scale normalization denominator; multiply the cumulative risk gradient value by the Euclidean distance between the target grid sampling point and the target cluster center as the path scale normalization numerator, and multiply the corresponding fraction by the path cost gain coefficient and add the constant 1 to calculate the result as the symmetric path cost gain factor.

[0033] Furthermore, the adaptive distance metric is obtained by performing path cost gain correction on the Euclidean distance, including:

[0034] The Euclidean distance between the target grid sampling point and the target cluster center is multiplied by the symmetric path cost gain factor between the target grid sampling point and the target cluster center as the adaptive distance metric between the target grid sampling point and the target cluster center.

[0035] Compared with the prior art, the present invention has the following advantages:

[0036] The present invention describes a straw return fertilization optimization method that integrates the soil carbon-nitrogen ratio. By introducing a risk potential field and constructing a path cost gain factor, it enables the identification of the nonlinear coupling effect between the amount of straw returned and the soil carbon-nitrogen ratio when zoning farmland. In actual straw return scenarios, when fields have both a high straw load and a high carbon-nitrogen ratio, the microbial decomposition process significantly intensifies nitrogen competition, making it difficult for traditional Euclidean distance-based clustering methods to accurately identify this risk. The technical approach proposed in this invention can form a sensitive response to risk thresholds in feature space and, through dynamic correction of risk gradients, enable the differences between different fields to be more agronomically characterized, thereby avoiding the misclassification of high-risk and low-risk areas. At the application level, this invention not only improves the scientific nature and accuracy of farmland management zoning, but also directly outputs differentiated fertilization recommendations. High-coupling risk areas can be identified and nitrogen fertilizer compensation strategies can be obtained, high-straw load areas can receive nitrogen guarantees during the start-up period, and areas with limited soil nitrogen supply can receive support from long-term improvement plans. By linking this zoning with fertilization strategies, the present invention provides farmers with a comprehensive fertilization optimization method that takes into account both short-term crop needs and long-term soil health, thereby improving crop yield and quality while promoting the sustainable use of soil resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The accompanying drawings, which constitute part of the present invention, are provided to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are provided to explain the present invention and do not constitute an undue limitation of the present invention. In the accompanying drawings:

[0038] Figure 1 A method flow chart of a straw returning fertilization optimization method fusing soil carbon-nitrogen ratio according to an embodiment of the present application. DETAILED DESCRIPTION

[0039] The present application will be described in detail below with reference to the accompanying drawings and embodiments.

[0040] Reference Figure 1 is a method flow chart of a straw returning fertilization optimization method fusing soil carbon-nitrogen ratio provided by the first embodiment of the present application, as shown in Figure 1 , a straw returning fertilization optimization method fusing soil carbon-nitrogen ratio can include:

[0041] Step S1, obtaining a farmland standardized feature vector and a feature standardization threshold by standardizing farmland static feature data.

[0042] This step aims to provide farmland feature data under a unified scale for subsequent risk assessment and clustering analysis. First, grid segmentation is performed in the target farmland area, and a sampling point is set in each grid. Soil samples are collected at each sampling point location, and the static soil feature data corresponding to the sampling location are recorded, including soil organic matter content, soil total nitrogen content, straw returning amount, and straw type. The soil total carbon content is obtained by approximate analysis of soil organic matter content. The soil carbon-nitrogen ratio is obtained by ratio evaluation of soil total carbon content and soil total nitrogen content. For any target grid sampling point in the target farmland area, the straw returning amount, soil carbon-nitrogen ratio, soil organic matter content, soil total nitrogen content, and soil total carbon content of the target grid sampling point are subjected to Z-score standardization in each dimension among all grid sampling points, and the corresponding multi-dimensional feature vector is formed as the multi-dimensional feature vector of the target grid sampling point.

[0043] The straw returning amount threshold and the soil carbon-nitrogen ratio threshold are set. In the present embodiment, the straw returning amount threshold is set to 600 kg per mu, and the soil carbon-nitrogen ratio threshold is set to 12. The straw returning amount threshold and the soil carbon-nitrogen ratio threshold can be adjusted according to the actual farmland scene, and are not required. The straw returning amount threshold and the soil carbon-nitrogen ratio threshold are subjected to Z-score standardization in the straw returning amount feature dimension and the soil carbon-nitrogen ratio feature dimension, respectively, to obtain a straw returning amount standardization threshold and a soil carbon-nitrogen ratio standardization threshold.

[0044] At this point, the farmland standardized feature vector and the feature standardization threshold are obtained by standardizing the farmland static feature data.

[0045] Step S2, obtaining a risk factor by coupling analysis of the straw returning amount and the soil carbon-nitrogen ratio in the farmland standardized feature vector.

[0046] When performing cluster calculations for farmland management zoning, the prior art usually directly uses the original static eigenvalues ​​of each field for distance measurement, which regards the multidimensional feature space as uniform. However, in the specific application scenario of straw return to the field, the potential possibility of a field undergoing decomposition and nitrogen loss is determined by the coupling relationship between the two characteristics of straw return amount and soil basic carbon-nitrogen ratio. The distance measurement of the prior art, because it regards the contribution of each feature dimension as independent of each other, cannot effectively identify the coupling effect generated by a specific combination of eigenvalues. Therefore, in order to construct a similarity metric that can better reflect agronomy, the first step of the present invention is to construct a risk potential field in the entire feature space. The construction logic of the risk potential field is that the potential energy value of any point in the space should be able to reflect the specific combination state formed when the two characteristics of straw return amount and soil basic carbon-nitrogen ratio corresponding to the point are at a high level at the same time. By constructing such a potential field function, a basic reference is provided for the subsequent distance calculation of risk terrain.

[0047] In summary, first, the standardized characteristic data of straw return amount and soil basic carbon-nitrogen ratio are processed by threshold integration to obtain nonlinear risk response assessment. Specifically, for any target grid sampling point in the target farmland area, the standardized threshold of straw return amount is divided by the square of the calculated result of the straw return amount of the target grid sampling point as the threshold difference assessment of the straw return amount of the target grid sampling point; the reciprocal of the sum of the constant 1 and the threshold difference assessment of the straw return amount of the target grid sampling point is used as the state conversion function of the straw return amount of the target grid sampling point; the state conversion function of the straw return amount of the target grid sampling point is used as the integral of the integrand in the numerical range of the constant 0 to the straw return amount of the target grid sampling point as the straw return amount risk of the target grid sampling point. Response assessment; the standardized threshold of soil carbon-nitrogen ratio is divided by the square of the calculated result of the soil carbon-nitrogen ratio of the target grid sampling point as the soil carbon-nitrogen ratio threshold difference assessment of the target grid sampling point; the reciprocal of the sum of the constant 1 and the soil carbon-nitrogen ratio threshold difference assessment of the target grid sampling point is used as the state conversion function of the soil carbon-nitrogen ratio of the target grid sampling point; the state conversion function of the soil carbon-nitrogen ratio of the target grid sampling point is used as the integral of the integrand in the numerical range from constant 0 to the soil carbon-nitrogen ratio of the target grid sampling point as the soil carbon-nitrogen ratio risk response assessment of the target grid sampling point; the calculation result of multiplying the straw return amount risk response assessment of the target grid sampling point by the soil carbon-nitrogen ratio risk response assessment of the target grid sampling point is used as the nonlinear risk response assessment of the target grid sampling point.

[0048] In one embodiment, assuming that The standardized value of the amount of straw returned to the field at each grid sampling point is ;No. The standardized value of soil carbon-nitrogen ratio at each grid sampling point is: The straw returning to field quantity standardization threshold value is The soil carbon nitrogen ratio standardization threshold value is The non-linear risk response evaluation of the first grid sampling point is calculated according to the following expression:

[0049]

[0050] Wherein, the non-linear risk response evaluation of the first grid sampling point is represented by The non-linear risk response evaluation of the first grid sampling point is represented by The straw returning to field quantity standardization value of the first grid sampling point is represented by The soil carbon nitrogen ratio standardization value of the first grid sampling point is represented by The straw returning to field quantity standardization value of the first grid sampling point is represented by The soil carbon nitrogen ratio standardization value of the first grid sampling point is represented by The straw returning to field quantity standardization threshold value is represented by The soil carbon nitrogen ratio standardization threshold value is represented by The straw returning to field quantity standardization value in the integral function is represented by The soil carbon nitrogen ratio standardization value in the integral function is represented by

[0051] After obtaining the non-linear risk response evaluation of the target grid sampling point, the feature coupling analysis is continued on the non-linear risk response evaluation to obtain the risk factor. Specifically, the risk gain coefficient is set, which is set to 1 in the embodiment of the present application, and can be adjusted according to the actual scene, and is not required. The calculation result of multiplying the risk gain coefficient and the non-linear risk response evaluation of the target grid sampling point and adding 1 is taken as the risk factor of the target grid sampling point.

[0052] In an embodiment, the risk gain coefficient is The risk factor of the first grid sampling point is calculated according to the following expression:

[0053]

[0054] Wherein, the risk factor of the first grid sampling point is represented by The risk factor of the first grid sampling point is represented by The risk gain coefficient is represented by The non-linear risk response evaluation of the first grid sampling point is represented by

[0055] ​​​​​It should be noted that in the actual scenario of returning straw to the field, the existing technology directly uses the original eigenvalue to calculate the distance, which brings two specific problems: first, it incorrectly assumes that the risk increases linearly with the increase of the eigenvalue; second, it cannot identify the coupling effect caused by the simultaneous existence of multiple risk factors. First, in order to solve the nonlinear problem of risk assessment, the present invention does not directly use the eigenvalue but constructs the characteristic state conversion term, namely and In actual scenarios, when the amount of straw returned to the field or the soil carbon-nitrogen ratio is lower than the safety threshold, its contribution to the risk of decomposition and nitrogen loss is negligible; once the threshold is exceeded, the risk will rise sharply. The construction of the integral term in the above formula is precisely to capture this nonlinear situation. The integrand in the integral term ensures that when the eigenvalue is much smaller than the threshold, its contribution to the integral area is extremely small, and when the eigenvalue exceeds the threshold, its contribution to the integral area will increase rapidly. Therefore, by calculating the cumulative area of ​​the integral, a linear original eigenvalue is converted into an indicator that can truly reflect the degree of risk accumulation, thereby solving the problem of inaccurate assessment of a single risk factor. Secondly, in order to solve the coupling problem of risk assessment, that is, to identify the increased risk brought about by the simultaneous occurrence of high straw amount and high soil carbon-nitrogen ratio, the existing technology regards the risks of different characteristics as independent and additive, which is inconsistent with the facts. The present invention adopts a feature coupling term, that is, by multiplying the results of the above-mentioned integral transformation of the two key characteristics, so that in a nitrogen-rich soil (low carbon-nitrogen ratio), even if a large amount of straw is added, the nitrogen deprivation risk is controllable; on the contrary, in a nitrogen-poor soil (high carbon-nitrogen ratio), the addition of a large amount of straw will trigger catastrophic nitrogen competition. This multiplication structure mathematically ensures that only when the values ​​of the two characteristics of straw return amount and soil basic carbon-nitrogen ratio corresponding to a grid sampling point simultaneously exceed their respective risk thresholds, the nitrogen deprivation risk will be controlled. and , so that only when the calculated results of both integral terms are large will their product yield a significantly higher value than in other cases. If only a single feature is at a high risk level, its corresponding integral term, although large, will still be low after multiplying it with the smaller integral term. Through this design, the final calculated risk factor for each grid sampling point is large enough to effectively identify and quantify the coupled risk resulting from the combined action of the two key risk factors.

[0056] At this point, the risk factors are obtained by coupling analysis of the amount of straw returned to the field and the soil carbon-nitrogen ratio in the standardized characteristic vector of farmland.

[0057] Step S3: Obtain a symmetric path cost gain factor by performing path gradient integration on the risk factor.

[0058] In step S2, the present invention defines a non-uniform risk potential field in the multidimensional feature space by constructing risk factors, which can reflect the inherent coupling risk level of each plot. However, if clustering is simply modified based on the risk factor of each point, two problems still exist: First, because the risk factor maps multidimensional features to a one-dimensional scalar, plots with different feature combinations may have the same calculated risk factor, resulting in potential information confusion; second, the standard distance metric is inherently spatially uniform, and it fails to reflect the different sensitivities of the similarity metric between plots in different regions of the risk potential field.

[0059] Therefore, the second optimization step of the present invention aims to address the aforementioned issues. Its core logic is that the true agronomic similarity between any target grid sampling point and any target cluster center should not be determined solely by their respective risk factors, but rather by the complexity of the terrain traversed by the path connecting the two points within the risk potential field. A path that traverses an area of ​​drastic risk variation (i.e., a large risk gradient) indicates greater differences in agronomic management between the two points, and their similarity should be correspondingly reduced (i.e., the distance should be amplified). To achieve this goal, the present invention constructs a symmetric path gain factor. This factor quantitatively measures the ruggedness of the path by integrating the gradient of the risk factor at each point along the straight path between any target grid sampling point and any target cluster center. This ruggedness is an inherent property of the path and is independent of direction, thus fundamentally ensuring the symmetry of the final distance metric. This design ensures that even if two plots of land with different origins happen to have the same risk factor, their different locations in feature space mean that the paths connecting them to the same cluster center are different, traversing different risk terrain and, consequently, different calculated path ruggedness. This not only solves the problem of information confusion with the same score but also, more importantly, enables the final distance metric to truly reflect the real differences between fields due to crossing different risk areas.

[0060] To summarize, first, the risk gradient vector of each sampling point on the path is obtained by performing gradient calculation on the risk factor. Specifically, for any target grid sampling point in the target farmland area, the partial derivative of the risk factor with respect to the characteristic dimension of straw return amount and the characteristic dimension of soil carbon-nitrogen ratio is calculated as the gradient component on the characteristic dimension; the gradient component in the characteristic dimensions of soil organic matter content, soil total nitrogen content, and soil total carbon content is set to 0; the gradient component of each characteristic dimension is combined according to the characteristic dimension to form a vector, which is used as the risk gradient vector of the target grid sampling point.

[0061] Then, the module of the risk gradient vector on the path is integrated to obtain the cumulative risk gradient value of the path. Specifically, an integration step size is set. In the embodiment of the present invention, the integration step size is set to the number of grid sampling points between the target grid sampling point and the target cluster center point, so that the integral approximate evaluation is performed by accumulating the module of the risk gradient vector of the grid sampling points between the target grid sampling point and the target cluster center point; in the process of clustering the grid sampling points, all cluster centers in the clustering iteration process are obtained; on the connection path between the target grid sampling point and any target cluster center, the path is divided into multiple continuous sampling positions according to the set integration step size; the module value of the risk gradient vector is calculated at each sampling position, and the calculation result of the cumulative sum of the module values ​​of the risk gradient vector at each sampling position is used as the cumulative risk gradient value of the path.

[0062] Finally, the cumulative risk gradient value is subjected to path scale normalization to obtain a symmetric path cost gain factor. Specifically, a path cost gain coefficient is set. In the embodiment of the present invention, the path cost gain coefficient is set to 1. It can be adjusted according to the actual scenario and is not required. The reference distance scale parameter is set by the Euclidean distance between non-repeated point pairs in all grid sampling points. In the embodiment of the present invention, the squares of the Euclidean distances between all grid sampling point pairs are accumulated and the arithmetic mean of the accumulated calculation results is used as the reference distance scale parameter. The square of the Euclidean distance between the target grid sampling point and the target cluster center is added to the reference distance scale parameter to obtain the result of the path scale normalization denominator. The cumulative risk gradient value is multiplied by the Euclidean distance between the target grid sampling point and the target cluster center to obtain the path scale normalization numerator. The corresponding fraction is multiplied by the path cost gain coefficient and the result of the addition to the constant 1 to obtain the symmetric path cost gain factor.

[0063] In one embodiment, it is assumed that the path cost gain coefficient is ;No. The risk gradient vector of the grid sampling point is ;No. The grid sampling points and The first The grid sampling points are ; The base distance scale parameter is , then The grid sampling points and The calculation expression of the symmetric path cost gain factor between cluster centers is:

[0064]

[0065] in, Indicates the The grid sampling points and The symmetric path cost gain factor between cluster centers; represents the path cost gain coefficient; Indicates the The grid sampling points and The first The risk gradient vector of each grid sampling point; Indicates the Multidimensional feature vector of grid sampling points; Indicates the The multidimensional feature vector of the cluster center points; Indicates the The cluster center and the The Euclidean distance between grid sampling points; Indicates the The grid sampling points and The first The magnitude of the risk gradient vector at each grid sampling point.

[0066] It should be noted that after step S2 solves the problem of quantifying the inherent risk of a single field, this step aims to solve the problem of how to correct the similarity measure between two points based on the risk terrain they are located in. The Euclidean distance in the prior art is like measuring on a flat map. It assumes that the feature space is uniform and cannot reflect the intrinsic agronomic differences between different regions. The symmetric path cost gain factor constructed by the present invention introduces the ability to perceive risk terrain into this similarity measure through the path integral structure. First, the core of the formula is the path gradient modulo integral term, that is, , which quantifies the The grid sampling points and The sum of the risk-changing regions traversed by the straight path of the cluster center point is calculated by discrete cumulative sum modulo integral of the path gradient in the embodiment of the present invention. The size of characterizes the steepness of the risk potential field. A path that passes through a flat risk area (small gradient modulus) will have a small integral value, which indicates that the two points are agronomically in a relatively homogeneous risk platform and their similarity is high. Conversely, a path that passes through an area where risk changes dramatically (large gradient modulus) will have a large integral value, which indicates that to classify the two points into one category, a significant risk gradient needs to be crossed, so their similarity should be reduced (i.e., the distance should be increased). Secondly, the cumulative risk gradient value is multiplied by the Euclidean distance between the target grid sampling point and the target cluster center to obtain the total path risk exposure. This ensures the measurement of similarity while taking into account the ruggedness and length of the path. Finally, the scale adjustment term introduced , normalizing the gain effect. In summary, the symmetric path cost gain factor enables the final distance assessment to move beyond a simple reflection of geometric distance. Instead, it incorporates information about the risk gradient along the path, providing a comprehensive measure that better reflects true agronomic differences. When comparing two points, if the path connecting them is flat, the symmetric path cost gain factor is close to 1, and the distance is primarily determined by Euclidean distance. If the path is rugged, the symmetric path cost gain factor will be significantly greater than 1, thereby amplifying the distance between them and reducing their similarity.

[0067] At this point, the symmetric path cost gain factor is obtained by performing path gradient integration on the risk factor.

[0068] Step S4, performing path cost gain correction on the Euclidean distance between the grid sampling points in the farmland to obtain an adaptive distance metric between the grid sampling points in the farmland.

[0069] After obtaining the symmetric path cost gain factor between the target grid sampling point and the target cluster center point, the path cost gain of the Euclidean distance is further corrected by the symmetric path cost gain factor to obtain an adaptive distance metric. Specifically, the Euclidean distance between the target grid sampling point and the target cluster center is multiplied by the symmetric path cost gain factor between the target grid sampling point and the target cluster center to obtain the adaptive distance metric between the target grid sampling point and the target cluster center.

[0070] At this point, the adaptive distance measurement between the grid sampling points in the farmland is obtained by performing path cost gain correction on the Euclidean distance between the grid sampling points in the farmland.

[0071] Step S5, performing fuzzy C-means clustering on the standardized feature vectors of the farmland by using the adaptive distance metric between grid sampling points in the farmland to obtain differentiated fertilization management zones.

[0072] The target number of clusters is set. In this embodiment of the present invention, the target number of clusters is set to 6. That is, the farmland grid sampling points are divided into 6 clusters through the clustering process. The membership matrix in the fuzzy C-means clustering process is updated by using the adaptive distance metric between the target grid sampling points and the target cluster centers, and the clustering process is completed to obtain the optimized partitioning results of the grid sampling points in the farmland.

[0073] After obtaining the optimized partitioning results of the grid sampling points in the farmland, firstly, a statistical analysis is performed on all the plots in each partition to calculate the representative characteristics of the partition. The average straw return amount and the average soil basic carbon-nitrogen ratio of all the plots in the partition are calculated. Subsequently, targeted fertilization management recommendations are generated based on the representative characteristics of each partition. For a partition whose internal members generally have the risk characteristics of coupling high straw return amount and high soil basic carbon-nitrogen ratio, fertilization should focus on adding a certain amount of fast-acting nitrogen fertilizer (such as urea) in the early stage of crop sowing to compensate for the temporary fixation of soil nitrogen by microorganisms when decomposing high carbon-nitrogen ratio organic materials, thereby avoiding nitrogen starvation in the crop seedling stage. For partitions that are identified as having insignificant risk characteristics and balanced soil fertility, a conventional balanced fertilization plan based on target yield is adopted. In addition, the method of the present invention can also identify two other important intermediate types of partitions:

[0074] First, high straw loading areas are characterized by significantly higher-than-average straw return rates but a low soil basal carbon-to-nitrogen ratio. This suggests that while the soil's basal nitrogen supply capacity is adequate, the excessive straw input still creates significant short-term nitrogen pressure. For this sub-area, a moderate increase in the amount of available nitrogen used as seed fertilizer or basal fertilizer is recommended to ensure that crop start-up is not impacted, but there is no need to significantly increase total nitrogen as in high-coupling risk areas.

[0075] Second, areas with limited soil nitrogen supply are characterized by a significantly higher-than-average soil basal carbon-to-nitrogen ratio and low-to-medium straw return rates. This suggests that the primary problem in this area isn't the straw returned that season, but rather the low nitrogen availability of the soil's organic matter and its poor basal nitrogen supply capacity. Fertilization in this zone should involve not only adjusting the seasonal nitrogen fertilizer application rate but also incorporating low-carbon-to-nitrogen ratio organic fertilizers or slow-release nitrogen fertilizers to improve soil properties and enhance basal nitrogen supply capacity over the long term.

[0076] At this point, the fuzzy C-means clustering of the standardized feature vectors of the farmland is completed through the adaptive distance measurement between the grid sampling points in the farmland to obtain the differentiated fertilization management zones.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A straw return fertilization optimization method integrating soil carbon-nitrogen ratio, characterized by: Step S1: obtaining the farmland standardized feature vector and feature standardization threshold by standardizing the farmland static feature data; Step S2: Obtain risk factors by coupling analysis of the amount of straw returned to the field and the soil carbon-nitrogen ratio in the standardized characteristic vector of the farmland; Step S3: Obtain a symmetric path cost gain factor by performing path gradient integration on the risk factor; Step S4: Obtaining an adaptive distance metric between grid sampling points in the farmland by performing path cost gain correction on the Euclidean distance between grid sampling points in the farmland; Step S5: Perform fuzzy C-means clustering on the standardized feature vectors of the farmland through the adaptive distance metric between the grid sampling points in the farmland to obtain differentiated fertilization management zones.

2. The method for optimizing straw return to field and fertilizing by integrating soil carbon-nitrogen ratio according to claim 1, characterized in that: The method of obtaining the farmland standardized feature vector and feature standardized threshold by standardizing the farmland static feature data includes: Gridding was performed within the target farmland area, and sampling points were set up in each grid. Soil samples were collected at each sampling point, and static soil characteristic data corresponding to the sampling location were recorded, including soil organic matter content, soil total nitrogen content, amount of straw returned to the field, and straw type. The soil total carbon content is obtained by approximate analysis of the soil organic matter content; the soil carbon-nitrogen ratio is obtained by evaluating the ratio of the soil total carbon content to the soil total nitrogen content; For any target grid sampling point in the target farmland area, the straw return amount, soil carbon-nitrogen ratio, soil organic matter content, soil total nitrogen content, and soil total carbon content of the target grid sampling point are Z-score standardized in all grid sampling points, and the corresponding multidimensional feature vector is used as the multidimensional feature vector of the target grid sampling point; The straw return amount threshold and the soil carbon-nitrogen ratio threshold were set, and the straw return amount threshold and the soil carbon-nitrogen ratio threshold were Z-score standardized in the straw return amount feature dimension and the soil carbon-nitrogen ratio feature dimension, respectively, to obtain the standardized straw return amount threshold and the soil carbon-nitrogen ratio threshold.

3. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 1, characterized in that: The risk factors are obtained by coupling analysis of the amount of straw returned to the field and the soil carbon-nitrogen ratio in the standardized characteristic vector sink of the farmland, including: By performing threshold integration processing on the standardized characteristic data of straw return amount and soil basic carbon-nitrogen ratio, a nonlinear risk response assessment is obtained; and by performing characteristic coupling analysis on the nonlinear risk response assessment, the risk factor is obtained.

4. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 3, characterized in that: The method performs threshold integration processing on the standardized characteristic data of the amount of straw returned to the field and the basic carbon-nitrogen ratio of the soil to obtain a nonlinear risk response assessment, including: For any target grid sampling point in the target farmland area, the standardized threshold value of straw return amount is divided by the square of the calculated result of the straw return amount at the target grid sampling point as the threshold difference assessment of the straw return amount at the target grid sampling point; the reciprocal of the sum of the constant 1 and the threshold difference assessment of the straw return amount at the target grid sampling point is used as the state conversion function of the straw return amount at the target grid sampling point; the state conversion function of the straw return amount at the target grid sampling point is used as the integral of the integrand in the numerical range from the constant 0 to the straw return amount at the target grid sampling point as the risk response assessment of the straw return amount at the target grid sampling point; The standardized threshold of soil carbon-nitrogen ratio is divided by the square of the calculated result of soil carbon-nitrogen ratio at the target grid sampling point as the soil carbon-nitrogen ratio threshold difference assessment at the target grid sampling point; the reciprocal of the sum of constant 1 and the soil carbon-nitrogen ratio threshold difference assessment at the target grid sampling point is used as the state conversion function of soil carbon-nitrogen ratio at the target grid sampling point; the state conversion function of soil carbon-nitrogen ratio at the target grid sampling point is used as the integral function in the range from constant 0 to the soil carbon-nitrogen ratio value of the target grid sampling point as the soil carbon-nitrogen ratio risk response assessment at the target grid sampling point; The result of multiplying the risk response assessment of the straw return amount at the target grid sampling point by the risk response assessment of the soil carbon-nitrogen ratio at the target grid sampling point is used as the nonlinear risk response assessment of the target grid sampling point.

5. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 3, characterized in that: The risk factors are obtained by performing characteristic coupling analysis on the nonlinear risk response assessment, including: Set a gain coefficient; multiply the gain coefficient by the nonlinear risk response assessment of the target grid sampling point and add the result to a constant 1 as the risk factor of the target grid sampling point; Set a risk gain coefficient; multiply the risk gain coefficient by the nonlinear risk response assessment of the target grid sampling point and add the result to a constant 1 as the risk factor of the target grid sampling point.

6. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 1, characterized in that: The method of obtaining the symmetric path cost gain factor by performing path gradient integration on the risk potential field function includes: By performing gradient calculation on the risk factor, the risk gradient vector of each sampling point on the path is obtained; by integrating the modulus of the risk gradient vector on the path, the cumulative risk gradient value of the path is obtained; The symmetric path cost gain factor is obtained by normalizing the cumulative risk gradient value to the path scale.

7. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 6, characterized in that: The risk gradient vector of each sampling point on the path is obtained by performing gradient calculation on the risk factor, including: For any target grid sampling point in the target farmland area, the partial derivatives of the risk factor with respect to the characteristic dimension of straw return amount and the characteristic dimension of soil carbon-nitrogen ratio are calculated as the gradient components in the characteristic dimension respectively; the gradient components in the characteristic dimensions of soil organic matter content, soil total nitrogen content and soil total carbon content are set to 0; the gradient components of each characteristic dimension are combined according to the characteristic dimension to form a vector, which is used as the risk gradient vector of the target grid sampling point.

8. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 6, characterized in that: The step of integrating the modulus of the risk gradient vector on the path to obtain the cumulative risk gradient value of the path includes: Set the integration step size; in the process of clustering the grid sampling points, obtain all cluster center points in the clustering iteration process; on the connection path between the target grid sampling point and any target cluster center, divide the path into multiple continuous sampling positions according to the set integration step size; calculate the modulus of the risk gradient vector at each sampling position, and take the cumulative risk gradient value of the path as the result of the sum of the modulus values ​​of the risk gradient vectors at each sampling position.

9. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 6, characterized in that: The step of performing path scale normalization on the cumulative risk gradient value to obtain a symmetric path cost gain factor includes: Set the path cost gain coefficient; set the benchmark distance scale parameter through the Euclidean distance between all non-repeated point pairs in all grid sampling points; add the square of the Euclidean distance between the target grid sampling point and the target cluster center to the benchmark distance scale parameter and use it as the path scale normalization denominator; multiply the cumulative risk gradient value by the Euclidean distance between the target grid sampling point and the target cluster center as the path scale normalization numerator, and multiply the corresponding fraction by the path cost gain coefficient and add the constant 1 to calculate the result as the symmetric path cost gain factor.

10. The method for optimizing straw return and fertilization by integrating soil carbon-nitrogen ratio according to claim 1, characterized in that: The adaptive distance metric is obtained by performing path cost gain correction on the Euclidean distance, including: The Euclidean distance between the target grid sampling point and the target cluster center is multiplied by the symmetric path cost gain factor between the target grid sampling point and the target cluster center as the adaptive distance metric between the target grid sampling point and the target cluster center.

Images