Soil fertility inversion method and system based on improved myxomycete algorithm

Abstract

The invention relates to a soil fertility inversion method and system based on an improved myxomycete algorithm, and the method comprises the steps: collecting a hyperspectral remote sensing image of a target region, and obtaining a remote sensing image only containing soil pixels through an SVM; performing spectral feature transformation to obtain a high-dimensional spectral feature set, performing spectral feature optimization based on an improved myxobacteria optimization algorithm to obtain an optimal feature subset, and inputting a pre-trained statistical regression model between spectral features and soil parameters to obtain a soil fertility spatial distribution diagram of the target area; according to the improved myxomycete optimization algorithm, a priori Chebyshev initialization strategy fusing PLS and Lasso coefficients is adopted, a multi-level position updating strategy and a reverse learning and sparrow search algorithm are adopted to obtain a high-quality feature subset, a PLS weight coefficient is utilized to perform preliminary screening to obtain a strong correlation feature subset, a composite fitness calculation function is utilized to perform quality evaluation, and a high-quality feature subset is obtained. And obtaining an optimal feature subset. Compared with the prior art, the method improves the applicability and accuracy of soil fertility inversion.

Description

Technical Field

[0001] This invention relates to the field of soil fertility detection technology, and in particular to a soil fertility inversion method and system based on an improved slime mold algorithm. Background Technology

[0002] Soil, as the most basic natural resource for human survival, plays a vital role in supporting crop production, providing ecological resources, and regulating environmental factors. It is an important element in the feedback cycle of the ecosystem and provides strong support for human survival and development.

[0003] In remote sensing estimation studies based on soil composition, external environment, sensor characteristics, and human operation can all introduce irrelevant information and noise, making it difficult to extract effective features. Therefore, it is necessary to further employ spectral feature dimensionality reduction methods to optimize feature representation, highlight soil spectral feature information, and improve the accuracy and reliability of soil parameter inversion results. Currently, feature selection and feature extraction are two main strategies for feature dimensionality reduction and have been widely applied in the field of hyperspectral remote sensing. Feature selection methods are suitable for processing high-dimensional datasets with few samples. They generate feature subsets of the original spectral dataset by removing irrelevant or redundant features from the original feature space. For example, recursive feature elimination repeatedly builds machine learning models to inversely filter spectral features, eliminating the least important features in each round to determine the optimal subset. However, traditional feature selection methods often struggle to solve NP-hard combinatorial optimization problems efficiently due to high computational complexity. Intelligent optimization strategies based on metaheuristic algorithms, through a global search mechanism, can effectively balance the dual optimization objectives of feature correlation and redundancy, providing a new solution to feature optimization problems.

[0004] However, current soil parameter feature extraction frameworks based on metaheuristic optimization are still in the exploratory stage, and most are applicable to feature selection problems of single soil parameters. For example, patent application CN118840661A discloses a soil parameter inversion method based on a heuristic intelligent selection algorithm, which is also an improved soil hyperspectral feature selection technology based on the marine predator algorithm. This method optimizes spectral feature selection by introducing weight information and a reverse learning strategy in the population initialization stage and introducing mutation operations in the later stage of iteration, and infers soil parameters based on a statistical regression model. Although this method claims to be applicable to generalized "soil parameters", its fitness function, weight calculation and band ratio strategy are all deeply coupled with the spectral characteristics of a single soil organic matter parameter. It fails to build a general framework that can be extended to the collaborative inversion of multiple parameters (such as water content, total nitrogen, heavy metals, etc.), and is essentially still limited to single parameter optimization problems, unable to break through the applicability bottleneck of existing metaheuristic algorithms in soil remote sensing inversion. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a soil fertility inversion method and system based on an improved slime mold algorithm, which significantly improves the applicability and accuracy of soil fertility inversion.

[0006] The objective of this invention can be achieved through the following technical solutions: A soil fertility inversion method based on an improved slime mold algorithm includes the following steps: Collect hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM; The remote sensing image containing only soil pixels is subjected to spectral feature transformation to obtain a high-dimensional spectral feature set; The improved slime mold optimization algorithm is used to optimize the spectral features of the high-dimensional spectral feature set to obtain the optimal feature subset. In the population initialization stage, the improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population. In the spectral feature encirclement stage, a multi-level position update strategy, reverse learning, and sparrow search algorithm are used to obtain a high-quality feature subset. The high-quality feature subset is initially screened using PLS weight coefficients to obtain a strongly correlated feature subset. The quality of each strongly correlated feature subset is quantitatively evaluated by a composite fitness calculation function to obtain the optimal feature subset. The optimal feature subset is input into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area.

[0007] Furthermore, the specific steps for acquiring hyperspectral remote sensing images of the target area and obtaining target images containing only soil pixels using SVM include: The hyperspectral remote sensing image is divided into two categories, soil and other land features, by using SVM, and a binary mask is generated to distinguish between soil areas and non-soil areas. The binary mask that distinguishes between soil and non-soil regions is overlaid on the hyperspectral remote sensing image to obtain a target image containing only soil pixels.

[0008] Furthermore, the specific steps for performing spectral feature transformation on the remote sensing image containing only soil pixels to obtain a high-dimensional spectral feature set include: A multidimensional mathematical transformation is performed on the original spectrum of the remote sensing image containing only soil pixels to obtain multidimensional spectral transformation features. The Pearson correlation coefficients between the multidimensional spectral transformation features and soil fertility indicators are calculated band by band in wavelength order. The feature combination with the highest Pearson correlation coefficient is retained to form a high-dimensional spectral feature set.

[0009] Furthermore, the mathematical transformations include: reciprocal transformation, reciprocal logarithmic transformation, first-order differential transformation, envelope removal, and band ratio method.

[0010] Furthermore, the soil fertility indicators include multiple indicators among total phosphorus content, total potassium content, total nitrogen content, and soil organic matter content.

[0011] Furthermore, the prior Chebyshev initialization strategy for fusing PLS and Lasso coefficients is as follows: In the formula, The initial solution vector for starting the iteration of the improved slime mold optimization algorithm. For Chebyshev chaotic mapping, For a randomly generated solution, For PLS weighting coefficients, These are the Lasso weighting coefficients.

[0012] Furthermore, the multi-level position update strategy is as follows: In the formula, The first in the population The individual in the first Position at the next iteration For inertial weights, As of the date In the next iteration, the entire population finds the globally optimal position. for Random numbers generated within the interval for Random numbers generated within the interval To search for oscillation parameters globally, The weighting factor for random individuals of slime mold. and Based on the golden ratio Calculated search parameters, In the first The position of individual A is randomly selected from the population in the next iteration. In the first The position of individual B is randomly selected from the population in the next iteration. The first in the population The individual in the first Position at the next iteration To control the oscillation parameters of the slime mold search direction, The golden sine factor, The maximum number of iterations, For Levy's flight random walk strategy, For adaptive probability threshold, For a fixed probability threshold, its empirical value is 0.9. A random number generated within the range [0,1]. The expression for the inertia weight is: In the formula, A random number generation function on [0,1]. This represents the current iteration number; The expression for the weighting factor of the random individuals of the slime mold is: In the formula, This represents the optimal fitness value in the current population. These are the sorted fitness values. is the worst fitness value in the current population, and N is the total number of individuals in the population.

[0013] Furthermore, the specific steps for initially screening the high-quality feature subset using PLS weighting coefficients to obtain the strongly correlated feature subset are as follows: Based on the partial least squares regression model, the PLS weight coefficients corresponding to each feature in the high-dimensional spectral feature set are calculated, and each feature is sorted in ascending order according to the PLS weight coefficients to form an ordered feature sequence. Based on the total number of features in the ordered feature sequence, the features are divided into high-weight features and low-weight features. The specific steps include: when the total number of features in the ordered feature sequence is less than or equal to 25, the first feature after sorting is classified as a low-weight feature and the remaining features are classified as high-weight features; when the total number of features in the ordered feature sequence is greater than 25, the features in the top 1% after sorting are classified as low-weight features and the remaining features are classified as high-weight features. The high-weight features and low-weight features are selected based on their retention probabilities. Low-weight features are removed from the high-dimensional spectral feature set, while high-weight features are retained to form a strongly correlated feature subset.

[0014] Furthermore, the expression for the composite fitness calculation function is: In the formula, For the first The fitness of a feature subset For precision weights, As the coefficient of determination, For simplicity weight, The total characteristic number, The number of features included in the strongly correlated feature subset selected for the currently evaluated slime mold individual. For adjustment coefficients, This represents the PLS weighting coefficient.

[0015] According to another aspect of the present invention, a soil fertility inversion system based on an improved slime mold algorithm is provided, comprising: The remote sensing image acquisition module is used to acquire hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM. The feature transformation module is used to perform spectral feature transformation on the remote sensing image containing only soil pixels to obtain a high-dimensional spectral feature set. The feature optimization module is used to perform spectral feature optimization on the high-dimensional spectral feature set based on the improved slime mold optimization algorithm to obtain the optimal feature subset. The improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population in the population initialization stage. In the spectral feature encirclement stage, a multi-level position update strategy, reverse learning, and sparrow search algorithm are used to obtain a high-quality feature subset. The high-quality feature subset is initially screened using PLS weight coefficients to obtain a strongly correlated feature subset. The merits of each strongly correlated feature subset are quantitatively evaluated by a composite fitness calculation function to obtain the optimal feature subset. The soil fertility inversion module is used to input the optimal feature subset into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention introduces an improved slime mold optimization algorithm. In the feature selection stage, it adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a high-quality initial population. It also enhances global exploration and local development capabilities by introducing multi-level position update mechanisms such as dynamic adjustment of inertial weights, Levy algorithm, and golden section coefficient. At the same time, it uses a composite fitness function to quantitatively evaluate the accuracy, simplicity, and weight of feature subsets. This invention solves the defects of traditional feature selection methods in processing high-dimensional spectral data, such as high computational complexity, low search efficiency, and easy getting trapped in local optima due to the NP-hard combinatorial optimization problem. It significantly improves the efficiency and accuracy of feature selection for soil fertility inversion.

[0017] 2. This invention constructs a high-dimensional spectral feature set and designs a multi-index compatible composite fitness calculation function to perform multi-dimensional mathematical transformations on soil pixel remote sensing images to generate spectral features covering multiple indicators such as total phosphorus, total potassium, total nitrogen, and organic matter. It uses a pre-trained statistical regression model to simultaneously invert multiple soil fertility parameters and output a comprehensive fertility spatial distribution map. This solves the problem that existing feature selection methods based on metaheuristic algorithms are mostly limited to the inversion of single soil parameters and have insufficient model practicality. It significantly improves the generalization ability and application efficiency of soil fertility inversion and supports a one-time comprehensive assessment of soil fertility status. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process for a soil fertility inversion method based on an improved slime mold algorithm proposed in this invention. Figure 2 This is a flowchart illustrating the preliminary screening of spectral features using PLS weighting coefficients. Figure 3 This is a spatial distribution map of total phosphorus content; Figure 4 This is a spatial distribution map of total potassium content; Figure 5 This is a spatial distribution map of total nitrogen content; Figure 6 This is a spatial distribution map of soil organic matter content. Detailed Implementation

[0019] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0020] The following English abbreviations are involved: Support Vector Machine (SVM) Partial Least Squares Regression (PLSR) Partial Least Squares (PLS) Improved Slime Mould Algorithm (ISMA) Example 1 This embodiment provides a soil fertility inversion method based on an improved slime mold algorithm, such as... Figure 1 As shown, it includes the following steps: S1. Acquire hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM.

[0021] Acquiring hyperspectral remote sensing images of the target area, including pixels of soil, vegetation, buildings, roads, and rivers, and obtaining a target image containing only soil pixels using SVM involves the following steps: The hyperspectral remote sensing image was divided into two categories, soil and other land cover, by using SVM, and a binary mask was generated to distinguish between soil areas and non-soil areas. A binary mask that distinguishes between soil and non-soil areas is overlaid on the hyperspectral remote sensing image to obtain a target image containing only soil pixels.

[0022] S2. Perform spectral feature transformation on the remote sensing image containing only soil pixels to obtain a high-dimensional spectral feature set.

[0023] The specific steps for performing spectral feature transformation on remote sensing images containing only soil pixels to obtain a high-dimensional spectral feature set include: Multidimensional mathematical transformation is performed on the original spectrum of a remote sensing image containing only soil pixels to obtain multidimensional spectral transformation features. The Pearson correlation coefficient between the multidimensional spectral transformation features and soil fertility indicators is calculated band by band in wavelength order. The feature combination with the highest Pearson correlation coefficient is retained to form a high-dimensional spectral feature set.

[0024] Mathematical transformations include: reciprocal transformation, reciprocal logarithmic transformation, first-order differential transformation, envelope removal, and band ratio method, as shown in the table below.

[0025] Table 1 Spectral Feature Transformation In the formula, Original reflectivity The corresponding value of the calculated envelope, , They represent wavelengths of and The reflectivity value at that location, and .

[0026] Soil fertility indicators include multiple indicators from total phosphorus content, total potassium content, total nitrogen content, and soil organic matter content.

[0027] S3. Based on the improved slime mold optimization algorithm, the high-dimensional spectral feature set is optimized to obtain the optimal feature subset.

[0028] The improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population during the population initialization stage.

[0029] The prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients is as follows: In the formula, The initial solution vector for starting the iteration of the improved slime mold optimization algorithm. For Chebyshev chaotic mapping, For a randomly generated solution, For PLS weighting coefficients, These are the Lasso weighting coefficients.

[0030] The improved slime mold optimization algorithm replaces the original algorithm's random initialization with a Chebyshev chaotic operator. By utilizing random numbers generated by a chaotic system to initialize the population, it improves population diversity. Simultaneously, to ensure the algorithm focuses more on important features during initialization, PLS and Lasso coefficients are incorporated. The PLS weighting coefficient effectively captures the nonlinear correlation between spectral bands and soil properties; however, this process may introduce weakly correlated features. The Lasso coefficient achieves sparsity by adding a penalty term (L1 regularization) to the optimization objective function, reducing the weights of weakly correlated bands that PLS might retain to zero, thus simplifying the model and improving its generalization ability.

[0031] In the original algorithm, the position update method is determined by the position of the current best individual and the sum of the positions of two random individuals. In this case, the slime mold randomly explores around the current best position. This randomness is aimless, and a single search strategy may cause the algorithm to converge prematurely, potentially missing important features. The improved slime mold optimization algorithm employs a multi-level position update strategy, back-learning, and sparrow search algorithm in the enclosing spectral feature stage to obtain a high-quality feature subset. The PLS weight coefficient is used to initially screen the high-quality feature subset to obtain a strongly correlated feature subset. Finally, a composite fitness calculation function is used to quantitatively evaluate the quality of each strongly correlated feature subset to obtain the optimal feature subset.

[0032] The multi-level location update strategy is as follows: In the formula, The first in the population The individual in the first Position at the next iteration For inertial weights, As of the date In the next iteration, the entire population finds the globally optimal position. for Random numbers generated within the interval for Random numbers generated within the interval To search for oscillation parameters globally, The weighting factor for random individuals of slime mold. and Based on the golden ratio Calculated search parameters, In the first The position of individual A is randomly selected from the population in the next iteration. In the first The position of individual B is randomly selected from the population in the next iteration. The first in the population The individual in the first Position at the next iteration To control the oscillation parameters of the slime mold search direction, The golden sine factor, The maximum number of iterations, For Levy's flight random walk strategy, For adaptive probability threshold, For a fixed probability threshold, its empirical value is 0.9. A random number generated within the range [0,1].

[0033] The expression for the weighting factor of random slime mold individuals is: In the formula, This represents the optimal fitness value in the current population. These are the sorted fitness values. is the worst fitness value in the current population, and N is the total number of individuals in the population.

[0034] Since the golden ratio does not require gradient information, each step only requires one iteration and has a definite shrinkage step size. Furthermore, the sine value of each point on the sine function corresponds to the ordinate value of a point on the unit circle centered at the origin. Scanning a unit circle with the sine function is similar to exploring a search space, possessing good global exploration capabilities.

[0035] To better coordinate the algorithm's global and local optimization capabilities, the ISMA algorithm uses improved inertia weights. Inertial weights represent the influence of the velocity of the previous generation of particles on the velocity of the current generation of particles. The larger the value, the stronger the ability to explore new regions and the stronger the global optimization ability. To enable the algorithm to have a high global search ability in the early stage and a high local search ability in the later stage to improve the convergence accuracy, so it should not be a fixed constant. However, for the more commonly used method is the linear reduction of the inertia weight. Generally, the optimization process of the spectral characteristics of slime molds is non-linear and very complex. The linear reduction method has too simple changes, resulting in limited adjustment and adaptation ability to the non-linear and complex optimization process, and it is easy to fall into local extrema. On this basis, the present invention uses a function to construct the evolution curve of the inertia weight. This is because the sine function can introduce periodic changes, enabling the inertia weight to be dynamically adjusted within a certain range. The expression of the inertia weight is shown as follows: In the formula, is a random number generation function on [0,1], is the current iteration number.

[0036] Among them, a high inertia weight value can enhance global exploration, allowing slime molds to search new regions with large steps, while a low inertia weight value is biased towards local development, finely adjusting the current position.

[0037] Since the Levy algorithm does not require too many parameters to be set in advance, has strong adaptability, and can automatically adjust the search strategy according to the actual problem, it is introduced into the position update strategy of the improved algorithm. For the complexity of spectral data, Levy can better adapt to different high-dimensional data distributions, thus improving the effect of feature selection. By setting q = 0.9, the ISMA algorithm is controlled: when r < q, Levy flight is used with a high probability for global exploration to expand the search range; when r >= q, the perturbation similar to the original algorithm is retained with a low probability to maintain the local development ability and adapt to the high-dimensional and discrete characteristics of the feature selection problem.

[0038] Because the individual's original motion direction is maintained through the inertia term and local exploration is combined with random factors. However, this may cause individuals to gather in the local optimal region. Therefore, random reverse learning is introduced to enhance the robustness of the algorithm through randomness elements and dynamics. The addition of reverse learning can effectively avoid over-reliance on certain features in the feature selection process, enhance the diversity of feature subsets, and thus improve the robustness of the model. Therefore, reverse learning is used to break the local convergence caused by inertial updates and the existing feature combination limitations, helping to discover features with higher discrimination and representativeness. The position update formula is as follows: In the formula, is the population position state after the update of the random reverse learning strategy, This is the upper bound of the search range for the slime mold algorithm. This is the lower bound of the search range for the slime mold algorithm. The value is a random value within the range [0,1]. The state of the population after the above multi-level position update strategy.

[0039] To further increase the breadth and depth of the algorithm search and thus explore the problem space more comprehensively, a sparrow search algorithm is introduced, which searches through multiple individual sparrows. The position update formula is as follows; In the formula, This represents the population position status after the sparrow search strategy has been updated. Based on the perturbation mechanism in the sparrow search algorithm, the updated position is calculated according to the current best position, worst position, and the individual's position before this step. It is a random probability value. It is a random number that follows a normal distribution. Indicates the first The position of the individual with the worst fitness value in the next iteration. Indicates the first The position of the individual with the best fitness value in the next iteration. This represents a matrix whose elements are randomly assigned the value 1 or -1, and , It is a matrix whose elements are all 1.

[0040] Because this algorithm has a high computational cost (requiring multiple iterations of training), to improve its efficiency, the original algorithm, after employing a multi-level position update strategy, utilizes PLS weight coefficients for preliminary screening of spectral features to obtain a subset of strongly correlated features, such as... Figure 2 As shown, the specific steps are as follows: Based on the partial least squares regression model, the PLS weight coefficients corresponding to each feature in the high-dimensional spectral feature set are calculated, and each feature is sorted in ascending order according to the PLS weight coefficients to form an ordered feature sequence. Based on the total number of features in the ordered feature sequence, the features are divided into high-weight features and low-weight features. The specific steps include: when the total number of features in the ordered feature sequence is less than or equal to 25, the first feature after sorting is classified as a low-weight feature and the remaining features are classified as high-weight features; when the total number of features in the ordered feature sequence is greater than 25, the features in the top 1% after sorting are classified as low-weight features and the remaining features are classified as high-weight features. The high-weighted and low-weighted features are selected based on their retention probabilities. The specific steps are as follows: If the retention probability of a high-weight feature is greater than 0.5, then its retention probability is further increased so that the high-weight feature has a greater probability of being retained. If the retention probability of low-weight features is less than 0.5, then further reduce their retention probability so that low-weight features have a greater probability of being discarded.

[0041] Low-weight features are removed from the high-dimensional spectral feature set, while high-weight features are retained to form a strongly correlated feature subset.

[0042] The optimal feature subset is obtained by quantitatively evaluating the merits of each strongly correlated feature subset using a composite fitness calculation function. The expression for the composite fitness calculation function is as follows: In the formula, For the first The fitness of a feature subset For precision weights, As the coefficient of determination, For simplicity weight, The total characteristic number, The number of features included in the strongly correlated feature subset selected for the currently evaluated slime mold individual. For adjustment coefficients, This represents the PLS weighting coefficient.

[0043] S4. Input the optimal feature subset into the pre-trained statistical regression model between spectral features and soil parameters to obtain the spatial distribution map of soil fertility in the target area.

[0044] First, based on ground-measured samples and their test data, surface soil samples were randomly divided into training and test sets. Based on the training set data, a statistical regression model between soil hyperspectral characteristics and key fertility indicators was constructed using the XGBoost machine learning algorithm.

[0045] The optimal feature subset is input into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area. The spatial distribution map of total phosphorus content is shown below. Figure 3 As shown in the figure, the spatial distribution of total potassium content is as follows: Figure 4 As shown, the spatial distribution map of total nitrogen content is as follows: Figure 5 As shown in the figure, the spatial distribution map of soil organic matter content is as follows: Figure 6 As shown.

[0046] Example 2 This embodiment provides a soil fertility inversion system based on an improved slime mold algorithm, including: The remote sensing image acquisition module is used to acquire hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM. The feature transformation module is used to perform spectral feature transformation on remote sensing images containing only soil pixels to obtain a high-dimensional spectral feature set. The feature optimization module is used to optimize the spectral features of a high-dimensional spectral feature set based on an improved slime mold optimization algorithm to obtain the optimal feature subset. In the population initialization stage, the improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population. In the spectral feature encirclement stage, a multi-level position update strategy, reverse learning, and sparrow search algorithm are used to obtain a high-quality feature subset. The high-quality feature subset is initially screened using PLS weight coefficients to obtain a strongly correlated feature subset. Finally, the quality of each strongly correlated feature subset is quantitatively evaluated using a composite fitness calculation function to obtain the optimal feature subset. The soil fertility inversion module is used to input the optimal feature subset into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area.

[0047] The rest is the same as in Example 1.

[0048] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A soil fertility inversion method based on an improved slime mold algorithm, characterized in that, Includes the following steps: Collect hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM; The remote sensing image containing only soil pixels is subjected to spectral feature transformation to obtain a high-dimensional spectral feature set; The improved slime mold optimization algorithm is used to optimize the spectral features of the high-dimensional spectral feature set to obtain the optimal feature subset. In the population initialization stage, the improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population. In the spectral feature encirclement stage, a multi-level position update strategy, reverse learning, and sparrow search algorithm are used to obtain a high-quality feature subset. The high-quality feature subset is initially screened using PLS weight coefficients to obtain a strongly correlated feature subset. The quality of each strongly correlated feature subset is quantitatively evaluated by a composite fitness calculation function to obtain the optimal feature subset. The optimal feature subset is input into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area.

2. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The specific steps for acquiring hyperspectral remote sensing images of the target area and obtaining a target image containing only soil pixels using SVM include: The hyperspectral remote sensing image is divided into two categories, soil and other land features, by using SVM, and a binary mask is generated to distinguish between soil areas and non-soil areas. The binary mask that distinguishes between soil and non-soil regions is overlaid on the hyperspectral remote sensing image to obtain a target image containing only soil pixels.

3. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The specific steps for performing spectral feature transformation on the remote sensing image containing only soil pixels to obtain a high-dimensional spectral feature set include: A multidimensional mathematical transformation is performed on the original spectrum of the remote sensing image containing only soil pixels to obtain multidimensional spectral transformation features. The Pearson correlation coefficients between the multidimensional spectral transformation features and soil fertility indicators are calculated band by band in wavelength order. The feature combination with the highest Pearson correlation coefficient is retained to form a high-dimensional spectral feature set.

4. The soil fertility inversion method based on the improved slime mold algorithm according to claim 3, characterized in that, The mathematical transformations include: reciprocal transformation, reciprocal logarithmic transformation, first-order differential transformation, envelope removal, and band ratio method.

5. The soil fertility inversion method based on the improved slime mold algorithm according to claim 3, characterized in that, The soil fertility indicators include multiple indicators from total phosphorus content, total potassium content, total nitrogen content, and soil organic matter content.

6. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The prior Chebyshev initialization strategy for fusing PLS and Lasso coefficients is as follows: In the formula, The initial solution vector for starting the iteration of the improved slime mold optimization algorithm. For Chebyshev chaotic mapping, For a randomly generated solution, For PLS weighting coefficients, These are the Lasso weighting coefficients.

7. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The multi-level location update strategy is as follows: In the formula, The first in the population The individual in the first Position at the next iteration For inertial weights, As of the date In the next iteration, the entire population finds the globally optimal position. for Random numbers generated within the interval for Random numbers generated within the interval To search for oscillation parameters globally, The weighting factor for random individuals of slime mold. and Based on the golden ratio Calculated search parameters, In the first The position of individual A is randomly selected from the population in the next iteration. In the first The position of individual B is randomly selected from the population in the next iteration. The first in the population The individual in the first Position at the next iteration To control the oscillation parameters of the slime mold search direction, The golden sine factor, The maximum number of iterations, For Levy's flight random walk strategy, For adaptive probability threshold, For a fixed probability threshold, its empirical value is 0.

9. A random number generated within the range [0,1]. The expression for the inertia weight is: In the formula, A random number generation function on [0,1]. This represents the current iteration number; The expression for the weighting factor of the random individuals of the slime mold is: In the formula, This represents the optimal fitness value in the current population. These are the sorted fitness values. is the worst fitness value in the current population, and N is the total number of individuals in the population.

8. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The specific steps for initially screening the high-quality feature subset using PLS weighting coefficients to obtain the strongly correlated feature subset are as follows: Based on the partial least squares regression model, the PLS weight coefficients corresponding to each feature in the high-dimensional spectral feature set are calculated, and each feature is sorted in ascending order according to the PLS weight coefficients to form an ordered feature sequence. Based on the total number of features in the ordered feature sequence, the features are divided into high-weight features and low-weight features. The specific steps include: when the total number of features in the ordered feature sequence is less than or equal to 25, the first feature after sorting is classified as a low-weight feature and the remaining features are classified as high-weight features; when the total number of features in the ordered feature sequence is greater than 25, the features in the top 1% after sorting are classified as low-weight features and the remaining features are classified as high-weight features. The high-weight features and low-weight features are selected based on their retention probabilities. Low-weight features are removed from the high-dimensional spectral feature set, while high-weight features are retained to form a strongly correlated feature subset.

9. The soil fertility inversion method based on the improved slime mold algorithm according to claim 1, characterized in that, The expression for the composite fitness calculation function is: In the formula, For the first The fitness of a feature subset For precision weights, As the coefficient of determination, For simplicity weight, The total characteristic number, The number of features included in the strongly correlated feature subset selected for the currently evaluated slime mold individual. For adjustment coefficients, This represents the PLS weighting coefficient.

10. A soil fertility inversion system based on an improved slime mold algorithm, characterized in that, include: The remote sensing image acquisition module is used to acquire hyperspectral remote sensing images of the target area and obtain remote sensing images containing only soil pixels through SVM. The feature transformation module is used to perform spectral feature transformation on the remote sensing image containing only soil pixels to obtain a high-dimensional spectral feature set. The feature optimization module is used to perform spectral feature optimization on the high-dimensional spectral feature set based on the improved slime mold optimization algorithm to obtain the optimal feature subset. The improved slime mold optimization algorithm adopts a prior Chebyshev initialization strategy that integrates PLS and Lasso coefficients to generate a diverse initial feature subset population in the population initialization stage. In the spectral feature encirclement stage, a multi-level position update strategy, reverse learning, and sparrow search algorithm are used to obtain a high-quality feature subset. The high-quality feature subset is initially screened using PLS weight coefficients to obtain a strongly correlated feature subset. The merits of each strongly correlated feature subset are quantitatively evaluated by a composite fitness calculation function to obtain the optimal feature subset. The soil fertility inversion module is used to input the optimal feature subset into a pre-trained statistical regression model between spectral features and soil parameters to obtain a spatial distribution map of soil fertility in the target area.

Images