Crop nutrient information inversion method based on ELM inversion model

Abstract

The invention discloses a crop nutrient information inversion method based on an ELM inversion model, and the method comprises the following steps: obtaining leaf physicochemical parameters, canopy structure parameters, ground parameters and angle information of related crops, and constructing a rice canopy reflectivity simulation data set through an RPIOSL-UBM model; constructing an initial lookup table; constraining the initial lookup table; constructing a comprehensive data set by combining actual measurement spectrum data of related crops; carrying out ELM inversion model training in combination with a whale optimization algorithm; and carrying out nutrient information inversion on crops to be detected through the optimized ELM inversion model to obtain corresponding leaf area index LAI, normalized vegetation index NDVI, chlorophyll, nitrogen and moisture nutrient indexes, and generating a nutrient map. According to the method, the RPIOSL-UBM model is coupled, and ELM parameters are optimized in combination with actual measurement data constraints and the whale algorithm, so that high-precision and strong-generalization inversion of multi-nutrient indexes of crops is realized, and the method can be suitable for multi-growth-stage monitoring requirements in a complex field environment.

Description

Technical Field

[0001] The present invention relates to the field of agricultural remote sensing monitoring, and more particularly to a crop nutrient information inversion method based on an ELM inversion model. Background Art

[0002] In agricultural remote sensing monitoring, retrieval of crop nutrient information is a core technology for precision agriculture management. By analyzing crop canopy spectral characteristics and combining them with physical models or machine learning methods to retrieve key nutrient indicators such as leaf area index (LAI), chlorophyll, and nitrogen, a scientific basis can be provided for crop growth assessment and fertilization decisions. However, complex field environments, characterized by canopy-soil interactions, multi-parameter coupling effects, and dynamic growth stage differences, pose significant challenges to the accuracy and applicability of traditional methods.

[0003] Existing physical inversion methods based on radiative transfer models (such as PROSAIL) can simulate canopy reflectance, but their simplified assumptions about soil background reflection and multiple scattering make it difficult to accurately depict the canopy-soil coupling effects of densely planted crops such as rice, resulting in significant deviations in inversion results in complex scenarios. Furthermore, traditional lookup table (LUT) methods rely on uniform parameter sampling and fail to incorporate dynamic constraints from measured data. This can lead to reduced inversion efficiency due to parameter redundancy and multicollinearity, especially during the middle and late stages of crop growth when canopy structure undergoes drastic changes, causing a sharp decline in model generalization.

[0004] On the other hand, while data-driven approaches based on machine learning can fit nonlinear relationships, traditional ELM model parameters are prone to falling into local optima due to random initialization, and they lack effective utilization of data generated by multiple physical models. This results in insufficient adaptability to small sample sizes and data from multiple growth stages. Existing technologies often rely on single models or static datasets, making it difficult to balance the advantages of physical mechanisms and data-driven approaches. This limits the accuracy and stability of simultaneous inversion of multiple nutrient indicators in complex environments.

[0005] Therefore, how to design a crop nutrient information inversion method based on the ELM inversion model to solve the problems of insufficient simulation of complex canopy-soil coupled reflections, redundant lookup table parameters, and limited optimization of machine learning models is an urgent problem that technicians in this field need to solve. Summary of the Invention

[0006] In view of this, the present invention provides a crop nutrient information inversion method based on the ELM inversion model, which accurately simulates the crop canopy reflectance through multi-model coupling, optimizes the lookup table to improve data matching accuracy, and uses the whale optimization algorithm to efficiently train the ELM inversion model, thereby achieving the goal of quickly and accurately obtaining crop nutrient information.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] A crop nutrient information inversion method based on the ELM inversion model comprises the following steps:

[0009] S1. Obtain leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information of relevant crops, and construct a rice canopy reflectance simulation dataset using the RPIOSL-UBM model.

[0010] S2. Construct an initial lookup table based on the rice canopy reflectance simulation dataset;

[0011] S3. Constraining the initial lookup table based on measured agronomic parameters of relevant crops to obtain an optimized lookup table;

[0012] S4. Constructing a comprehensive data set by combining the optimized lookup table with the measured spectral data of relevant crops;

[0013] S5. Based on the comprehensive data set, the ELM inversion model is trained in combination with the whale optimization algorithm to obtain an optimized ELM inversion model;

[0014] S6. The nutrient information of the tested crops is inverted through the optimized ELM inversion model to obtain the corresponding leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll, nitrogen and water nutrient indicators, and generate a nutrient map.

[0015] Furthermore, in S1, a rice canopy reflectance simulation dataset is constructed using the RPIOSL-UBM model, including:

[0016] S11. Input leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information into the RPIOSL model to simulate the multiple scattering reflectance of the rice canopy in a specific spectral band.

[0017] S12, inputting ground parameters into the UBM model, calculating soil background reflectance and its interaction with the canopy, and generating background reflectance components;

[0018] S13, based on the multiple scattering reflectance under the RPIOSL model and the background reflectance component under the UBM model, the total reflectance of the canopy-soil system is coupled through the radiation transfer equation;

[0019] S14. Use a parameter space uniform sampling strategy to generate multiple sets of parameter combinations within the preset parameter range, and repeat S11-S13 to generate a rice canopy reflectance simulation data set covering different parameter scenarios.

[0020] Furthermore, the S2 includes:

[0021] S21, discretizing leaf physical and chemical parameters, canopy structural parameters, ground parameters and angle information, and generating a parameter combination list using the Latin hypercube sampling method;

[0022] S22. Associate each parameter combination with the corresponding spectral reflectance data to construct an initial lookup table containing all possible parameter combinations and their corresponding spectral reflectances; wherein each row in the initial lookup table represents an independent parameter combination and its corresponding spectral reflectance feature.

[0023] Furthermore, the S3 includes:

[0024] S31, collecting measured agronomic parameters of relevant crops corresponding to the parameter combinations in the initial lookup table;

[0025] S32, using Mahalanobis distance to measure the multidimensional similarity between the parameter combination in the initial lookup table and the measured data, and screening parameter combinations with a distance less than a threshold;

[0026] S33, cross-validating the filtered lookup table, dividing the measured data into a training subset and a validation subset, and adjusting the parameter step size through K-fold cross-validation until the root mean square error (RMSE) of the validation set drops to a preset stable range;

[0027] S34. Performing sparse processing on the cross-validated lookup table, using the K-means clustering algorithm to merge the parameter combinations of similar spectral features, and retaining the cluster center point to compress the lookup table size, to obtain an optimized lookup table.

[0028] Furthermore, the S4 includes:

[0029] S41, using a dynamic time warping algorithm to match the measured spectrum with the spectrum curve in the optimized lookup table to obtain a matching result;

[0030] S42, performing confidence scoring on the matching results, retaining data pairs with confidence greater than a confidence threshold, and obtaining an initial comprehensive data set;

[0031] S43. Start the feedback optimization mechanism for the data pairs with confidence less than the confidence threshold, re-input the corresponding parameter combination into the RPIOSL-UBM model for local parameter fine-tuning, generate supplementary data and incorporate it into the initial comprehensive data set to obtain a comprehensive data set.

[0032] Furthermore, the S5 includes:

[0033] S51, initializing the parameters of the ELM inversion model;

[0034] S52, dividing the training set and the validation set according to the measured spectral data and the corresponding nutrient information in the comprehensive data set;

[0035] S53. Initialize a group of whale individuals using the whale optimization algorithm, where each individual represents a set of weights and bias parameters of the ELM model;

[0036] S54, calculating the model prediction error of each individual whale under the current parameters as the fitness function value;

[0037] S55. According to the fitness function value, the optimal whale individual parameters are selected as the best solution for the current iteration, and the global optimal solution is updated;

[0038] S56. Combine the rules of the whale optimization algorithm to update the position of each individual whale and simulate the foraging behavior of the whale group to explore a better solution;

[0039] S57. Repeat steps S54 to S56 until a preset stop condition is met, and obtain the ELM inversion model with the optimal whale individual parameter configuration.

[0040] Furthermore, the S6 includes:

[0041] S61. Obtain measured spectral data of the crops to be tested, ensuring that the data covers different growth stages and environmental conditions of the crops;

[0042] S62, inputting the measured spectral data into the optimized ELM inversion model, and using the model to analyze the spectral data and invert nutrient information;

[0043] S63, extracting leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll content, nitrogen content and water content nutrient indicators from the model output;

[0044] S64, performing a spatiotemporal consistency check on the nutrient index; the spatiotemporal consistency check includes: a spatial check and a temporal check;

[0045] S65. Based on the nutrient indicators after spatiotemporal consistency verification, a spatial interpolation or mapping algorithm is used to generate a nutrient distribution map of the crops to be tested.

[0046] It can be seen from the above technical solution that compared with the prior art, the technical solution of the present invention has the following advantages:

[0047] Beneficial effects:

[0048] 1. By coupling the RPIOSL model with the UBM model and integrating the radiative transfer equations for canopy multiple scattering reflectance and soil background reflectance, the spectral characteristics of the complex rice canopy-soil system can be more realistically simulated. Furthermore, by introducing measured agronomic parameters to perform multidimensional similarity screening and cross-validation constraints on the lookup table, this effectively reduces multicollinearity between model parameters and significantly improves the physical plausibility and accuracy of nutrient inversion, particularly during growth stages with complex crop structures.

[0049] 2. Latin hypercube sampling is used to generate the initial parameter space. A highly representative sparse lookup table is constructed by combining Mahalanobis distance screening, K-means clustering compression, and a feedback-based local fine-tuning mechanism. This method, while preserving key spectral features, matches measured data through a dynamic time warping algorithm and adaptively supplements local parameter combinations. This allows the lookup table to cover a wide parameter range while focusing on sensitive areas in real-world scenarios, significantly improving the model's spectral resolution and generalization performance for complex field environments.

[0050] 3. The whale optimization algorithm is incorporated into the extreme learning machine parameter optimization process, performing a global search for ELM weights and biases by simulating the foraging behavior of whale groups. Compared to traditional random initialization methods, this strategy significantly accelerates model convergence and avoids falling into local optima. This makes the ELM inversion model more robust in predicting highly nonlinear nutrient indicators such as chlorophyll and nitrogen, while reducing reliance on large-scale training data, making it suitable for rapid modeling of multiple crop varieties across multiple growth stages. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0052] Figure 1 A flow chart of a crop nutrient information inversion method based on the ELM inversion model provided by an embodiment of the present invention;

[0053] Figure 2 Schematic diagram of the process of extracting relevant nutrient indicators based on the ELM inversion model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] like Figure 1 As shown, this embodiment provides a crop nutrient information inversion method based on the ELM inversion model, comprising the following steps:

[0056] S1. Obtain leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information of relevant crops, and construct a rice canopy reflectance simulation dataset using the RPIOSL-UBM model.

[0057] S2. Construct an initial lookup table based on the rice canopy reflectance simulation dataset;

[0058] S3. Constraining the initial lookup table based on measured agronomic parameters of relevant crops to obtain an optimized lookup table;

[0059] S4. Constructing a comprehensive data set by combining the optimized lookup table with the measured spectral data of relevant crops;

[0060] S5. Based on the comprehensive data set, the ELM inversion model is trained in combination with the whale optimization algorithm to obtain an optimized ELM inversion model;

[0061] S6. The nutrient information of the tested crops is inverted through the optimized ELM inversion model to obtain the corresponding leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll, nitrogen and water nutrient indicators, and generate a nutrient map.

[0062] This method improves the accuracy of crop nutrient inversion through multi-source model coupling and parameter constraints. It enhances model generalization using a dynamically optimized hybrid lookup table and incorporates a whale optimization algorithm to drive efficient training of an extreme learning machine. This method more realistically simulates the spectral characteristics of complex canopy-soil systems, effectively addresses multicollinearity, and adaptively supplements local parameter combinations, resulting in greater robustness and faster convergence in complex field environments. It is particularly well-suited for rapid and accurate nutrient inversion for crops across multiple growth stages and varieties.

[0063] The following is a detailed description of each step in the above method;

[0064] In this embodiment S1, leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information of relevant crops are obtained, and a rice canopy reflectance simulation dataset is constructed using the RPIOSL-UBM model. Specifically, the following steps are involved:

[0065] S11. Inputting leaf physicochemical parameters, canopy structural parameters, ground parameters, and angular information into the RPIOSL model to simulate the multiple scattering reflectance of the rice canopy in a specific spectral band; wherein the leaf physicochemical parameters include chlorophyll content, leaf water content, and leaf nitrogen content; the canopy structural parameters include leaf area index (LAI), leaf inclination distribution, and canopy height; the ground parameters include soil reflectivity, soil moisture, and background vegetation cover; and the angular information includes solar zenith angle, sensor observation zenith angle, and relative azimuth angle;

[0066] S12, inputting ground parameters into the UBM model, calculating soil background reflectance and its interaction with the canopy, and generating background reflectance components;

[0067] S13, based on the multiple scattering reflectance under the RPIOSL model and the background reflectance component under the UBM model, the total reflectance of the canopy-soil system is coupled through the radiation transfer equation;

[0068] S14. Using a uniform sampling strategy in parameter space, generate multiple parameter combinations within the preset parameter range. Repeat S11-S13 to generate a simulated rice canopy reflectance dataset covering different parameter scenarios. The dataset includes parameter combinations and their corresponding spectral reflectance curves. The spectral range covers the visible to near-infrared band (400-2500nm), with a spectral resolution consistent with the target sensor.

[0069] The RPIOSL-UBM model in this step consists of two parts: RPIOSL and UBM;

[0070] The RPIOSL model is based on an improved radiation transfer theory and adopts a layered leaf scattering algorithm. It combines leaf physical and chemical parameters (such as chlorophyll content and leaf moisture) with canopy structural parameters (such as leaf area index and leaf inclination distribution). It simulates the multiple scattering process within the canopy through Monte Carlo ray tracing and outputs high-resolution canopy reflectance in the 400-2500nm band.

[0071] The UBM model adopts a semi-empirical bidirectional reflectance distribution function based on the Hapke soil reflectance model framework. It inputs parameters such as soil moisture, organic matter content and surface roughness to calculate the directional reflectance of the soil background in the visible-near infrared band, and introduces a canopy gap ratio weight factor to characterize the interactive contribution of soil-canopy reflectance.

[0072] Furthermore, the canopy reflectance output by RPIOSL and the soil reflectance output by UBM are weighted and superimposed through the radiation transfer equation. The weight is dynamically adjusted by the canopy coverage and the observation zenith angle. Finally, the total reflectance spectrum of the canopy-soil system is output with a resolution of up to 1 nm, covering the typical band range of active and passive remote sensing sensors.

[0073] The coupling of RPIOSL and the UBM model enables a detailed characterization of the complex interactions between the canopy and soil of densely planted crops (such as rice). This particularly addresses the errors caused by the simplification of soil background reflectance in traditional models (such as PROSAIL). A uniform sampling strategy in parameter space ensures that the dataset covers multiple parameter combinations, providing comprehensive physical mechanism support for subsequent lookup table construction and significantly improving the model's adaptability to heterogeneous field environments.

[0074] In this embodiment S2, an initial lookup table is constructed based on a rice canopy reflectance simulation dataset, specifically including:

[0075] S21, discretizing leaf physical and chemical parameters, canopy structural parameters, ground parameters and angle information, and generating a parameter combination list using the Latin hypercube sampling method;

[0076] S22. Associate each parameter combination with the corresponding spectral reflectance data to construct an initial lookup table containing all possible parameter combinations and their corresponding spectral reflectances; wherein each row in the initial lookup table represents an independent parameter combination and its corresponding spectral reflectance feature.

[0077] In this step, Latin hypercube sampling ensures efficient coverage of the parameter space, avoiding the combinatorial explosion problem caused by traditional uniform sampling and achieving a representative distribution of multidimensional parameters with limited data. Combining discretization with the LUT structure facilitates rapid retrieval and matching, providing a structured data foundation for subsequent optimization and inversion, and reducing computational complexity.

[0078] In this embodiment S3, the initial lookup table is constrained in combination with the measured agronomic parameters of the relevant crops to obtain an optimized lookup table; specifically, the following steps are performed:

[0079] S31. Collecting measured agronomic parameters of relevant crops corresponding to the parameter combinations in the initial lookup table; the measured agronomic parameters include chlorophyll fluorescence parameters (Fv / Fm), laboratory-measured values ​​of leaf nitrogen content, and NDVI verification data based on UAV multispectral imagery;

[0080] S32, using Mahalanobis distance to measure the multidimensional similarity between the parameter combination in the initial lookup table and the measured data, and screening parameter combinations with a distance less than a threshold;

[0081] S33, cross-validating the filtered lookup table, dividing the measured data into a training subset and a validation subset, and adjusting the parameter step size through K-fold cross-validation until the root mean square error (RMSE) of the validation set drops to a preset stable range;

[0082] S34. Performing sparse processing on the cross-validated lookup table, using the K-means clustering algorithm to merge the parameter combinations of similar spectral features, and retaining the cluster center point to compress the lookup table size, to obtain an optimized lookup table.

[0083] Mahalanobis distance filtering combined with measured data constraints effectively eliminates parameter redundancy and multicollinearity, improving the physical rationality of the LUT. K-means clustering sparsification significantly reduces data size while retaining key spectral features, enhancing inversion efficiency, making it particularly suitable for real-time field monitoring scenarios.

[0084] In this embodiment S4, a comprehensive data set is constructed by combining the optimized lookup table with the measured spectral data of relevant crops. Specifically, the following steps are performed:

[0085] S41, using a dynamic time warping algorithm to match the measured spectrum with the spectrum curve in the optimized lookup table to obtain a matching result;

[0086] S42, performing confidence scoring on the matching results, retaining data pairs with confidence greater than a confidence threshold, and obtaining an initial comprehensive data set;

[0087] S43. Initiate a feedback optimization mechanism for the data pairs with a confidence level less than a confidence threshold, re-input the corresponding parameter combination into the RPIOSL-UBM model for local parameter fine-tuning, generate supplementary data and incorporate it into the initial comprehensive data set to obtain a comprehensive data set; the local parameter fine-tuning is performed by reducing the step size to 10% of the original value;

[0088] The DTW algorithm addresses spectral time misalignment caused by differences in growth stages, improving matching accuracy. The feedback optimization mechanism dynamically supplements sensitive parameter combinations missing from the LUT, enhancing the dataset's ability to represent complex real-world scenarios and preventing model overfitting.

[0089] In this embodiment S5, based on the comprehensive data set, the ELM inversion model is trained in combination with the whale optimization algorithm to obtain an optimized ELM inversion model; specifically, the following steps are performed:

[0090] S51, initializing the parameters of the ELM inversion model;

[0091] S52, dividing the training set and the validation set according to the measured spectral data and the corresponding nutrient information in the comprehensive data set;

[0092] S53. Initialize a group of whale individuals using the whale optimization algorithm, where each individual represents a set of weights and bias parameters of the ELM model;

[0093] S54, calculating the model prediction error of each individual whale under the current parameters as the fitness function value;

[0094] S55. According to the fitness function value, the optimal whale individual parameters are selected as the best solution for the current iteration, and the global optimal solution is updated;

[0095] S56. Combine the rules of the whale optimization algorithm to update the position of each individual whale and simulate the foraging behavior of the whale group to explore a better solution;

[0096] S57. Repeat steps S54 to S56 until a preset stop condition is met, and obtain the ELM inversion model with the optimal whale individual parameter configuration.

[0097] The ELM inversion model adopts a single hidden layer feedforward neural network architecture, which consists of an input layer, a hidden layer, and an output layer. The specific structure is as follows:

[0098] Input layer: The number of nodes is equal to the spectral feature dimension, and the input data is the preprocessed spectral reflectance; hidden layer: The number of nodes is determined by dynamic search through the whale optimization algorithm, the activation function uses Sigmoid or ReLU, and the weights and biases are globally optimized by WOA rather than traditional random initialization to avoid local optimality; output layer: The number of nodes corresponds to the number of target nutrient indicators, and its five nodes output LAI, NDVI, chlorophyll, nitrogen, and water, respectively, using a linear activation function.

[0099] WOA encodes the ELM weight matrix and bias vector into the position vector of an individual whale. By simulating the spiral update and random search behavior of whales swarming prey, it globally optimizes in parameter space, using the validation set RMSE as the fitness function. The final model achieves simultaneous high-precision inversion of multiple indicators by minimizing the joint loss function of spectral reconstruction error and nutrient prediction error.

[0100] In this embodiment S6, the nutrient information of the tested crop is inverted using the optimized ELM inversion model to obtain the corresponding leaf area index (LAI), normalized difference vegetation index (NDVI), chlorophyll, nitrogen, and water nutrient indicators, and generate a nutrient map. Specifically, this includes:

[0101] S61. Obtain measured spectral data of the crops to be tested, ensuring that the data covers different growth stages and environmental conditions of the crops;

[0102] S62, inputting the measured spectral data into the optimized ELM inversion model, and using the model to analyze the spectral data and invert nutrient information;

[0103] S63, extracting leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll content, nitrogen content and water content nutrient indicators from the model output;

[0104] like Figure 2 As shown, it specifically includes:

[0105] 1) Original spectrum input;

[0106] Input the measured spectral data of the crops to be tested into the input layer of the model. This data should cover different growth stages and environmental conditions of the crops to ensure comprehensiveness.

[0107] 2) Nonlinear transformation of hidden layer;

[0108] The input data is processed through the hidden layer. In the hidden layer, each node performs a nonlinear transformation on the input data based on its weight and activation function (such as Sigmoid or ReLU). This process can capture complex features and patterns in the data.

[0109] 3) Linear combination of output layer;

[0110] After processing in the hidden layer, the data enters the output layer. This layer uses a linear activation function to directly generate predictions. At this point, the data has been mapped to the target nutrient index space through model learning.

[0111] 4) Extracting various nutrient indicators;

[0112] The five nodes in the output layer each extract a corresponding nutrient indicator: LAI node: The output value of the first node is an estimate of the leaf area index (LAI). NDVI node: The output value of the second node represents an estimate of the normalized difference vegetation index (NDVI). Chlorophyll node: The output value of the third node represents an estimate of chlorophyll content. Nitrogen node: The output value of the fourth node corresponds to an estimate of nitrogen content. Water node: The output value of the fifth node reflects an estimate of water content.

[0113] S64, performing a spatiotemporal consistency check on the nutrient index; the spatiotemporal consistency check includes: a spatial check and a temporal check;

[0114] S65. Based on the nutrient indicators after spatiotemporal consistency verification, a spatial interpolation or mapping algorithm is used to generate a nutrient distribution map of the crops to be tested.

[0115] In this step, spatiotemporal validation eliminates outlier interference and ensures the agronomic plausibility of the inversion results. Spatial interpolation, combined with high-density sampling points, generates refined nutrient distribution maps, providing intuitive guidance for variable-rate fertilization and precise management, particularly suitable for field-scale crop growth monitoring.

[0116] In this embodiment, the crop nutrient information inversion method based on the ELM inversion model constructs a rice canopy reflectance simulation dataset through the RPIOSL-UBM model, optimizes the lookup table based on measured agronomic parameters, and then uses the whale optimization algorithm to train the ELM inversion model. Ultimately, it achieves high-precision inversion of crop leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll, nitrogen and water nutrient indicators, and generates a nutrient map, thereby improving the accuracy and efficiency of nutrient information inversion.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. References to the same or similar parts between the various embodiments are sufficient. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For relevant parts, refer to the method description.

[0118] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A crop nutrient information inversion method based on the ELM inversion model, characterized in that: The following steps are involved: S1. Obtain leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information of relevant crops, and construct a rice canopy reflectance simulation dataset using the RPIOSL-UBM model. S2. Construct an initial lookup table based on the rice canopy reflectance simulation dataset; S3. Constraining the initial lookup table based on measured agronomic parameters of relevant crops to obtain an optimized lookup table; S4. Constructing a comprehensive data set by combining the optimized lookup table with the measured spectral data of relevant crops; S5. Based on the comprehensive data set, the ELM inversion model is trained in combination with the whale optimization algorithm to obtain an optimized ELM inversion model; S6. The nutrient information of the tested crops is inverted through the optimized ELM inversion model to obtain the corresponding leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll, nitrogen and water nutrient indicators, and generate a nutrient map.

2. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: In S1, a rice canopy reflectance simulation dataset was constructed using the RPIOSL-UBM model, including: S11. Input leaf physicochemical parameters, canopy structural parameters, ground parameters, and angle information into the RPIOSL model to simulate the multiple scattering reflectance of the rice canopy in a specific spectral band; S12, inputting ground parameters into the UBM model, calculating soil background reflectance and its interaction with the canopy, and generating background reflectance components; S13, based on the multiple scattering reflectance under the RPIOSL model and the background reflectance component under the UBM model, the total reflectance of the canopy-soil system is coupled through the radiation transfer equation; S14. Use a parameter space uniform sampling strategy to generate multiple sets of parameter combinations within the preset parameter range, and repeat S11-S13 to generate a rice canopy reflectance simulation data set covering different parameter scenarios.

3. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: Said S2 comprises: S21, discretizing leaf physical and chemical parameters, canopy structural parameters, ground parameters, and angle information, and generating a parameter combination list using the Latin hypercube sampling method; S22. Associate each parameter combination with the corresponding spectral reflectance data to construct an initial lookup table containing all possible parameter combinations and their corresponding spectral reflectances; wherein each row in the initial lookup table represents an independent parameter combination and its corresponding spectral reflectance feature.

4. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: The S3 includes: S31, collecting measured agronomic parameters of relevant crops corresponding to the parameter combinations in the initial lookup table; S32, using Mahalanobis distance to measure the multidimensional similarity between the parameter combination in the initial lookup table and the measured data, and screening parameter combinations with a distance less than a threshold; S33, cross-validating the filtered lookup table, dividing the measured data into a training subset and a validation subset, and adjusting the parameter step size through K-fold cross-validation until the root mean square error (RMSE) of the validation set drops to a preset stable range; S34. Performing sparse processing on the cross-validated lookup table, using the K-means clustering algorithm to merge the parameter combinations of similar spectral features, and retaining the cluster center point to compress the lookup table size, to obtain an optimized lookup table.

5. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: Said S4 comprises: S41, using a dynamic time warping algorithm to match the measured spectrum with the spectrum curve in the optimized lookup table to obtain a matching result; S42, performing confidence scoring on the matching results, retaining data pairs with confidence greater than a confidence threshold, and obtaining an initial comprehensive data set; S43. Start the feedback optimization mechanism for the data pairs with confidence less than the confidence threshold, re-input the corresponding parameter combination into the RPIOSL-UBM model for local parameter fine-tuning, generate supplementary data and incorporate it into the initial comprehensive data set to obtain a comprehensive data set.

6. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: Said S5 comprises: S51, initializing the parameters of the ELM inversion model; S52, dividing the training set and the validation set according to the measured spectral data and the corresponding nutrient information in the comprehensive data set; S53. Initialize a group of whale individuals using the whale optimization algorithm, where each individual represents a set of weights and bias parameters of the ELM model; S54, calculating the model prediction error of each individual whale under the current parameters as the fitness function value; S55. According to the fitness function value, the optimal whale individual parameters are selected as the best solution for the current iteration, and the global optimal solution is updated; S56. Combine the rules of the whale optimization algorithm to update the position of each individual whale and simulate the foraging behavior of the whale group to explore a better solution; S57. Repeat steps S54 to S56 until a preset stop condition is met, and obtain the ELM inversion model with the optimal whale individual parameter configuration.

7. The crop nutrient information inversion method based on the ELM inversion model according to claim 1, characterized in that: Said S6 comprises: S61. Obtain measured spectral data of the crops to be tested, ensuring that the data covers different growth stages and environmental conditions of the crops; S62, inputting the measured spectral data into the optimized ELM inversion model, and using the model to analyze the spectral data and invert nutrient information; S63, extracting leaf area index LAI, normalized difference vegetation index NDVI, chlorophyll content, nitrogen content and water content nutrient indicators from the model output; S64: Performing a spatiotemporal consistency check on the nutrient index; the spatiotemporal consistency check includes: Spatial checksum and time checksum; S65. Based on the nutrient indicators after spatiotemporal consistency verification, a spatial interpolation or mapping algorithm is used to generate a nutrient distribution map of the crops to be tested.

Images