Tobacco nitrogen deficiency risk prevention and control method

Abstract

The invention discloses a tobacco nitrogen deficiency risk prevention and control method, which comprises the following steps: S1, collecting and preprocessing multi-source data of a tobacco planting area, and constructing a data set; s2, constructing a support vector machine model, and setting a penalty coefficient and kernel function parameters; s3, introducing a grey wolf optimization algorithm, and carrying out global optimization on support vector machine model parameters; s4, performing nitrogen deficiency risk prediction on new data by using the optimized model; s5, calling a nitrogen application rule base to generate an intervention strategy according to the prediction result and the plot planting condition; s6, deploying the intervention strategy in the intelligent agricultural control system, and driving the fertilization device to execute intervention operation; and S7, collecting feedback data after fertilization, and performing incremental updating and optimization on the model. The method has the effect of realizing closed-loop management of accurate identification of the tobacco nitrogen deficiency risk, differential fertilization strategy generation and intelligent intervention control.

Description

Technical Field

[0001] The present invention relates to the field of agricultural informatization and intelligent decision-making and control technology, and in particular to a method for preventing and controlling nitrogen deficiency risks in tobacco. Background Art

[0002] In the context of the rapid development of modern agriculture, tobacco, as an important economic crop, has its yield and quality significantly affected by a variety of environmental factors and cultivation management measures. Among them, nitrogen, as one of the key nutrients for crop growth and development, plays a core role in promoting tobacco leaf expansion, improving photosynthetic efficiency and anabolism. Insufficient nitrogen supply, especially nitrogen deficiency during the critical growth period, can lead to stunted growth of tobacco plants, yellowing of leaves, thinning of leaves, and decreased nicotine content, which in turn seriously affects the grade structure and processing quality of tobacco leaves. Therefore, how to achieve early identification, precise intervention and intelligent regulation of tobacco nitrogen deficiency risks has become a key technical problem in the refined cultivation and management of tobacco.

[0003] In existing technologies, the identification of nitrogen deficiency in tobacco mainly relies on manual field observation, soil testing, and simple plant physiological index assessment methods. Although traditional methods have certain reference value, they suffer from defects in actual application, such as strong data lag, high subjectivity in identification, slow response speed, and poor adaptability. On the one hand, manual observation has a clear dependence on experience and lacks an efficient and standardized assessment system. On the other hand, although laboratory soil sample analysis has high accuracy, the process is complex, the cost is high, and it is difficult to achieve real-time dynamic monitoring, especially difficult to meet the rapid response requirements for nitrogen deficiency in large tobacco areas. In addition, due to the "average control" method of traditional agricultural fertilization programs, it is difficult to achieve differentiated control based on the actual fertilizer requirements of the crop, which often results in insufficient or excessive fertilization, which not only affects the quality of tobacco leaves, but also leads to resource waste and environmental burden.

[0004] With the development of intelligent and information-based agricultural technologies, machine learning models have begun to be introduced for crop nutrition diagnosis and risk prediction. Among them, the support vector machine (SVM), a small-sample learning model with strong nonlinear classification capabilities, has been widely used in crop disease and pest diagnosis, remote sensing classification, and soil type discrimination. Based on the kernel function, it constructs nonlinear mapping relationships and can find the optimal hyperplane in high-dimensional space, thereby achieving accurate classification. However, in practical applications, the SVM model still suffers from strong hyperparameter dependence, limited modeling stability, and poor adaptability to different types of data structures. Its performance is largely limited by the rationality of parameter selection and kernel function configuration. For agricultural applications with complex multi-source data structures (such as remote sensing indices, image features, soil properties, fertilization records, and other comprehensive inputs), a single SVM model configuration is difficult to fully exploit the nonlinear feature coupling relationships between data.

[0005] To improve the adaptability and accuracy of the SVM model in tobacco nitrogen deficiency risk prediction, recent research has introduced intelligent optimization algorithms, such as particle swarm optimization, genetic algorithms, and ant colony algorithms, for global search and optimization of model parameters. However, these traditional intelligent optimization algorithms are prone to premature convergence and local optimality in high-dimensional parameter spaces and lack adaptability to the specific characteristics of agricultural tasks. In particular, when faced with complex factors such as differentiated planting conditions, regional soil variations, and dynamic meteorological influences, the deep integration of optimization algorithms with agronomic models remains insufficient, lacking a systematic mechanism for linking parameter optimization, risk output, and control strategies.

[0006] Furthermore, existing research, after identifying nitrogen deficiency, often stops at outputting risk classification results, lacking a closed-loop technical approach from identification to response. Even with risk identification models, these approaches fail to develop truly actionable agricultural strategies based on the identified results, lacking differentiated intervention logic and a mechanism for integrating equipment control. Therefore, current nitrogen deficiency prediction solutions cannot meet the integrated closed-loop control requirements of "prediction-intervention-feedback" within intelligent agricultural operation systems.

[0007] Therefore, how to provide a method for preventing and controlling nitrogen deficiency risks in tobacco is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0008] One purpose of the present invention is to propose a method for preventing and controlling nitrogen deficiency risks in tobacco. The present invention fully integrates intelligent optimization algorithms and machine learning models, constructs a nitrogen deficiency risk identification model with a composite kernel support vector machine as the core, and uses the gray wolf optimization algorithm to perform global search optimization on the penalty coefficient, kernel function parameters and kernel weights of the model. At the same time, combined with the nitrogen deficiency prediction results and the planting conditions of the plots, the nitrogen application rule library is automatically called to generate differentiated intervention strategies, and the strategies are further deployed to the intelligent agricultural control system to drive the fertilization device to implement precise intervention operations, and a multi-dimensional dynamic monitoring and model feedback update mechanism is constructed, which has the advantages of high recognition accuracy, executable strategy output, complete operation closed loop and strong adaptability.

[0009] A method for preventing and controlling nitrogen deficiency risk in tobacco according to an embodiment of the present invention includes the following steps:

[0010] S1. Collect multi-source data of tobacco growing areas, pre-process the multi-source data, and construct a dataset;

[0011] S2. Construct a support vector machine model, using the constructed dataset as input and the tobacco nitrogen deficiency risk level as the output label, and define the penalty coefficient and kernel function parameters of the support vector machine model as adjustable variables;

[0012] S3. Introducing the gray wolf optimization algorithm to perform global optimization search on the penalty coefficient and kernel function parameters of the support vector machine model, using the cross-validation recognition accuracy as the fitness function, executing the gray wolf population initialization, position update, fitness evaluation and population iteration process to obtain the optimal parameter combination;

[0013] S4. Use the support vector machine model optimized by the Gray Wolf algorithm to predict nitrogen deficiency risk for the newly collected tobacco planting area data, and output the corresponding risk level label and risk score value;

[0014] S5. Based on the nitrogen deficiency risk prediction results and the plot planting conditions, the preset nitrogen application rule library is called to generate differentiated intervention strategies and output as fertilization instructions that can be executed by agricultural equipment;

[0015] S6. Deploy differentiated intervention strategies in smart agricultural control systems to drive fertilization devices to implement nitrogen deficiency intervention operations in target areas, while also dynamically monitoring plant growth, soil nitrogen concentration, and remote sensing indicators after fertilization.

[0016] S7. Input the feedback data after fertilization into the dataset as the incremental update sample of the support vector machine model, retrain the support vector machine model and dynamically adjust the parameters of the gray wolf optimization algorithm to build a closed-loop feedback tobacco nitrogen deficiency risk prevention and control mechanism.

[0017] Optionally, the multi-source data specifically includes soil nitrogen content data, meteorological environment data, plant growth image data, remote sensing vegetation index data and historical fertilization record data.

[0018] Optionally, the preprocessing of multi-source data specifically includes data denoising, null value filling, standardization, feature extraction and principal component analysis dimensionality reduction operations.

[0019] Optionally, the S2 specifically includes:

[0020] S21. Set the input data set to:

[0021] X={x1,x2,...,x n}, x i ∈R d ;

[0022] Among them, x i represents the feature vector of the i-th sample, d is the feature dimension, R is a real number set, i = 1, 2, ..., n, n is the number of data sets, and the sample label set is:

[0023] Y={y1,y2,...,y n},y i ∈{0,1,2,3};

[0024] Four classification labels corresponding to tobacco nitrogen deficiency risk levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency and severe nitrogen deficiency;

[0025] S22. Construct a support vector machine model, wherein the first layer is a two-class support vector machine for determining whether the input sample has a risk of nitrogen deficiency, and the second layer is a three-class support vector machine for further subdividing the samples at risk of nitrogen deficiency into mild, moderate, or severe levels. Each layer of support vector machine uses an independently trained data subset;

[0026] S23, define the composite kernel function K(x i ,x j ) is in the form of:

[0027]

[0028] Among them, the first term exp(-γ‖x i -x j ‖ 2 ) is the radial basis kernel function, the second is a polynomial kernel function, λ1 and λ2 are combined weight parameters, γ, r, and d are the kernel width parameter, constant term, and polynomial order of the kernel function, respectively. exp is an exponential function, and x j Represents the feature vector of the jth sample;

[0029] S24. After the support vector machine model training is completed, the support vector is dynamically screened based on the boundary contribution of the support vector. The screening index includes the absolute value of the decision function |f(x i )| and the Lagrange multiplier α i The size of the support vectors with low contribution that meet the set threshold conditions is eliminated;

[0030] S25. The constructed support vector machine model supported by the composite kernel function, having a hierarchical structure and a dynamic support vector screening mechanism is used as the basic structure of the tobacco nitrogen deficiency risk identification model, and the parameters are used as adjustable variables for the gray wolf optimization algorithm to perform joint optimization.

[0031] Optionally, the S3 specifically includes:

[0032] S31. Construct a dynamic encoding optimization structure to encode each individual gray wolf into a hyperparameter combination vector of the support vector machine model:

[0033] X k =(C k ,γ k ,λ 1k ,λ 2k );

[0034] Among them, C k is the penalty coefficient, γ k is the kernel width parameter of the radial basis kernel function, λ 1k and λ 2k are the combined weights of RBF kernel and polynomial kernel in the composite kernel function, and N is the population size;

[0035] S32. Set the maximum number of iterations T max , initialize the position vector of each individual in the gray wolf population, set the initial evolution factor a = 2, and record the initial global optimal fitness value F best (0);

[0036] S33, perform five-fold cross validation on the support vector machine model represented by each individual, obtain the predicted label and the true label sequence, and calculate the fuzzy risk loss function L k for:

[0037]

[0038] in, is the predicted label of the i-th sample, y i is the true label, is a 0-1 indicator function, is the risk penalty weight corresponding to each nitrogen deficiency level;

[0039] Further calculate the accuracy Acc k, prediction stability variance σ k , construct a multi-objective fitness function:

[0040]

[0041] Among them, α, β and η are weighted coefficients, Fitness(X k ) represents the position vector X of the kth individual k The overall fitness value, L k represents the normalized fuzzy risk loss value, and n represents the total number of samples in the validation set;

[0042] S34. Select the three individuals with the best fitness values ​​in the current iteration as guide individuals, denoted as α, β, and δ respectively, and build an elite inheritance pool to store the best individuals in the past m generations for local guidance;

[0043] S35. Define the adaptive evolution factor a based on the fitness convergence rate t :

[0044]

[0045] Where ΔF(t)=F best (t)-F best (t-1), ρ>0 is the search enhancement coefficient, ∈>0 is the convergence smoothing factor, t is the current iteration number, T max is the preset maximum number of iterations, exp is the exponential function;

[0046] S36. Update the individual position vector according to the adaptive evolution factor and execute the search strategy:

[0047]

[0048] Among them, r k is the individual disturbance factor, X elite is the reference individual in the elite inheritance pool, is the position vector of the k-th gray wolf individual in the t+1th iteration, X α 、X β and X δ is the position vector of the three leading individuals with the best fitness value in the current population;

[0049] S37, update the population fitness ranking and the global optimal solution record, iteratively execute steps S33 to S36 until the termination condition t≥T is met max , output the final optimal parameter combination to build the optimized support vector machine classification model.

[0050] Optionally, the S4 specifically includes:

[0051] S41, receiving newly collected tobacco planting area data, wherein the data includes pre-processed multi-dimensional feature information and is consistent with the data structure used in the support vector machine model training stage;

[0052] S42, inputting the new data into a support vector machine model obtained by the gray wolf optimization algorithm, where the support vector machine model includes an optimal hyperparameter configuration and a trained classification structure;

[0053] S43, calling the support vector machine model to perform classification reasoning and perform nitrogen deficiency risk identification processing on each input sample;

[0054] S44. Output a corresponding risk level label for each input sample, where the risk level includes four levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency, and severe nitrogen deficiency;

[0055] S45. Generate a set of risk score values ​​reflecting the degree of prediction credibility for each input sample based on the calculation results of the internal discriminant function of the support vector machine;

[0056] S46. Use the predicted labels and risk scores as model outputs for intervention strategy formulation and control instruction generation.

[0057] Optionally, the S5 specifically includes:

[0058] S51, receiving the nitrogen deficiency risk prediction result output by the support vector machine model optimized by the gray wolf, outputting the corresponding nitrogen deficiency risk level label and risk score value for each target plot, representing the prediction and judgment of the nitrogen status of crops in the current agricultural area;

[0059] S52: Synchronously call a plot information database to obtain planting condition information corresponding to each target plot, wherein the plot planting conditions include environmental and management variables such as plot area, soil texture type, tobacco crop variety, growth cycle stage, crop health status assessment indicators, and historical fertilization records;

[0060] S53. Based on the risk level label and the plot planting conditions, a pre-built nitrogen application rule library is called to perform a rule matching and condition screening process, and a nitrogen application strategy template that best matches the target conditions is selected from the nitrogen application rule library. The rule library presets responsive fertilization strategy parameters under different risk levels based on agronomic experience and a crop fertilizer requirement model.

[0061] S54. Based on the nitrogen application strategy template and taking into account the actual conditions of each plot, a differentiated intervention strategy generation process is implemented to ultimately form an operational nitrogen application intervention plan that includes nitrogen application dosage, nitrogen application method, and nitrogen application time window, achieving the goal of precise nitrogen application regulation in multiple regions and types of plots;

[0062] S55, converting the differentiated intervention strategy into an operation instruction structure recognizable by agricultural equipment through a parameter mapping and format conversion module, generating a complete fertilization control instruction set including fertilization operation path planning information, specific equipment control parameters, operation crop identification, and execution timing logic;

[0063] S56. Output the generated fertilization instructions that can be executed by agricultural equipment, and issue the instructions to the agricultural operation terminal through the control system interface to complete the control preparation of the nitrogen deficiency intervention task and ensure the effective linkage implementation of the prediction results and the intelligent control strategy.

[0064] Optionally, the S6 specifically includes:

[0065] S61. Transmitting the differentiated intervention strategy generated based on the nitrogen deficiency risk prediction results and the planting conditions of the plot to the intelligent agricultural control system via a network interface. The intelligent agricultural control system has the ability to receive, analyze, and dispatch agricultural operation instructions.

[0066] S62. Complete the analysis of the intervention strategy and the allocation of execution tasks in the intelligent agricultural control system, and send the fertilization operation parameters and control instructions to specific fertilization devices, such as intelligent fertilizer spreaders, automatic irrigation systems, or agricultural operation terminal equipment such as plant protection drones;

[0067] S63: driving the fertilizing device in the target area to perform nitrogen deficiency intervention operations according to the fertilizing instruction, completing the processes of nitrogen dosage delivery, fertilizing path movement, and operation cycle control;

[0068] S64. During the nitrogen deficiency intervention operation, the synchronous monitoring module is activated to collect data on plant growth changes in the intervention area, including leaf area index, color index, canopy height, and physiological growth parameters of growth potential images;

[0069] S65. Synchronously obtain information on changes in soil nitrogen concentration within the fertilization area, collect shallow and deep soil sample data, evaluate the trend of nitrogen level changes, and form a dynamic soil nitrogen monitoring sequence;

[0070] S66. Use remote sensing observation equipment or drone platforms to obtain spectral images and vegetation index information of the intervention area, conduct a macroscopic evaluation of the intervention effect at the spatial level, and ultimately form multi-dimensional feedback data on plant growth, soil nitrogen concentration, and remote sensing indicators.

[0071] The beneficial effects of the present invention are:

[0072] The present invention overcomes the problems of low prediction accuracy, strong dependence on model parameters, and delayed control response in existing tobacco nitrogen deficiency identification technologies by introducing a deep fusion mechanism of the Gray Wolf Optimization Algorithm and the Support Vector Machine, and significantly improves the accuracy of risk identification and the intelligence level of the system. As a core classification model, the support vector machine has good small sample learning and nonlinear discrimination capabilities. After combining with the composite kernel function structure, it can take into account both local similarity and global feature distribution, making the discrimination of tobacco nitrogen deficiency status more robust and generalizable. The Gray Wolf Optimization Algorithm has excellent global search capabilities and population coordination mechanisms, and can efficiently optimize the multi-dimensional hyperparameter combination of the support vector machine, significantly enhancing the adaptability and accuracy of the model, and solving technical bottlenecks such as the traditional parameter setting dependence on experience and model instability.

[0073] This invention not only achieves multi-level, high-precision predictions of tobacco nitrogen deficiency risks, but also establishes a complete, differentiated intervention mechanism based on this. By calling a preset nitrogen application rule library and combining the prediction results with planting conditions, a highly adaptable nitrogen application strategy is generated, enabling the fertilization process to move from "empirical" to "data-driven," achieving the goal of precise fertilization tailored to local conditions and the times. By converting intervention strategies into recognizable operating instructions for agricultural equipment and deploying them in intelligent agricultural control systems, the entire process from identification, decision-making to control execution is effectively opened up, improving the automation and precision of agricultural operations.

[0074] In addition, the present invention also constructs a real-time dynamic monitoring and model closed-loop feedback mechanism. After the fertilization operation is implemented, the system can continuously collect key parameters such as plant growth, soil nitrogen concentration and remote sensing indicators, conduct a quantitative assessment of the fertilization effect, and use the feedback results for incremental updates and optimization iterations of the support vector machine model, further improving the system's adaptive and dynamic control capabilities. Overall, the present invention has formed a closed-loop tobacco nitrogen deficiency risk prevention and control system of "intelligent identification - precise intervention - dynamic monitoring - model self-evolution", which has the comprehensive advantages of high prediction accuracy, strong real-time response, good control execution implementation, and perfect feedback capabilities. It has effectively promoted the transformation and upgrading of tobacco planting management from traditional experience-driven to intelligent data decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0076] Figure 1 This is a flow chart of a tobacco nitrogen deficiency risk prevention and control method proposed by the present invention;

[0077] Figure 2This is a schematic diagram of the fusion structure of the support vector machine model and the gray wolf optimization algorithm for a tobacco nitrogen deficiency risk prevention and control method proposed in the present invention. DETAILED DESCRIPTION

[0078] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0079] refer to Figure 1 and Figure 2 , a method for preventing and controlling nitrogen deficiency risk in tobacco, comprising the following steps:

[0080] S1. Collect multi-source data of tobacco growing areas, pre-process the multi-source data, and construct a dataset;

[0081] S2. Construct a support vector machine model, using the constructed dataset as input and the tobacco nitrogen deficiency risk level as the output label, and define the penalty coefficient and kernel function parameters of the support vector machine model as adjustable variables;

[0082] S3. Introducing the gray wolf optimization algorithm to perform global optimization search on the penalty coefficient and kernel function parameters of the support vector machine model, using the cross-validation recognition accuracy as the fitness function, executing the gray wolf population initialization, position update, fitness evaluation and population iteration process to obtain the optimal parameter combination;

[0083] S4. Use the support vector machine model optimized by the Gray Wolf algorithm to predict nitrogen deficiency risk for the newly collected tobacco planting area data, and output the corresponding risk level label and risk score value;

[0084] S5. Based on the nitrogen deficiency risk prediction results and the plot planting conditions, the preset nitrogen application rule library is called to generate differentiated intervention strategies and output as fertilization instructions that can be executed by agricultural equipment;

[0085] S6. Deploy differentiated intervention strategies in smart agricultural control systems to drive fertilization devices to implement nitrogen deficiency intervention operations in target areas, while also dynamically monitoring plant growth, soil nitrogen concentration, and remote sensing indicators after fertilization.

[0086] S7. Input the feedback data after fertilization into the dataset as the incremental update sample of the support vector machine model, retrain the support vector machine model and dynamically adjust the parameters of the gray wolf optimization algorithm to build a closed-loop feedback tobacco nitrogen deficiency risk prevention and control mechanism.

[0087] In this embodiment, the multi-source data specifically includes soil nitrogen content data, meteorological environment data, plant growth image data, remote sensing vegetation index data, and historical fertilization record data.

[0088] In this embodiment, the preprocessing of multi-source data specifically includes data denoising, null value filling, standardization, feature extraction and principal component analysis dimensionality reduction operations.

[0089] In this embodiment, S2 specifically includes:

[0090] S21. Set the input data set to:

[0091] X={x1,x2,...,x n}, x i ∈R d ;

[0092] Among them, x i represents the feature vector of the i-th sample, d is the feature dimension, R is a real number set, i = 1, 2, ..., n, n is the number of data sets, and the sample label set is:

[0093] Y={y1,y2,...,y n},y i ∈{0,1,2,3};

[0094] Four classification labels corresponding to tobacco nitrogen deficiency risk levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency and severe nitrogen deficiency;

[0095] S22. Construct a support vector machine model, wherein the first layer is a two-class support vector machine for determining whether the input sample has a risk of nitrogen deficiency, and the second layer is a three-class support vector machine for further subdividing the samples at risk of nitrogen deficiency into mild, moderate, or severe levels. Each layer of support vector machine uses an independently trained data subset;

[0096] S23, define the composite kernel function K(x i ,x j ) is in the form of:

[0097]

[0098] Among them, the first term exp(-γ‖x i -x j ‖ 2 ) is the radial basis kernel function, the second is a polynomial kernel function, λ1 and λ2 are combined weight parameters, γ, r, and d are the kernel width parameter, constant term, and polynomial order of the kernel function, respectively. exp is an exponential function, and x j Represents the feature vector of the jth sample;

[0099] S24. After the support vector machine model training is completed, the support vector is dynamically screened based on the boundary contribution of the support vector. The screening index includes the absolute value of the decision function |f(x i)| and the Lagrange multiplier α i The size of the support vectors with low contribution that meet the set threshold conditions is eliminated;

[0100] S25. The constructed support vector machine model supported by the composite kernel function, having a hierarchical structure and a dynamic support vector screening mechanism is used as the basic structure of the tobacco nitrogen deficiency risk identification model, and the parameters are used as adjustable variables for the gray wolf optimization algorithm to perform joint optimization.

[0101] In this embodiment, S3 specifically includes:

[0102] S31. Construct a dynamic encoding optimization structure to encode each individual gray wolf into a hyperparameter combination vector of the support vector machine model:

[0103] X k =(C k ,γ k ,λ 1k ,λ 2k );

[0104] Among them, C k is the penalty coefficient, γ k is the kernel width parameter of the radial basis kernel function, λ 1k and λ 2k are the combined weights of RBF kernel and polynomial kernel in the composite kernel function, and N is the population size;

[0105] S32. Set the maximum number of iterations T max , initialize the position vector of each individual in the gray wolf population, set the initial evolution factor a = 2, and record the initial global optimal fitness value F best (0);

[0106] S33, perform five-fold cross validation on the support vector machine model represented by each individual, obtain the predicted label and the true label sequence, and calculate the fuzzy risk loss function L k for:

[0107]

[0108] in, is the predicted label of the i-th sample, y i is the true label, is a 0-1 indicator function, wy i is the risk penalty weight corresponding to each nitrogen deficiency level;

[0109] Further calculate the accuracy Acc k , prediction stability variance σ k , construct a multi-objective fitness function:

[0110]

[0111] Among them, α, β and η are weighted coefficients, Fitness(X k ) represents the position vector X of the kth individual k The overall fitness value, L k represents the normalized fuzzy risk loss value, and n represents the total number of samples in the validation set;

[0112] S34. Select the three individuals with the best fitness values ​​in the current iteration as guide individuals, denoted as α, β, and δ respectively, and build an elite inheritance pool to store the best individuals in the past m generations for local guidance;

[0113] S35. Define the adaptive evolution factor a based on the fitness convergence rate t :

[0114]

[0115] Where ΔF(t)=F best (t)-F best (t-1), ρ>0 is the search enhancement coefficient, ∈>0 is the convergence smoothing factor, t is the current iteration number, T max is the preset maximum number of iterations, exp is the exponential function;

[0116] S36. Update the individual position vector according to the adaptive evolution factor and execute the search strategy:

[0117]

[0118] Among them, r k is the individual disturbance factor, X elite is the reference individual in the elite inheritance pool, is the position vector of the k-th gray wolf individual in the t+1th iteration, X α 、X β and X δ is the position vector of the three leading individuals with the best fitness value in the current population;

[0119] S37, update the population fitness ranking and the global optimal solution record, iteratively execute steps S33 to S36 until the termination condition t≥T is met max , output the final optimal parameter combination to build the optimized support vector machine classification model.

[0120] In this embodiment, the S4 specifically includes:

[0121] S41, receiving newly collected tobacco planting area data, wherein the data includes pre-processed multi-dimensional feature information and is consistent with the data structure used in the support vector machine model training stage;

[0122] S42, inputting the new data into a support vector machine model obtained by the gray wolf optimization algorithm, where the support vector machine model includes an optimal hyperparameter configuration and a trained classification structure;

[0123] S43, calling the support vector machine model to perform classification reasoning and perform nitrogen deficiency risk identification processing on each input sample;

[0124] S44. Output a corresponding risk level label for each input sample, where the risk level includes four levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency, and severe nitrogen deficiency;

[0125] S45. Generate a set of risk score values ​​reflecting the degree of prediction credibility for each input sample based on the calculation results of the internal discriminant function of the support vector machine;

[0126] S46. Use the predicted labels and risk scores as model outputs for intervention strategy formulation and control instruction generation.

[0127] In this embodiment, the S5 specifically includes:

[0128] S51, receiving the nitrogen deficiency risk prediction result output by the support vector machine model optimized by the gray wolf, outputting the corresponding nitrogen deficiency risk level label and risk score value for each target plot, representing the prediction and judgment of the nitrogen status of crops in the current agricultural area;

[0129] S52: Synchronously call a plot information database to obtain planting condition information corresponding to each target plot, wherein the plot planting conditions include environmental and management variables such as plot area, soil texture type, tobacco crop variety, growth cycle stage, crop health status assessment indicators, and historical fertilization records;

[0130] S53. Based on the risk level label and the plot planting conditions, a pre-built nitrogen application rule library is called to perform a rule matching and condition screening process, and a nitrogen application strategy template that best matches the target conditions is selected from the nitrogen application rule library. The rule library presets responsive fertilization strategy parameters under different risk levels based on agronomic experience and a crop fertilizer requirement model.

[0131] S54. Based on the nitrogen application strategy template and taking into account the actual conditions of each plot, a differentiated intervention strategy generation process is implemented to ultimately form an operational nitrogen application intervention plan that includes nitrogen application dosage, nitrogen application method, and nitrogen application time window, achieving the goal of precise nitrogen application regulation in multiple regions and types of plots;

[0132] S55, converting the differentiated intervention strategy into an operation instruction structure recognizable by agricultural equipment through a parameter mapping and format conversion module, generating a complete fertilization control instruction set including fertilization operation path planning information, specific equipment control parameters, operation crop identification, and execution timing logic;

[0133] S56. Output the generated fertilization instructions that can be executed by agricultural equipment, and issue the instructions to the agricultural operation terminal through the control system interface to complete the control preparation of the nitrogen deficiency intervention task and ensure the effective linkage implementation of the prediction results and the intelligent control strategy.

[0134] In this embodiment, S6 specifically includes:

[0135] S61. Transmitting the differentiated intervention strategy generated based on the nitrogen deficiency risk prediction results and the planting conditions of the plot to the intelligent agricultural control system via a network interface. The intelligent agricultural control system has the ability to receive, analyze, and dispatch agricultural operation instructions.

[0136] S62. Complete the analysis of the intervention strategy and the allocation of execution tasks in the intelligent agricultural control system, and send the fertilization operation parameters and control instructions to specific fertilization devices, such as intelligent fertilizer spreaders, automatic irrigation systems, or agricultural operation terminal equipment such as plant protection drones;

[0137] S63: driving the fertilizing device in the target area to perform nitrogen deficiency intervention operations according to the fertilizing instruction, completing the processes of nitrogen dosage delivery, fertilizing path movement, and operation cycle control;

[0138] S64. During the nitrogen deficiency intervention operation, the synchronous monitoring module is activated to collect data on plant growth changes in the intervention area, including leaf area index, color index, canopy height, and physiological growth parameters of growth potential images;

[0139] S65. Synchronously obtain information on changes in soil nitrogen concentration within the fertilization area, collect shallow and deep soil sample data, evaluate the trend of nitrogen level changes, and form a dynamic soil nitrogen monitoring sequence;

[0140] S66. Use remote sensing observation equipment or drone platforms to obtain spectral images and vegetation index information of the intervention area, conduct a macroscopic evaluation of the intervention effect at the spatial level, and ultimately form multi-dimensional feedback data on plant growth, soil nitrogen concentration, and remote sensing indicators.

[0141] Example 1:

[0142] To verify the feasibility of this invention, a field trial was conducted in a typical tobacco-growing area, located in hilly terrain at an altitude of 1,200 to 1,400 meters. The soil is primarily yellow, prone to nitrogen fluctuations and a high risk of nitrogen deficiency. From April to July 2024, the experimental team selected two tobacco-growing areas of comparable size and consistent management as experimental subjects: one area, designated the "experimental group," fully applied the method provided by this invention; the other, designated the "control group," continued with the existing empirical fertilization and manual inspection methods.

[0143] In practice, the experimental group continuously collected multi-source data, including soil nitrate nitrogen concentration, ammonium nitrogen, NDVI, and plant canopy color index, using soil nitrogen sensors, air temperature and humidity monitors, and drone remote sensing platforms deployed throughout the plots. All data were normalized and processed for principal component dimensionality reduction before being fed into a composite kernel support vector machine classification model. The model used the Grey Wolf Optimization algorithm to automatically search and dynamically optimize the penalty coefficient C, radial basis kernel width parameter γ, and kernel combination weight λ to generate the optimal classifier for nitrogen deficiency risk level determination.

[0144] After each prediction round, the system automatically reads historical fertilization information, soil type, crop growth period, and other planting conditions for the corresponding plots. Through rule matching and parameter adjustment mechanisms, it generates differentiated intervention strategies. The system converts these strategies into operational instructions that can be read by intelligent fertilization equipment. The system then uses local agricultural control systems to drive automated fertilization vehicles for precise, on-demand application. After fertilization is complete, the system continuously tracks crop growth, soil nitrogen changes, and remote sensing indices in the target area to generate feedback data and update the model.

[0145] In contrast, the control group continued to rely on farmers to inspect fields every seven days to observe leaf color and area to estimate nitrogen status, and grassroots agricultural technicians then manually issued fertilizer dosage recommendations. Fertilization was carried out by manually controlled spraying equipment, lacking a feedback mechanism, which affected both the accuracy and timeliness of fertilization.

[0146] During the 60-day monitoring period, the experimental group's system processed 3,820 data collection records, completed 54 fertilization strategy outputs, and activated fertilization equipment 38 times. The control group, on the other hand, executed 42 fertilization operations, with some instances of duplicate and missed fertilization. After the experiment, five key indicators were compared.

[0147] Table 1 Comparison of application effects of tobacco nitrogen deficiency prevention and control intelligent system

[0148]

[0149] According to the data results in Table 1, it can be clearly seen that the method of the present invention has significant advantages over the traditional method in multiple key indicators, reflecting its comprehensive superiority in recognition accuracy, operation efficiency, resource conservation and control effect.

[0150] First, in terms of nitrogen deficiency identification accuracy, the control group, using traditional empirical judgment and manual observation, achieved only 76.4% accuracy, prone to identification delays and misjudgments. However, the experimental group using the "Gray Wolf Optimized Support Vector Machine Model" of the present invention achieved an accuracy rate of 92.3%, an increase of 15.9 percentage points. This demonstrates that the present invention can more accurately distinguish the nitrogen status of tobacco plants and improve the reliability of risk identification.

[0151] Secondly, in terms of fertilization efficiency, the experimental group's average fertilization time was 11.2 minutes per mu, far lower than the control group's 18.5 minutes per mu, representing an efficiency improvement of approximately 39.5%. This significant time saving is primarily due to the efficient collaboration between the system's automatically generated fertilization instructions and the intelligent operation equipment, which avoids inefficient manual intervention such as repeated paths and uneven fertilization.

[0152] In terms of resource conservation, the experimental group, while maintaining the same yield target, used an average of 17.1 kg of nitrogen fertilizer per mu, a 15.8% reduction compared to the control group's 20.3 kg per mu. This demonstrates that through precise nitrogen deficiency prediction and differentiated intervention strategies, the system can achieve "on-demand fertilization," ensuring healthy crop growth while avoiding the economic waste and environmental risks associated with excessive nitrogen application.

[0153] In terms of crop growth performance, using the NDVI increase as a reference indicator, the experimental group saw a 0.15 increase from 0.63 to 0.78, significantly higher than the control group's mere 0.10. This result demonstrates that intervention measures guided by the present method are not only more timely and effective, but also directly reflect improvements in the plant's physiological state and photosynthetic capacity.

[0154] Finally, in terms of the number of manual interventions, the experimental group required only 1.6 manual operations per mu, significantly lower than the control group's 3.2 operations, a 50% reduction. This data demonstrates the practical effectiveness of this invention in promoting intelligent agriculture and reducing labor dependence, and also provides a practical basis for large-scale promotion.

[0155] In summary, the data in this table fully demonstrates the effectiveness and practicality of this invention in agricultural production practice. By integrating intelligent algorithm optimization with machine learning models, and coordinating with agricultural control systems and operating equipment, this invention achieves a closed-loop control system from "data-driven identification" to "strategy output execution" to "dynamic feedback update." This truly solves the pain points of traditional methods, such as inaccurate identification, slow response, extensive fertilization, and inconsistent control methods, and has significant promotional value and industrial application prospects.

[0156] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for preventing and controlling nitrogen deficiency risk in tobacco, characterized in that: The steps include: S1. Collect multi-source data of tobacco growing areas, pre-process the multi-source data, and construct a dataset; S2. Construct a support vector machine model, using the constructed dataset as input and the tobacco nitrogen deficiency risk level as the output label, and define the penalty coefficient and kernel function parameters of the support vector machine model as adjustable variables; S3. Introducing the gray wolf optimization algorithm to perform global optimization search on the penalty coefficient and kernel function parameters of the support vector machine model, using the cross-validation recognition accuracy as the fitness function, executing the gray wolf population initialization, position update, fitness evaluation and population iteration process to obtain the optimal parameter combination; S4. Use the support vector machine model optimized by the Gray Wolf algorithm to predict nitrogen deficiency risk for the newly collected tobacco planting area data, and output the corresponding risk level label and risk score value; S5. Based on the nitrogen deficiency risk prediction results and the plot planting conditions, the preset nitrogen application rule library is called to generate differentiated intervention strategies and output as fertilization instructions that can be executed by agricultural equipment; S6. Deploy differentiated intervention strategies in smart agricultural control systems to drive fertilization devices to implement nitrogen deficiency intervention operations in target areas, while also dynamically monitoring plant growth, soil nitrogen concentration, and remote sensing indicators after fertilization. S7. Input the feedback data after fertilization into the dataset as the incremental update sample of the support vector machine model, retrain the support vector machine model and dynamically adjust the parameters of the gray wolf optimization algorithm to build a closed-loop feedback tobacco nitrogen deficiency risk prevention and control mechanism.

2. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The multi-source data specifically include soil nitrogen content data, meteorological environment data, plant growth image data, remote sensing vegetation index data and historical fertilization record data.

3. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The preprocessing of multi-source data specifically includes data denoising, null value filling, standardization, feature extraction and principal component analysis dimensionality reduction operations.

4. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The S2 specifically includes: S21. Set the input data set to: X={x1,x2,…,x n },x i ∈R d 4 Among them, x i represents the feature vector of the i-th sample, d is the feature dimension, R is a real number set, i = 1, 2, ..., n, n is the number of data sets, and the sample label set is: Y={y1,y2,...,y n },y i ∈{0,1,2,3}; Four classification labels corresponding to tobacco nitrogen deficiency risk levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency and severe nitrogen deficiency; S22. Construct a support vector machine model, wherein the first layer is a two-class support vector machine for determining whether the input sample has a risk of nitrogen deficiency, and the second layer is a three-class support vector machine for further subdividing the samples at risk of nitrogen deficiency into mild, moderate, or severe levels. Each layer of support vector machine uses an independently trained data subset; S23, define the composite kernel function K(x i ,x j ) is in the form of: Among them, the first term exp(-γ‖x i -x j ‖ 2 ) is the radial basis kernel function, the second is a polynomial kernel function, λ1 and λ2 are combined weight parameters, γ, r, and d are the kernel width parameter, constant term, and polynomial order of the kernel function, respectively. exp is an exponential function, and x j Represents the feature vector of the jth sample; S24. After the support vector machine model training is completed, the support vector is dynamically screened based on the boundary contribution of the support vector. The screening index includes the absolute value of the decision function |f(x i )| and the Lagrange multiplier α i The size of the support vectors with low contribution that meet the set threshold conditions is eliminated; S25. The constructed support vector machine model supported by the composite kernel function, having a hierarchical structure and a dynamic support vector screening mechanism is used as the basic structure of the tobacco nitrogen deficiency risk identification model, and the parameters are used as adjustable variables for the gray wolf optimization algorithm to perform joint optimization.

5. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The S3 specifically includes: S31. Construct a dynamic encoding optimization structure to encode each individual gray wolf into a hyperparameter combination vector of the support vector machine model: X k =(C k ,c k ,l 1k ,l 2k ); Among them, C k is the penalty coefficient, γ k is the kernel width parameter of the radial basis kernel function, λ 1k and λ 2k are the combined weights of RBF kernel and polynomial kernel in the composite kernel function, and N is the population size; S32. Set the maximum number of iterations T max , initialize the position vector of each individual in the gray wolf population, set the initial evolution factor a = 2, and record the initial global optimal fitness value F best (0); S33, perform five-fold cross validation on the support vector machine model represented by each individual, obtain the predicted label and the true label sequence, and calculate the fuzzy risk loss function L k for: in, is the predicted label of the i-th sample, y i is the true label, is a 0-1 indicator function, is the risk penalty weight corresponding to each nitrogen deficiency level; Further calculate the accuracy Acc k , prediction stability variance σ k , construct a multi-objective fitness function: Among them, α, β and η are weighted coefficients, Fitness(X k ) represents the position vector X of the kth individual k The overall fitness value, L k represents the normalized fuzzy risk loss value, and n represents the total number of samples in the validation set; S34. Select the three individuals with the best fitness values ​​in the current iteration as guide individuals, denoted as α, β, and δ respectively, and build an elite inheritance pool to store the best individuals in the past m generations for local guidance; S35. Define the adaptive evolution factor a based on the fitness convergence rate t : Where ΔF(t)=F best (t)-F best (t-1), ρ>0 is the search enhancement coefficient, ∈>0 is the convergence smoothing factor, t is the current iteration number, T max is the preset maximum number of iterations, exp is the exponential function; S36. Update the individual position vector according to the adaptive evolution factor and execute the search strategy: Among them, r k is the individual disturbance factor, X elite is the reference individual in the elite inheritance pool, is the position vector of the k-th gray wolf individual in the t+1th iteration, X α 、X β and X δ is the position vector of the three leading individuals with the best fitness value in the current population; S37, update the population fitness ranking and the global optimal solution record, iteratively execute steps S33 to S36 until the termination condition t≥T is met max , output the final optimal parameter combination to build the optimized support vector machine classification model.

6. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The S4 specifically includes: S41, receiving newly collected tobacco planting area data, wherein the data includes pre-processed multi-dimensional feature information and is consistent with the data structure used in the support vector machine model training stage; S42, inputting the new data into a support vector machine model obtained by the gray wolf optimization algorithm, where the support vector machine model includes an optimal hyperparameter configuration and a trained classification structure; S43, calling the support vector machine model to perform classification reasoning and perform nitrogen deficiency risk identification processing on each input sample; S44. Output a corresponding risk level label for each input sample, where the risk level includes four levels: no risk, mild nitrogen deficiency, moderate nitrogen deficiency, and severe nitrogen deficiency; S45. Generate a set of risk score values ​​reflecting the degree of prediction credibility for each input sample based on the calculation results of the internal discriminant function of the support vector machine; S46. Use the predicted labels and risk scores as model outputs for intervention strategy formulation and control instruction generation.

7. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The S5 specifically includes: S51, receiving the nitrogen deficiency risk prediction result output by the support vector machine model optimized by the gray wolf, outputting the corresponding nitrogen deficiency risk level label and risk score value for each target plot, representing the prediction and judgment of the nitrogen status of crops in the current agricultural area; S52: Synchronously call a plot information database to obtain planting condition information corresponding to each target plot, wherein the plot planting conditions include environmental and management variables such as plot area, soil texture type, tobacco crop variety, growth cycle stage, crop health status assessment indicators, and historical fertilization records; S53. Based on the risk level label and the plot planting conditions, a pre-built nitrogen application rule library is called to perform a rule matching and condition screening process, and a nitrogen application strategy template that best matches the target conditions is selected from the nitrogen application rule library. The rule library presets responsive fertilization strategy parameters under different risk levels based on agronomic experience and a crop fertilizer requirement model. S54. Based on the nitrogen application strategy template and taking into account the actual conditions of each plot, a differentiated intervention strategy generation process is implemented to ultimately form an operational nitrogen application intervention plan that includes nitrogen application dosage, nitrogen application method, and nitrogen application time window, achieving the goal of precise nitrogen application regulation in multiple regions and types of plots; S55, converting the differentiated intervention strategy into an operation instruction structure recognizable by agricultural equipment through a parameter mapping and format conversion module, generating a complete fertilization control instruction set including fertilization operation path planning information, specific equipment control parameters, operation crop identification, and execution timing logic; S56. Output the generated fertilization instructions that can be executed by agricultural equipment, and issue the instructions to the agricultural operation terminal through the control system interface to complete the control preparation of the nitrogen deficiency intervention task and ensure the effective linkage implementation of the prediction results and the intelligent control strategy.

8. A tobacco nitrogen deficiency risk prevention and control method according to claim 1, characterized in that: The S6 specifically includes: S61. Transmitting the differentiated intervention strategy generated based on the nitrogen deficiency risk prediction results and the planting conditions of the plot to the intelligent agricultural control system via a network interface. The intelligent agricultural control system has the ability to receive, analyze, and dispatch agricultural operation instructions. S62. Complete the analysis of the intervention strategy and the allocation of execution tasks in the intelligent agricultural control system, and send the fertilization operation parameters and control instructions to specific fertilization devices, such as intelligent fertilizer spreaders, automatic irrigation systems, or agricultural operation terminal equipment such as plant protection drones; S63: driving the fertilizing device in the target area to perform nitrogen deficiency intervention operations according to the fertilizing instruction, completing the processes of nitrogen dosage delivery, fertilizing path movement, and operation cycle control; S64. During the nitrogen deficiency intervention operation, the synchronous monitoring module is activated to collect data on plant growth changes in the intervention area, including leaf area index, color index, canopy height, and physiological growth parameters of growth potential images; S65. Synchronously obtain information on changes in soil nitrogen concentration within the fertilization area, collect shallow and deep soil sample data, evaluate the trend of nitrogen level changes, and form a dynamic soil nitrogen monitoring sequence; S66. Use remote sensing observation equipment or drone platforms to obtain spectral images and vegetation index information of the intervention area, conduct a macroscopic evaluation of the intervention effect at the spatial level, and ultimately form multi-dimensional feedback data on plant growth, soil nitrogen concentration, and remote sensing indicators.

Images