Method, medium and system for judging influence of climate soil on tobacco comprehensive quality

Abstract

The invention provides a method, medium and system for judging the influence of climate soil on the comprehensive quality of tobacco, and belongs to the technical field of use of neural network models.The method includes the steps that firstly, tobacco leaf and soil sample records and related geographic information of sampling points arranged at multiple points are obtained; obtaining soil physicochemical property parameters, growing season climate parameters and tobacco leaf quality evaluation indexes, and calculating tobacco comprehensive quality indexes; constructing a climate soil element Bayesian network model and an environment element incidence matrix, and converting the environment element incidence matrix into a sparse incidence matrix through threshold screening; identifying a plurality of correlation structures among the environmental elements; analyzing the influence of climate and geographical factors on soil characteristics by using a tobacco soil element interaction pre-training model; key ecological factors are determined through a mixed effect model and variance decomposition; and establishing a segmented structure equation model and analyzing influence mechanism differences under different conditions according to soil pH value grouping, thereby realizing systematic analysis of the tobacco quality formation mechanism.

Description

Technical Field

[0001] This invention belongs to the technical field of neural network model application, specifically, it relates to a method, medium and system for determining the impact of climate and soil on the overall quality of tobacco. Background Technology

[0002] Tobacco quality is influenced by a combination of ecological factors, including geographical environment, climate conditions, and soil characteristics, and is a key indicator determining tobacco value. Traditional tobacco quality evaluation mainly relies on sensory evaluation, chemical composition analysis, and assessment of tobacco appearance characteristics, employing single-factor analysis or simple correlation analysis to study the relationship between environmental factors and tobacco quality. While these methods can reflect certain aspects of tobacco quality, they struggle to reveal the intrinsic mechanisms underlying quality formation.

[0003] Current research on factors influencing tobacco quality often focuses on analyzing the impact of single environmental factors, such as soil nutrients, climate elements, or geographical location, on a particular quality indicator. This approach ignores the complex interactions between environmental factors and the differences in the impact of climate factors under different soil conditions. This leads to one-sided analytical results, making it difficult to explain the complete process of tobacco quality formation and providing precise technical support for tobacco cultivation under different environmental conditions.

[0004] Existing technologies struggle to establish conditional probability networks among climate and soil elements, accurately identify unidirectional, bidirectional, and cyclical correlations among environmental elements, and quantify the direct and indirect impacts of ecological factors on tobacco quality. Particularly, the differences in the mechanisms by which environmental factors affect tobacco quality under varying soil pH conditions are difficult to reveal, severely hindering the scientific evaluation and precise regulation of tobacco quality. In other words, existing technologies suffer from a lack of comprehensive and systematic analysis of the multi-level impact mechanisms of climate and soil factors on tobacco quality. Summary of the Invention

[0005] In view of this, the present invention provides a method, medium and system for determining the impact of climate and soil on the overall quality of tobacco, which can solve the technical problem in the prior art of lacking a comprehensive and systematic analysis of the multi-level impact mechanism of climate and soil factors on the overall quality of tobacco.

[0006] The present invention is implemented as follows: The first aspect of the present invention provides a method for determining the impact of climate and soil on the comprehensive quality of tobacco, comprising: acquiring records of tobacco leaf and soil samples from sampling points, as well as geographical coordinates and altitude data; acquiring soil physicochemical properties parameters; acquiring climate parameters during the growing season; acquiring tobacco leaf appearance and sensory characteristic indicators; applying an entropy integration function to calculate the comprehensive quality index of tobacco; constructing a Bayesian network model of climate and soil elements, and establishing a cross-correlation probability matrix between climate parameters during the growing season and soil physicochemical properties parameters; constructing a dense correlation matrix of environmental elements using the Pearson correlation coefficient method; converting the dense correlation matrix of environmental elements into a sparse correlation matrix of environmental elements using a threshold screening method; identifying the correlation structure between environmental elements based on the sparse correlation matrix of environmental elements; analyzing the impact of climate parameters and geographical data on soil parameters using a tobacco-soil element interaction pre-training model; determining key ecological factors affecting the comprehensive quality index of tobacco using a mixed-effects model and variance decomposition method; establishing a piecewise structural equation model to reveal the mechanism by which ecological parameters affect the comprehensive quality index of tobacco; and grouping the data according to soil pH value to analyze the differences in the impact of ecological factors on the comprehensive quality index of tobacco under different conditions.

[0007] The entropy integration function is used to integrate tobacco leaf appearance characteristics, tobacco leaf chemical composition, and tobacco leaf sensory characteristics into a comprehensive tobacco quality index. The inputs include the tobacco leaf appearance characteristics index scoring matrix, the tobacco leaf chemical composition index measurement result matrix, and the tobacco leaf sensory characteristics index scoring matrix, as well as the standardized limit parameters of each index. The output is the comprehensive tobacco quality index.

[0008] Among them, the Bayesian network model of climate and soil elements refers to a network model based on probabilistic graph theory to establish conditional probability relationships between environmental variables. It uses a directed acyclic graph to represent the dependencies between variables, with nodes representing random variables and edges representing conditional dependencies, and is used to infer the probability distribution of the influence of climate parameters on soil parameters.

[0009] Among them, the cross-correlation probability matrix refers to the conditional probability distribution matrix of soil physicochemical property parameters on the average temperature during the growing season, the average air humidity during the growing season, the total precipitation during the growing season, and the number of sunshine hours per ten days during the growing season.

[0010] Among them, the dense correlation matrix of environmental elements refers to the correlation matrix between various elements such as climate parameters, geographical coordinates, altitude data, and soil physicochemical properties parameters during the growing season. Each element in the matrix represents the correlation strength between the corresponding environmental elements in the row and column, forming a fully connected network structure.

[0011] Among them, the sparse correlation matrix of environmental elements refers to the simplified matrix obtained by filtering the dense correlation matrix of environmental elements by setting a correlation degree threshold. Only the connections with a correlation degree higher than the threshold are retained, weak correlations are excluded, and the interpretability of the model is improved.

[0012] Among them, the relationship structure between environmental elements includes: unidirectional relationship structure, which refers to the asymmetric influence relationship between environmental elements, where a change in one element leads to a change in another element, but the reverse influence is not significant; bidirectional relationship structure, which refers to the mutual influence relationship between environmental elements, where the two elements are mutually causal and form a feedback mechanism; and cyclic relationship structure, which refers to multiple environmental elements forming a closed-loop influence chain, where the first element influences the second element, the second element influences the third element, and finally returns to the first element through a series of intermediate elements to form a cycle.

[0013] The pre-trained model for element interaction between tobacco and soil is a multi-branch deep neural network architecture, which includes a climate data processing branch, a spatial data processing branch, and a soil basic parameter processing branch. Each branch uses different convolution kernel sizes to capture features at different scales. The features of the three branches are weighted and fused through an attention mechanism. The output layer predicts the content of macro- and micro-elements in the soil. The parameters of the pre-trained model for element interaction between tobacco and soil are adaptively adjusted through three key parameters: soil pH, soil organic matter content, and total precipitation during the growing season.

[0014] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described method for determining the impact of climate and soil on the overall quality of tobacco.

[0015] A third aspect of the present invention provides a system for determining the impact of climate and soil on the overall quality of tobacco, comprising the aforementioned computer-readable storage medium. The system is any one of a computer, a server, or a microcontroller. The computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

[0016] Compared with existing technologies, this invention provides a method, medium, and system for determining the impact of climate and soil on the overall quality of tobacco. This invention proposes a method for determining the impact of climate and soil on the overall quality of tobacco, which integrates the appearance characteristics, chemical composition, and sensory characteristics of tobacco leaves into an overall quality index through an entropy integration function, establishes a Bayesian network model of climate and soil elements, constructs a sparse correlation matrix of environmental elements, analyzes the various correlation structures between environmental elements, and achieves a holistic understanding of the complex system of the tobacco production environment.

[0017] This invention successfully identified key ecological factors affecting tobacco quality using a pre-trained model of tobacco-soil element interactions, a mixed-effects model, and variance decomposition. Furthermore, it quantified the direct and indirect impact mechanisms of environmental factors on tobacco quality using a piecewise structural equation model. In addition, this invention grouped data according to soil pH values, precisely revealing the differentiated impact patterns of ecological factors on tobacco quality under different pH conditions, thus overcoming the shortcomings of existing technologies.

[0018] The present invention solves the technical problem in the prior art that the multi-level influence mechanism of climate and soil on the comprehensive quality of tobacco cannot be systematically analyzed. By constructing an environmental factor association network and hierarchically analyzing the influence mechanism, the systematic analysis of the formation mechanism of tobacco quality is realized, providing a scientific basis for precise regulation of tobacco planting in different ecological regions, and effectively improving the scientificity and pertinence of tobacco quality management. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is a flowchart of the method of the present invention.

[0020] Figure 2 Random forest model diagram, showing the relative importance of environmental variables to the macro and trace elements in tobacco planting soil. *0.01 < P < 0.05; **0.001 < P < 0.01. Each sub-diagram is specifically (a) total nitrogen, (b) total phosphorus, (c) total potassium, (d) available phosphorus, (e) available potassium, (f) available copper, (g) available manganese, (h) available zinc, (i) available iron.

[0021] Figure 3 It is a coefficient diagram of the mixed effect model, showing the influence of environmental variables on the macro and trace elements in tobacco planting soil, including error bars representing 95% confidence intervals, and markers at different significance levels (p < 0.05; p < 0.01; p < 0.001). Each sub-diagram is specifically (a) total nitrogen, (b) total phosphorus, (c) total potassium, (d) available phosphorus, (e) available potassium, (f) available copper, (g) available manganese, (h) available zinc, (i) available iron.

[0022] Figure 4 It is a relative contribution diagram of climate and soil factors to the change of tobacco comprehensive quality. The sub-diagrams include (a) showing the contribution ratio of each factor, and the relationship diagrams between tobacco comprehensive quality and (b) total precipitation in the growing season, (c) total potassium and (d) available copper, where the shaded part represents 95% confidence intervals.

[0023] Figure 5 It is a structural equation model (SEM) diagram, showing the direct and indirect effects of environmental variables on tobacco quality. Among them, the red arrows represent significant positive correlation relationships, the black arrows represent significant negative correlation relationships, the standardized path coefficients are marked beside the arrows, and the variance explanation ratio (R Figure 6 ,

[0024] ) is marked beside each response variable, and different significance levels (p < 0.001, p < 0.01, p < 0.05) are indicated.

[0024] Figure 6It is a relationship diagram between environmental variables and tobacco quality under different pH conditions. The soil pH values are divided into three categories: acidic (pH < 6.5), neutral (6.5 < pH < 7.4), and alkaline (pH > 7.4). The differences in the effects of environmental factors including soil total potassium on tobacco quality under different pH conditions are analyzed. Specific implementation manners

[0025] To make the purpose, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0026] As Figure 1 shown, it is a flowchart of a method for determining the impact of climate and soil on the comprehensive quality of tobacco provided by the first aspect of the present invention. This method includes the following steps:

[0027] S01. Obtain the records of collecting tobacco leaf samples and soil samples at sampling points arranged in multiple layers and at multiple points in the test area, including the precise geographical coordinates and elevation data of each sampling point;

[0028] S02. Obtain the measured soil physical and chemical property parameters;

[0029] S03. Obtain the climate parameters of the growing season at the sampling points;

[0030] S04. Obtain the appearance characteristic indexes of tobacco leaves evaluated according to national standards, and obtain the sensory characteristic indexes of tobacco leaves evaluated by an expert group;

[0031] S05. Calculate the comprehensive quality index of tobacco by applying an entropy integration function;

[0032] S06. Construct a Bayesian network model of climate and soil elements, and establish a cross-correlation probability matrix between the climate parameters of the growing season and the soil physical and chemical property parameters;

[0033] S07. Construct a dense association matrix of environmental elements by using the Pearson correlation coefficient method;

[0034] S08. Convert the dense association matrix of environmental elements into a sparse association matrix of environmental elements by using the threshold screening method;

[0035] S09. Identify the one-way association structure, two-way association structure, and cyclic association structure between environmental elements based on the sparse association matrix of environmental elements;

[0036] S10. Use the pre-trained model of tobacco-soil element interaction to analyze the influence of the climate parameters of the growing season, geographical coordinates, and elevation data on the soil physical and chemical property parameters;

[0037] S11. Use the mixed effect model and variance decomposition method to determine the key ecological factors affecting the comprehensive quality index of tobacco

[0038] S12. Establish a piecewise structural equation model to reveal the direct and indirect influence mechanisms of growing season climate parameters, geographical coordinates, altitude data, and soil physicochemical properties on the comprehensive quality indicators of tobacco.

[0039] S13. Based on soil pH, the data were divided into acidic soil group, neutral soil group and alkaline soil group, and the differences in the influence of ecological factors on the comprehensive quality index of tobacco under different soil pH conditions were analyzed.

[0040] Among them, the climate parameters of the growing season include the average temperature of the growing season, the average air humidity of the growing season, the total precipitation of the growing season, and the sunshine hours per ten days of the growing season;

[0041] Among them, the soil physicochemical properties parameters include soil pH value, soil organic matter content, soil total nitrogen content, soil total phosphorus content, soil total potassium content, soil available phosphorus content, soil available potassium content, and the contents of four soil trace elements: soil available copper content, soil available manganese content, soil available zinc content, and soil available iron content.

[0042] Among them, the appearance characteristics of tobacco leaves include tobacco leaf color, tobacco leaf maturity, tobacco leaf identity, tobacco leaf structure, tobacco leaf oil content, and tobacco leaf chroma.

[0043] Among them, the chemical composition indicators of tobacco leaves include total alkaloid content, total sugar content, reducing sugar content, total nitrogen content, potassium content, chlorine content, and sugar-alkaloid ratio of tobacco leaves;

[0044] Among them, the sensory characteristics indicators of tobacco leaves include tobacco aroma quality indicators, tobacco aroma quantity indicators, tobacco permeability indicators, tobacco fineness indicators, tobacco smoothness indicators, tobacco aftertaste indicators, tobacco roundness indicators, tobacco aroma and off-flavor indicators, tobacco irritation indicators, and tobacco dryness indicators.

[0045] The entropy integration function is used to integrate tobacco leaf appearance characteristics, tobacco leaf chemical composition, and tobacco leaf sensory characteristics into a comprehensive tobacco quality index. The input includes a scoring matrix of six tobacco leaf appearance characteristics, a matrix of measurement results of seven tobacco leaf chemical composition, a scoring matrix of ten tobacco leaf sensory characteristics, and standardized limit parameters for each index. The output is the comprehensive tobacco quality index.

[0046] Among them, the Bayesian network model of climate and soil elements refers to a network model based on probabilistic graph theory to establish conditional probability relationships between environmental variables. It uses a directed acyclic graph to represent the dependencies between variables, with nodes representing random variables and edges representing conditional dependencies, and is used to infer the probability distribution of the influence of climate parameters on soil parameters.

[0047] Among them, the cross-correlation probability matrix refers to the conditional probability distribution matrix of the average temperature during the growing season, the average air humidity during the growing season, the total precipitation during the growing season, the number of sunshine hours per ten days during the growing season, the soil pH value, and the soil organic matter content on the soil total nitrogen content, soil total phosphorus content, soil total potassium content, soil available phosphorus content, soil available potassium content, soil available copper content, soil available manganese content, soil available zinc content, and soil available iron content.

[0048] Among them, the dense correlation matrix of environmental elements refers to the correlation matrix between various elements such as average temperature during the growing season, average air humidity during the growing season, total precipitation during the growing season, sunshine hours per ten days during the growing season, geographical coordinates, altitude data, soil pH value, soil organic matter content, soil total nitrogen content, soil total phosphorus content, soil total potassium content, soil available phosphorus content, soil available potassium content, soil copper content, soil manganese content, soil zinc content, and soil iron content. Each element in the matrix represents the correlation strength between the corresponding environmental elements in the row and column, forming a fully connected network structure.

[0049] Among them, correlation degree refers to the quantitative indicator of the degree of mutual influence between two environmental elements. It is calculated by statistical measures such as Pearson correlation coefficient, partial correlation coefficient or mutual information. The correlation degree ranges from 0 to 1, and the larger the value, the stronger the correlation.

[0050] Among them, the sparse correlation matrix of environmental elements refers to the simplified matrix obtained by filtering the dense correlation matrix of environmental elements by setting a correlation degree threshold. Only the connections with a correlation degree higher than the threshold are retained, weak correlations are excluded, and the interpretability of the model is improved.

[0051] Among them, the one-way correlation structure refers to the asymmetric influence relationship between environmental elements, where a change in one element leads to a change in another element, but the reverse influence is not significant;

[0052] Among them, the two-way correlation structure refers to the mutual influence relationship between environmental elements, where the two elements are mutually causal and form a feedback mechanism.

[0053] Among them, the circular correlation structure refers to multiple environmental elements forming a closed-loop influence chain, where the first element affects the second element, the second element affects the third element, and finally returns to the first element through a series of intermediate elements to form a cycle;

[0054] The specific structure of the pre-trained model for element interaction between tobacco and soil is a multi-branch deep neural network architecture, which includes a climate data processing branch, a spatial data processing branch, and a soil basic parameter processing branch. Each branch uses different convolution kernel sizes to capture features at different scales. The features of the three branches are weighted and fused through an attention mechanism, and the output layer predicts the content of macro- and micro-elements in the soil. The parameters of the pre-trained model for element interaction between tobacco and soil are adaptively adjusted through three key parameters: soil pH, soil organic matter content, and total precipitation during the growing season.

[0055] The mixed-effects model refers to a model that includes both fixed-effects and random-effects variables. The fixed-effects represent the factors of interest in the study, while the random-effects are used to control for correlations introduced by the sampling design. The mixed-effects model includes counties as random-effects.

[0056] Among them, variance decomposition refers to a statistical method for assessing the relative importance of variables by calculating the contribution rate of independent variables and their interactions to the total variance of the dependent variable;

[0057] Among them, the piecewise structural equation model refers to a statistical method used to analyze complex causal networks, which can quantify the direct and indirect influence paths between variables.

[0058] Among them, the growing season refers to the entire growth cycle of tobacco from transplanting to harvest, which is usually from March to September of the same year;

[0059] Among them, the entropy weight method refers to the method of calculating the weight of indicators based on the inherent structural characteristics of data. The weight is determined by measuring the information entropy of each evaluation indicator. The smaller the information entropy, the greater the variability of the indicator and the higher the weight.

[0060] The specific implementation methods of the above steps are described in detail below.

[0061] The specific implementation of step S01 involves using stratified random sampling to select representative sampling areas, ensuring that the samples cover different climate zones and soil types. First, the tobacco planting area is divided into several regional units according to administrative divisions. Then, a second level of stratification is performed based on climate characteristics and topographic features, and sampling points are randomly selected from each sub-region. At each sampling point, precise geographic coordinates, including longitude and latitude values, are recorded using a Global Positioning System (GPS), accurate to six decimal places. Simultaneously, an altimeter is used to record the altitude, with accuracy controlled within ±1m. Topsoil samples (0–30cm) and tobacco leaf samples are collected from each sampling point. Sampling is conducted within two weeks before the tobacco maturity period to ensure data representativeness. This step provides basic geospatial information and sample data for subsequent analysis.

[0062] The specific implementation of step S02 involves determining the physicochemical properties of the collected soil samples. First, the soil samples are air-dried, ground, and passed through a 2mm sieve. The soil pH is determined using the potentiometric method, specifically by mixing soil and water at a 1:2.5 ratio, shaking thoroughly for 30 minutes, allowing to stand for 10 minutes, and then measuring using a calibrated pH meter. Soil organic matter content is determined using the potassium dichromate oxidation method; total nitrogen content is determined using the Kjeldahl method; total phosphorus content is determined using hydrofluoric acid-perchloric acid digestion and molybdenum-antimony colorimetric method; total potassium content is determined using hydrofluoric acid-perchloric acid digestion and flame photometry; available phosphorus content is determined using sodium bicarbonate extraction and molybdenum-antimony colorimetric method; available potassium content is determined using ammonium acetate extraction and flame photometry; and the contents of four trace elements—available copper, available manganese, available zinc, and available iron—are determined using DTPA extraction and atomic absorption spectrophotometry. All measurements are performed three times, and the average value is taken as the final result.

[0063] The specific implementation of step S03 involves acquiring climate data for the tobacco growing season at the sampling points. Complete climate information is obtained through three channels: first, using historical data from the meteorological station closest to the sampling point to collect daily meteorological records during the tobacco growing season (March to September); second, deploying automatic meteorological monitoring equipment at the sampling site to record temperature, humidity, precipitation, and sunshine data in real time; and third, using meteorological satellite and radar data for spatial interpolation to compensate for the spatial limitations of ground observation stations. The acquired raw climate data undergoes quality control and homogeneity checks, outliers are removed, and missing values ​​are supplemented. Finally, the average temperature during the growing season (obtained by summing the daily average temperatures and dividing by the number of days), the average air humidity during the growing season (obtained by summing the daily average relative humidity and dividing by the number of days), the total precipitation during the growing season (obtained by summing the daily precipitation), and the sunshine hours per ten-day period during the growing season (obtained by summing the sunshine hours for each 10-day period during the growing season) are calculated.

[0064] The specific implementation method of step S04 is to evaluate the appearance characteristics of the collected tobacco leaf samples according to the Chinese national standard GB 2635-1992. The specific evaluation process is as follows: First, the tobacco leaf samples are visually evaluated under standard lighting conditions (standard light source with a color temperature of 5500K). At least three experienced reviewers score six indicators, namely tobacco leaf color, tobacco leaf maturity, tobacco leaf identity, tobacco leaf structure, tobacco leaf oil content, and tobacco leaf chroma, on a 10-point scale, and the average value is taken as the final score. At the same time, the chemical composition indicators of the tobacco leaves are determined using standard analytical methods. Among them, the total alkaloid content and potassium content of the tobacco leaves are determined using continuous flow analysis. The total sugar content and reducing sugar content of tobacco leaves were determined using the phenol-sulfuric acid method, the total nitrogen content was determined using the Kjeldahl method, and the chlorine content was determined using the potentiometric titration method. Finally, at least five professional tasters evaluated the sensory characteristics of the tobacco leaves, scoring 10 indicators on a 5-point scale, including aroma quality, aroma quantity, transparency, fineness, smoothness, aftertaste, roundness, off-flavors, irritation, and dryness. The average score was taken as the final score for each indicator.

[0065] The specific implementation of step S05 involves applying an entropy integration function to calculate the comprehensive quality index of tobacco. First, the appearance characteristics, chemical composition, and sensory characteristics of tobacco leaves are standardized to eliminate the influence of different dimensions. For positive indicators, a maximum standardization method is used; for negative indicators, a minimum standardization method is used; and for intermediate indicators, an interval standardization method is used. Then, the weights of each indicator are calculated based on the entropy weight method, and the information entropy of each indicator is calculated. The smaller the information entropy, the greater the variability of the indicator, and the greater its contribution to the comprehensive evaluation; therefore, the higher the weight should be. Finally, the standardized indicator values ​​are multiplied by their corresponding weights and summed to obtain the comprehensive quality index of tobacco. This index ranges from 0 to 100; a higher value indicates better comprehensive quality. Experiments show that the comprehensive quality index of high-quality tobacco is typically above 80, medium-quality tobacco is between 60 and 80, and low-quality tobacco is below 60.

[0066] The specific implementation of step S06 involves constructing a Bayesian network model of climate and soil elements. First, data preprocessing is performed on the growing season climate parameters and soil physicochemical properties, including missing value handling, outlier detection and removal, and data standardization. Then, a structure learning algorithm is used to construct the Bayesian network structure. Specifically, a search-based method is employed, starting with a blank network and evaluating the network's score function value (e.g., BIC score) after each operation (adding, deleting, or reversing edges), selecting the network structure that optimizes the score function value. After the network structure is determined, the conditional probability distribution parameters are learned using maximum likelihood estimation, resulting in a complete Bayesian network model. Based on the Bayesian network, the conditional probabilities between nodes are calculated, and a cross-correlation probability matrix is ​​constructed to quantify the probability distribution of the influence of growing season climate parameters on soil physicochemical properties. Each element of this matrix represents the conditional probability that the soil parameter falls within a certain range of climate parameter values, providing a probabilistic basis for subsequent analysis.

[0067] The specific implementation of step S07 involves constructing a dense correlation matrix of environmental elements using the Pearson correlation coefficient method. First, the environmental element data, including average temperature during the growing season, average air humidity during the growing season, total precipitation during the growing season, sunshine hours per ten-day period during the growing season, geographical coordinates, altitude, soil pH, soil organic matter content, total nitrogen content, total phosphorus content, total potassium content, available phosphorus content, available potassium content, copper content, manganese content, zinc content, and iron content, are standardized. Then, the Pearson correlation coefficient between any two environmental elements is calculated, forming an n×n dense correlation matrix of environmental elements, where n is the total number of environmental elements. The closer the absolute value of the correlation coefficient is to 1, the stronger the correlation between the two elements; a positive correlation coefficient indicates a positive correlation, and a negative correlation indicates a negative correlation. To improve the reliability of the results, a random resampling technique is used to test the significance of the correlation coefficients, retaining only correlation coefficients with a p-value less than 0.05, and setting the rest to 0. This matrix comprehensively reflects the correlation strength between various environmental elements, providing a data foundation for subsequent screening of key correlation paths.

[0068] The specific implementation of step S08 involves converting the dense correlation matrix of environmental elements into a sparse correlation matrix using a threshold screening method. Based on statistical principles and experience in environmental science, a correlation coefficient threshold of ±0.5 is set. That is, when the absolute value of an element in the dense correlation matrix is ​​greater than 0.5, a strong correlation is considered to exist between the two elements, and the connection is retained; when the absolute value is less than or equal to 0.5, the corresponding matrix element is set to 0, and the connection is deleted. Simultaneously, the threshold can be dynamically adjusted according to data complexity and research objectives, simplifying the network structure while retaining important information. In the sparse correlation matrix of environmental elements obtained through threshold screening, non-zero elements represent strong correlation paths, simplifying the complex network structure, highlighting the main influencing paths, and providing a clear foundation for subsequent analysis.

[0069] The specific implementation of step S09 is based on identifying unidirectional, bidirectional, and cyclical association structures among environmental elements using a sparse association matrix. First, the sparse association matrix is ​​represented as a directed graph, where nodes represent environmental elements, edges represent the relationships between elements, and the direction of the edges is determined by Granger causality tests. For any two environmental elements A and B, if the Granger causality test p-value for A to B is less than 0.05, while the p-value for B to A is greater than or equal to 0.05, it is determined to be a unidirectional association structure, meaning A unidirectionally influences B. If the Granger causality test p-values ​​for both A to B and B to A are less than 0.05, it is determined to be a bidirectional association structure. A depth-first search algorithm is used to detect whether there exists a path starting from a node and ultimately returning to that node in the directed graph; if such a path exists, it is identified as a cyclical association structure. The identified association structures are then labeled and visualized to clarify the causal network structure among environmental elements, providing a scientific basis for understanding the complex interactions in ecosystems.

[0070] The specific implementation of step S10 involves using a pre-trained model of soil element interactions to analyze the impact of growing season climate parameters, geographic coordinates, and altitude data on soil physicochemical properties. This model employs a multi-branch deep neural network architecture, including a one-dimensional convolutional neural network branch for processing time-series climate data, a two-dimensional convolutional neural network branch for processing spatial geographic data, and a multilayer perceptron branch for processing basic soil parameters. Each branch uses convolutional kernels of different sizes to capture multi-scale features. The climate data processing branch uses one-dimensional convolutional kernels of 1×3, 1×5, and 1×7 sizes, while the spatial data processing branch uses two-dimensional convolutional kernels of 3×3 and 5×5 sizes. The feature outputs of the three branches are fused through a self-attention mechanism, with the attention weights adaptively adjusted by three key parameters: soil pH, soil organic matter content, and total precipitation during the growing season. The model output layer is optimized using a mean squared error loss function to predict the content of soil elements such as total nitrogen, total phosphorus, total potassium, available phosphorus, available potassium, copper, manganese, zinc, and iron. The model was trained using stochastic gradient descent with a batch size of 64. The initial learning rate was set to 0.001, and a learning rate decay strategy was used to train for 200 epochs or until the loss on the validation set no longer decreased.

[0071] The specific implementation of step S11 involves using a mixed-effects model and variance decomposition to determine the key ecological factors affecting the overall quality index of tobacco. First, a multicollinearity test is performed on all environmental factors, and the variance inflation factor is calculated, eliminating variables with a variance inflation factor greater than 5. Then, a mixed-effects model is constructed with the overall quality index of tobacco as the dependent variable and environmental factors as independent variables. The fixed effects include growing season climate parameters, geographical coordinates, altitude data, and soil physicochemical properties, while the random effect is the county level. Maximum likelihood estimation is used to estimate the model parameters, and the model significance is determined using the likelihood ratio test. Subsequently, variance decomposition is applied to calculate the contribution rate of each environmental factor to the total variance of the overall quality index of tobacco. Environmental factors with higher contribution rates have a more significant impact on the overall quality of tobacco. Based on the variance contribution rate ranking, environmental factors with a cumulative contribution rate exceeding 80% are selected as key ecological factors, providing a scientific basis for subsequent precise regulation of the tobacco planting environment.

[0072] The specific implementation of step S12 involves establishing a piecewise structural equation model to reveal the influence mechanism of environmental factors on the overall quality index of tobacco. First, based on the key ecological factors identified in step S11 and the correlation structure identified in step S09, an initial path map is constructed, including direct and indirect paths. Then, sample data required for model fitting is collected, including growing season climate parameters, geographical coordinates, altitude data, soil physicochemical properties, and overall tobacco quality index. Path coefficients are estimated using the piecewise structural equation model, and the model's goodness of fit is evaluated using methods such as chi-square tests and comparison indices. During model optimization, a correction index is used to guide model adjustments, deleting insignificant paths and adding theoretically supported new paths until a final model with satisfactory goodness of fit is obtained. In the final model, the path coefficients represent the intensity of the direct influence of each factor on the overall quality of tobacco. The indirect influence intensity is calculated by multiplying the path coefficients, and the sum of the two is the total effect, revealing the complete mechanism by which environmental factors influence the overall quality of tobacco.

[0073] The specific implementation of step S13 involves dividing all sample data into three groups based on soil pH: acidic soil (pH less than 6.5), neutral soil (pH between 6.5 and 7.4), and alkaline soil (pH greater than 7.4). Steps S11 and S12 are repeated for each of the three groups to construct intra-group mixed effect models and piecewise structural equation models. By comparing the differences in the influence coefficients of environmental factors on the overall quality indicators of tobacco across the three soil pH groups, the regulatory effect of soil pH on the intensity of ecological factors is analyzed. Specific comparisons include: differences in the composition of key ecological factors within each group; differences in the direction and intensity of the same ecological factors under different pH conditions; and differences in the main pathways by which ecological factors affect the overall quality of tobacco under different pH conditions. Through inter-group difference analysis, differentiated strategies for regulating tobacco quality under different soil pH conditions are identified, providing scientific guidance for achieving precision tobacco planting management.

[0074] The specific implementation method for the pre-trained model of tobacco element interaction and the establishment of the training dataset is as follows: First, a multi-branch deep neural network architecture is constructed. The climate data processing branch uses a combination of one-dimensional convolutional layers, batch normalization layers, and max pooling layers. The one-dimensional convolutional layer is designed with three parallel channels, using convolutional kernels of 1×3, 1×5, and 1×7 sizes respectively, to capture climate feature patterns at different time scales. The spatial data processing branch uses two-dimensional convolutional layers to process latitude and longitude gridded data, using parallel channels with convolutional kernel sizes of 3×3 and 5×5. The soil basic parameter processing branch uses a three-layer fully connected network with 128 and 64 hidden layer neurons respectively, and the activation function is ReLU. The features of the three branches are fused through a multi-head self-attention mechanism, using 8 attention heads, each with a dimension of 64. The attention weights are adaptively calculated through a parameter adjustment network, which takes soil pH, soil organic matter content, and total precipitation during the growing season as inputs and outputs adjustment coefficients. The training dataset was built using historical data from 205 tobacco planting sites across the country, covering a wide area from latitude 24°18′N to 44°22′N and longitude 103°48′E to 129°50′E. Data preprocessing included outlier detection and handling, missing value imputation, and data standardization. A transfer learning strategy was adopted, first pre-training the encoder on large-scale soil data in an unsupervised manner, and then performing supervised fine-tuning using labeled data, which effectively improved model performance.

[0075] The specific implementation method for establishing the Bayesian network model of climate and soil elements and the training dataset is as follows: First, the climate and soil data are discretized, dividing continuous variables into a finite number of intervals. For temperature and humidity parameters, five equidistant intervals are used; for precipitation, five intervals are based on quantiles; and for soil parameters, intervals are based on their suitability for agricultural production. Then, the enhanced Hill-Climbing algorithm is used to learn the network structure. This algorithm starts with a blank network and iteratively adds, deletes, or reverses edges, using the Bayesian information criterion to evaluate the quality of the network structure. To avoid local optima, the algorithm performs 100 random restarts and selects the network structure with the highest score. In the parameter learning phase, Bayesian estimation is used to introduce prior knowledge to alleviate the data sparsity problem. The training dataset contains multi-year observation records of climate, spatial, and soil parameters, and 10-fold cross-validation is used to evaluate model performance. Finally, the model calculates the conditional independence relationship between any two nodes, generating a cross-correlation probability matrix. The diagonal elements of this matrix are 1, and the off-diagonal elements are the conditional probability values ​​between nodes, intuitively reflecting the degree of influence of climate elements on soil elements.

[0076] A second aspect of the present invention provides a computer-readable storage medium storing program instructions, which, when executed in a computer, are used to perform the above-described method for determining the impact of climate and soil on the overall quality of tobacco.

[0077] A third aspect of the present invention provides a system for determining the impact of climate and soil on the overall quality of tobacco, comprising the aforementioned computer-readable storage medium. The system is any one of a computer, a server, or a microcontroller. The computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

[0078] The mathematical model or calculation process involved in this invention will be described in detail below.

[0079] The process of calculating the comprehensive quality index of tobacco using the entropy integration function in step S05 is specifically represented as follows:

[0080] First, the appearance characteristics, chemical composition, and sensory characteristics of tobacco leaves were standardized:

[0081]

[0082] In the formula, Z ij X is the standardized value of the j-th indicator for the i-th sample; ij Let X be the original value of the j-th indicator for the i-th sample; min(X) j ) represents the minimum value of the j-th index; max(X) j ) represents the maximum value of the j-th indicator.

[0083] Then, calculate the information entropy of the j-th indicator:

[0084]

[0085] In the formula, E j Let n be the information entropy of the j-th indicator; n be the number of samples. Let represent the weight of the j-th indicator in the i-th sample.

[0086] Next, calculate the weight of the j-th indicator:

[0087]

[0088] In the formula, W j is the weight of the j-th indicator; m is the total number of indicators.

[0089] Finally, calculate the overall quality index of tobacco:

[0090]

[0091] In the formula, Q is the comprehensive quality index of tobacco, with a value range of 0 to 100.

[0092] The specific process of calculating the cross-correlation probability matrix of the climate and soil element Bayesian network model in step S06 is as follows:

[0093] First, define the set of climate parameters C = {c1, c2, c3, c4} and the set of soil parameters S = {s1, s2, ..., s4}. 11}, where c1 is the average temperature during the growing season, c2 is the average air humidity during the growing season, c3 is the total precipitation during the growing season, and c4 is the sunshine hours per ten-day period during the growing season; s1 to s 11 These are soil pH, soil organic matter content, soil total nitrogen content, soil total phosphorus content, soil total potassium content, soil available phosphorus content, soil available potassium content, soil available copper content, soil available manganese content, soil available zinc content, and soil available iron content.

[0094] Then, the conditional probability distribution is calculated using a Bayesian network:

[0095]

[0096] In the formula, P(s) j |c i Given climate parameter c i Soil parameters s under the conditions j The conditional probability; P(s) j c i ) represents the climate parameter c i and soil parameters s j The joint probability; P(c i ) represents the climate parameter c i Marginal probability; P(c i |s j ) is a given soil parameter s j Climate parameter c under the condition i The conditional probability; P(s) j ) represents soil parameter s j The marginal probability.

[0097] Construct a cross-correlation probability matrix M, with the following elements:

[0098] M ij =P(s) j |c i );

[0099] In the formula, M ij This represents the element in the i-th row and j-th column of the cross-correlation probability matrix, i.e., given the climate parameter c. i Soil parameters s under the conditions jThe conditional probability.

[0100] The calculation process for constructing the dense correlation matrix of environmental elements using the Pearson correlation coefficient method in step S07 is specifically represented as follows:

[0101] Define the environmental element vector V = {v1, v2, ..., v...} n}, where n is the total number of environmental elements, including growing season climate parameters, geographic coordinates, altitude data, and soil physicochemical properties.

[0102] Calculate the Pearson correlation coefficient among environmental factors:

[0103]

[0104] In the formula, r ij For environmental factors v i With v j The Pearson correlation coefficient between them; m is the number of sampling points; v ik For the environmental element v at the k-th sampling point i The value; For environmental factors v i The average value; v jk For the environmental element v at the k-th sampling point j The value; For environmental factors v j The average value.

[0105] Construct a dense correlation matrix R of environmental elements:

[0106]

[0107] In the formula, R is the dense correlation matrix of environmental elements; r ij For environmental factors v i With v j The Pearson correlation coefficient between them.

[0108] The calculation process in step S08, which uses a threshold filtering method to convert the dense correlation matrix of environmental elements into a sparse correlation matrix of environmental elements, is specifically represented as follows:

[0109] Setting the correlation coefficient threshold θ = 0.5, construct the sparse correlation matrix R′ of environmental elements:

[0110]

[0111] In the formula, R′ ij r is the element in the i-th row and j-th column of the sparse correlation matrix of environmental elements; ij θ represents the element in the i-th row and j-th column of the dense correlation matrix of environmental elements; θ is the correlation coefficient threshold, with a value of 0.5.

[0112] The specific calculation process involved in identifying the correlation structure between environmental elements based on the sparse correlation matrix of environmental elements in step S09 is as follows:

[0113] For any two environmental factors v i and v j Calculate the Granger causality test F-statistic:

[0114]

[0115] In the formula, F i→j For environmental factors v i For v j Granger causality test F-statistic; RSS r For the constrained model (excluding v) i The sum of squared residuals (lagging terms); RSS ur For unrestricted models (including v) i The sum of squared residuals (lagging terms); p is the lag order; T is the sample size.

[0116] Calculate the p-value based on the F-statistic. i→j Determine the association type:

[0117] One-way association structure: if p i→j <0.05 and p j→i If ≥0.05, then v exists. i to v j One-way association;

[0118] Bidirectional association structure: If p i→j <0.05 and p j→i If v < 0.05, then v exists. i and v j Two-way association between them;

[0119] Circular association structure: Detecting the existence of a slave node v in a directed graph using a depth-first search algorithm. i Can you eventually return to v after setting off? i The path. The specific algorithm is as follows:

[0120]

[0121] In the formula, DFS is the depth-first search function; v i `i` represents the currently visited node; `visited` represents the set of visited nodes; `path` represents the set of nodes in the current path; `E` represents the set of edges in the directed graph; `(i, j) ∈ E` indicates that there exists a follower node `v`. i to node v j The edge.

[0122] The calculation process for determining the key ecological factors affecting the overall quality index of tobacco using the mixed-effects model and variance decomposition method in step S11 is specifically represented as follows:

[0123] First, calculate the variance inflation factor for each environmental element:

[0124]

[0125] In the formula, VIF j Let be the variance inflation factor of the j-th environmental element; The determination coefficient is obtained by performing regression analysis with the j-th environmental element as the dependent variable and all other environmental elements as independent variables.

[0126] Constructing a mixed-effects model:

[0127]

[0128] In the formula, Q ij βi represents the comprehensive tobacco quality index of the j-th county and the i-th sampling point; β0 is the intercept term; βi k X is the fixed effect coefficient for the k-th environmental factor; ijk Let p be the value of the k-th environmental element at the i-th sampling point in the j-th county; p is the number of environmental elements; u j Let be the random effect in the j-th county, following a normal distribution. ε ij The residual term follows a normal distribution.

[0129] The contribution rate of each environmental factor to the overall quality index of tobacco was calculated by variance decomposition:

[0130]

[0131] In the formula, C k β represents the contribution rate of the k-th environmental factor to the overall quality index of tobacco. k Var(X) is the fixed effect coefficient for the k-th environmental factor; k Let ) represent the variance of the k-th environmental element; The variance of the random effects at the county level; denoted as the variance of the residual term.

[0132] The calculation process for establishing a piecewise structural equation model in step S12 to reveal the direct and indirect influence mechanisms of environmental factors on the overall quality indicators of tobacco is specifically represented as follows:

[0133] The structural equations consist of two parts: the measurement equations and the structural equations.

[0134] Measurement equation:

[0135] X = Λ x ξ+δ;

[0136] Y = Λ y η+ε;

[0137] In the formula, X is the vector of observed environmental element variables; Y is the observed variable of comprehensive tobacco quality index; ξ is the vector of latent environmental element variables; η is the latent variable of comprehensive tobacco quality; Λ x For environmental factor loading matrix; Λ y δ is the tobacco comprehensive quality factor loading matrix; δ is the environmental factor measurement error vector; ε is the tobacco comprehensive quality measurement error.

[0138] Structural equations:

[0139] η = Bη + Γξ + ζ;

[0140] In the formula, B is the path coefficient matrix between endogenous latent variables; Γ is the path coefficient matrix between exogenous latent variables and endogenous latent variables; and ζ is the structural equation residual vector.

[0141] Calculation of direct effects, indirect effects, and total effects:

[0142] The direct effects matrix DE = Γ;

[0143] Indirect effects matrix: IE = (IB) -1 Γ-Γ;

[0144] Total effects matrix TE = (IB) -1 Γ;

[0145] In the formula, I is the identity matrix; (IB) -1 It is the inverse of matrix (IB).

[0146] The functional and variable relationships used in the above equations are based on statistical theory and ecological principles. The standardization process employs the minimax method to eliminate the influence of different dimensions and ensure comparability among the indicators. The information entropy calculation formula originates from information theory and is used to quantify the uncertainty and information content of indicators; the smaller the information entropy, the greater the indicator variability and its contribution to the evaluation. The weight calculation formula ensures that the weights sum to 1 and are inversely proportional to the information entropy, reflecting the contribution of indicator discrimination to the evaluation.

[0147] Conditional probability calculations in Bayesian networks, based on Bayes' theorem, effectively describe the probabilistic dependencies between variables. The Pearson correlation coefficient measures the degree of linear correlation between two variables, ranging from -1 to 1; the closer the absolute value is to 1, the stronger the correlation. The Granger causality test, based on time series analysis theory, determines causal relationships between variables by comparing the sum of squared residuals of restricted and unrestricted models.

[0148] Mixed-effects models, incorporating both fixed and random effects, can handle nested data structures and are suitable for analyzing hierarchical data from sampling points within a county. Variance decomposition methods, based on linear regression theory, quantify the relative importance of each variable by calculating its contribution to the total variance. Piecewise structural equation modeling combines the advantages of factor analysis and path analysis, enabling the simultaneous handling of multiple dependent and independent variables, analysis of complex causal networks, and differentiation between direct and indirect effects.

[0149] Specifically, the principle of this invention is as follows: Based on the principles of systems science and multi-level analysis, this invention decomposes the complex relationship between tobacco quality and environmental factors into multiple interrelated analytical levels. First, this invention employs an entropy integration function, based on information entropy theory, to adaptively allocate weights according to the degree of variation of each evaluation indicator, scientifically integrating 22 scattered tobacco quality indicators into a comprehensive indicator. This provides a unified quantitative standard for quality evaluation, avoiding the problems of strong subjectivity and independent indicators in traditional evaluation methods.

[0150] Secondly, this invention employs Bayesian network theory to construct a conditional probability relationship model among climate and soil elements. A directed acyclic graph (DAG) represents the dependencies between variables. Compared to traditional correlation analysis, this method reveals the directionality of causal relationships. Furthermore, a dense correlation matrix of environmental elements is constructed using the Pearson correlation coefficient method, and then converted into a sparse correlation matrix through threshold filtering, effectively identifying complex interaction patterns among environmental elements. This graph-based network structure analysis method overcomes the limitations of traditional methods in handling multivariate interactions.

[0151] This invention also innovatively designs a pre-trained model for elemental interactions in tobacco soil, employing a multi-branch deep neural network architecture with different data processing branches. By using an attention mechanism to weightedly fuse features, it effectively captures the complex influences of climate and geographical factors on soil properties. Simultaneously, a mixed-effects model is used to control the spatial correlation introduced by the sampling design, variance decomposition is used to assess the relative importance of each variable, and a piecewise structural equation model is used to quantify direct and indirect influence paths. Finally, grouped analysis based on soil pH values ​​reveals the differences in influence mechanisms under different pH conditions.

[0152] This multi-level analytical framework, which moves from the whole to the part, from correlation to causation, and from direct to indirect analysis, enables the present invention to systematically reveal the impact mechanism of climate and soil on the overall quality of tobacco. It provides a theoretical basis for precise regulation of tobacco quality in different ecological environments and solves the technical problem of existing technologies that analyze tobacco quality factors in a single and one-sided manner.

[0153] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0154] The specific implementation methods of steps S01-S04 in this embodiment are the same as those described above, and will not be repeated here.

[0155] The specific implementation of step S05 involves applying an entropy integration function to calculate the comprehensive quality index of tobacco. First, the appearance characteristics, chemical composition, and sensory characteristics of tobacco leaves are standardized to eliminate the influence of different dimensions. The specific calculation formula is as follows: In the formula, Z ij X is the standardized value of the j-th indicator for the i-th sample; ij Let X be the original value of the j-th indicator for the i-th sample; min(X) j ) represents the minimum value of the j-th index; max(X) j Let be the maximum value of the j-th indicator. Then, calculate the weight of each indicator based on the entropy weight method, and calculate the information entropy of each indicator: In the formula, E j Let n be the information entropy of the j-th indicator; n be the number of samples. Let j represent the weight of the j-th indicator for the i-th sample. Then, calculate the weight of the j-th indicator: In the formula, W j Let be the weight of the j-th indicator; m be the total number of indicators. Finally, calculate the overall quality index of tobacco: In the formula, Q represents the comprehensive quality index of tobacco, with a value ranging from 0 to 100. This method effectively integrates various characteristics of tobacco leaves and provides an objective and quantitative comprehensive quality evaluation index.

[0156] The specific implementation of step S06 involves constructing a Bayesian network model of climate and soil elements. First, data preprocessing is performed on the growing season climate parameters and soil physicochemical properties parameters, including missing value handling, outlier detection and removal, and data standardization. Then, a structure learning algorithm is used to construct the Bayesian network structure, specifically the enhanced Hill-Climbing algorithm. Starting from a blank network, the network's BIC score is evaluated after each operation (adding, deleting, or reversing edges), and the network structure that optimizes the score function value is selected. After the network structure is determined, the conditional probability distribution parameters are learned using maximum likelihood estimation to obtain the complete Bayesian network model. Based on the Bayesian network, the conditional probabilities between each node are calculated, and a cross-correlation probability matrix is ​​constructed. The calculation formula is as follows: In the formula, P(s) j |c i Given climate parameter c i Soil parameters s under the conditions j The conditional probability; P(s) j c i ) represents the climate parameter c i and soil parameters s j The joint probability; P(ci ) represents the climate parameter c i Marginal probability; P(c i |s j ) is a given soil parameter s j Climate parameter c under the condition i The conditional probability; P(s) j ) represents soil parameter s j The marginal probabilities. Construct a cross-correlation probability matrix M, with matrix elements as: M ij =P(s) j |c i In the formula, M ij This represents the element in the i-th row and j-th column of the cross-correlation probability matrix, i.e., given the climate parameter c. i Soil parameters s under the conditions j The conditional probabilities are calculated. This matrix quantifies the probabilistic impact of climate factors on soil parameters, providing a probabilistic basis for subsequent analysis.

[0157] The specific implementation of step S07 involves constructing a dense correlation matrix of environmental elements using the Pearson correlation coefficient method. First, the environmental element data, including average temperature during the growing season, average air humidity during the growing season, total precipitation during the growing season, sunshine hours per ten-day period during the growing season, geographical coordinates, altitude data, soil pH, soil organic matter content, total nitrogen content, total phosphorus content, total potassium content, available phosphorus content, available potassium content, copper content, manganese content, zinc content, and iron content, are standardized. Then, the Pearson correlation coefficient between any two environmental elements is calculated using the following formula: In the formula, r ij For environmental factors v i With v j The Pearson correlation coefficient between them; m is the number of sampling points; v ik For the environmental element v at the k-th sampling point i The value; For environmental factors v i The average value; v jk For the environmental element v at the k-th sampling point j The value; Let vj be the average value of environmental element vj. Construct a dense correlation matrix R for environmental elements: In the formula, R is the dense correlation matrix of environmental elements; r ij For environmental factors v i With v j The Pearson correlation coefficients between the elements were used. To improve the reliability of the results, a random resampling technique was employed to test the significance of the correlation coefficients, retaining only those with p-values ​​less than 0.05 and setting the rest to 0. This matrix comprehensively reflects the correlation strength between various environmental elements, providing a data foundation for subsequent screening of key correlation paths.

[0158] The specific implementation of step S08 involves converting the dense correlation matrix of environmental elements into a sparse correlation matrix using a threshold screening method. Based on statistical principles and experience in environmental science, a correlation coefficient threshold of ±0.5 is set. That is, when the absolute value of an element in the dense correlation matrix is ​​greater than 0.5, a strong correlation is considered to exist between the two elements, and the connection is retained; when the absolute value is less than or equal to 0.5, the corresponding matrix element is set to 0, and the connection is deleted. The specific calculation formula is as follows: In the formula, R′ ij r is the element in the i-th row and j-th column of the sparse correlation matrix of environmental elements; ij Let be the element in the i-th row and j-th column of the dense correlation matrix of environmental elements; θ is the correlation coefficient threshold, with a value of 0.5. By filtering with the threshold, the original dense correlation matrix is ​​transformed into a sparse matrix that highlights the main correlations, simplifying the complex network structure, highlighting the main influencing paths, and providing a clear correlation network foundation for subsequent analysis.

[0159] The specific implementation of step S09 is based on identifying unidirectional, bidirectional, and cyclic association structures between environmental elements using the sparse association matrix of environmental elements. First, the sparse association matrix of environmental elements is represented as a directed graph, where nodes represent environmental elements, edges represent the association relationships between elements, and the direction of the edges is determined by the Granger causality test. For any two environmental elements v... i and v j Calculate the Granger causality test F-statistic: In the formula, F i→j For environmental factors v i For v j Granger causality test F-statistic; RSS r For the constrained model (excluding v) i The sum of squared residuals (lagging terms); RSS ur For unrestricted models (including v) i The sum of squared residuals (lagging terms); p is the lag order; T is the sample size. The p-value is calculated using the F-statistic. i→j Determine the association type: a one-way association structure is p i→j <0.05 and p j→i ≥0.05, bidirectional association structure is p i→j <0.05 and p j→i <0.05. For circularly associated structures, a depth-first search algorithm is used to detect whether a slave node v exists in the directed graph. i Can you eventually return to v after setting off? i The path, specifically the algorithm is as follows: In the formula, DFS is the depth-first search function; vi `i` represents the currently visited node; `visited` represents the set of visited nodes; `path` represents the set of nodes in the current path; `E` represents the set of edges in the directed graph; `(i, j) ∈ E` indicates that there exists a follower node `v`. i to node v j This method identifies various relational structures, clarifies the causal network structure among environmental elements, and provides a scientific basis for understanding the complex interactions in ecosystems.

[0160] The specific implementation of step S10 involves using a pre-trained model of soil element interactions to analyze the impact of growing season climate parameters, geographic coordinates, and altitude data on soil physicochemical properties. This model employs a multi-branch deep neural network architecture, including a one-dimensional convolutional neural network branch for processing time-series climate data, a two-dimensional convolutional neural network branch for processing spatial geographic data, and a multilayer perceptron branch for processing basic soil parameters. Each branch uses convolutional kernels of different sizes to capture multi-scale features. The climate data processing branch uses one-dimensional convolutional kernels of 1×3, 1×5, and 1×7 sizes, while the spatial data processing branch uses two-dimensional convolutional kernels of 3×3 and 5×5 sizes. The feature outputs of the three branches are fused through a self-attention mechanism, with the attention weights adaptively adjusted by three key parameters: soil pH, soil organic matter content, and total precipitation during the growing season. The model output layer is optimized using a mean squared error loss function to predict the content of soil elements such as total nitrogen, total phosphorus, total potassium, available phosphorus, available potassium, copper, manganese, zinc, and iron. The model was trained using stochastic gradient descent with a batch size of 64. The initial learning rate was set to 0.001, and a learning rate decay strategy was used to train for 200 epochs or until the loss on the validation set no longer decreased.

[0161] The specific implementation of step S11 involves using a mixed-effects model and variance decomposition method to determine the key ecological factors affecting the overall quality indicators of tobacco. First, a multicollinearity test is performed on all environmental factors, and the variance inflation factor is calculated. In the formula, VIF j Let be the variance inflation factor of the j-th environmental element; The determination coefficient is obtained by performing a regression analysis with the j-th environmental element as the dependent variable and all other environmental elements as independent variables. Variables with a variance inflation factor greater than 5 are removed. Then, a mixed-effects model is constructed with the comprehensive quality index of tobacco as the dependent variable and environmental elements as independent variables. The specific formula is as follows: In the formula, Q ij βi represents the comprehensive tobacco quality index of the j-th county and the i-th sampling point; β0 is the intercept term; βi k X is the fixed effect coefficient for the k-th environmental factor; ijkLet p be the value of the k-th environmental element at the i-th sampling point in the j-th county; p is the number of environmental elements; u j Let be the random effect in the j-th county, following a normal distribution. ε ij The residual term follows a normal distribution. The contribution rate of each environmental factor to the overall quality index of tobacco was calculated by variance decomposition: In the formula, C k β represents the contribution rate of the k-th environmental factor to the overall quality index of tobacco. k Let be the fixed effect coefficient of the k-th environmental element;

[0162] Var(X k Let ) represent the variance of the k-th environmental element; The variance of the random effects at the county level; This represents the variance of the residual term. Based on the variance contribution rate, environmental elements with a cumulative contribution rate exceeding 80% are selected as key ecological factors, providing a scientific basis for subsequent precise regulation of the tobacco planting environment.

[0163] The specific implementation of step S12 involves establishing a piecewise structural equation model to reveal the influence mechanism of environmental factors on the overall quality indicators of tobacco. First, based on the key ecological factors identified in step S11 and the correlation structure identified in step S09, an initial path diagram is constructed, including direct and indirect paths. The structural equation model consists of two parts: a measurement equation and a structural equation. The measurement equation is: X = Λ x ξ+δ,Y=Λ y η+ε; where X is the vector of observed environmental element variables; Y is the observed variable of comprehensive tobacco quality index; ξ is the vector of latent environmental element variables; η is the latent variable of comprehensive tobacco quality; Λ x For environmental factor loading matrix; Λ y Let η be the factor loading matrix for the overall quality of tobacco; δ be the measurement error vector for environmental factors; and ε be the measurement error for the overall quality of tobacco. Structural equation: η = Bη + Γξ + ζ; where B is the path coefficient matrix between endogenous latent variables; Γ is the path coefficient matrix between exogenous latent variables and endogenous latent variables; and ζ is the residual vector of the structural equation. Calculation of direct effects, indirect effects, and total effects: Direct effects matrix DE = Γ; Indirect effects matrix IE = (IB) -1 Γ-Γ; Total effects matrix TE=(Ib) -1 Γ; where I is the identity matrix; (IB) -1 This is the inverse matrix of the matrix (IB). These formulas calculate the direct, indirect, and total impacts of various environmental factors on the overall quality of tobacco, revealing the complete mechanism by which environmental factors influence the overall quality of tobacco.

[0164] The specific implementation of step S13 in this embodiment is the same as that described above, and will not be repeated here.

[0165] To better understand and implement this invention, a specific application scenario is provided below as Example 2: Researchers selected 41 counties (cities, districts) in 13 major tobacco-growing areas as the research region, setting up a total of 205 sampling points. The latitude range of the sampling points was 24°18'~44°22'N, and the longitude range was 103°48'~129°50'E, covering a wide range of climate zones from the North Temperate Zone to the Subtropical Zone. The sampling points were mainly distributed in Yunnan, Guizhou, Sichuan, Hunan, Hubei, Henan, Shandong, Anhui, Jiangsu, Zhejiang, Fujian, Guangdong, and Heilongjiang provinces. The basic information of each sampling point is shown in Table 1.

[0166] Table 1 Basic Information of Sampling Points in the Study Area

[0167]

[0168] Researchers conducted a systematic study on the method for determining the impact of climate and soil on the overall quality of tobacco according to the present invention. Precise geographical coordinates and elevation data were recorded at each sampling point using GPS, and topsoil samples (0–30 cm) and tobacco leaf samples from the middle section were collected. During sampling, a "Z"-shaped sampling layout was used for rectangular plots, and a diagonal or grid layout was used for approximately square plots. Ten to fifteen subsamples from each plot were combined into a single comprehensive soil sample of approximately 4 kg.

[0169] A comprehensive physicochemical analysis was conducted on the collected soil samples, measuring soil pH, organic matter content, total nitrogen content, total phosphorus content, total potassium content, available phosphorus content, available potassium content, copper content, manganese content, zinc content, and iron content. The statistical results of the soil physicochemical parameters collected from each study area are shown in Table 2.

[0170] Table 2 Statistical results of soil physicochemical properties in the study area

[0171]

[0172]

[0173] Researchers obtained growing season climate parameters for each sampling point using data from the China Meteorological Administration, small-scale weather stations in the experimental fields, and spatial interpolation methods. These parameters included average growing season temperature, average growing season humidity, total growing season precipitation, and sunshine hours per ten-day period during the growing season. Statistical results of climate parameters for each study area are shown in Table 3.

[0174] Table 3 Statistical results of climate parameters during the growing season in the study area

[0175] Climate parameters Minimum value Maximum value average value Standard deviation Average temperature (°C) 15.6 28.3 23.5 3.2 Average air humidity (%) 58.3 86.7 73.6 7.5 Total precipitation (mm) 362.5 1586.3 876.4 294.8 Sunshine hours per ten-day period (h) 36.2 82.5 57.8 12.3

[0176] Researchers evaluated the appearance characteristics of tobacco leaves according to the national standard GB 2635-1992, measured the chemical composition indicators of tobacco leaves, and organized an expert panel to evaluate the sensory characteristics of tobacco leaves. The overall quality index of tobacco was calculated using an entropy integration function; the specific calculation process is as follows:

[0177] First, the appearance characteristics, chemical composition, and sensory characteristics of tobacco leaves were standardized. For appearance indicators, the maximum value standardization method was used. Z ij Let X be the standardized value of the j-th indicator for the i-th sample. ij For the original value, min(X) j ) is the minimum value, max(X) j The maximum value is ). For chemical composition indicators, the membership degree is determined according to the membership function type and critical value in Table S1. For sensory characteristic indicators, positive indicators are processed using Formula 4(a), and negative indicators are processed using Formula 4(b).

[0178] Then calculate the information entropy of each indicator: in Let represent the weight of the j-th indicator in the i-th sample.

[0179] Next, calculate the weights of each indicator: Where m represents the total number of indicators.

[0180] Finally, calculate the overall quality index:

[0181] The dense correlation matrix of environmental elements was constructed using the Pearson correlation coefficient method. The calculation formula is as follows: The results are shown in Table 4.

[0182] Table 4. Dense Correlation Matrix of Environmental Elements (Partial Results)

[0183]

[0184]

[0185] The dense correlation matrix of environmental elements is transformed into a sparse correlation matrix of environmental elements using a threshold screening method (threshold set at ±0.5): Based on the sparse correlation matrix of environmental elements, researchers identified unidirectional, bidirectional, and cyclic correlation structures among environmental elements. Specific results are shown in Table 5.

[0186] Table 5. Main Relationships Among Environmental Elements

[0187]

[0188]

[0189] According to the appendix Figure 2 As shown, mixed-effects model analysis indicates that environmental variables have a significant impact on macro- and micro-elements in tobacco-growing soil. Researchers used mixed-effects modeling and variance decomposition to identify key ecological factors affecting the overall quality indicators of tobacco, calculated using the following formula: Q ij Let β0 be the comprehensive tobacco quality index of the j-th county and the i-th sampling point, and let β0 be the intercept term. k Let X be the fixed effect coefficient of the k-th environmental factor. ijk For the value of environmental element, u j For county-level random effects, ε ij This represents the residual term. The contribution rate of each factor is calculated through variance decomposition: The results of the mixed-effects model analysis are shown in Table 6.

[0190] Table 6. Results of the mixed-effects model analysis

[0191] Environmental factors Regression coefficient Standard error p-value Variance contribution rate (%) Total precipitation during the growing season 0.058 0.012 <0.001 41.49 Soil pH 5.236 1.253 <0.001 35.82 Soil total potassium content -0.843 0.267 0.005 6.82 available copper content in soil -4.358 1.582 0.012 5.08 available manganese content in soil 0.326 0.143 0.032 5.82 intercept 42.675 10.234 <0.001 -

[0192] According to the appendix Figure 3 As shown, variance decomposition revealed that precipitation explained 41.49% of the variation in overall tobacco quality, soil pH explained 35.82%, soil macroelements explained 11.90%, and soil microelements explained 10.90%. Overall tobacco quality improved with increasing precipitation and decreased with increasing total potassium and copper content.

[0193] Researchers constructed a piecewise structural equation model to reveal the direct and indirect influence mechanisms of environmental factors on the overall quality indicators of tobacco. The structural equation includes the measurement equation: X = Λ x ξ+δ,Y=Λ y η+ε and the structural equation: η=Bη+Γξ+ζ. The direct effects matrix DE=Γ and the indirect effects matrix IE=(IB) are calculated. -1 Γ-Γ and the total effects matrix TE=(IB) -1 Γ, analyzed the impact pathways of various environmental factors. According to Appendix Figure 4 and Figure 6 As shown in the figure, SEM analysis indicates that the overall quality of tobacco is positively influenced by factors such as precipitation, pH, and manganese, and negatively correlated with total potassium and copper. Precipitation indirectly affects the overall quality of tobacco by negatively impacting pH. The results of the structural equation model analysis are shown in Table 7.

[0194] Table 7. Structural Equation Modeling Analysis Results (Standardized Path Coefficients)

[0195]

[0196] Researchers divided all sample data into three groups based on soil pH: acidic soil (pH less than 6.5, 83 samples), neutral soil (pH between 6.5 and 7.4, 76 samples), and alkaline soil (pH greater than 7.4, 46 samples). (See attached...) Figure 5 As shown, the relationship between environmental variables and tobacco quality differs significantly under different pH conditions. In neutral soils, the positive effect of precipitation on overall tobacco quality is stronger than in acidic soils, while the negative effect of total potassium on overall tobacco quality is weaker in neutral soils than in acidic soils. The results of the mixed-effects model analysis under different pH groups are shown in Table 8.

[0197] Table 8. Results of the mixed effect model analysis under different pH groups

[0198]

[0199] Traditional tobacco quality evaluation methods are mainly based on single-factor or simple linear relationships, failing to comprehensively reflect the complex influence mechanisms of climate, space, and soil factors on the overall quality of tobacco. Traditional methods typically focus only on surface phenomena, lacking exploration of essential causal relationships, leading to inaccurate quality assessment results and failing to provide effective guidance for actual production. The method for determining the impact of climate and soil on the overall quality of tobacco in this invention has the following advantages: First, it constructs a Bayesian network model of climate and soil elements, quantifying the conditional probability distribution of growing season climate parameters on soil physicochemical properties, which more accurately describes nonlinear relationships compared to traditional linear models; second, it identifies unidirectional, bidirectional, and cyclical correlation structures through dense and sparse correlation matrices of environmental elements, revealing a complex interaction network among environmental elements; third, it accurately quantifies the contribution rate of each environmental factor to the overall quality of tobacco by applying mixed-effects models and variance decomposition methods; and fourth, it establishes a piecewise structural equation model to systematically analyze the direct and indirect influence paths of environmental elements on the overall quality of tobacco. Using this method, researchers discovered that precipitation and soil pH are the two most critical factors affecting the overall quality of tobacco, with the former contributing 41.49% and the latter 35.82%. The study also revealed the differentiated impact of environmental factors under different pH conditions, providing a scientific basis for precision tobacco planting management in different regions. This method has been validated in major tobacco-producing areas of China, significantly improving the accuracy of tobacco quality prediction and enhancing actual production efficiency.

[0200] It should be noted that the variables involved in this invention are explained in detail in Tables 9 and 10 below.

[0201] Table 9. Variable Explanation Table (Part 1)

[0202]

[0203]

[0204] Table 10 Variable Explanation Table (Part Two)

[0205]

[0206] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the impact of climate and soil on the overall quality of tobacco, characterized in that, include: This study aims to acquire records of tobacco leaf and soil samples, along with their geographic coordinates and altitude data. It also involves obtaining soil physicochemical parameters, climate parameters during the growing season, and tobacco leaf appearance and sensory characteristics. The study calculates the overall tobacco quality index using an entropy integration function. A Bayesian network model of climate and soil elements is constructed, establishing a cross-correlation probability matrix between climate parameters and soil physicochemical parameters during the growing season. A dense correlation matrix of environmental elements is constructed using the Pearson correlation coefficient method. This dense correlation matrix is ​​then converted into a sparse correlation matrix using a threshold screening method. The study identifies the correlation structure between environmental elements based on the sparse correlation matrix. A pre-trained model of tobacco-soil element interactions is used to analyze the impact of climate parameters and geographic data on soil parameters. A mixed-effects model and variance decomposition method are used to identify key ecological factors affecting the overall tobacco quality index. A piecewise structural equation model is established to reveal the mechanism by which ecological parameters influence the overall tobacco quality index. Finally, data are grouped according to soil pH value to analyze the differences in the impact of ecological factors on the overall tobacco quality index under different conditions.

2. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 1, characterized in that, The entropy integration function is used to integrate tobacco leaf appearance characteristics, tobacco leaf chemical composition, and tobacco leaf sensory characteristics into a comprehensive tobacco quality index. The inputs include the tobacco leaf appearance characteristics index scoring matrix, the tobacco leaf chemical composition index measurement result matrix, and the tobacco leaf sensory characteristics index scoring matrix, as well as the standardized limit parameters of each index. The output is the comprehensive tobacco quality index.

3. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 2, characterized in that, The Bayesian network model of climate and soil elements refers to a network model based on probabilistic graph theory to establish conditional probabilistic relationships between environmental variables. It uses a directed acyclic graph to represent the dependencies between variables, with nodes representing random variables and edges representing conditional dependencies, and is used to infer the probability distribution of the influence of climate parameters on soil parameters.

4. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 3, characterized in that, The cross-correlation probability matrix refers to the conditional probability distribution matrix of soil physicochemical property parameters on the average temperature, average air humidity, total precipitation, and sunshine hours per ten-day period during the growing season.

5. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 4, characterized in that, The dense correlation matrix of environmental elements refers to the correlation matrix among various elements such as climate parameters, geographical coordinates, altitude data, and soil physicochemical properties during the growing season. Each element in the matrix represents the correlation strength between the corresponding environmental elements in the row and column, forming a fully connected network structure.

6. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 5, characterized in that, The sparse correlation matrix of environmental elements refers to a simplified matrix obtained by filtering the dense correlation matrix of environmental elements by setting a correlation degree threshold. Only connections with a correlation degree higher than the threshold are retained, weak correlations are excluded, and the interpretability of the model is improved.

7. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 6, characterized in that, The interrelationship structure among environmental elements includes: a one-way interrelationship structure, which refers to an asymmetric influence relationship between environmental elements, where a change in one element leads to a change in another element, but the reverse influence is not significant; a two-way interrelationship structure, which refers to a mutual influence relationship between environmental elements, where the two elements are mutually causal and form a feedback mechanism; and a cyclical interrelationship structure, which refers to multiple environmental elements forming a closed-loop influence chain, where the first element influences the second element, the second element influences the third element, and finally returns to the first element through a series of intermediate elements to form a cycle.

8. The method for determining the impact of climate and soil on the overall quality of tobacco according to claim 7, characterized in that, The pre-trained model for element interaction in tobacco soil adopts a multi-branch deep neural network architecture, which includes a climate data processing branch, a spatial data processing branch, and a soil basic parameter processing branch. Each branch uses different convolution kernel sizes to capture features at different scales. The features of the three branches are weighted and fused through an attention mechanism. The output layer predicts the content of macro- and micro-elements in the soil. The parameters of the pre-trained model for element interaction in tobacco soil are adaptively adjusted through three key parameters: soil pH, soil organic matter content, and total precipitation during the growing season.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform the method for determining the impact of climate and soil on the overall quality of tobacco as described in any one of claims 1-8.

10. A system for determining the impact of climate and soil on the overall quality of tobacco, characterized in that, The system includes the computer-readable storage medium of claim 9, wherein the system is any one of a computer, a server, or a microcontroller, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

Images