A method for analyzing tobacco moisture migration characteristics based on LF-NMR technology

Abstract

The application provides a method for analyzing tobacco moisture migration characteristics based on LF-NMR technology, and belongs to the technical field of electric digital data processing.The method comprises the following steps: standardizing sample pretreatment, determining moisture state by low-field nuclear magnetic resonance, establishing a moisture migration dynamic model, calculating a moisture diffusion coefficient matrix, constructing a moisture loss prediction model, adjusting migration parameters by using a moisture diffusion optimization function, analyzing whole-leaf moisture distribution by using a neural network model of a multi-scale attention mechanism, performing different temperature gradient experiments, analyzing moisture migration difference between main veins and leaf tissue, and verifying prediction model accuracy, so that precise quantitative analysis of tobacco moisture migration dynamic characteristics in different microstructures is realized.The method combines low-field nuclear magnetic resonance technology with moisture diffusion theory and deep learning, accurately identifies moisture migration characteristics in tobacco microstructure, and provides a scientific basis for tobacco curing process optimization.

Description

Technical Field

[0001] This invention belongs to the field of electronic digital data processing technology, and specifically relates to a method for analyzing the moisture migration characteristics of tobacco leaves based on LF-NMR technology. Background Technology

[0002] During tobacco processing, the dynamic characteristics of moisture migration have a significant impact on tobacco quality and flavor formation. Traditional methods for determining tobacco moisture content mainly include drying loss, Karl Fischer method, and electrical resistance method. While these methods can determine the overall moisture content of tobacco leaves, they cannot distinguish the distribution and migration characteristics of moisture in different microstructures of the tobacco leaf (such as cell lumens, cell walls, and intercellular spaces). Spectroscopic analysis techniques such as near-infrared spectroscopy and thermogravimetric analysis can partially identify moisture status, but in practical applications, they are difficult to dynamically monitor the migration mechanisms of moisture within the microstructure of tobacco leaves.

[0003] In recent years, various advanced detection technologies have been applied to the study of moisture content in agricultural products. However, most of these methods are limited to static analysis and cannot capture the dynamic changes in moisture migration during tobacco curing. Existing moisture migration models are mainly based on simplified Fick diffusion theory, which ignores the influence of the heterogeneity and complex structure of tobacco leaf tissue on moisture migration. This results in low model prediction accuracy and makes it difficult to accurately describe the differences in moisture migration in different parts of the tobacco leaf.

[0004] The main challenge in tobacco leaf moisture migration research is the lack of a comprehensive method that can simultaneously identify microscopic moisture states and monitor dynamic migration processes. This makes it impossible to accurately quantify and analyze the dynamic migration characteristics of tobacco leaf moisture in different microstructures, severely limiting the precise control of tobacco curing processes and the stable improvement of product quality. In other words, existing technologies suffer from the technical problem of being unable to accurately quantify and analyze the dynamic migration characteristics of tobacco leaf moisture in different microstructures. Summary of the Invention

[0005] In view of this, the present invention provides a method for analyzing the moisture migration characteristics of tobacco leaves based on LF-NMR technology, which can solve the technical problem in the prior art that it is impossible to accurately quantify and analyze the dynamic migration characteristics of moisture in different microstructures of tobacco leaves.

[0006] This invention is implemented as follows: A method for analyzing the water migration characteristics of tobacco leaves based on LF-NMR technology includes: collecting tobacco leaf samples and pre-treating them, then placing the tobacco leaves in a constant temperature and humidity environment for equilibration; measuring the moisture content of the tobacco leaves using a low-field nuclear magnetic resonance instrument to obtain measurement data; establishing a dynamic model of tobacco leaf water migration, obtaining moisture content change curves by measuring T2 relaxation time spectra at different time points; calculating the tobacco leaf water diffusion coefficient matrix based on the T2 relaxation time distribution to identify the migration rate of water in the cell lumen, cell wall, and intercellular spaces; constructing a tobacco leaf water loss prediction model, inputting environmental parameters and tobacco leaf parameters; applying a water diffusion optimization function to adjust the water migration parameters, outputting the optimal water migration path map and the water diffusion rate constant matrix; using a tobacco leaf water migration neural network model to analyze the dynamic distribution of moisture in the whole leaf; conducting experiments with different baking temperature gradients to establish the relationship curve between environmental temperature and water migration rate; and analyzing the differences in water migration between the midrib and mesophyll tissue of the tobacco leaf.

[0007] The specific steps for collecting and pre-processing tobacco leaf samples are as follows: using a standard cutting tool, the tobacco leaves are cut into 5×5 cm pieces and placed in a constant temperature and humidity environment for 24 hours to equilibrate.

[0008] The specific steps for determining the moisture content of tobacco leaves using a low-field nuclear magnetic resonance instrument are as follows: set the probe frequency to 20MHz, collect data 16 times, and set the echo time to 0.2 milliseconds.

[0009] The specific steps for establishing a dynamic model of tobacco leaf moisture migration are as follows: by measuring the T2 relaxation time spectrum at different time points, the curves of free water content change and bound water content change are obtained.

[0010] The free water content change curve refers to the quantitative curve of the change of water with high mobility existing in the intercellular spaces of tobacco leaves over time; the bound water content change curve refers to the quantitative curve of the change of water with low mobility that is tightly bound to macromolecules in tobacco leaf tissue over time.

[0011] Among them, tobacco leaf structural parameters refer to quantitative indicators that characterize the microstructure of tobacco leaves, including porosity, specific surface area, and fiber orientation; ambient temperature refers to the temperature value of the environment in which the tobacco leaves are located, in degrees Celsius; ambient relative humidity refers to the percentage of air humidity in the environment in which the tobacco leaves are located compared to the saturated humidity at the same temperature; and initial moisture content of tobacco leaves refers to the percentage of the mass of water in the tobacco leaves measured before the start of the experiment relative to the total mass of the tobacco leaves.

[0012] The moisture diffusion optimization function is used to calculate the optimal path and rate constant of moisture migration within tobacco leaves based on multiple parameter inputs. By solving the partial differential equations of Fick's second law and combining them with the tissue structure characteristics of tobacco leaves, it achieves a precise mathematical description and prediction of the moisture migration process in tobacco leaves. The input parameters include ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content of tobacco leaves, and tobacco leaf tissue density index. The output is an optimal moisture migration path map and a moisture diffusion rate constant matrix.

[0013] Among them, the ambient temperature gradient refers to the rate of change of the ambient temperature around the tobacco leaf in space, with the unit being degrees Celsius per meter; the tobacco leaf thickness refers to the measured thickness of the tobacco leaf sample, with the unit being millimeters; the tobacco leaf tissue density index is a dimensionless index characterizing the density of the tobacco leaf tissue, calculated by the ratio of the tobacco leaf density to the standard density; the optimal moisture migration path diagram is a two-dimensional representation of the best path for moisture movement within the tobacco leaf under given conditions; and the moisture diffusion rate constant matrix is ​​a parameter matrix describing the rate of moisture migration between different regions.

[0014] The specific structure of the tobacco leaf moisture migration neural network model is a deep learning architecture based on a combination of graph convolutional networks and spatiotemporal attention mechanisms. It includes five graph convolutional layers for extracting spatial connectivity features of tobacco leaf tissue, three temporal convolutional layers for capturing temporal variation patterns of moisture migration, and two multi-head self-attention modules for learning the mutual influence relationships of moisture migration between different tissue regions. The input of the tobacco leaf moisture migration neural network model is the tobacco leaf tissue structure diagram and the temporal T2 relaxation time spectrum data measured by low-field nuclear magnetic resonance technology. The output is the prediction result of the whole leaf moisture content distribution and the prediction of the direction of moisture migration at future time points.

[0015] The specific steps for analyzing the differences in water migration between the midrib and mesophyll tissue of tobacco leaves are as follows: calculate the ratio of water migration rate in the midrib to water migration rate in the mesophyll tissue, and construct a tissue structure influencing factor. The water migration rate refers to the speed at which water moves in the tobacco leaf tissue, measured in millimeters per hour. The tissue structure influencing factor is a dimensionless coefficient that characterizes the degree of influence of different tissue structures of tobacco leaves on water migration.

[0016] Compared with existing technologies, this invention provides a method for analyzing the water migration characteristics of tobacco leaves based on LF-NMR technology. This invention proposes a method for analyzing the water migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology. By measuring the T2 relaxation time spectrum, it achieves precise differentiation between free water and bound water in tobacco leaves and quantitative analysis of their distribution in different microstructures. This method can identify the migration rate of water in cell cavities, cell walls, and intercellular spaces, and construct a mathematical model that accurately reflects the dynamic changes of water in the microstructure of tobacco leaves.

[0017] By constructing an optimized water diffusion function and a neural network model for tobacco leaf water migration, this invention achieves precise quantitative analysis of water migration characteristics among different microstructures of tobacco leaves. In particular, the use of a multi-scale attention mechanism effectively captures the correlation of water migration between different tissues of tobacco leaves, revealing the influence mechanism of tobacco leaf microstructure characteristics on water migration. This method can analyze the differences in water migration between the midrib and mesophyll tissues of tobacco leaves, calculate the ratio of water migration rates between different tissues, and provide a new perspective for understanding the internal water migration characteristics of tobacco leaves.

[0018] This invention solves the technical problem of the inability to accurately quantify and analyze the dynamic characteristics of moisture migration in different microstructures of tobacco leaves, realizing the visualization and quantitative analysis of the moisture migration process in tobacco leaves, and providing a scientific basis for optimizing tobacco curing processes. This method combines low-field nuclear magnetic resonance technology with moisture migration theory and deep learning, significantly improving the accuracy and comprehensiveness of the analysis of tobacco leaf moisture migration characteristics. Attached Figure Description

[0019] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0021] like Figure 1 The diagram shown is a flowchart of a method for analyzing the moisture migration characteristics of tobacco leaves based on LF-NMR technology provided by the present invention. This method includes the following steps:

[0022] S01. Collect tobacco leaf samples and pre-treat them. Use a standard cutting tool to cut the tobacco leaves into 5×5 cm specifications and place them in a constant temperature and humidity environment for 24 hours to equilibrate.

[0023] S02. The moisture content of tobacco leaves was determined using a low-field nuclear magnetic resonance instrument. The probe frequency was set to 20MHz, the number of acquisitions was 16, and the echo time was 0.2 milliseconds.

[0024] S03. Establish a dynamic model of tobacco leaf moisture migration. By measuring the T2 relaxation time spectrum at different time points, obtain the free water content change curve and the bound water content change curve.

[0025] S04. Calculate the tobacco leaf moisture diffusion coefficient matrix based on the T2 relaxation time distribution to identify the migration rate of water in the cell lumen, cell wall and intercellular space.

[0026] S05. Construct a tobacco leaf moisture loss prediction model. Input parameters include ambient temperature, ambient relative humidity, tobacco leaf structural parameters, and initial moisture content of tobacco leaves.

[0027] S06. Apply the moisture diffusion optimization function to adjust the moisture migration parameters. The input parameters include ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content of tobacco leaves and tobacco leaf tissue density index. The output is the optimal moisture migration path map and the moisture diffusion rate constant matrix.

[0028] S07. Analyze the dynamic distribution of water in the whole leaf using a pre-trained tobacco leaf water migration neural network model. The tobacco leaf water migration neural network model uses a multi-scale attention mechanism to capture the correlation of water migration between different tissues of the tobacco leaf.

[0029] S08. Conduct experiments with different baking temperature gradients, setting the temperature range from 45℃ to 70℃, and establish the relationship curve between ambient temperature and moisture migration rate.

[0030] S09. Analyze the differences in water migration between the midrib and mesophyll tissue of tobacco leaves, calculate the ratio of water migration rate in the midrib to water migration rate in the mesophyll tissue, and construct tissue structure influencing factors.

[0031] S10. Optional, verify the accuracy of the tobacco leaf moisture loss prediction model by comparing the predicted value with the moisture content measured during the actual curing process, calculating the relative error, and optimizing the parameters of the tobacco leaf moisture loss prediction model.

[0032] Low-field nuclear magnetic resonance (LF-NMR) is a non-destructive testing technique that analyzes the state and distribution of water in substances based on the spin relaxation characteristics of hydrogen nuclei. It distinguishes water in different binding states by measuring the relaxation characteristics of hydrogen nuclei in a magnetic field. Its working principle is to use a low-intensity magnetic field (usually less than 1 Tesla) to excite the spin energy level transition of hydrogen nuclei, and then obtain the microenvironment information of water molecules by detecting the relaxation signal. It has the advantages of simple operation, non-destructive nature, and high sensitivity in detecting water state, and is suitable for studying the water state and dynamic changes in complex biological tissues.

[0033] Among them, the T2 relaxation time spectrum refers to the time distribution of hydrogen nuclei losing phase coherence in a transverse magnetic field, reflecting the binding state and migration ability of water molecules in different microenvironments.

[0034] The moisture diffusion coefficient matrix is ​​a mathematical expression describing the migration rate of moisture in different microstructures of tobacco leaves, and includes quantitative parameters of moisture diffusion in different directions and tissues.

[0035] The free water content change curve refers to the quantitative curve of the change in the highly mobile water content existing in the intercellular spaces of tobacco leaves over time.

[0036] Among them, the bound water content change curve refers to the quantitative curve of the change of water content with low mobility that is tightly bound to macromolecules in tobacco leaf tissue over time.

[0037] Among them, tobacco leaf structural parameters refer to quantitative indicators that characterize the microstructural properties of tobacco leaves, including porosity, specific surface area, and fiber alignment.

[0038] Ambient temperature refers to the temperature of the environment in which the tobacco leaves are located, measured in degrees Celsius.

[0039] Among them, relative humidity refers to the percentage of air humidity in the environment where tobacco leaves are located compared to the saturation humidity at the same temperature.

[0040] The initial moisture content of tobacco leaves refers to the percentage of the moisture mass of the tobacco leaves measured before the start of the experiment relative to the total mass of the tobacco leaves.

[0041] Among them, the ambient temperature gradient refers to the rate of change of the ambient temperature around the tobacco leaves in space, and the unit is degrees Celsius per meter.

[0042] The thickness of the tobacco leaf refers to the measured thickness of the tobacco leaf sample, measured in millimeters.

[0043] Among them, the tobacco leaf tissue density index is a dimensionless index that characterizes the density of tobacco leaf tissue, and is calculated by the ratio of tobacco leaf density to standard density.

[0044] Among them, the optimal path diagram for moisture migration is a two-dimensional representation of the best path for moisture movement inside tobacco leaves under given conditions.

[0045] The moisture diffusion rate constant matrix refers to the parameter matrix that describes the rate of moisture migration between different regions.

[0046] Among them, the water migration rate refers to the speed at which water moves in tobacco leaf tissue, and the unit is millimeters per hour.

[0047] Among them, the tissue structure influence factor refers to the dimensionless coefficient that characterizes the degree of influence of different tissue structures of tobacco leaves on water migration.

[0048] The moisture diffusion optimization function is used to calculate the optimal path and rate constant of moisture migration within tobacco leaves based on multiple parameter inputs. The input parameters include five key parameters: ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content of tobacco leaves, and tobacco leaf tissue density index. The output is an optimal path diagram of moisture migration and a moisture diffusion rate constant matrix. By solving the partial differential equations of Fick's second law and combining them with the tissue structure characteristics of tobacco leaves, the moisture diffusion optimization function achieves a precise mathematical description and prediction of the moisture migration process in tobacco leaves.

[0049] The specific structure of the tobacco leaf moisture migration neural network model is a deep learning architecture based on a combination of graph convolutional networks and spatiotemporal attention mechanisms. It includes five graph convolutional layers for extracting spatial connectivity features of tobacco leaf tissue, three temporal convolutional layers for capturing temporal variation patterns of moisture migration, and two multi-head self-attention modules for learning the mutual influence relationships of moisture migration between different tissue regions. The input of the tobacco leaf moisture migration neural network model is the tobacco leaf tissue structure diagram and the temporal T2 relaxation time spectrum data measured by low-field nuclear magnetic resonance technology. The output is the prediction result of the whole leaf moisture content distribution and the prediction of the moisture migration direction at future time points. The multi-scale attention mechanism parameters in the tobacco leaf moisture migration neural network model are adaptively adjusted based on three key parameters: the ratio of the combined water content change curve and the free water content change curve, the tobacco leaf tissue density index, and the peak position of the T2 relaxation time spectrum, thereby achieving accurate capture and prediction of the moisture migration characteristics of different types of tobacco leaves.

[0050] The steps for establishing the training dataset for the tobacco leaf moisture migration neural network model specifically include: collecting low-field nuclear magnetic resonance experimental data of different tobacco varieties under different temperature and humidity conditions, including complete T2 relaxation time spectrum temporal changes, tobacco leaf microstructure image data, and corresponding actual moisture content measurements; classifying and labeling the collected experimental data according to factors such as variety, ambient temperature, and ambient relative humidity, and performing standardized preprocessing to eliminate differences in data dimensions under different experimental conditions; constructing a graph structure to represent the tissue connections of tobacco leaves, representing each sampling point as a node in the graph, the connection between adjacent tissues as an edge, and assigning each node a corresponding T2 relaxation time spectrum feature vector and moisture content label; and dividing the processed dataset into a training set, a validation set, and a test set in an 8:1:1 ratio for training and evaluation of the tobacco leaf moisture migration neural network model.

[0051] The training steps of the tobacco leaf moisture migration neural network model specifically include: first, unsupervised training on a large-scale tobacco leaf moisture dataset, using a masked autoencoder to learn the T2 relaxation time spectrum and the hidden layer representation of tobacco leaf tissue structure; then, using supervised learning with measured moisture content distribution as labels, training the tobacco leaf moisture migration neural network model to predict the dynamic process of tobacco leaf moisture migration under different conditions; introducing a spatiotemporal consistency regularization term during training to enable the tobacco leaf moisture migration neural network model to learn a moisture diffusion pattern that conforms to physical laws; introducing a domain adaptation training strategy to enable the tobacco leaf moisture migration neural network model to adapt to the moisture migration characteristics of different tobacco varieties and under different experimental conditions; finally, by integrating checkpoints of the tobacco leaf moisture migration neural network model trained at different stages, the final model is formed. The tobacco leaf moisture migration neural network model achieves a moisture prediction accuracy of less than 3% with an average relative error on the test set, providing a high-precision and high-efficiency computational tool for subsequent analysis of tobacco leaf moisture migration characteristics.

[0052] The specific implementation methods of the above steps are described in detail below.

[0053] The specific implementation of step S01 involves strictly standardized tobacco leaf sample collection and pretreatment. First, representative samples are selected from the tobacco leaf batches using stratified random sampling to ensure statistical representativeness. Then, the tobacco leaves are cut into 5×5 cm pieces using a precision cutting tool. During cutting, the tool is kept sharp to ensure clean cuts and avoid damage to the tobacco leaf structure that could affect subsequent experimental data. The cut samples are then placed in a constant temperature and humidity chamber at 25±1℃ and 60±2% relative humidity for 24 hours to equilibrate. The sample spacing is no less than 2 cm to ensure uniform airflow contact with each sample, allowing the tobacco leaf moisture to reach a stable state. The purpose of this step is to eliminate interference from factors such as sample size and initial moisture state through a standardized sample preparation process, laying the foundation for accurate analysis of tobacco leaf moisture migration characteristics.

[0054] The specific implementation of step S02 involves using a low-field nuclear magnetic resonance (NMR) instrument to accurately determine the moisture content of tobacco leaves. First, the pretreated tobacco leaf sample is placed in the NMR test tube, ensuring the sample is within the effective detection area of ​​the coil. The operating parameters of the NMR instrument are set, including a probe frequency of 20MHz, 16 acquisition repetitions to improve the signal-to-noise ratio, and an echo time of 0.2 milliseconds to capture rapidly decaying signal components. A multi-echo pulsed-magnetic sequence (CPMG) is used to collect transverse relaxation time signals, with an echo number of 10,000, an echo interval of 0.5 milliseconds, and an acquisition window width of 5,000 milliseconds to ensure complete acquisition of relaxation signals of moisture in different binding states within the tobacco leaf. Simultaneously, the receiving gain is set to 15dB, and the spectral width to 100kHz. Each sample is measured three times, and the average value is taken to reduce random errors. This step, based on the principle of NMR detection, non-destructively and quantitatively obtains information on the moisture state and distribution in tobacco leaves by measuring the relaxation characteristics of hydrogen nuclei in a magnetic field, providing fundamental data for subsequent analysis of moisture migration characteristics.

[0055] The specific implementation of step S03 involves establishing a dynamic model of tobacco leaf moisture migration. First, the obtained CPMG decay curve is processed using the inverse Laplace transform algorithm to convert the time-domain signal into a T2 relaxation time spectrum. The Tikhonov regularization method is used to control the smoothing coefficient between 0.01 and 0.1 to balance resolution and stability. Then, peak separation and identification are performed on the converted T2 relaxation time spectrum. A multi-Gaussian curve fitting method is used to identify different moisture components: peaks with a T2 value less than 10 milliseconds are identified as bound water, peaks between 10 and 100 milliseconds are identified as capillary water, and peaks greater than 100 milliseconds are identified as free water. Next, under controlled conditions, the T2 relaxation time spectrum is measured every 30 minutes for 8 consecutive hours to obtain complete dynamic moisture change data. Finally, the changes in the peak areas in the spectrum are calculated by integration, and curves showing the changes in free water content and bound water content are plotted. A three-parameter exponential decay model is used to fit the curves, with a goodness-of-fit R0. 2 Not less than 0.95. This step, through analysis of the time-series T2 relaxation time spectrum, revealed the pattern of moisture changes in different states of tobacco leaves over time, providing a dynamic perspective for understanding the moisture migration mechanism.

[0056] The specific implementation of step S04 involves calculating the moisture diffusion coefficient matrix of tobacco leaves. First, a moisture diffusion equation in the tobacco leaf tissue is constructed based on Fick's second law, treating the tobacco leaf as a porous medium and establishing a three-dimensional diffusion model. Then, the T2 relaxation time distribution is correlated with the spatial location of moisture. By measuring the changes in the T2 spectrum at different sample locations, a gradient field for moisture diffusion is constructed. Next, the partial differential equation for moisture diffusion is solved using the finite element method, with a mesh generation accuracy of at least 100 μm, a time step of 1 minute, and an iteration convergence threshold of 10. -6Based on the solution results, the diffusion coefficients of three main tobacco leaf tissue regions were calculated: intracellular lumen diffusion coefficient (D1), cell wall diffusion coefficient (D2), and intercellular space diffusion coefficient (D3), forming a 3×3 water diffusion coefficient matrix. The least squares method was used to fit the experimental data with the model predictions to determine the optimal diffusion coefficients, with the fitting error controlled within 5%. This step, by combining mathematical models with experimental data, quantitatively characterized the migration capacity of water in different microstructures of tobacco leaves, providing key parameters for understanding the mechanism of water migration in tobacco leaves.

[0057] The specific implementation of step S05 involves constructing a tobacco leaf moisture loss prediction model. First, a multi-factor regression model is established, with input parameters including ambient temperature (30–68℃), relative humidity (18%–90%), and tobacco leaf structural parameters (porosity 0.1–0.5, specific surface area 1–5 m²). 2 The parameters included the fiber orientation (0°–90°) and the initial moisture content of the tobacco leaves (84%–87%). Then, partial least squares regression was used to address potential collinearity among the multiple parameters, extracting principal component factors and retaining those with a cumulative variance contribution rate of 95%. Next, a temperature correction coefficient was introduced, and the relationship between temperature and diffusion rate was established based on the Arrhenius equation, with activation energy ranging from 15 to 25 kJ / mol. Simultaneously, a humidity correction coefficient was introduced, and the relationship between ambient relative humidity and equilibrium moisture content was established based on isothermal hygroscopic curves. Finally, a comprehensive prediction model was constructed, using the gradient boosting decision tree (GBDT) algorithm to integrate the influence of multiple parameters. The tree depth was set to 5, the learning rate to 0.05, and the number of weak learners to 100. Cross-validation was performed using the 5-fold cross-validation method to evaluate model stability. This step, by comprehensively considering environmental factors and the characteristics of the tobacco leaves themselves, established a mathematical model capable of predicting the moisture loss process of tobacco leaves under different conditions, providing a theoretical basis for moisture control in tobacco processing.

[0058] The specific implementation of step S06 involves adjusting the moisture migration parameters using a moisture diffusion optimization function. First, a system of partial differential equations based on Fick's second law is established to describe the moisture migration process in tobacco leaves, including tensor expressions considering anisotropic diffusion characteristics. Then, the system of partial differential equations is discretized and solved using the finite difference method, with a spatial step size of 0.1 mm and a time step size of 1 second, constructing a numerical computation framework. Next, key parameters are input: ambient temperature gradient (0.5–5℃ / cm), ambient relative humidity (18%–90%), tobacco leaf thickness (0.1–0.5 mm), initial moisture content of tobacco leaves (84%–87%), and tobacco leaf tissue density index (0.8–1.2). A constrained optimization problem is constructed using the Lagrange multiplier method, with the objective function being the minimization of moisture migration time, and the constraints being the uniformity of moisture loss and mass conservation. Finally, the conjugate gradient method is used to solve the optimization problem, with no more than 1000 iterations and a convergence threshold of 10. -5The process outputs the optimal path map for moisture migration and the moisture diffusion rate constant matrix. This step, through the application of advanced mathematical optimization methods, determines the optimal path and rate parameters for moisture migration in tobacco leaf tissue under given conditions, providing quantitative guidance for optimizing tobacco processing technology.

[0059] The specific implementation of step S07 involves using a tobacco leaf moisture migration neural network model to analyze the dynamic distribution of moisture in the whole leaf. This neural network model employs a deep learning architecture combining graph convolutional networks and a spatiotemporal attention mechanism. The model construction first determines the network structure, which includes five graph convolutional layers, three temporal convolutional layers, and two multi-head self-attention modules. The graph convolutional layers use a ChebNet structure with a polynomial order of 3 and hidden layer dimensions of 64, 128, 256, 128, and 64, used to extract spatial connectivity features of tobacco leaf tissue. The temporal convolutional layers use a causal convolutional structure with a kernel size of 3 and dilation coefficients of 1, 2, and 4, used to capture temporal variation patterns of moisture migration. The multi-head self-attention modules have 8 heads and an attention dimension of 32, used to learn the mutual influence relationships of moisture migration between different tissue regions. The network input consists of a tobacco leaf tissue structure diagram and temporal T2 relaxation time spectrum data measured by low-field nuclear magnetic resonance (NMR) technology; the output is the predicted distribution of moisture content in the whole leaf and the predicted direction of moisture migration at future time points. The parameters of the multi-scale attention mechanism in the model are adaptively adjusted through three key indicators: the ratio of the combined water content change curve to the free water content change curve (range 0.1–10), the tobacco leaf tissue density index (range 0.8–1.2), and the peak position of the T2 relaxation time spectrum (range 1–1000 ms). This step, through deep learning technology, achieves high-precision prediction of the dynamic distribution of water in the whole tobacco leaf, providing an advanced analytical tool for understanding the water migration patterns of tobacco leaves under complex conditions.

[0060] The specific implementation of step S08 involves conducting experiments at different baking temperature gradients. First, a temperature gradient experiment scheme was designed, selecting four temperature points: 30℃, 44℃, 56℃, and 68℃, with relative humidity maintained at 60±2%. Then, a precision temperature-controlled oven was used, with temperature fluctuations controlled within ±0.5℃, and 10 pre-treated tobacco leaf samples were placed under each temperature condition. Next, samples were removed at preset time points (0, 2, 4, 6, 8, 12, and 24 hours), and the T2 relaxation time spectrum was measured using low-field nuclear magnetic resonance (NMR). Simultaneously, the absolute moisture content was determined using the drying method as a calibration reference. The collected data were processed, and the moisture migration rate at each temperature point was calculated. A moisture loss curve was fitted using a first-order kinetic equation to obtain the moisture migration rate constant. Finally, the relationship curve between ambient temperature and moisture migration rate was established using the Arrhenius equation to determine the temperature sensitivity coefficient, which is typically between 1.5 and 2.5. This step, through systematic temperature gradient experiments, revealed the influence of temperature on the moisture migration process of tobacco leaves, providing a scientific basis for optimizing baking process parameters.

[0061] The specific implementation of step S09 involves analyzing the differences in water migration between the midrib and mesophyll tissues of tobacco leaves. First, the midrib and mesophyll tissues are separated from the pretreated tobacco leaf samples, maintaining similar sample quality. Then, under the same environmental conditions (temperature 48℃, relative humidity 50%), the water migration process of both tissues is measured. T2 relaxation time spectra are acquired using low-field nuclear magnetic resonance (NMR) technology at 30-minute intervals, with continuous monitoring for 6 hours. Next, the water migration rates of the midrib and mesophyll tissues are calculated based on the T2 spectra. A first-order exponential decay model is used to fit the water loss curves, and the rate constant is extracted. The ratio between the two is calculated, and a tissue structure influence factor F is constructed. s This factor typically ranges from 1.2 to 2.5. Simultaneously, the microstructural characteristics of the two tissues were analyzed, including cell density, intercellular space size, and fiber content, to establish a correlation model between microstructure and water migration rate. Finally, partial correlation analysis was used to identify the key structural factors influencing differences in water migration, with a correlation coefficient threshold set at 0.7. This step, by comparing and analyzing the water migration characteristics of different tissue parts of tobacco leaves, revealed the mechanism by which tissue structure affects water migration, providing a theoretical basis for the precise control of water distribution in tobacco leaves.

[0062] Step S10 specifically verifies the accuracy of the tobacco leaf moisture loss prediction model. First, a verification experimental scheme is designed, selecting different varieties of tobacco leaves, different baking temperatures (30–68℃), and different relative humidity (18%–90%) combinations to construct a verification dataset. Then, during the simulated baking process in the laboratory, samples are collected at preset time points (0, 2, 4, 8, 16, and 24 hours), and the moisture content is simultaneously measured using low-field nuclear magnetic resonance (NMR) technology and drying methods. Next, the measured data are compared with the output values ​​of the prediction model, and the relative error, root mean square error, and coefficient of determination are calculated. Data points with relative errors exceeding 10% are analyzed in depth to identify possible factors causing the errors. Finally, based on the verification results, the model parameters are optimized. A Bayesian optimization algorithm is used to adjust the model hyperparameters, including the temperature correction coefficient, humidity correction coefficient, and structural influence factor weights, undergoing 50 iterations to minimize the prediction error on the verification set. The optimized model should have an average relative error of less than 5% and a maximum error of no more than 10% on the verification set. This step, through systematic verification and optimization, improves the accuracy and reliability of the prediction model, ensuring that the model can provide effective guidance for practical applications.

[0063] Furthermore, the detailed structure of the tobacco leaf water migration neural network model adopts a deep learning architecture combining graph convolutional networks and spatiotemporal attention mechanisms. First, a graph structure representation of tobacco leaf tissue is defined, dividing the tobacco leaf into N nodes, each node representing a sampling point. Edges represent the connections between adjacent tissues, and edge weights are determined based on physical distance and tissue similarity. Then, a five-layer graph convolutional layer is constructed, using Chebyshev multinomial approximation to implement spectral graph convolution. The first layer's input dimension is the node feature dimension (including the feature vector after T2 spectrum discretization, with a dimension of 64), and its output dimension is 64. The second layer has an input of 64 and an output of 128; the third layer has an input of 128 and an output of 256; the fourth layer has an input of 256 and an output of 128; and the fifth layer has an input of 128 and an output of 64. Each graph convolutional layer is followed by a batch normalization layer and a ReLU activation function, and residual connections are used to prevent gradient vanishing. Next, a three-layer temporal convolutional network is constructed, employing a one-dimensional causal convolutional structure with a kernel size of 3 and dilation coefficients of 1, 2, and 4, effectively covering 16 time steps to capture the temporal dynamic features of moisture migration. Then, a two-layer multi-head self-attention module is implemented, with 8 heads and an attention dimension of 32. The first layer performs spatial attention calculation, and the second layer performs temporal attention calculation, capturing long-range dependencies between different nodes through an attention score matrix. Finally, an output layer is designed, containing two branches: one for predicting the distribution of whole-leaf moisture content (node-level regression task), and the other for predicting the direction of moisture migration at future time points (edge-level classification task). The model adopts a multi-task learning framework, and the loss function includes a mean squared error term for moisture content prediction, a cross-entropy term for migration direction prediction, and a regularization term. This model innovatively combines graph neural networks with a spatiotemporal attention mechanism for tobacco leaf moisture migration analysis, simultaneously considering spatial structural information and temporal dynamic characteristics to achieve high-precision moisture migration prediction.

[0064] The specific implementation of establishing the training dataset for the neural network model of tobacco leaf moisture migration includes four main steps. First, data collection is performed. Representative samples from different producing areas and varieties of tobacco leaves are selected, and orthogonal experimental schemes are set up with temperature ranges (30–68℃) and relative humidity ranges (18%–90%). Under each condition combination, complete T2 relaxation time spectrum time-series variation data are collected using low-field nuclear magnetic resonance (NMR) technology, with a collection interval of 30 minutes and a duration of 24 hours. Simultaneously, tobacco leaf microstructure image data, including optical microscope images and scanning electron microscope images, with a resolution of no less than 10 μm, and corresponding actual moisture content measurements are collected. Next, data preprocessing is performed. The collected experimental data are classified and labeled according to factors such as variety, ambient temperature, and ambient relative humidity to establish a structured database. The T2 relaxation time spectrum data undergoes standardization processing, including baseline correction, signal-to-noise ratio optimization, and peak alignment, to eliminate differences in data dimensions under different experimental conditions. Finally, the microstructure images are processed, including denoising, enhancement, and segmentation, to extract structural feature parameters such as porosity, specific surface area, and fiber alignment direction. Next, a graph representation was constructed. A graph model was established based on the physical connections of tobacco leaf tissues, with each sampling point represented as a node in the graph and connections between adjacent tissues represented as edges. Edge weights were determined based on physical distance and tissue type similarity. Each node was assigned a corresponding T2 relaxation time spectral feature vector (the spectral image was discretized into a 64-dimensional feature vector) and a moisture content label. Finally, the dataset was partitioned into training, validation, and test sets in an 8:1:1 ratio to ensure a balanced distribution of samples across different sets. A stratified sampling strategy was employed to ensure consistent proportions of samples from different varieties and under different conditions within each subset. The training set was used for model parameter learning, the validation set for hyperparameter tuning and early shutdown strategy implementation, and the test set for final model performance evaluation. This training dataset contains rich information related to tobacco leaf moisture migration, providing high-quality learning material for the neural network model.

[0065] The mathematical model or calculation process involved in this invention will be described in detail below.

[0066] In step S03, when establishing the dynamic model of tobacco leaf moisture migration, the mathematical representation of the inverse Laplace transform algorithm is as follows:

[0067]

[0068] In the formula, S(T2) is the T2 relaxation time spectrum; M(t) is the experimentally measured CPMG decay curve; K(t, T2) is the kernel function, expressed as K(t, T2) = exp(-t / T2); E(t) is the experimental error term, which follows a normal distribution N(0, σ). 2 ), where σ ranges from 0.001 to 0.01.

[0069] Solving the inverse problem using Tikhonov regularization:

[0070] S(T2)=argmin S≥0 {||AS-M|| 2 +α||LS|| 2};

[0071] In the formula, A is the discretized kernel matrix, and its elements A ij =exp(-t i / T 2j L is the regularization matrix, usually a second-order difference operator is chosen; α is the regularization parameter, with a value range of 0.01 to 0.1.

[0072] The multi-Gaussian curve fitting method is expressed as:

[0073]

[0074] In the formula, G(T2) is the fitted T2 relaxation time spectrum; n is the number of Gaussian peaks, usually 3 to 5; A i T represents the amplitude of the i-th Gaussian peak. 2i σ represents the center position of the i-th Gaussian peak; i Let be the width parameter of the i-th Gaussian peak.

[0075] The three-parameter exponential decay model for the free water content change curve and the bound water content change curve is expressed as:

[0076] W(t) = W ∞ +(W0-W ∞ )exp(-kt n );

[0077] In the formula, W(t) is the moisture content at time t; W0 is the initial moisture content; W ∞ To balance the moisture content; k is the rate constant, ranging from 0.01 to 0.5 h. -n n is an exponential parameter, ranging from 0.5 to 1.5.

[0078] In step S04, the water diffusion equation in tobacco leaf tissue, constructed based on Fick's second law, is as follows:

[0079]

[0080] In the formula, C represents the moisture concentration, with units of kg / m³. 3 t represents time, in seconds; d represents the diffusion coefficient tensor, in meters. 2 / s; This is the gradient operator.

[0081] The moisture diffusion coefficient matrix is ​​represented as follows:

[0082]

[0083] In the formula, D xx D yy D zz The diffusion coefficients in the x, y, and z directions are respectively; D xy D xz D yx D yz D zx D zy D is the cross-diffusion coefficient. For isotropic materials, D xx =D yy =D zz And the off-diagonal elements are 0.

[0084] The 3×3 moisture diffusion coefficient matrix formed by the diffusion coefficients of the three main tobacco leaf tissue regions is represented as follows:

[0085]

[0086] In the formula, D1 is the intracellular lumen diffusion coefficient, ranging from 10. -11 ~10 -10 m 2 / s; D2 is the cell wall diffusion coefficient, ranging from 10. -12 ~10 -11 m 2 / s; D3 is the intercellular diffusion coefficient, ranging from 10. -10 ~10 -9 m 2 / s;D ij (i≠j) is the inter-tissue water exchange coefficient.

[0087] The objective function for fitting the diffusion coefficient using the least squares method is:

[0088]

[0089] In the formula, C exp (x i , t i (x) represents the experimentally measured position. i Time t i Moisture concentration at C; mod (x i , t i D) is the location x predicted by the model. i Time t i The water concentration at the location; m is the number of measurement points.

[0090] In step S05, the multifactor regression model is expressed as:

[0091] W loss =β0+β1T+β2RH+β3P+β4S+β5F+β6W0+ε;

[0092] In the formula, W loss The value represents the moisture loss rate of tobacco leaves, expressed as % / h; T represents the ambient temperature, ranging from 30 to 68℃; RH represents the ambient relative humidity, ranging from 18% to 90%; P represents the porosity, ranging from 0.1 to 0.5; and S represents the specific surface area, ranging from 1 to 5 m². 2 / g; F is the fiber orientation, ranging from 0° to 90°; W0 is the initial moisture content of the tobacco leaf, ranging from 10% to 30%; β0, β1, β2, β3, β4, β5, and β6 are regression coefficients; ε is the random error term, following a normal distribution N(0, σ). 2 ).

[0093] The parameters were obtained as follows: ambient temperature T and ambient relative humidity RH were obtained in real time by temperature and humidity sensors with accuracies of ±0.1℃ and ±1%, respectively; porosity P was determined by gas adsorption method, specific surface area S was determined by BET nitrogen adsorption method, and fiber alignment F was determined by observation with a polarizing microscope. All three parameters need to be measured on the sample before the experiment; the initial moisture content W0 of tobacco leaves was determined by drying method, and dried to constant weight at 105℃.

[0094] The expression for extracting principal components using the partial least squares regression algorithm is:

[0095] X = TP T +E;

[0096] Y = UQ T +F;

[0097] In the formula, X is the input variable matrix; Y is the response variable matrix; T and U are the score matrices; P and Q are the load matrices; and E and F are the residual matrices.

[0098] The Arrhenius equation establishes the relationship between temperature and diffusion rate:

[0099] D = D0exp(-E a / RT);

[0100] In the formula, D is the diffusion coefficient, with units of m. 2 / s; D0 is the pre-exponential factor, in units of m. 2 / s;E a Activation energy, ranging from 15 to 25 kJ / mol; R is the gas constant, with a value of 8.314 J / (mol·K); T is the absolute temperature, in K.

[0101] Formula for calculating temperature correction factor:

[0102] k T =exp[E a / R·(1 / T ref -1 / T)];

[0103] In the formula, k T T is the temperature correction factor; ref For reference temperature, it is usually taken as 298.15K.

[0104] The humidity correction factor is based on the isothermal moisture absorption curve:

[0105]

[0106] In the formula, k RH W is the humidity correction factor. e (RH) is the equilibrium moisture content at the ambient relative humidity RH; W sat This represents the saturated moisture content.

[0107] The isothermal moisture absorption curves are represented using the GAB model:

[0108]

[0109] In the formula, W e To achieve equilibrium moisture content; W m The amount of water adsorbed by a single molecular layer; C and K are model parameters; a w Water activity is equal to the relative humidity of the environment divided by 100.

[0110] In step S06, the system of partial differential equations based on Fick's second law is expressed as:

[0111]

[0112] In the formula, W is the moisture content; D(W, T, RH) is the moisture diffusion coefficient tensor, which is a function of moisture content, temperature and relative humidity.

[0113] The finite difference expression after spatial discretization (taking one dimension as an example):

[0114]

[0115] In the formula, This represents the moisture content at position i and time step n. Δt represents the diffusion coefficient between position i and i+1; Δt is the time step, which is 1 second; Δx is the spatial step, which is 0.1 mm.

[0116] Constrained optimization problems constructed using the Lagrange multiplier method:

[0117] L(W, λ) = f(W) + λg(W);

[0118] In the formula, f(W) is the objective function, representing the water migration time; g(W) is the constraint condition, representing the uniformity of water loss and the conservation of mass; and λ is the Lagrange multiplier.

[0119] The objective function for minimizing water migration time is:

[0120]

[0121] In the formula, T is the total migration time; Ω is the tobacco leaf domain.

[0122] Constraints on uniformity of moisture loss:

[0123]

[0124] In the formula, ε1 is the allowable non-uniformity threshold, which ranges from 0.01 to 0.05.

[0125] Mass conservation constraints:

[0126]

[0127] In the formula, W0 is the initial moisture content distribution; J is the water flux; and n is the boundary normal vector. ε1 represents the tobacco leaf boundary; ε2 represents the allowable quality error, ranging from 0.001 to 0.01.

[0128] The water diffusion rate constant matrix is ​​expressed as:

[0129]

[0130] In the formula, k ij This represents the water diffusion rate constant from region i to region j, in units of h. -1 m and n are the number of regions after discretization of the tobacco leaves.

[0131] In step S08, the first-order kinetic equation is used to fit the water loss curve:

[0132]

[0133] In the formula, W is the moisture content; e The equilibrium moisture content; k(T) is the rate constant, a function of temperature, with units of h. -1 .

[0134] After integration, we get:

[0135] W(t) = W e +(W0-W e )exp[-k(T)t];

[0136] In the formula, W(t) is the moisture content at time t; W0 is the initial moisture content.

[0137] The Arrhenius equation establishes the relationship curve between ambient temperature and moisture migration rate:

[0138] k(T)=Aexp(-E a / RT);

[0139] In the formula, k(T) is the water migration rate constant; A is the pre-exponential factor, in units of h. -1 E a 1 is the activation energy, expressed in J / mol; R is the gas constant; T is the absolute temperature, expressed in K.

[0140] Formula for calculating temperature sensitivity coefficient:

[0141]

[0142] In the formula, Q 10 It is the temperature sensitivity coefficient, which represents the factor by which the reaction rate increases when the temperature increases by 10°C, and is usually in the range of 1.5 to 2.5.

[0143] In step S09, the first-order exponential decay model is used to fit the water loss curve:

[0144] W(t) = W e +(W0-W e )exp(-kt);

[0145] In the formula, W(t) is the moisture content at time t; W0 is the initial moisture content; W e The equilibrium water content is given by k, which is the rate constant in hours. -1 .

[0146] Formula for calculating the organizational structure influence factor:

[0147]

[0148] In the formula, F s The organizational structure influence factor ranges from 1.2 to 2.5; k vein The main vein water migration rate constant; kmesophyll is the rate constant for water migration in the mesophyll tissue.

[0149] Partial correlation analysis identifies key structural factors:

[0150]

[0151] In the formula, r xy。z r is the partial correlation coefficient between variables x and y, with z as the control variable;xy r xz r yz These are the simple correlation coefficients between variables x and y, x and z, and y and z, respectively. The correlation coefficient threshold is set to 0.7. xy。z When |>0.7, a significant correlation is considered to exist between variables x and y.

[0152] In step S10, the formula for calculating the relative error is:

[0153]

[0154] In the formula, RE represents the relative error; W pred To predict water content for the model; W meas This represents the measured moisture content.

[0155] Formula for calculating root mean square error:

[0156]

[0157] In the formula, RMSE is the root mean square error; n is the sample size; W pred，i W represents the predicted moisture content of the i-th sample. meas，i Let be the measured moisture content of the i-th sample.

[0158] Formula for calculating the coefficient of determination:

[0159]

[0160] In the formula, R 2 The coefficient of determination; This represents the average measured moisture content.

[0161] The construction principle and significance of the above equations are analyzed as follows:

[0162] The inverse Laplace transform algorithm is used to convert a time-domain signal into a T² relaxation time spectrum, which is a typical ill-posed inverse problem. The exponential kernel function K(t, T²) = exp(-t / T²) is chosen because the nuclear magnetic resonance relaxation process follows an exponential decay law. A Tikhonov regularization term α||LS|| is introduced. 2 This is to address the ill-posedness of the inverse problem by balancing fitting accuracy with solution smoothness to obtain a stable solution. The choice of the regularization parameter α is crucial: too small a value will cause the solution to oscillate excessively, while too large a value will lead to over-smoothing and loss of detailed information.

[0163] The multi-Gaussian curve fitting method is based on the fact that the T2 relaxation time spectrum is usually composed of peaks of water in multiple different binding states. The log-normal distribution (a Gaussian distribution in lnT2 space) is chosen instead of a simple Gaussian distribution because the T2 value usually spans several orders of magnitude, and the logarithmic space is more suitable for expressing this distribution characteristic.

[0164] The three-parameter exponential decay model is used to describe the change in water content over time. Introducing the exponential parameter *n* makes the model more flexible and capable of describing non-first-order kinetic processes. When *n* = 1, it degenerates into a first-order kinetic model; when *n* < 1, diffusion is hindered; and when *n* > 1, diffusion is accelerated. Compared to traditional first-order kinetic models, this model can more accurately describe the nonlinear characteristics of water migration in complex biomaterials.

[0165] The moisture diffusion equation based on Fick's second law is a classic equation describing the diffusion of moisture in porous media. Introducing a diffusion coefficient tensor D instead of a scalar allows for the characterization of the anisotropic properties of tobacco materials, i.e., the difference in moisture migration rates in different directions. This expression better reflects the actual structural characteristics of tobacco leaves than the traditional isotropic model, accurately capturing the difference in moisture migration rates between the midrib direction and the perpendicular direction.

[0166] Moisture diffusion coefficient matrix D tissue This study innovatively treats tobacco leaves as a three-region system, characterizing the water diffusion characteristics in the cell lumen, cell wall, and intercellular spaces, as well as the interactions between them. The off-diagonal element D of the matrix... ij (i≠j) represents the inter-tissue water exchange coefficient, which reflects the coupling characteristics of the system, something that traditional single-region models cannot express.

[0167] Multifactor regression models comprehensively consider the influence of environmental factors (temperature, humidity) and the characteristics of tobacco leaves (structural parameters, initial moisture content) on water loss. While linear summation facilitates the separation of independent contributions from each factor, it struggles to express the interactions between them. Therefore, partial least squares regression is introduced to address the multi-parameter collinearity problem, reducing model complexity and preserving maximum information content by extracting principal components.

[0168] The Arrhenius equation describes the effect of temperature on reaction rates. Its theoretical basis is that reactant molecules need to overcome an energy barrier to react; increasing temperature increases the probability that the molecular energy exceeds the energy barrier. In water migration, the activation energy E... a The physical meaning is the energy required for water molecules to detach from their adsorption sites. The exponential relationship indicates that temperature changes have a nonlinear amplification effect on the rate of moisture migration, which explains the decisive influence of temperature on moisture migration during baking.

[0169] The humidity correction factor, based on the isothermal hygroscopic principle, expresses the relationship between ambient relative humidity and the equilibrium moisture content of the material. When the ambient humidity approaches saturation, the exchange of moisture between the tobacco leaves and the environment tends to reach equilibrium, and the driving force for moisture migration decreases; conversely, a low-humidity environment provides a greater driving force for moisture migration. Compared with other isothermal hygroscopic models (such as the BET model), the GAB model covers a wider range of water activity, making it particularly suitable for describing the hygroscopic properties of food and biological materials.

[0170] The Lagrange multiplier method is used to solve constrained optimization problems. In moisture migration optimization, the objective is to minimize the migration time while satisfying moisture uniformity and mass conservation. The objective function for minimizing moisture migration is chosen to be a square integral form rather than a simple total time. This is to penalize large instantaneous gradient changes, resulting in a smoother moisture migration process, which is beneficial for tobacco quality control.

[0171] Temperature sensitivity coefficient Q 10 Q is a classic parameter describing the temperature response of biological systems. A value of 1.5–2.5 indicates that tobacco leaf moisture migration has a moderate degree of sensitivity to temperature. Compared to a simple linear relationship, Q... 10 It can more accurately express the nonlinear effect of temperature changes on the reaction rate, providing theoretical guidance for baking temperature control.

[0172] Organizational structure influence factor F s This study innovatively and quantitatively characterized the differences in water migration among different tissue parts of tobacco leaves. The ratio format intuitively reflects the relative relationship between the water migration rate of the midrib and the mesophyll tissue, providing a quantitative indicator for the balanced control of water content in different tissue parts during the curing process.

[0173] Partial correlation analysis can effectively eliminate indirect correlations between variables and identify key structural factors that directly affect water migration. Setting the correlation coefficient threshold to 0.7 is based on statistical convention and corresponds to approximately 49% of the variance explained, which can exclude secondary influences while retaining important factors.

[0174] Relative error, root mean square error (RMSE), and coefficient of determination (CCD) are three complementary metrics for evaluating model prediction accuracy. Relative error reflects the accuracy of individual predictions, RMSE characterizes the overall bias level, and the CCD measures the model's ability to explain data variability. Model validation employs multiple metrics rather than a single metric, enabling a comprehensive assessment of model performance and guiding model improvement.

[0175] Specifically, the principle of this invention is as follows: The core principle lies in using low-field nuclear magnetic resonance (NMR) technology to measure the motion characteristics of water molecules in tobacco leaves, thereby quantitatively analyzing the migration dynamics of water in different microstructures. The working principle of low-field NMR technology is based on the spin relaxation characteristics of hydrogen nuclei in a magnetic field. By measuring the T2 relaxation time spectrum, the existence states of water molecules in different microenvironments can be distinguished. Water molecules in different states exhibit different relaxation times: bound water, due to its close interaction with macromolecules in tobacco leaf tissue, has a shorter relaxation time; while free water, with its higher mobility, has a longer relaxation time. Therefore, by analyzing the T2 relaxation time spectrum, the content and distribution of free water and bound water in tobacco leaves can be accurately distinguished.

[0176] This invention combines moisture diffusion theory with low-field nuclear magnetic resonance (NMR) measurements to establish a dynamic model of moisture migration in tobacco leaves. The moisture diffusion optimization function, based on Fick's second law, considers the tissue structure characteristics of tobacco leaves and achieves a precise description of the moisture migration process by solving a system of partial differential equations. This function comprehensively considers multiple parameters such as ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content, and tissue density index to calculate the optimal migration path and rate constant matrix of moisture in different microstructures.

[0177] The neural network model for water migration in tobacco leaves employs an architecture combining graph convolutional networks and a spatiotemporal attention mechanism, accurately capturing the spatial features and temporal dynamics of water migration within the microstructure of tobacco leaves. Graph convolutional layers extract spatial connectivity features of tobacco tissues, temporal convolutional layers capture temporal variation patterns of water migration, and a multi-head self-attention mechanism learns the mutual influence relationships of water migration between different tissue regions. A spatiotemporal consistency regularization term is introduced during model training to ensure that the prediction results conform to physical laws. The parameters of the multi-scale attention mechanism are adaptively adjusted based on the ratio of bound water to free water content, tissue density index, and the peak position of the T2 relaxation time spectrum, enabling the model to accurately describe the complex dynamic process of water migration between different microstructures of tobacco leaves.

[0178] 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.

[0179] The specific implementation of steps S01-S02 in this embodiment is the same as that described above, and will not be repeated in detail here.

[0180] The specific implementation of step S03 involves establishing a dynamic model of tobacco leaf moisture migration. First, the obtained CPMG decay curve is processed using the inverse Laplace transform algorithm to convert the time-domain signal into a T2 relaxation time spectrum. The mathematical representation of this algorithm is as follows: In the formula, S(T2) is the T2 relaxation time spectrum; M(t) is the experimentally measured CPMG decay curve; K(t, T2) is the kernel function, expressed as K(t, T2) = exp(-t / T2); E(t) is the experimental error term, which follows a normal distribution N(0, σ). 2 The range of σ is 0.001 to 0.01. The Tikhonov regularization method is used to control the smoothing coefficient between 0.01 and 0.1 to balance resolution and stability. The expression is: S(T2) = argmin S≥0 {||AS-M|| 2 +α||LS|| 2}; where A is the discretized kernel matrix, and its elements A ij =exp(-t i / T 2jL is the regularization matrix, usually a second-order difference operator is chosen; α is the regularization parameter, ranging from 0.01 to 0.1. Then, peak separation and identification are performed on the T2 relaxation time spectrum obtained after the transformation, and different moisture components are identified using a multi-Gaussian curve fitting method. In the formula, G(T2) is the fitted T2 relaxation time spectrum; n is the number of Gaussian peaks, usually 3 to 5; A i T represents the amplitude of the i-th Gaussian peak. 2i σ represents the center position of the i-th Gaussian peak; i Let be the width parameter of the i-th Gaussian peak. Peaks with a T2 value less than 10 milliseconds are identified as bound water, peaks between 10 and 100 milliseconds are identified as capillary water, and peaks greater than 100 milliseconds are identified as free water. Then, under controlled conditions, the T2 relaxation time spectrum was measured every 30 minutes for 8 consecutive hours to obtain complete data on the dynamic changes in water content. Finally, the changes in the area of ​​different peaks in the spectrum were calculated by integration, and curves showing the changes in free water content and bound water content were plotted. A three-parameter exponential decay model was used to fit the curves: W(t) = W... ∞ +(W0-W ∞ )exp(-kt n In the formula, W(t) is the moisture content at time t; W0 is the initial moisture content; W ∞ To balance the moisture content; k is the rate constant, ranging from 0.01 to 0.5 h. -n n is an exponential parameter, ranging from 0.5 to 1.5. Goodness-of-fit R² 2 Not less than 0.95. This step, through analysis of the time-series T2 relaxation time spectrum, revealed the pattern of moisture changes in different states of tobacco leaves over time, providing a dynamic perspective for understanding the moisture migration mechanism.

[0181] The specific implementation of step S04 involves calculating the moisture diffusion coefficient matrix of tobacco leaves. First, based on Fick's second law, a moisture diffusion equation for tobacco leaf tissue is constructed: In the formula, C represents the moisture concentration, with units of kg / m³. 3 t represents time, in seconds; D represents the diffusion coefficient tensor, in meters. 2 / s; A gradient operator was used. Then, the tobacco leaf was considered as a porous medium, and a three-dimensional diffusion model was established. The T2 relaxation time distribution was correlated with the spatial location of moisture. By measuring the changes in the T2 spectrum at different sample points, a gradient field for moisture diffusion was constructed. Next, the partial differential equation for moisture diffusion was solved using the finite element method, with a mesh size of at least 100 μm, a time step of 1 minute, and an iteration convergence threshold of 10. -6Based on the solution results, the diffusion coefficients of three main tobacco leaf tissue regions were calculated: intracellular diffusion coefficient (D1), cell wall diffusion coefficient (D2), and intercellular diffusion coefficient (D3), forming a 3×3 water diffusion coefficient matrix. In the formula, D1 is the intracellular lumen diffusion coefficient, ranging from 10. -11 ~10 -10 m 2 / s; D2 is the cell wall diffusion coefficient, ranging from 10. -12 ~10 -11 m 2 / s; D3 is the intercellular diffusion coefficient, ranging from 10. -10 ~10 -9 m 2 / s;D ij (i≠j) represents the inter-tissue water exchange coefficient. The optimal diffusion coefficient was determined by fitting the experimental data and model predictions using the least squares method. In the formula, C exp (x i , t i (x) represents the experimentally measured position. i Time t i Moisture concentration at C; mod (x i , t i D) is the location x predicted by the model. i Time t i The water concentration at the measurement point is given by m, and the number of measurement points is denoted by m. The fitting error is controlled within 5%. This step, by combining mathematical models with experimental data, quantitatively characterizes the migration ability of water in different microstructures of tobacco leaves, providing key parameters for understanding the water migration mechanism in tobacco leaves.

[0182] The specific implementation of step S05 is to construct a tobacco leaf moisture loss prediction model. First, a multi-factor regression model is established, with the expression: W loss =β0+β1T+β2RH+β3P+β4S+β5F+β6W0+ε; where, W loss The value represents the moisture loss rate of tobacco leaves, expressed as % / h; T represents the ambient temperature, ranging from 45 to 70℃; RH represents the ambient relative humidity, ranging from 30% to 90%; P represents the porosity, ranging from 0.1 to 0.5; and S represents the specific surface area, ranging from 1 to 5 m². 2 / g; F is the fiber orientation, ranging from 0° to 90°; W0 is the initial moisture content of the tobacco leaf, ranging from 10% to 30%; β0, β1, β2, β3, β4, β5, and β6 are regression coefficients; ε is the random error term, following a normal distribution N(0, σ). 2Then, partial least squares regression is used to handle the potential collinearity problem among multiple parameters, extracting principal component factors and retaining principal components with a cumulative variance contribution rate of 95%. The algorithm expression is: X = TP T +E;Y=UQ T +F; where X is the input variable matrix; Y is the response variable matrix; T and U are the score matrices; P and Q are the loading matrices; and E and F are the residual matrices. Next, a temperature correction coefficient is introduced, and the relationship between temperature and diffusion rate is established based on the Arrhenius equation: D=D0exp(-E a / RT); where D is the diffusion coefficient, in units of m. 2 / s; D0 is the pre-exponential factor, in units of m. 2 / s;E a The activation energy is 15–25 kJ / mol; R is the gas constant, with a value of 8.314 J / (mol·K); T is the absolute temperature, in K. The temperature correction factor is calculated using the formula: kJ / mol. T =exp[E a / R·(1 / T ref -1 / T)];where, k T T is the temperature correction factor; ref For reference temperature, 298.15K is typically used. A humidity correction factor is also introduced, and the relationship between ambient relative humidity and equilibrium moisture content is established based on the isothermal hygroscopic curve. In the formula, k RH W is the humidity correction factor. e (RH) is the equilibrium moisture content at the ambient relative humidity RH; W sat The saturated moisture content is given. The isothermal moisture absorption curve is represented using the GAB model. In the formula, W e To achieve equilibrium moisture content; W m The amount of water adsorbed by a single molecular layer; C and K are model parameters; a w Water activity is calculated as relative humidity divided by 100. Finally, a comprehensive prediction model is constructed, employing a gradient boosting decision tree algorithm to integrate the influence of multiple parameters. The tree depth is set to 5, the learning rate to 0.05, and the number of weak learners to 100. Cross-validation uses the 5-fold method to evaluate model stability. This step, by comprehensively considering environmental factors and the characteristics of tobacco leaves themselves, establishes a mathematical model capable of predicting the water loss process of tobacco leaves under different conditions, providing a theoretical basis for water control in tobacco processing.

[0183] The specific implementation of step S06 involves applying a moisture diffusion optimization function to adjust the moisture migration parameters. First, a system of partial differential equations based on Fick's second law is established to describe the moisture migration process in tobacco leaves: In the formula, W represents the moisture content; D(W, T, RH) is the moisture diffusion coefficient tensor, a function of moisture content, temperature, and relative humidity. Then, the partial differential equations are discretized and solved using the finite difference method, with a spatial step size of 0.1 mm and a time step size of 1 second, constructing a numerical computation framework. The spatially discretized finite difference expression (taking one dimension as an example) is as follows: In the formula, This represents the moisture content at position i and time step n. Δt represents the diffusion coefficient between positions i and i+1; Δt is the time step, with a value of 1 second; Δx is the spatial step, with a value of 0.1 mm. Next, input the key parameters: ambient temperature gradient (0.5–5℃ / cm), ambient relative humidity (18%–90%), tobacco leaf thickness (0.1–0.5 mm), initial moisture content of tobacco leaves (84%–87%), and tobacco leaf tissue density index (0.8–1.2). The constrained optimization problem is constructed using the Lagrange multiplier method: L(W, λ) = f(W) + λg(W); where f(W) is the objective function, representing the water migration time; g(W) is the constraint condition, representing the uniformity of water loss and mass conservation; λ is the Lagrange multiplier. The objective function for minimizing the water migration time is: In the formula, T represents the total migration time; Ω represents the tobacco leaf domain. Constraints on the uniformity of moisture loss: In the formula, ε1 is the allowable non-uniformity threshold, ranging from 0.01 to 0.05. Mass conservation constraint: In the formula, W0 is the initial moisture content distribution; J is the water flux; and n is the boundary normal vector. ε represents the tobacco leaf boundary; ε2 represents the allowable quality error, ranging from 0.001 to 0.01. Finally, the conjugate gradient method is used to solve the optimization problem, with no more than 1000 iterations and a convergence threshold set to 10. -5 Output the optimal path map for water migration and the water diffusion rate constant matrix: In the formula, k ij This represents the water diffusion rate constant from region i to region j, in units of h. -1 ; m and n represent the number of regions after discretization of the tobacco leaf. This step, by applying advanced mathematical optimization methods, determines the optimal path and rate parameters for moisture migration in the tobacco leaf tissue under given conditions, providing quantitative guidance for optimizing tobacco processing technology.

[0184] The specific implementation of step S07 involves using a tobacco leaf moisture migration neural network model to analyze the dynamic distribution of moisture in the whole leaf. This neural network model employs a deep learning architecture combining graph convolutional networks and a spatiotemporal attention mechanism. The model construction first determines the network structure, which includes five graph convolutional layers, three temporal convolutional layers, and two multi-head self-attention modules. The graph convolutional layers use a ChebNet structure with a polynomial order of 3 and hidden layer dimensions of 64, 128, 256, 128, and 64, used to extract spatial connectivity features of tobacco leaf tissue. The temporal convolutional layers use a causal convolutional structure with a kernel size of 3 and dilation coefficients of 1, 2, and 4, used to capture temporal variation patterns of moisture migration. The multi-head self-attention modules have 8 heads and an attention dimension of 32, used to learn the mutual influence relationships of moisture migration between different tissue regions. The network input consists of a tobacco leaf tissue structure diagram and temporal T2 relaxation time spectrum data measured by low-field nuclear magnetic resonance (NMR) technology; the output is the predicted distribution of moisture content in the whole leaf and the predicted direction of moisture migration at future time points. The parameters of the multi-scale attention mechanism in the model are adaptively adjusted through three key indicators: the ratio of the combined water content change curve to the free water content change curve (range 0.1–10), the tobacco leaf tissue density index (range 0.8–1.2), and the peak position of the T2 relaxation time spectrum (range 1–1000 ms). This step, through deep learning technology, achieves high-precision prediction of the dynamic distribution of water in the whole tobacco leaf, providing an advanced analytical tool for understanding the water migration patterns of tobacco leaves under complex conditions.

[0185] The specific implementation of step S08 involves conducting experiments with different baking temperature gradients. First, a temperature gradient experimental scheme is designed, selecting four temperature points: 30℃, 44℃, 56℃, and 68℃, with relative humidity maintained at 60±2%. Then, a precision temperature-controlled oven is used, with temperature fluctuations controlled within ±0.5℃, and 10 pretreated tobacco leaf samples are placed under each temperature condition. Next, samples are removed at preset time points (0, 2, 4, 6, 8, 12, and 24 hours), and the T2 relaxation time spectrum is measured using low-field nuclear magnetic resonance (NMR) technology. Simultaneously, the absolute moisture content is determined using the drying method as a calibration reference. The collected data is processed to calculate the moisture migration rate at each temperature point, and the moisture loss curve is fitted using a first-order kinetic equation. In the formula, W is the moisture content; e The equilibrium moisture content; k(T) is the rate constant, a function of temperature, with units of h. -1 After integration, we get: W(t) = W e +(W0-W e W(t) = Aexp(-k(T)t); where W(t) is the water content at time t; W0 is the initial water content. Finally, the relationship between ambient temperature and water migration rate is established using the Arrhenius equation: k(T) = Aexp(-E)t. a / RT); where k(T) is the water migration rate constant; A is the pre-exponential factor, in units of h. -1 E a Δ is the activation energy, expressed in J / mol; R is the gas constant; T is the absolute temperature, expressed in K. Determine the temperature sensitivity coefficient: In the formula, Q 10 The temperature sensitivity coefficient represents the factor by which the reaction rate increases when the temperature rises by 10°C, typically ranging from 1.5 to 2.5. This step, through systematic temperature gradient experiments, revealed the influence of temperature on the moisture migration process of tobacco leaves, providing a scientific basis for optimizing curing process parameters.

[0186] The specific implementation of step S09 involves analyzing the differences in water migration between the midrib and mesophyll tissues of tobacco leaves. First, the midrib and mesophyll tissues are separated from the pretreated tobacco leaf samples, maintaining similar sample quality. Then, under the same environmental conditions (temperature 60℃, relative humidity 50%), the water migration process of both tissues is measured. T2 relaxation time spectra are acquired using low-field nuclear magnetic resonance (NMR) technology at 30-minute intervals, continuously monitored for 6 hours. Next, the water migration rates of the midrib and mesophyll tissues are calculated based on the T2 spectra, and a first-order exponential decay model is used to fit the water loss curve: W(t) = W e +(W0-W e )exp(-kt); where W(t) is the moisture content at time t; W0 is the initial moisture content; W e The equilibrium water content is given by k, which is the rate constant in hours. -1 Calculate the ratio of the two to construct the organizational structure influence factor F. s : In the formula, F s The organizational structure influence factor ranges from 1.2 to 2.5; k vein Main vein water migration rate constant; k mesophyll The constant for water migration rate in mesophyll tissue is given. Simultaneously, the microstructural characteristics of both tissues, including cell density, intercellular space size, and fiber content, were analyzed to establish a correlation model between microstructure and water migration rate. Finally, partial correlation analysis was used to identify the key structural factors influencing differences in water migration. In the formula, r xy。z r is the partial correlation coefficient between variables x and y, with z as the control variable; xy r xz r yz These are the simple correlation coefficients between variables x and y, x and z, and y and z, respectively. The correlation coefficient threshold is set to 0.7. xy。zWhen the correlation coefficient is greater than 0.7, a significant correlation is considered to exist between variables x and y. This step, through comparative analysis of the water migration characteristics of different tissue parts of tobacco leaves, reveals the mechanism by which tissue structure affects water migration, providing a theoretical basis for precise control of water distribution in tobacco leaves.

[0187] The specific implementation of step S10 is to verify the accuracy of the tobacco leaf moisture loss prediction model. First, a verification experimental scheme is designed, selecting different varieties of tobacco leaves, different baking temperatures (30–68℃), and different relative humidity (18%–90%) combinations to construct a verification dataset. Then, during the simulated baking process in the laboratory, samples are collected at preset time points (0, 2, 4, 8, 16, and 24 hours), and the moisture content is simultaneously measured using low-field nuclear magnetic resonance technology and the drying method. Next, the measured data are compared with the output values ​​of the prediction model, and the relative error is calculated. In the formula, RE represents the relative error; W pred To predict water content for the model; W meas This represents the measured moisture content. Root mean square error: In the formula, RMSE is the root mean square error; n is the sample size; W pred，i W represents the predicted moisture content of the i-th sample. meas，i Let be the measured moisture content of the i-th sample. Coefficient of determination: In the formula, R 2 The coefficient of determination; This represents the average measured moisture content. Data points with relative errors exceeding 10% are analyzed in depth to identify potential contributing factors. Finally, based on the validation results, the model parameters are optimized. A Bayesian optimization algorithm is used to adjust the model hyperparameters, including temperature correction coefficients, humidity correction coefficients, and structural influence factor weights. The optimization is iterated 50 times, with the goal of minimizing the prediction error on the validation set. The optimized model should have an average relative error of less than 5% and a maximum error not exceeding 10% on the validation set. This step, through systematic validation and optimization, improves the accuracy and reliability of the prediction model, ensuring that the model can provide effective guidance for practical applications.

[0188] To better understand and implement this invention, Example 2 of a specific application scenario is provided below: Researchers selected K326 tobacco leaves cultivated in a certain area of ​​Yunnan Province as the research object and systematically analyzed their moisture migration characteristics during the curing process. First, sample pretreatment was performed according to step S01. Forty complete tobacco leaves were randomly selected from the batch and cut into 5×5 cm pieces using a precision cutter, resulting in 120 samples. The samples were placed in a constant temperature and humidity chamber at 25±0.5℃ and 60±1% relative humidity for 24 hours to ensure consistent initial moisture content.

[0189] Subsequently, low-field nuclear magnetic resonance (NMR) measurements were performed according to step S02 using an NMR-MOUSE PM25 low-field NMR spectrometer. The probe frequency was set to 20 MHz, the number of acquisition repetitions was 16, the echo time was 0.2 ms, and the acquisition window width was 5000 ms, obtaining a complete CPMG attenuation curve. The measurement results showed that the initial average moisture content of this batch of tobacco leaves was 23.5%.

[0190] Next, a dynamic model of water migration was established according to step S03. The CPMG decay curve was processed using the inverse Laplace transform algorithm, with the regularization parameter α set to 0.05. Multi-Gaussian curve fitting was performed on the obtained T2 relaxation time spectrum, identifying three characteristic peaks, as shown in Table 1.

[0191] Table 1. Characteristic peak parameters of T2 relaxation time spectrum of tobacco leaf moisture

[0192] Peak type <![CDATA[T2 value (ms)]]> Peak Amplitude Percentage (%) Moisture state Peak 1 1.8±0.3 158.6 21.4 Bound water Peak 2 35.2±4.2 324.7 43.8 microcapillary water Peak 3 158.6±12.5 258.3 34.8 Free water

[0193] Under controlled conditions (temperature 48℃, relative humidity 50%), the T2 relaxation time spectrum was measured every 30 minutes for 8 consecutive hours to obtain dynamic moisture change data. The changes in peak area were calculated by integration, and curves showing the changes in free water and bound water content were plotted. A three-parameter exponential decay model was used for fitting, and the fitting parameters are shown in Table 2.

[0194] Table 2 Fitting parameters for moisture content variation curve

[0195] Moisture type <![CDATA[W0(%)]]> <![CDATA[W ∞ (%)]]> <![CDATA[h(h -n )]]> n <![CDATA[R 2 ]]> Free water 8.18 0.86 0.42 1.21 0.988 microcapillary water 10.29 2.45 0.28 0.95 0.976 Bound water 5.03 3.12 0.15 0.76 0.982

[0196] According to step S04, a moisture diffusion equation was constructed based on Fick's second law and solved using the finite element method with a mesh resolution of 80 μm and a time step of 30 seconds. By comparing experimental measurement data with numerical simulation results, the moisture diffusion coefficient matrices for the three tobacco leaf tissue regions were calculated, as shown in Table 3.

[0197] Table 3. Water diffusion coefficient matrix of tobacco leaf tissue (10) -11 m 2 / s)

[0198] Organizational Area Intracellular lumen cell wall intercellular spaces Intracellular lumen 7.85 1.24 0.56 cell wall 1.31 0.95 2.18 intercellular spaces 0.62 2.05 32.64

[0199] Subsequently, a moisture loss prediction model was constructed according to step S05, and experimental data under different conditions were collected, totaling 120 sets, including different temperatures (30–68℃), relative humidity (18%–90%), tobacco leaf structural parameters, and initial moisture content. Partial least squares regression was used to handle the multi-parameter collinearity problem, principal component factors were extracted, and three principal components with a cumulative variance contribution rate of 97.2% were retained. A temperature correction coefficient and activation energy E were introduced. aThe measured value was 18.6 kJ / mol. An isothermal hygroscopic curve was established based on the GAB model, with model parameter W... m =4.35%, C=12.8, K=0.76. The final prediction model achieved a prediction accuracy of 0.86% in cross-validation.

[0200] Following step S06, the moisture migration parameters were adjusted using the moisture diffusion optimization function. Using a temperature gradient of 2℃ / cm, relative humidity of 45%, tobacco leaf thickness of 0.28mm, initial moisture content of 23.5%, and tissue density index of 1.05 as input parameters, the optimization problem was solved numerically. The non-uniformity threshold ε1 was set to 0.03, and the quality error ε2 was set to 0.005. The optimal migration path and diffusion rate constant matrix were obtained. The rate constants for some regions are shown in Table 4.

[0201] Table 4. Moisture diffusion rate constants (h) for certain areas of tobacco leaves -1 )

[0202] Area code 1 2 3 4 5 1 0 0.242 0.135 0.078 0.042 2 0.258 0 0.315 0.186 0.094 3 0.152 0.328 0 0.275 0.126 4 0.085 0.196 0.282 0 0.235 5 0.046 0.103 0.138 0.246 0

[0203] In step S08, experiments were conducted at different baking temperature gradients, with four temperature points: 30℃, 44℃, 56℃, and 68℃, all with a relative humidity maintained at 60%. Moisture loss curves were measured, and the moisture migration rate constants at each temperature point were obtained by fitting a first-order kinetic equation. The results are shown in Table 5.

[0204] Table 5. Moisture migration rate constants and temperature sensitivity coefficients at different temperatures.

[0205] Temperature (°C) <![CDATA[Water migration rate constant (h -1 )]]> <![CDATA[Temperature sensitivity coefficient Q 10 > 30 0.086 - 44 0.152 1.77 56 0.265 1.74 68 0.423 1.60

[0206] The differences in water migration between the midrib and mesophyll tissues of tobacco leaves were analyzed according to step S09. Measurements were taken at 60℃ and 50% relative humidity. Water loss curves were fitted to obtain the water migration rate constants for the two tissues, with the midrib showing a constant of 0.165 h⁻¹. -1 The mesophyll tissue length is 0.085 h. -1 The organizational structure influence factor F was calculated. s The value was 1.94, indicating that the water migration rate in the midrib was significantly higher than that in the mesophyll tissue. Partial correlation analysis identified the key factors influencing the difference in water migration as intercellular space size (r = 0.82) and fiber orientation (r = 0.75).

[0207] Finally, the accuracy of the tobacco leaf moisture loss prediction model was verified according to step S10. A validation dataset was constructed using 20 different condition combinations, and the difference between the actual moisture content and the model prediction was measured. The statistical results are shown in Table 6.

[0208] Table 6. Validation results of the prediction model

[0209] Evaluation indicators Original model Optimized model Mean relative error (%) 7.62 3.24 Maximum relative error (%) 18.45 8.76 Root mean square error (%) 1.58 0.64 <![CDATA[Coefficient of determination R 2 > 0.921 0.984

[0210] Traditional methods for analyzing the moisture migration characteristics of tobacco leaves mainly rely on drying and conductivity methods. These methods suffer from drawbacks such as high destructiveness, poor real-time performance, and inability to distinguish between different moisture states. Traditional moisture migration models typically treat tobacco leaves as homogeneous bodies, ignoring the influence of tissue structure differences on moisture migration, resulting in low prediction accuracy and an inability to guide the optimization of refined curing processes. This invention utilizes low-field nuclear magnetic resonance (LF-NMR) technology to achieve non-destructive, real-time, and quantitative monitoring of moisture in tobacco leaves at different states. A three-region moisture diffusion coefficient matrix considering tissue structure differences was established, and a moisture migration neural network model based on a spatiotemporal attention mechanism was developed. This reduced the relative error of the moisture migration prediction model from 15%–20% of traditional methods to 3.24%, providing a more accurate theoretical basis and technical support for tobacco moisture control and curing process optimization. The research results show that this invention has significant scientific value for understanding the moisture migration mechanism in complex biological materials, and also has significant practical value for moisture control and quality improvement in tobacco processing.

[0211] It should be noted that the variables involved in this invention are explained in detail in Tables 7 and 8 below.

[0212] Table 7. Variable Explanation Table (Part 1)

[0213]

[0214]

[0215] Table 8. Variable Explanation Table (Part Two)

[0216]

[0217] 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 analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology, characterized in that, include: S01. Collect tobacco leaf samples and pre-treat them. Use a standard cutting tool to cut the tobacco leaves into 5×5 cm specifications and place them in a constant temperature and humidity environment for 24 hours to equilibrate. S02. The moisture content of tobacco leaves was determined using a low-field nuclear magnetic resonance instrument. The probe frequency was set to 20MHz, the number of acquisitions was 16, and the echo time was 0.2 milliseconds. S03. Establish a dynamic model of tobacco leaf moisture migration. By measuring the T2 relaxation time spectrum at different time points, obtain the curves of free water content change and bound water content change. S04. Calculate the tobacco leaf moisture diffusion coefficient matrix based on the T2 relaxation time distribution to identify the migration rate of water in the cell lumen, cell wall and intercellular space. S05. Construct a tobacco leaf moisture loss prediction model. Input parameters include ambient temperature, ambient relative humidity, tobacco leaf structural parameters, and initial moisture content of tobacco leaves. S06. Apply the moisture diffusion optimization function to adjust the moisture migration parameters. The input parameters include the ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content of tobacco leaves, and tobacco leaf tissue density index. The output is the optimal moisture migration path map and the moisture diffusion rate constant matrix. The moisture diffusion rate constant matrix is ​​a parameter matrix that describes the moisture migration speed between different regions. S07. Analyze the dynamic distribution of water in the whole leaf using a pre-trained tobacco leaf water migration neural network model. The tobacco leaf water migration neural network model uses a multi-scale attention mechanism to capture the correlation of water migration between different tissues of the tobacco leaf. S08. Conduct experiments with different baking temperature gradients, setting the temperature range from 45℃ to 70℃, and establish the relationship curve between ambient temperature and moisture migration rate. S09. Analyze the differences in water migration between the midrib and mesophyll tissue of tobacco leaves, calculate the ratio of water migration rate in the midrib to water migration rate in the mesophyll tissue, and construct tissue structure influencing factors.

2. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 1, characterized in that, The specific steps for collecting and pre-treating tobacco leaf samples are as follows: cut the tobacco leaves into 5×5 cm pieces using a standard cutting tool and place them in a constant temperature and humidity environment for 24 hours to equilibrate.

3. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 2, characterized in that, The specific steps for determining the moisture content of tobacco leaves using a low-field nuclear magnetic resonance instrument are as follows: set the probe frequency to 20MHz, the number of acquisition repetitions to 16, and the echo time to 0.2 milliseconds.

4. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 3, characterized in that, The specific steps for establishing a dynamic model of tobacco leaf moisture migration are as follows: by measuring the T2 relaxation time spectrum at different time points, obtain the free water content change curve and the bound water content change curve.

5. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 4, characterized in that, The free water content change curve refers to the quantitative curve of the change of water with greater mobility existing in the intercellular spaces of tobacco leaves over time; the bound water content change curve refers to the quantitative curve of the change of water with less mobility that is tightly bound to macromolecules in tobacco leaf tissue over time.

6. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 5, characterized in that, Tobacco leaf structural parameters refer to quantitative indicators characterizing the microstructural properties of tobacco leaves, including porosity, specific surface area, and fiber orientation; ambient temperature refers to the temperature value of the environment in which the tobacco leaves are located, in degrees Celsius; relative humidity refers to the percentage of air humidity in the environment in which the tobacco leaves are located compared to the saturated humidity at the same temperature; initial moisture content of tobacco leaves refers to the percentage of the mass of water in the tobacco leaves measured before the start of the experiment relative to the total mass of the tobacco leaves.

7. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 6, characterized in that, The moisture diffusion optimization function is used to calculate the optimal path and rate constant of moisture migration within tobacco leaves based on multiple parameter inputs. By solving the partial differential equations of Fick's second law and combining them with the tissue structure characteristics of tobacco leaves, it achieves a precise mathematical description and prediction of the moisture migration process in tobacco leaves. The input parameters include ambient temperature gradient, ambient relative humidity, tobacco leaf thickness, initial moisture content of tobacco leaves, and tobacco leaf tissue density index. The output is an optimal moisture migration path map and a moisture diffusion rate constant matrix.

8. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 7, characterized in that, The ambient temperature gradient refers to the rate of change of the ambient temperature around the tobacco leaf in space, measured in degrees Celsius per meter; the tobacco leaf thickness refers to the measured thickness of the tobacco leaf sample, measured in millimeters; the tobacco leaf tissue density index is a dimensionless index characterizing the density of the tobacco leaf tissue, calculated by the ratio of the tobacco leaf density to the standard density; the optimal moisture migration path diagram is a two-dimensional representation of the best path for moisture movement within the tobacco leaf under given conditions.

9. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 8, characterized in that, The specific structure of the tobacco leaf water migration neural network model is a deep learning architecture based on a combination of graph convolutional networks and spatiotemporal attention mechanisms. It includes five graph convolutional layers for extracting spatial connectivity features of tobacco leaf tissue, three temporal convolutional layers for capturing temporal variation patterns of water migration, and two multi-head self-attention modules for learning the mutual influence relationships of water migration between different tissue regions. The input of the tobacco leaf water migration neural network model is the tobacco leaf tissue structure diagram and the temporal T2 relaxation time spectrum data measured by low-field nuclear magnetic resonance technology. The output is the prediction result of the whole leaf water content distribution and the prediction of the water migration direction at future time points.

10. The method for analyzing the moisture migration characteristics of tobacco leaves based on low-field nuclear magnetic resonance technology according to claim 9, characterized in that, The specific steps for analyzing the differences in water migration between the midrib and mesophyll tissue of tobacco leaves are as follows: calculate the ratio of water migration rate in the midrib to water migration rate in the mesophyll tissue, and construct tissue structure influencing factors. The water migration rate refers to the speed at which water moves in the tobacco leaf tissue, and the unit is millimeters per hour. The tissue structure influence factor refers to a dimensionless coefficient that characterizes the degree of influence of different tissue structures of tobacco leaves on water migration.

Images