Baking process based on quality of Xiangyan No.7 fresh tobacco leaves

Abstract

The invention relates to the technical field of tobacco baking, and provides a baking process based on the quality of fresh tobacco leaves of Xiangyan No.7. The baking process comprises the following steps: S1, carrying out near infrared spectrum scanning on harvested fresh tobacco leaves to obtain near infrared spectrum data of each tobacco leaf; s2, dividing the near infrared spectrum data into a training set and a test set; s3, taking the near infrared spectrum data of each tobacco leaf in the training set as an input vector, inputting the input vector into a self-organizing neural network model for clustering, and dividing the tobacco leaves into three subsets; s4, aiming at the three divided subsets, baking process parameters are optimized according to a three-section baking process, and the three-section baking process comprises a yellowing period, a color fixing period and a stem drying period; and S5, performing near infrared spectrum scanning on newly harvested tobacco leaves, classifying the newly harvested tobacco leaves through the self-organizing neural network model, determining subsets to which the newly harvested tobacco leaves belong, and according to the baking process curves corresponding to the subsets, sequentially executing the baking processes of a yellowing period, a color fixing period and a stem drying period.

Description

Technical Field

[0001] This invention relates to the field of tobacco curing technology, and in particular to a curing process based on the quality of fresh tobacco leaves of Xiangyan No. 7. Background Technology

[0002] In the field of tobacco curing, the design of the curing process directly affects curing efficiency and tobacco quality. The curing process typically includes stages such as yellowing, color fixing, and drying, with temperature, humidity, and ventilation parameters controlling moisture evaporation and chemical component transformation. However, due to significant differences in chemical composition, moisture content, and physicochemical properties among different tobacco varieties, especially high-end varieties (such as Xiangyan No. 7), existing curing technologies cannot precisely control the process according to the specific characteristics of the tobacco leaves, resulting in inconsistent curing effects, a high proportion of low-quality tobacco, and difficulty in bringing out the desired flavor.

[0003] For example, Chinese patent CN102499424A discloses a method for curing tobacco leaves, which includes placing the tobacco leaves in a curing barn and setting multiple temperature control schemes according to dry-bulb and wet-bulb temperatures, so that the tobacco leaves experience different temperature and humidity conditions at stages such as yellowing, color fixing, and drying. This method improves the automation and production capacity of tobacco leaf curing by controlling the curing process through temperature and humidity gradients. However, it still uses fixed temperature and humidity parameters uniform for all tobacco leaves, without classifying and dynamically adjusting them according to the differences in the chemical composition of the tobacco leaves themselves, making it difficult to achieve differentiated optimization control of tobacco leaves of different qualities.

[0004] Another typical Chinese patent, CN101579144B, discloses a method for curing tobacco leaves and a curing barn. By adjusting the airflow direction and hot air flow, the tobacco leaves are cured at different temperature and humidity stages within the barn. This solution makes some improvements to the process flow, but it still fails to classify and optimize parameters to address the physicochemical differences of different tobacco varieties, and thus cannot significantly improve the curing quality and uniformity of high-quality tobacco leaves.

[0005] In addition, the existing technology has the following shortcomings: 1. Most existing roasting methods use uniform temperature and humidity parameters for control, ignoring the significant impact of differences in tobacco varieties and chemical composition on the roasting process; 2. Traditional processes lack intelligent classification and dynamic parameter adjustment mechanisms based on the quality of the tobacco leaves themselves, resulting in insufficient material transformation of tobacco leaves of different qualities during the roasting process; 3. Existing technologies focus on staged temperature control, but have not formed a control system for tobacco leaf composition, classification judgment, and dynamic optimization of process parameters, making it difficult to meet the high requirements of high-end tobacco leaves for flavor, aroma, and texture; 4. In the existing roasting process, it is difficult to simultaneously improve the quality of high-quality tobacco leaves and reduce the quality of low-quality tobacco leaves, failing to effectively improve the overall product quality stability.

[0006] This invention was made to address the common problems in the field, such as fixed baking parameters, inability to classify and control, insufficient material conversion, and uneven flavor expression. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of current methods by proposing a curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves.

[0008] To overcome the shortcomings of the prior art, the present invention adopts the following technical solution: A curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves, the curing process including the following steps: S1. Near-infrared spectral scanning is performed on the harvested fresh tobacco leaves to obtain the near-infrared spectral data of each tobacco leaf; S2. Divide the near-infrared spectral data into a training set and a test set; S3. Use the near-infrared spectral data of each tobacco leaf in the training set as the input vector, input it into the self-organizing neural network model for clustering, and divide the tobacco leaves into 3 subsets; S4. For the three subsets, optimize the baking process parameters according to the three-stage baking process, which includes the yellowing period, the color-fixing period, and the drying period. The optimized parameters include temperature, humidity, ventilation, and temperature stabilization time at each stage. With the quality of the cured tobacco leaves as the target, the curing process curves for the corresponding subset are established. S5. The newly harvested tobacco leaves are scanned by near-infrared spectroscopy, classified by the self-organizing neural network model, and their respective subsets are determined. The yellowing period, color fixing period and dry rib period are executed sequentially according to the baking process curve corresponding to the subset.

[0009] Optionally, the baking process further includes: in step S1, the near-infrared spectral scanning uses a wavelength range of 900~1100nm.

[0010] Optionally, the baking process further includes: in step S2, dividing the near-infrared spectral data of each tobacco leaf into a training set and a test set in a ratio of 7:3.

[0011] Optionally, the baking process further includes: in step S3, the number of neurons in the competitive layer of the self-organizing neural network model is 3, and the tobacco leaves are divided into 3 subsets.

[0012] Optionally, the working process of the self-organizing neural network model includes the following steps: S31: Initialization, using random numbers to set the initial weights between the input layer and the competition layer; S32: Euclidean distance is calculated between each neuron in the input layer and the competition layer. The distance between the j-th neuron in the competition layer and the input vector xi is calculated as follows: ; In the formula, x i For the input vector, W j Let j* be the weight vector of the j-th neuron in the competition layer. The neuron with the smallest distance is calculated and selected as the winning neuron. S33: The weights of the winning neuron j* and its neighboring neurons are corrected using the following formula: ; In the formula, W j (t+1) represents the corrected weight, W j (t) represents the current weights, and α(t) represents the learning rate. It is a neighborhood function; S34: Repeat steps S32~S33 until the preset number of iterations is reached or the weight change is less than the preset threshold.

[0013] Optionally, the baking process further includes: in step S34, the preset number of iterations is 500.

[0014] Optionally, the baking process further includes: in step S4: Yellowing period: The temperature is gradually increased from the initial temperature to the yellowing end temperature. Humidity and ventilation are controlled to degrade the chlorophyll in the tobacco leaves, cause the tobacco leaves to turn yellow, and hydrolyze the starch into reducing sugars. Color fixation period: The temperature continues to rise to the color fixation temperature, the humidity is reduced, and the ventilation is increased so that the tobacco leaves can fix their color, complete the material transformation, and form aroma substances; Drying stage: The temperature rises to the drying temperature, the humidity is further reduced, and the maximum ventilation is used to dry the main vein of the tobacco leaves and reduce the overall moisture content to 12-14%.

[0015] Optionally, the baking process further includes: establishing a chemical composition prediction model using the near-infrared spectral data to predict the chemical composition content of the tobacco leaves, wherein the chemical composition includes at least reducing sugar, total sugar, nicotine, total nitrogen, and potassium; The baking difficulty index (HDI) is calculated based on the content of the chemical components. HDI = (reducing sugar × α+potassium × β-nicotine × γ) / (total nitrogen × δ); In the formula, the value of α ranges from 0.40 to 0.50, the value of β ranges from 0.30 to 0.40, the value of γ ranges from 0.20 to 0.30, and the value of δ ranges from 0.25 to 0.35. The three training sets are distinguished based on HDI values: First subset: HDI > 8.5, indicating easy-to-roast tobacco leaves; The second subset: 6.0≤HDI≤8.5, represents suitable flue-cured tobacco leaves; Third subset: HDI < 6.0, indicating difficult-to-cook tobacco leaves.

[0016] Optionally, further optimization of the baking process for different subsets includes a segmented temperature and humidity control strategy based on biochemical reaction kinetics: The yellowing period was divided into four stages: chlorophyll degradation, starch hydrolysis, protein hydrolysis, and polyphenol oxidation. The temperature control range for each stage corresponds to the optimal activity temperature range for chlorophyllase, amylase, protease, and polyphenol oxidase, respectively. Chlorophyll degradation stage: temperature 35~40℃, residence time accounts for 30~35% of the total yellowing period; Starch hydrolysis stage: temperature 40~45℃, residence time accounts for 35~40% of the total yellowing period; Protein hydrolysis stage: temperature 42~48℃, residence time accounts for 15~20% of the total yellowing period; Polyphenol oxidation stage: temperature 45~50℃, residence time accounts for 10~15% of the total yellowing period.

[0017] Optionally, the baking process in step S5 may also include a real-time monitoring and dynamic adjustment step: The degree of yellowing, moisture content and / or concentration of volatile substances in tobacco leaves are monitored in real time using sensors. Calculate the actual yellowing rate The expected rate of yellowing Deviation, and / or actual dehydration rate With the expected dehydration rate Deviation; When the deviation exceeds a preset threshold, the temperature, humidity, and / or ventilation volume are dynamically adjusted according to preset adjustment rules, which include: When the actual yellowing rate When the value is greater than 1.5 times the expected value, lower the temperature by 1-2°C, increase the humidity by 1°C, and reduce the ventilation by 10-20%. When the actual yellowing rate If the humidity is less than 0.7 times the expected value, increase the temperature by 1-2°C, decrease the humidity by 0.5°C, and increase the ventilation by 5-10%. When the actual dehydration rate When the humidity is >3% / h, lower the temperature by 2℃, increase the humidity by 2℃, and reduce the ventilation volume by 20-30%. When the actual dehydration rate When the humidity is less than 0.5% / h, increase the temperature by 2℃, decrease the humidity by 1℃, and increase the ventilation by 15~20%.

[0018] The beneficial effects achieved by this invention are: 1. By combining near-infrared spectral scanning with a self-organizing neural network to perform unsupervised clustering of tobacco leaves, tobacco leaves of different qualities can be objectively and quickly divided into three subsets: easy-to-grill, suitable-to-grill, and difficult-to-grill, ensuring the accuracy and consistency of tobacco leaf classification. 2. By optimizing parameters such as temperature, humidity, and ventilation during the yellowing, color-fixing, and drying stages for different subsets, the tobacco leaves of each subset can fully complete the physiological and biochemical reactions of chlorophyll degradation, starch hydrolysis, and protein decomposition under suitable process conditions, thus ensuring the sufficiency and coordination of tobacco material transformation. 3. By using differentiated roasting processes, easy-to-roast tobacco leaves are prevented from being over-roasted, and difficult-to-roast tobacco leaves are prevented from being under-roasted. This significantly increases the proportion of high-quality tobacco and greatly reduces the proportion of low-quality tobacco, ensuring that the sensory qualities of the roasted tobacco leaves, such as color, aroma, and oil content, meet the standards for high-quality tobacco. This solves the problem of high proportion of low-quality tobacco and difficulty in highlighting flavor in existing technologies.

[0019] 4. By employing a four-level refined control strategy based on biochemical reaction kinetics, reactions such as chlorophyll degradation, starch hydrolysis, and polyphenol oxidation can proceed within their respective optimal temperature ranges. This ensures the refinement and optimization of the internal material transformation of tobacco leaves, further enhancing the quality and intrinsic properties of the cured tobacco leaves.

[0020] 5. By constructing a complete technology chain encompassing spectral scanning, intelligent classification, process optimization, parameter matching, and automated roasting, the analysis of tobacco leaf components, classification judgment, and optimization of process parameters are organically combined, ensuring the systematic and scientific nature of the roasting process; 6. By establishing a standardized curing process curve based on training set optimization, tobacco leaves from the same subset are cured according to the same process parameters, ensuring the stability and repeatability of curing quality between batches. 7. Through real-time monitoring and dynamic adjustment functions, the yellowing rate and dehydration rate during the baking process can be automatically adjusted according to the actual deviation, ensuring precise control and adaptive optimization of the baking process. This solves the problem of lacking dynamic parameter adjustment in traditional methods and further improves the stability of the process mechanism. Attached Figure Description

[0021] The invention will be further understood from the following description taken in conjunction with the accompanying drawings. The components in the drawings are not necessarily drawn to scale, but rather the emphasis is on illustrating the principles of the embodiments. In different views, the same reference numerals designate the same parts.

[0022] Figure 1This is a schematic diagram of the baking process of the present invention.

[0023] Figure 2 This is a diagram of the SOM neural network structure of the present invention.

[0024] Figure 3 This is a diagram of the three-stage baking process of the present invention.

[0025] Figure 4 This is a flowchart of the orthogonal experiment optimization process of the present invention.

[0026] Figure 5 This is a schematic diagram of the structure for real-time monitoring and dynamic adjustment according to the present invention.

[0027] Figure 6 This is a comparison chart showing the effect of the differentiated baking process (three-stage baking process) of the present invention and the traditional uniform baking process on the proportion of high-grade tobacco.

[0028] Figure 7 This diagram illustrates the comparison of the effects of the differentiated baking process (three-stage baking process) of the present invention and the traditional uniform baking process on the overall quality score index. Detailed Implementation

[0029] The following specific embodiments illustrate the implementation of the present invention. Those skilled in the art can understand the advantages and effects of the present invention from the content disclosed in this specification. The present invention can be implemented or applied through other different specific embodiments, and various details in this specification can also be modified and changed based on different viewpoints and applications without departing from the spirit of the present invention. Furthermore, the accompanying drawings of the present invention are for simple illustrative purposes only and are not depictions of actual dimensions; this is stated beforehand. The following embodiments will further describe the relevant technical content of the present invention in detail, but the disclosed content is not intended to limit the scope of protection of the present invention.

[0030] according to Figure 1 , Figure 2 , Figure 3 , Figure 4 , Figure 5 , Figure 6 ,as well as Figure 7 As shown, this embodiment provides a drying process based on the quality of Xiangyan No. 7 fresh tobacco leaves. The drying process includes the following steps: S1. Near-infrared spectral scanning is performed on the harvested fresh tobacco leaves to obtain the near-infrared spectral data of each tobacco leaf; In this embodiment, near-infrared spectral scanning was performed on the harvested fresh Xiangyan No. 7 tobacco leaves. The specific operation was as follows: Lower leaves (5th-8th leaves), middle leaves (9th-14th leaves), and upper leaves (15th-20th leaves) were harvested in the field according to the maturity and location of the leaves. Spectral scanning was performed immediately after harvesting to ensure the original state of the fresh tobacco leaves. 30-50 representative leaves from each location were randomly selected as samples. Before scanning, the surface of the tobacco leaves was gently wiped with a clean, soft cloth to remove surface dust and impurities without damaging the leaf surface structure. The tobacco leaves were laid flat on the scanning table, ensuring they were smooth and wrinkle-free to avoid spectral signal distortion caused by leaf bending.

[0031] A portable near-infrared spectrometer (e.g., the FieldSpec series from ASD, USA, or the MPA II from Bruker, Germany) was used, with a scanning band of 900-1100 nm, a preferred spectral resolution of 2 nm, a preferred scanning interval of 1 nm, and an integration time of 100-200 ms for each wavelength point.

[0032] Optionally, the baking process further includes: in step S1, the near-infrared spectral scanning uses a wavelength range of 900~1100nm.

[0033] Scanning environment requirements: Room temperature controlled at 20~25℃, relative humidity controlled at 40~60%, and stable light conditions. Three measurement points were selected on each side of the midrib of each tobacco leaf, and the average spectrum was taken as the representative spectrum of that tobacco leaf to improve data representativeness and stability. Each sample was scanned three times, and the average value was taken as the final spectral data to reduce random errors. The spectral data obtained for each tobacco leaf sample was a 201-dimensional vector (900~1100nm, 1nm interval, totaling 201 data points).

[0034] To eliminate noise and baseline drift in the spectral data, the raw spectral data underwent the following preprocessing: ① Standard Normal Variable Transform (SNV): This eliminates spectral baseline drift and nonlinear effects caused by scattering from the sample surface. An SNV transformation is performed on each spectral vector x: In the formula, The mean of the spectral vector. σ is the standard deviation of the spectral vector, and its value is calculated according to the following formula: ②Savitzky-Golay smoothing filter: Removes high-frequency noise from the spectrum while preserving the main spectral features. The following parameters are used: polynomial order: 2 (quadratic polynomial), window width: 11 points (corresponding to a 5nm range), derivative order: 0 (i.e., smoothing, no differentiation); for the 11 data points within the window, at the center point j, the smoothed value is calculated as follows: In the formula, cᵢ represents the Savitzky-Golay convolution coefficients. For an 11-point window and a quadratic polynomial, the coefficients are: c = [-36, 9, 44, 69, 84, 89, 84, 69, 44, 9, -36] / 429. For the first 5 and last 5 data points at both ends of the spectrum, a mirror continuation method is used to form a complete window.

[0035] The spectral data of all samples are normalized to the [0,1] interval. Global minimum and maximum values ​​are calculated from all samples and wavelengths in the training set: x_min = min{absorbance values ​​of all samples and wavelengths in the training set}; x_max = max{absorbance values ​​of all samples and wavelengths in the training set}. For each wavelength in both the training and test sets, the same x_min and x_max are used for normalization: x_norm = (x - x_min) / (x_max - x_min); where x is the absorbance value of the i-th wavelength after SNV transformation and SG smoothing, and x_min and x_max are the minimum and maximum values ​​among all sample spectral data, respectively. After preprocessing, a 201-dimensional normalized spectral vector for each tobacco leaf sample is obtained, which serves as input data for subsequent analysis.

[0036] S2. Divide the near-infrared spectral data into a training set and a test set; The specific method is as follows: To ensure that the proportions of tobacco leaves from different parts (lower leaves, middle leaves, and upper leaves) in the training and test sets are consistent with the overall population, a stratified random sampling method is adopted. The specific steps are: ① Divide all samples into three strata according to the tobacco leaf parts; ② Within each stratum, randomly select 70% of the samples as the training set and the remaining 30% as the test set; ③ Combine the training and test set samples from each stratum to obtain the final training and test sets.

[0037] Optionally, the baking process further includes: in step S2, dividing the near-infrared spectral data of each tobacco leaf into a training set and a test set in a ratio of 7:3.

[0038] Assuming a total of 120 tobacco leaf samples are collected, including 40 lower leaves, 50 middle leaves, and 30 upper leaves, the partitioning result is: training set: 28 lower leaves, 35 middle leaves, and 21 upper leaves, totaling 84 samples; Test set: 12 lower leaves, 15 middle leaves, and 9 upper leaves, for a total of 36 samples; The training set is used to train the self-organizing neural network model and optimize the baking process parameters; the test set is used to verify the model's generalization performance and the effect of process optimization.

[0039] During the partitioning process, a random seed (e.g., seed=42) is set to ensure that the data partitioning results are consistent each time the program is run, which facilitates the repeatability of the experiment and the comparison of results.

[0040] In this embodiment, unsupervised clustering of tobacco leaves is performed by combining near-infrared spectral scanning with a self-organizing neural network. This allows tobacco leaves of different qualities to be objectively and quickly divided into three subsets: easy-to-roast, suitable-to-roast, and difficult-to-roast. This ensures the accuracy and consistency of tobacco leaf classification and solves the problems of uniform parameter control and ignoring differences in tobacco varieties and chemical components in the prior art.

[0041] S3. Use the near-infrared spectral data of each tobacco leaf in the training set as the input vector, input it into the self-organizing neural network model for clustering, and divide the tobacco leaves into 3 subsets; Optionally, the baking process further includes: in step S3, the number of neurons in the competitive layer of the self-organizing neural network model is 3, and the tobacco leaves are divided into 3 subsets. That is, in this embodiment, it is preferable to divide the tobacco leaves into 3 subsets.

[0042] like Figure 2 As shown, the SOM neural network consists of two layers: an input layer and a competition layer (also known as an output layer or Kohonen layer). Input layer: 201 neurons, corresponding to the dimensions of the preprocessed spectral data (900-1100nm, 1nm interval). Competition layer: There are 3 neurons arranged in a 1×3 one-dimensional linear structure, and each neuron corresponds to a set of tobacco leaves; Meanwhile, the adjacency relationship is: d 1,2 = 1, d 2,3 = 1, d 1,3 = 2; The competition layer adopts a one-dimensional topology with clear adjacency relationships: neuron 1 is adjacent to neuron 2, neuron 2 is adjacent to neuron 1 and neuron 3, and neuron 3 is adjacent to neuron 2.

[0043] Each neuron in the competition layer has a 201-dimensional weight vector W. j (j=1,2,3), fully connected to the input layer.

[0044] Optionally, the working process of the self-organizing neural network model includes the following steps: S31: Initialization, using random numbers to set the initial weights between the input layer and the competition layer; In this method, a random initialization method is used to set the initial weight values ​​between the input layer and the competition layer. Specifically, for each neuron j (j=1,2,3) in the competition layer, its weight vector W j Each component Wj,k (k=1,2,...,201) Randomly draw W from a uniform distribution U(0,1). j,k After initializing ~U(0,1), normalize each weight vector: Ensure that all value vectors have a magnitude of 1, which is on the same scale as the normalized input vector.

[0045] S32: Euclidean distance is calculated between each neuron in the input layer and the competition layer. The j-th neuron in the competition layer and the input vector x... i The distance is calculated as follows: Determine the winning neuron: Calculate the Euclidean distance between the input vector and the weight vectors of neurons in each competing layer. ; In the formula, x i For the input vector, W j Let j* be the weight vector of the j-th neuron in the competition layer. The neuron with the smallest distance is calculated and selected as the winning neuron. Find the neuron with the smallest distance as the winning neuron j*: ; S33: The weights of the winning neuron j* and its neighboring neurons are corrected using the following formula: ; In the formula, W j (t+1) represents the corrected weight, W j (t) represents the current weights, and α(t) represents the learning rate. It is a neighborhood function; The learning rate uses exponential decay: In this embodiment, α0 = 0.7. =150.

[0046] In addition, the neighborhood function uses a Gaussian function: ;in, Let σ(t) be the distance between neuron j and the winning neuron j* in the competition layer; σ(t) is the neighborhood radius. In this embodiment, σ0 = 1.5. = 150.

[0047] S34: Repeat steps S32~S33 until the preset number of iterations is reached or the weight change is less than the preset threshold.

[0048] In this embodiment, the iteration count or weight change being less than a preset threshold is defined as meeting one of the following termination conditions: 1) reaching a preset iteration count of 500; 2) the maximum weight change over 10 consecutive iterations being less than 10.-4 In this embodiment, the baking process further includes: in step S34, the preset number of iterations is 500.

[0049] After training, the 84 samples in the training set were classified into three subsets: subset 1: 28 samples; subset 2: 32 samples; subset 3: 24 samples.

[0050] After training is complete, all training set samples are re-input into the trained SOM network, the distance between each sample and each neuron is calculated, and the sample is assigned to the subset corresponding to the nearest neuron.

[0051] After training is completed, all training set samples are re-input into the trained SOM network, the distance between each sample and each neuron is calculated, and the sample is assigned to the subset corresponding to the nearest neuron.

[0052] Calculate the distance between the centers of different subsets; the larger the distance, the better the inter-class separation.

[0053] The clustering quality is comprehensively evaluated, with a smaller DBI value indicating better clustering results. In this embodiment, the average intra-cluster distances of the three subsets are 0.23, 0.28, and 0.31, respectively, the minimum inter-cluster distance is 0.85, and the DBI index is 0.67, indicating good clustering results.

[0054] In this embodiment, chemical composition was detected for three subsets, and the results are shown in Table 1 below: Table 1: chemical composition Subset 1 Subset 2 Subset 3 Reducing sugar (%) 22.3±2.1 17.5±1.8 12.4±2.3 Total nitrogen (%) 1.75±0.15 2.15±0.18 2.85±0.25 Nicotine (%) 2.25±0.18 2.68±0.22 3.45±0.35 Potassium (%) 2.75±0.20 2.25±0.18 1.85±0.22 Sugar-base ratio 10.0±1.2 6.5±0.8 3.6±0.5 As can be seen from the table above, subset 1 (easy to roast): high reducing sugar content, low total nitrogen content, high sugar-to-alkaloid ratio, mainly consisting of fully mature middle and lower leaves; subset 2 (suitable to roast): moderate chemical composition, mainly consisting of normally mature middle and upper leaves; subset 3 (difficult to roast): low reducing sugar content, high total nitrogen and nicotine content, low sugar-to-alkaloid ratio, mainly consisting of underripe or overripe upper leaves.

[0055] By using unsupervised clustering with SOM neural network, the fresh tobacco leaves of Xiangyan No. 7 were successfully divided into three subsets with clear differences in chemical composition based on their near-infrared spectral characteristics, laying the foundation for subsequent optimization of differentiated roasting process.

[0056] In addition, in this embodiment, the baking process further includes: establishing a chemical composition prediction model using the near-infrared spectral data to predict the chemical composition content of the tobacco leaves, wherein the chemical composition includes at least reducing sugar, total sugar, nicotine, total nitrogen, and potassium; The baking difficulty index (HDI) is calculated based on the content of the chemical components. HDI = (reducing sugar × α+potassium × β-nicotine × γ) / (total nitrogen × δ); In the formula, the value of α ranges from 0.40 to 0.50, the value of β ranges from 0.30 to 0.40, the value of γ ranges from 0.20 to 0.30, and the value of δ ranges from 0.25 to 0.35. The three training sets are distinguished based on HDI values: First subset: HDI > 8.5, indicating easy-to-roast tobacco leaves; The second subset: 6.0≤HDI≤8.5, represents suitable flue-cured tobacco leaves; Third subset: HDI < 6.0, indicating difficult-to-cook tobacco leaves.

[0057] S4. For the three subsets, optimize the baking process parameters according to the three-stage baking process, which includes the yellowing stage, color-fixing stage, and dry-fiber stage. Each stage has different physiological and biochemical characteristics and process requirements: 1) Yellowing stage: The main physiological and biochemical reactions are chlorophyll degradation, starch hydrolysis into reducing sugars, and partial decomposition of proteins, causing the tobacco leaves to change from green to yellow. This stage requires appropriate temperature (35~48℃) and high humidity (relative humidity 80~95%) to maintain the physiological activity of the tobacco leaves and promote material transformation.

[0058] 2) Color Fixing Period: This stage mainly involves fixing the color of the tobacco leaves, forming aroma substances, and partially evaporating moisture. During this stage, the temperature is gradually increased (48~68℃) and the humidity is decreased (relative humidity 60~75%) to fix the shape and color of the tobacco leaves and complete the main chemical transformations.

[0059] 3) Drying stage: This stage mainly involves drying the main veins and stems of the tobacco leaves to reduce the moisture content to a suitable range (12-14%). Higher temperatures (68-72℃), lower humidity (40-50% relative humidity), and higher ventilation are used during this stage to accelerate moisture evaporation.

[0060] In this embodiment, by optimizing parameters such as temperature, humidity, and ventilation during the yellowing, color-fixing, and drying stages for different subsets, the tobacco leaves of each subset can fully complete the physiological and biochemical reactions of chlorophyll degradation, starch hydrolysis, and protein decomposition under suitable process conditions. This ensures the sufficiency and coordination of tobacco leaf material transformation and solves the problem of insufficient material transformation of tobacco leaves of different qualities in traditional processes.

[0061] Optionally, the baking process further includes: in step S4: 1) Yellowing period: the temperature is gradually increased from the initial temperature to the yellowing termination temperature, and the humidity and ventilation are controlled to degrade the chlorophyll in the tobacco leaves, turn the tobacco leaves yellow, and hydrolyze the starch into reducing sugars; 2) Color fixing period: the temperature continues to rise to the color fixing temperature, the humidity is reduced, and the ventilation is increased to fix the color of the tobacco leaves, complete the material transformation, and form aroma substances; 3) Drying period: the temperature is raised to the drying temperature, the humidity is further reduced, and the maximum ventilation is applied to dry the main vein of the tobacco leaves and reduce the overall moisture content to 12-14%.

[0062] like Figure 3 As shown, the optimization parameters include temperature, humidity, ventilation volume and temperature stabilization time at each stage. With the quality of the cured tobacco leaves as the target, the curing process curves of the corresponding subset are established. For each subset, orthogonal experimental design was used to optimize the baking process parameters. The optimized parameters included: Yellowing period: initial temperature, final temperature, heating rate, initial humidity, and temperature stabilization time; Color fixation period: initial temperature, final temperature, heating rate, humidity control, and temperature stabilization time; Drying period: temperature, humidity, ventilation, and temperature stabilization time; Based on the physiological and biochemical characteristics of Xiangyan No. 7 tobacco leaves and the characteristic differences of the three subsets, the adjustable range of process parameters at each stage was determined, as shown in Table 2 for the parameter range during the yellowing stage: Table 2: parameter Easy-to-bake type (subset 1) Suitable for baking (subset 2) Difficult to bake (subset 3) Initial temperature (°C) 35~38 36~39 37~40 Termination temperature (°C) 46~48 47~49 48~50 Heating rate (°C / 12h) 0.8~1.5 0.6~1.2 0.3~0.8 Initial humidity ΔT (°C) 0~1 0~1.5 0.5~2 Total duration (h) 55~75 70~90 85~110 In the table above, ΔT is the difference between dry-bulb temperature and wet-bulb temperature. The smaller ΔT is, the higher the relative humidity.

[0063] The range of color fixation period parameters is shown in Table 3: Table 3: parameter Easy-to-bake type (subset 1) Suitable for baking (subset 2) Difficult to bake (subset 3) Initial temperature (°C) 48~50 48~50 48~50 Termination temperature (°C) 54~56 55~58 56~60 Heating rate (°C / 12h) 1.5~2.5 1.2~2.0 1.0~1.8 Initial humidity ΔT (°C) 3~8 4~9 5~10 Total duration (h) 36~48 42~54 48~60 The parameter ranges for the dry reinforcement period are shown in Table 4: Table 4: parameter Easy-to-bake type (subset 1) Suitable for baking (subset 2) Difficult to bake (subset 3) Temperature (°C) 68~70 69~71 70~72 Humidity ΔT (°C) 12~14 12~15 13~15 Ventilation volume (opening degree of dehumidification window) 80~100% 85~100% 90~100% Total duration (h) 12~20 15~24 18~28 Taking subset 1 (easy-to-bake type) as an example, five main factors are selected, each with three levels, using L9(3) 4 An orthogonal array was used for experimental design. Key factors included the initial temperature of the yellowing period, the heating rate, the total duration, and the termination temperature of the color-fixing period. An L9(3)2 orthogonal array was employed. 4 An orthogonal array was used to design the experiment. Each group of experiments was cured for 3 batches, with 2500-3000 kg of tobacco leaves per batch. The results are shown in Table 4 (factors and levels) and Table 5 (orthogonal experiment results). As shown in Table 5: factor Level 1 Level 2 Level 3 A - Initial temperature during the yellowing stage (°C) 35 36.5 38 B - Warming rate during the yellowing stage (°C / 12h) 0.8 1.1 1.4 C - Total duration of the yellowing period (h) 60 66 72 D - Color fixation period termination temperature (°C) 54 55 56 like Figure 4As shown, nine groups of experiments were arranged according to an orthogonal array. Each group of experiments was cured according to the corresponding parameter combination (each group was cured 3 times, with 2500-3000 kg of tobacco per batch). The comprehensive quality evaluation of the cured tobacco leaves was carried out to obtain a comprehensive score. The comprehensive score was calculated by weighting chemical indicators (weight 25%), physical indicators (weight 20%), sensory indicators (weight 35%), and economic indicators (weight 20%), as shown in Table 6. Table 6: Test No. A B C D Overall score 1 1 1 1 1 83.3 2 1 2 2 2 88.5 3 1 3 3 3 85.6 4 2 1 2 3 86.8 5 2 2 3 1 92.4 6 2 3 1 2 83.9 7 3 1 3 2 80.5 8 3 2 1 3 86.2 9 3 3 2 1 88.4 The comprehensive score is calculated using the following formula: ; In the formula, the chemical index S 化学 To test the reducing sugar, total nitrogen, nicotine, potassium, and sugar-to-alkali ratio of flue-cured tobacco leaves, national standards such as GB / T 23203-2008 were used, and scores were awarded based on the degree of deviation of each indicator from the target range (out of 100 points). Physical index S 物理 To evaluate and score the color, oil content, maturity, and grade of flue-cured tobacco leaves according to GB 2635-1992 standard (out of 100 points). Sensory index S 感官 The evaluation and scoring of the aroma, aroma intensity, off-flavors, irritation, and aftertaste of the flue-cured tobacco leaves were conducted by 5-7 evaluators according to the GB / T 23356-2009 standard (out of 100 points). Economic indicator S 经济 The score is based on the proportion of premium cigarettes and the average price (out of 100).

[0064] Calculate the average score K for each level of each factor. ji The average score of all trials at the i-th level of factor j and the range R (R = max(K) - min(K)) are obtained in Table 7; Table 7: factor <![CDATA[K1]]> <![CDATA[K2]]> <![CDATA[K3]]> Range R optimal level A - Initial temperature 85.8 87.7 85.0 2.7 Level 2 (36.5℃) B - Heating rate 83.5 89.0 86.0 5.5 Level 2 (1.1℃ / 12h) C - Duration of yellowing 84.4 87.9 86.2 3.5 Level 2 (66h) D-Fixing Temperature 88 84.3 86.2 3.7 Level 1 (54℃) As can be seen from the table above, the optimal parameter combination is: A2B2C2D1, which means the initial temperature of the yellowing period is 36.5℃, the heating rate is 1.1℃ / 12h, the total duration is 66h, and the ending temperature of the color fixation period is 54℃.

[0065] The order of influence factors is as follows: heating rate (R=5.5) > color fixation temperature (R=3.7) > yellowing time (R=3.5) > initial temperature (R=2.7).

[0066] Based on the optimal parameters determined by orthogonal experiments, the complete baking process curves are shown in Table 8 below: Table 8: stage Time period Dry bulb temperature (°C) △T（℃） Ventilation status Key control points Yellowing period 0~12h 36.5 0.5 closure Stable temperature turns yellow 12~24h 37.5 1.0 closure Promotes yellowing 24~36h 39.0 1.5 closure Accelerates yellowing 36~48h 41.0 2.0 Young Master 70% of the leaves turned yellow 48~60h 43.5 2.5 Young Master 85% of the leaves turned yellow 60~66h 46.0 3.0 Zhongkai 90% of the leaves turned yellow Color fixation period 66~72h 48.0 4.0 Zhongkai Start color fixing 72~78h 50.0 5.0 Zhongkai Leaf color fixation 78~84h 52.0 6.0 Open Main vein 1 / 3 trunk 84~90h 54.0 7.0 Open Main vein 1 / 2 trunk 90~96h 54.0 8.0 Open Main vein 2 / 3 trunk Drying period 96~102h 68.0 12.0 Full open Rapid dehydration of the main vein 102~108h 69.0 13.0 Full open The main vein is basically dry 108~114h 69.0 14.0 Full open The main vein is completely dry As shown in the table above, the total baking time is 114 hours. After baking using this process curve, the quality indicators of the tobacco leaves are: a comprehensive score of 88.6 points, a proportion of top-grade tobacco of 68.5%, a proportion of medium-grade tobacco of 26.3%, a proportion of low-grade tobacco of 5.2%, and an average price of 45.8 yuan / kg.

[0067] Among them, the differentiated roasting process avoids over-roasting of easy-to-roast tobacco leaves and under-roasting of difficult-to-roast tobacco leaves, which significantly increases the proportion of high-quality tobacco and greatly reduces the proportion of low-quality tobacco. This ensures that the sensory quality of the roasted tobacco leaves, such as color, aroma, and oil content, meets the standards of high-quality tobacco, and solves the problem of high proportion of low-quality tobacco and difficulty in highlighting flavor in existing technologies.

[0068] Using the same orthogonal experimental method (similar to subset 1), the optimal process parameters for subset 2 were obtained as follows: initial temperature during the yellowing stage: 37.5℃, heating rate: 0.9℃ / 12h, total duration: 78h; final temperature during the color-fixing stage: 56℃, total duration: 48h; temperature during the drying stage: 70℃, total duration: 20h. The total curing time was 146 hours. After curing using this process, the tobacco leaf quality indicators were: a comprehensive score of 86.4 points, and a proportion of top-grade tobacco of 63.8%.

[0069] Using the same orthogonal experimental method (similar to subset 1), the optimal process parameters for subset 3 were obtained as follows: initial temperature of yellowing period 38.5℃, heating rate 0.6℃ / 12h, total duration 96h; ending temperature of color setting period 58℃, total duration 54h; drying temperature 71℃, total duration 24h. The total baking time was 174 hours.

[0070] After being cured using this process, the quality indicators of the tobacco leaves are as follows: a comprehensive score of 83.7 points, with 58.2% being of superior quality.

[0071] The stability and generalization ability of the process curves were verified using 36 tobacco leaf samples from the test set. After classifying the test set samples into corresponding subsets using the SOM network, the corresponding curing process curves were used for curing, and the results are shown in Table 8.

[0072] Table 9: subset index training set test set deviation Subset 1 Overall score 88.6 86.8 -2.0% Percentage of premium cigarettes (%) 68.5 65.2 -4.8% Subset 2 Overall score 86.4 84.9 -1.7% Percentage of premium cigarettes (%) 63.8 61.5 -3.6% Subset 3 Overall score 83.7 82.1 -1.9% Percentage of premium cigarettes (%) 58.2 56.8 -2.4% As can be seen from the table above, the deviation of all indicators is less than 5%, indicating that the process curve has good stability and generalization ability.

[0073] Optionally, further optimization of the baking process for different subsets includes a segmented temperature and humidity control strategy based on biochemical reaction kinetics: The yellowing period was divided into four stages: chlorophyll degradation, starch hydrolysis, protein hydrolysis, and polyphenol oxidation. The temperature control range for each stage corresponds to the optimal activity temperature range for chlorophyllase, amylase, protease, and polyphenol oxidase, respectively. Chlorophyll degradation stage: temperature 35~40℃, residence time accounts for 30~35% of the total yellowing period; Starch hydrolysis stage: temperature 40~45℃, residence time accounts for 35~40% of the total yellowing period; Protein hydrolysis stage: temperature 42~48℃, residence time accounts for 15~20% of the total yellowing period; Polyphenol oxidation stage: temperature 45~50℃, residence time accounts for 10~15% of the total yellowing period.

[0074] By employing a four-level refined control strategy based on biochemical reaction kinetics, reactions such as chlorophyll degradation, starch hydrolysis, and polyphenol oxidation can proceed within their respective optimal temperature ranges. This ensures the refinement and optimization of the internal material transformation of tobacco leaves, further enhancing the quality and intrinsic properties of the cured tobacco leaves.

[0075] To verify the technical effect of the differentiated roasting process, a control group was set up: the tobacco leaves of the three subsets were mixed and roasted using the traditional unified three-stage roasting process, as shown in Table 10. Table 10: index Differentiated baking Traditional uniform baking Increase Overall score 86.4 78.2 +10.5% Percentage of premium cigarettes (%) 64.2 48.5 +15.7pp Low-grade smoke ratio (%) 6.8 18.3 -11.5pp Average price (RMB / kg) 45.1 38.6 +16.8% As can be seen from the table above, differentiated baking processes significantly improve baking quality and economic benefits.

[0076] Through step S4, differentiated baking process curves corresponding to the three subsets were successfully established, providing a scientific basis and operable technical solutions for actual production.

[0077] By constructing a complete technology chain encompassing spectral scanning, intelligent classification, process optimization, parameter matching, and automatic roasting, the analysis of tobacco leaf components, classification judgment, and optimization of process parameters are organically combined, ensuring the systematic and scientific nature of the roasting process and solving the problem that existing technologies focus on stage-based temperature control and have not formed a complete control system.

[0078] S5. The newly harvested tobacco leaves are scanned by near-infrared spectroscopy, classified by the self-organizing neural network model, and their respective subsets are determined. The yellowing period, color fixing period and dry rib period are executed sequentially according to the baking process curve corresponding to the subset.

[0079] Optionally, the baking process in step S5 also includes a real-time monitoring and dynamic adjustment step: real-time monitoring of the yellowing degree, moisture content and / or volatile substance concentration of the tobacco leaves through sensors; In this embodiment, a dense curing barn is used for curing. The internal dimensions of the barn are 6.5m long × 2.9m wide × 3.5m high. The barn is equipped with multiple horizontal shelves (usually 8-12 layers) for hanging tobacco leaves.

[0080] Tobacco leaf loading method: The tobacco leaves are hung using tobacco clips, with each clip (approximately 1.2m long) holding 15-20 tobacco leaves, hanging down with the leaf tips pointing downwards. The woven tobacco clips are then hung horizontally on each shelf, with a spacing of 8-10cm between clips. The verticality between each shelf is 30-40cm. The tobacco is loaded into the curing oven according to the principle of lower leaves on the upper shelf and upper leaves on the lower shelf (Note: "lower leaves on the upper shelf" refers to the lower part of the tobacco plant, specifically the upper shelf of the curing barn).

[0081] After the tobacco is installed, a multi-layered tobacco leaf structure is formed inside the curing barn from bottom to top. Due to the physical property of hot air rising, there is a humidity gradient in the vertical direction inside the curing barn: the upper layer has a higher temperature and lower humidity, while the lower layer has a lower temperature and higher humidity.

[0082] Near-infrared spectral sensors (wavelengths 550nm and 650nm) or RGB color sensors are employed. Specifically, three near-infrared spectral sensors (or RGB color sensors) are installed on the inner wall of the curing barn to monitor tobacco leaves at different vertical heights: Upper layer sensor: installed 3.0m above the ground (0.5m from the top of the barn), monitoring tobacco leaves on the upper shelves (layers 10-12); Middle layer sensor: installed 1.8m above the ground, monitoring tobacco leaves on the middle shelves (layers 5-7); Lower layer sensor: installed 0.8m above the ground, monitoring tobacco leaves on the lower shelves (layers 2-3). Simultaneously, the probes of the near-infrared spectral sensors (wavelengths 550nm and 650nm) or RGB color sensors are positioned 20-30cm above the tobacco leaf layer, with the probes pointing horizontally towards the center of the tobacco leaves. The yellowing index YI = R is calculated by measuring reflectance. 650 / R 550 Thus, the yellowing index of different layers can be obtained; In addition, a weight sensor (weighing method) was used to monitor the thickness of the tobacco leaves. Two to three tobacco clips were selected as key monitoring samples at each of the three height levels: upper layer: 2-3 tobacco clips from shelves 10-12; middle layer: 2-3 tobacco clips from shelves 5-7; lower layer: 2-3 tobacco clips from shelves 2-3. Position sensors were installed between the tobacco sticks and the shelves of these monitoring clips (or the clips were suspended from the hooks of an electronic scale) to measure the weight of the clips in real time. The moisture content was calculated based on the weight change: W(%) = [(m_t - m_d) / m_d] × 100%, where m_t is the current time and m_d is the dry matter weight (based on the initial outdoor material content guideline).

[0083] All sensor data is transmitted to the automatic control system via timers or wireless means, with a sampling frequency of once every 5-10 minutes.

[0084] In this embodiment, after collecting data from the upper, middle, and lower layers of sensors, the speed of data from these three layers is calculated as the yellowing speed and advance speed of the entire drying chamber: The target is compared with the expected value, and dynamic adjustments are performed when the deviation exceeds a threshold.

[0085] Calculate the actual yellowing rate The expected rate of yellowing Deviation, and / or actual dehydration rate With the expected dehydration rate Deviation; The expected yellowing rate is determined by the following formula: In this embodiment, different parameters are set for different subsets. When determining the expected yellowing speed of different subsets, the relevant parameters can be queried through Table 11 and substituted into the calculation. Table 11: subset Maximum YI k <![CDATA[t0]]> Subset 1 (Easy-to-bake type) 6.5 0.12 40 Subset 2 (Suitable for baking) 6.2 0.10 48 Subset 3 (Difficult to bake) 5.8 0.08 58 Meanwhile, the expected yellowing rate In the baking test optimization process in step S4, the yellowing rate data of tobacco leaves in each subset at different time periods are recorded, and the average value of multiple tests of the same subset is taken as the expected yellowing rate of that subset.

[0086] For example, the expected yellowing rates of subset 1 (easy-to-bake type) in the time periods of 0~12h, 12~24h, and 24~36h are 2.5% / h, 4.2% / h, and 8.3% / h, respectively (based on the experimental data of the training set).

[0087] Expected dehydration rate Similarly, for the determination of the advance rate, the advance rate data recorded from multiple baking tests of the same subset (e.g., 9 sets of tests for subset 1, 3 batches per set, for a total of 27 batches) are statistically analyzed, and the prediction for each stage is calculated as the expected advance rate for that subset. For example, the expected dehydration rates for subset 1 in the yellowing stage, color setting stage, and dry tack stage are 0.38% / h, 1.33% / h, and 0.39% / h, respectively.

[0088] Similarly, subsets 2 and 3 are determined using the same method to determine their respective expected dehydration rates. In this embodiment, the method of subset 1 can be used to determine the subset, which will not be elaborated here.

[0089] Every 10-15 minutes, the control system calculates the deviation between the actual speed and the expected speed: Actual yellowing rate: In the formula, Δt is the time interval. The current rate at which it turns yellow. The rate at which it turned yellow in the previous moment; Actual dehydration rate: In the formula, Δt is the time interval. The dehydration rate at the current moment. The dehydration rate at the previous moment; When the deviation exceeds a preset threshold, the temperature, humidity, and / or ventilation volume are dynamically adjusted according to preset adjustment rules, which include: When the actual yellowing rate When the value is greater than 1.5 times the expected value, lower the temperature by 1-2°C, increase the humidity by 1°C, and reduce the ventilation by 10-20%. When the actual yellowing rate If the humidity is less than 0.7 times the expected value, increase the temperature by 1-2°C, decrease the humidity by 0.5°C, and increase the ventilation by 5-10%. When the actual dehydration rate When the humidity is >3% / h, lower the temperature by 2℃, increase the humidity by 2℃, and reduce the ventilation volume by 20-30%. When the actual dehydration rate When the humidity is less than 0.5% / h, increase the temperature by 2℃, decrease the humidity by 1℃, and increase the ventilation by 15~20%.

[0090] In this embodiment, at least 30 minutes are allowed between two adjustments to avoid parameter fluctuations caused by frequent adjustments.

[0091] In addition, by establishing a standardized baking process curve based on training set optimization, tobacco leaves of the same subset are baked according to the same process parameters, which ensures the stability and repeatability of baking quality between batches and solves the problems of large human factors and inconsistent baking effects in traditional baking processes.

[0092] In this embodiment, through real-time monitoring and dynamic adjustment functions, the yellowing rate and dehydration rate during the baking process can be automatically adjusted according to the actual deviation, ensuring precise control and adaptive optimization of the baking process. This solves the problem of lacking dynamic parameter adjustment in traditional methods and further improves the stability of the process mechanism.

[0093] The content disclosed above is only a preferred and feasible embodiment of the present invention, and is not intended to limit the scope of protection of the present invention. Therefore, all equivalent technical changes made based on the content of the present invention specification and drawings are included within the scope of protection of the present invention. Furthermore, the elements therein can be updated as technology develops.

Claims

1. A curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves, characterized in that, The baking process includes the following steps: S1. Near-infrared spectral scanning is performed on the harvested fresh tobacco leaves to obtain the near-infrared spectral data of each tobacco leaf; S2. Divide the near-infrared spectral data into a training set and a test set; S3. Use the near-infrared spectral data of each tobacco leaf in the training set as the input vector, input it into the self-organizing neural network model for clustering, and divide the tobacco leaves into 3 subsets; S4. For the three subsets, optimize the baking process parameters according to the three-stage baking process, which includes the yellowing period, the color-fixing period, and the drying period. The optimized parameters include temperature, humidity, ventilation, and temperature stabilization time at each stage. With the quality of the cured tobacco leaves as the target, the curing process curves for the corresponding subset are established. S5. The newly harvested tobacco leaves are scanned by near-infrared spectroscopy, classified by the self-organizing neural network model, and their respective subsets are determined. The yellowing period, color fixing period and dry rib period are executed sequentially according to the baking process curve corresponding to the subset.

2. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 1, characterized in that, The baking process further includes: in step S1, the near-infrared spectral scanning uses a wavelength range of 900~1100nm.

3. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 2, characterized in that, The baking process further includes: in step S2, dividing the near-infrared spectral data of each tobacco leaf into a training set and a test set in a ratio of 7:

3.

4. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 3, characterized in that, The baking process further includes: in step S3, the number of neurons in the competitive layer of the self-organizing neural network model is 3, and the tobacco leaves are divided into 3 subsets.

5. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 4, characterized in that, The working process of the self-organizing neural network model includes the following steps: S31: Initialization, using random numbers to set the initial weights between the input layer and the competition layer; S32: Euclidean distance is calculated between each neuron in the input layer and the competition layer. The distance between the j-th neuron in the competition layer and the input vector xi is calculated as follows: ； In the formula, x i For the input vector, W j Let j* be the weight vector of the j-th neuron in the competition layer. The neuron with the smallest distance is calculated and selected as the winning neuron. S33: The weights of the winning neuron j* and its neighboring neurons are corrected using the following formula: ； In the formula, W j (t+1) represents the corrected weight, W j (t) represents the current weights, and α(t) represents the learning rate. It is a neighborhood function; S34: Repeat steps S32~S33 until the preset number of iterations is reached or the weight change is less than the preset threshold.

6. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 5, characterized in that, The baking process further includes: in step S34, the preset number of iterations is 500.

7. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 6, characterized in that, The baking process further includes: in step S4: Yellowing period: The temperature is gradually increased from the initial temperature to the yellowing end temperature. Humidity and ventilation are controlled to degrade the chlorophyll in the tobacco leaves, cause the tobacco leaves to turn yellow, and hydrolyze the starch into reducing sugars. Color fixation period: The temperature continues to rise to the color fixation temperature, the humidity is reduced, and the ventilation is increased so that the tobacco leaves can fix their color, complete the material transformation, and form aroma substances; Drying stage: The temperature rises to the drying temperature, the humidity is further reduced, and the maximum ventilation is used to dry the main vein of the tobacco leaves and reduce the overall moisture content to 12-14%.

8. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 7, characterized in that, The roasting process further includes: establishing a chemical composition prediction model based on the near-infrared spectral data to predict the chemical composition content of tobacco leaves, wherein the chemical composition includes at least reducing sugar, total sugar, nicotine, total nitrogen, and potassium; The baking difficulty index (HDI) is calculated based on the content of the chemical components. The baking difficulty index (HDI) is calculated based on the content of the chemical components. HDI = (reducing sugar × α+potassium × β-nicotine × γ) / (total nitrogen × δ); In the formula, the value of α ranges from 0.40 to 0.50, the value of β ranges from 0.30 to 0.40, the value of γ ranges from 0.20 to 0.30, and the value of δ ranges from 0.25 to 0.

35. The three training sets are distinguished based on HDI values: First subset: HDI > 8.5, indicating easy-to-roast tobacco leaves; Second subset: 6.0 ≤ HDI ≤ 8.5, which is suitable for curing tobacco leaves; Third subset: HDI < 6.0, indicating difficult-to-cook tobacco leaves.

9. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 8, characterized in that, Further optimization of the baking process for different subsets includes a segmented temperature and humidity control strategy based on biochemical reaction kinetics: The yellowing period was divided into four stages: chlorophyll degradation, starch hydrolysis, protein hydrolysis, and polyphenol oxidation. The temperature control range for each stage corresponds to the optimal activity temperature range for chlorophyllase, amylase, protease, and polyphenol oxidase, respectively. Chlorophyll degradation stage: temperature 35~40℃, residence time accounts for 30~35% of the total yellowing period; Starch hydrolysis stage: temperature 40~45℃, residence time accounts for 35~40% of the total yellowing period; Protein hydrolysis stage: temperature 42~48℃, residence time accounts for 15~20% of the total yellowing period; Polyphenol oxidation stage: temperature 45~50℃, residence time accounts for 10~15% of the total yellowing period.

10. The curing process based on the quality of Xiangyan No. 7 fresh tobacco leaves according to claim 9, characterized in that, The baking process in step S5 also includes real-time monitoring and dynamic adjustment steps: The degree of yellowing, moisture content and / or concentration of volatile substances in tobacco leaves are monitored in real time using sensors. Calculate the actual yellowing rate The expected rate of yellowing Deviation, and / or actual dehydration rate With the expected dehydration rate Deviation; When the deviation exceeds a preset threshold, the temperature, humidity, and / or ventilation volume are dynamically adjusted according to preset adjustment rules, which include: When the actual yellowing rate When the value is greater than 1.5 times the expected value, lower the temperature by 1-2°C, increase the humidity by 1°C, and reduce the ventilation by 10-20%. When the actual yellowing rate If the humidity is less than 0.7 times the expected value, increase the temperature by 1-2°C, decrease the humidity by 0.5°C, and increase the ventilation by 5-10%. When the actual dehydration rate When the humidity is >3% / h, lower the temperature by 2℃, increase the humidity by 2℃, and reduce the ventilation volume by 20-30%. When the actual dehydration rate When the humidity is less than 0.5% / h, increase the temperature by 2℃, decrease the humidity by 1℃, and increase the ventilation by 15~20%.

Images