Sensory quality prediction model based on main internal components of flue-cured tobacco leaves as well as construction method and application of sensory quality prediction model

Abstract

The invention discloses a sensory quality prediction model based on main internal components of flue-cured tobacco leaves as well as a construction method and application thereof. The construction method of the prediction model comprises the following steps: selecting sample flue-cured tobacco leaves and collecting internal component data and sensory quality evaluation data of the sample flue-cured tobacco leaves; a recursive feature elimination algorithm is adopted, and main feature components related to sensory quality are extracted from the internal component data; classifying the extracted characteristic components related to the sensory quality according to the classification condition of the sensory quality evaluation data of the sample flue-cured tobacco leaves to obtain a smoke sensory characteristic component, an aroma sensory characteristic component and a taste sensory characteristic component; and obtaining a trained flue-cured tobacco leaf sensory quality prediction model according to the smoke sensory characteristic components, the aroma sensory characteristic components, the taste sensory characteristic components and the sensory quality evaluation data of the sample flue-cured tobacco leaves. According to the method, the influence of the comprehensive action of multiple components in the flue-cured tobacco leaves on the sensory quality is excavated, the prediction model is established, the accuracy of tobacco leaf sensory quality evaluation is improved, good interpretability is achieved, and a powerful tool is provided for evaluation and production of the tobacco leaf sensory quality.

Description

Technical Field

[0001] The present invention relates to the field of sensory quality prediction of tobacco leaves, and in particular to a sensory quality prediction model based on main internal components of freshly cured tobacco leaves, and a construction method and application thereof. Background Art

[0002] The quality assessment of tobacco leaves is a complex process, which involves chemical composition, appearance, physical properties, sensory quality, etc. Although the evaluation of sensory quality occupies an important position in the industry, this evaluation method is often subjective. Therefore, the industry has been seeking more objective and scientific evaluation methods.

[0003] The chemical composition of tobacco leaves is a key factor in determining their quality. These components not only determine the appearance and physical properties of tobacco leaves, but also directly affect the sensory quality. Research in recent years has shown that the sugars, alkaloids, organic acids, amino acids and other components in tobacco leaves, as well as the degree of balance and coordination between them, have a decisive influence on the quality of tobacco leaves. In addition, the quality of tobacco leaves is also affected by the growing region. Tobacco leaves from different regions differ in chemical composition and quality style.

[0004] Although some studies have been conducted to predict the sensory quality of tobacco leaves by analyzing their chemical composition, most of these studies focused on one or several types of chemicals and ignored the combined effects of multiple components in tobacco leaves.

[0005] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0006] In order to solve the above problems, the present invention provides a sensory quality prediction model based on the main internal components of primary flue-cured tobacco leaves, and a construction method and application thereof.

[0007] The technical solution adopted by the present invention to solve the technical problem is: The sensory quality prediction model based on the main internal components of first-cured tobacco leaves includes A feature input layer, for receiving the intrinsic component data of the target first-cured tobacco leaves; wherein the intrinsic component data includes protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content and starch content; A feature processing layer, for performing feature extraction and feature conversion on the input intrinsic component data to identify feature patterns related to sensory quality; A prediction output layer, used for outputting a prediction result of the sensory quality of the target freshly-cured tobacco leaves based on the characteristic pattern related to the sensory quality; Among them, the sensory quality prediction model based on the main internal components of freshly cured tobacco leaves is obtained through training and optimization based on a machine learning algorithm.

[0008] The present invention adopts the following technical solutions to solve the technical problems: The method for constructing a sensory quality prediction model based on the main internal components of first-cured tobacco leaves includes: According to the administrative division, topography and soil data of the tobacco planting area, the sample first-cured tobacco leaves are selected; wherein the sample first-cured tobacco leaves are obtained by baking fresh tobacco leaves; Collecting the intrinsic component data and sensory quality evaluation data of the sample first-cured tobacco leaves; extracting main characteristic components related to sensory quality from the intrinsic component data using a recursive feature elimination algorithm; According to the classification of the sensory quality evaluation data of the sample freshly cured tobacco leaves, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components; According to the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly-cured tobacco leaves, a sensory quality prediction model for freshly-cured tobacco leaves is obtained based on machine learning algorithm training and optimization.

[0009] Furthermore, the collection of the intrinsic component data and sensory quality evaluation data of the sampled first-cured tobacco leaves includes: Collect the following intrinsic component data of the sample flue-cured tobacco leaves: protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content, and starch content; The following sensory quality evaluation data of the sample freshly cured tobacco leaves are collected: aroma quality data, aroma intensity data, aroma persistence, taste dryness data, smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data.

[0010] Furthermore, the recursive feature elimination algorithm is used to extract the main characteristic components related to the sensory quality from the intrinsic component data, including: Determine a component characteristic data set of the intrinsic component data; Use random forest as the base model to build multiple decision trees; Calculating an average impurity reduction for each feature in the component feature data set; According to the first Features Sensory quality evaluation data of the sample first-cured tobacco leaves , a MIC-based nonlinear association test was performed to evaluate the characteristics Sensory quality evaluation data The relationship between them; According to the average impurity reduction of each feature in the component feature data set and the evaluation result of the correlation between each feature and the sensory quality evaluation data, characteristic components related to the sensory quality are extracted from the intrinsic component data.

[0011] Furthermore, according to the classification of the sensory quality evaluation data of the sample freshly cured tobacco leaves, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components, including: Classifying the sensory quality evaluation data of the sample freshly-cured tobacco leaves to obtain a classification result of the sensory quality evaluation data; wherein the classification result of the sensory quality evaluation data includes smoke sensory quality evaluation data, aroma sensory quality evaluation data, and taste sensory quality evaluation data; According to the smoke sensory quality evaluation data, the aroma sensory quality evaluation data and the taste sensory quality evaluation data, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components.

[0012] Furthermore, the sensory quality prediction model of the freshly cured tobacco leaves is obtained based on the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly cured tobacco leaves, and is optimized based on machine learning algorithm training, including: Based on the LAFF algorithm, the first LAFF block is used to perform feature fusion on the smoke sensory characteristics and the smoke sensory quality evaluation data to obtain a first fused feature vector representation; the second LAFF block is used to perform feature fusion on the aroma sensory characteristics and the aroma sensory quality evaluation data to obtain a second fused feature vector representation; the third LAFF block is used to perform feature fusion on the taste sensory characteristics and the taste sensory quality evaluation data to obtain a third fused feature vector representation; Integrate the first LAFF block, the second LAFF block and the third LAFF block in parallel through a multi-head strategy to obtain an integrated LAFF block; Based on the integrated LAFF block and the deep learning framework, the network structure of the deep belief network is initialized; wherein the network structure of the deep belief network includes an input layer and an RBM layer; the RBM layer includes an original visible layer and a hidden layer; the input layer has a first number of neurons, and the hidden layer has a second number of neurons; For each RBM layer, the following pre-training steps are performed: calculating the probability distribution of the hidden layer; sampling the hidden state from the probability distribution of the hidden layer; reconstructing the original visible layer using the hidden state; updating the connection weights between the reconstructed visible layer and the hidden layer according to the difference between the reconstructed visible layer and the original visible layer; Connect the pre-trained RBM layers in series; add a softmax layer based on the pre-trained deep belief network; The first fused feature vector representation, the second fused feature vector representation and the third fused feature vector representation are used to train and optimize the network parameters of the pre-trained deep belief network until the loss function converges, thereby obtaining a trained sensory quality prediction model for early-cured tobacco leaves.

[0013] Furthermore, the selecting of sample primary-cured tobacco leaves according to the administrative division, topography and soil data of the tobacco planting area includes: Analyze the topography of tobacco-growing areas in different administrative regions using geographic information systems and digital elevation model data; wherein the topography includes slope, slope aspect and altitude; Soil samples were collected from tobacco-growing areas in different administrative divisions to determine soil pH, organic matter content, and nutrient levels; Integrate the topographic analysis and soil data of tobacco-growing areas in different administrative divisions; The sample first-cured tobacco leaves are selected according to the integration results.

[0014] Furthermore, the free amino acids include the following 21 amino acids: lysine, histidine, arginine, phenylalanine, tryptophan, tyrosine, aspartic acid, glutamic acid, alanine, glycine, isoleucine, leucine, valine, threonine, serine, cystine, glutamine, methionine, 4-aminobutyric acid, asparagine and proline.

[0015] Furthermore, the dichloromethane extract mainly includes: carotenoid degradation products, browning reaction products, aromatic amino acid cracking products, cedarwood degradation products and chlorophyll degradation products.

[0016] Furthermore, the polyphenols mainly include: chlorogenic acid, scopoletin, and rutin.

[0017] Furthermore, the smoke sensory quality evaluation data includes smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data; the aroma sensory quality evaluation data includes aroma quality data, aroma intensity data and aroma persistence; the taste sensory quality evaluation data includes taste dryness data.

[0018] The present invention also adopts the following technical solutions to solve the technical problems: Application of sensory quality prediction model based on the main internal components of first-cured tobacco leaves in tobacco leaf sensory quality evaluation, including Collecting the internal component data of the target first-cured tobacco leaves; wherein the internal component data include protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content and starch content; inputting the intrinsic component data into the sensory quality prediction model; Output the sensory quality prediction result of the target freshly-cured tobacco leaves.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The present invention realizes the mining of the comprehensive effects of multiple components in the first-cured tobacco leaves on the sensory quality and establishes a prediction model. This model not only improves the accuracy of tobacco leaf sensory quality evaluation, but also has good interpretability, providing a powerful tool for tobacco leaf sensory quality evaluation and production. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments. The accompanying drawings described herein are used to provide a further understanding of the embodiments of the present invention and constitute a part of this application. The exemplary embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application.

[0021] Figure 1 is a structural diagram of an optional sensory quality prediction model based on the main internal components of freshly cured tobacco leaves according to an embodiment of the present application; Figure 2 The present invention is a flowchart of another optional method for constructing a sensory quality prediction model based on the main internal components of freshly cured tobacco leaves according to an embodiment of the present application. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Example 1 Reference Figure 1 , a sensory quality prediction model based on the main internal components of first-cured tobacco leaves, including The feature input layer 101 is used to receive the intrinsic component data of the target first-cured tobacco leaves; wherein the intrinsic component data includes protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content and starch content; The feature processing layer 102 is used to extract and transform the input intrinsic component data to identify the characteristic patterns related to the sensory quality; The prediction output layer 103 is used to output the sensory quality prediction result of the target freshly-cured tobacco leaves based on the characteristic pattern related to the sensory quality; Among them, the sensory quality prediction model based on the main internal components of freshly cured tobacco leaves is obtained through training and optimization based on a machine learning algorithm.

[0024] Based on the embodiments provided in the present application, a comprehensive sensory quality prediction model based on the main internal components of freshly cured tobacco leaves is constructed by integrating the feature input layer, the feature processing layer and the prediction output layer. By extracting and transforming the features of the internal component data of the target freshly cured tobacco leaves based on the sensory quality prediction model of the main internal components of freshly cured tobacco leaves, characteristic patterns related to sensory quality can be identified. The application of the sensory quality prediction model based on the main internal components of freshly cured tobacco leaves provides a quality control tool for the tobacco industry. Example 2 Reference Figure 2 , a method for constructing a sensory quality prediction model based on the main internal components of first-cured tobacco leaves, including: S201, selecting sample freshly-cured tobacco leaves according to the administrative division, topography and soil data of the tobacco-growing area; wherein the sample freshly-cured tobacco leaves are obtained by baking fresh tobacco leaves; S202, collecting the internal component data and sensory quality evaluation data of the sample freshly cured tobacco leaves; S203, extracting main characteristic components related to sensory quality from the intrinsic component data using a recursive feature elimination algorithm; S204, classifying the extracted characteristic components related to the sensory quality according to the classification of the sensory quality evaluation data of the sample freshly cured tobacco leaves, to obtain the smoke sensory characteristic components, the aroma sensory characteristic components, and the taste sensory characteristic components; S205, obtaining a sensory quality prediction model for freshly cured tobacco leaves based on machine learning algorithm training and optimization according to the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly cured tobacco leaves.

[0025] Based on the embodiments provided by the present application, by combining the administrative divisions, topography and soil data of the tobacco planting area to select sample first-cured tobacco leaves, the representativeness and diversity of the samples can be ensured, thereby improving the accuracy of sensory quality prediction; the recursive feature elimination algorithm is used to extract the main characteristic components related to sensory quality from the intrinsic component data, which can reduce unnecessary data noise and improve the relevance of features and the efficiency of the prediction model; the characteristic components are divided into smoke sensory characteristic components, aroma sensory characteristic components and taste sensory characteristic components. This refined classification helps to more accurately predict and regulate the sensory quality of tobacco leaves and meet the needs of different consumers; the sensory quality prediction model of first-cured tobacco leaves obtained by training and optimizing the machine learning algorithm has practical application value for tobacco leaf processing and quality control; through the sensory quality prediction model of first-cured tobacco leaves, the sensory quality of tobacco leaves can be evaluated during the production process, and the processing parameters can be adjusted in time, thereby improving production efficiency and the consistency and quality of the final product; the traditional sensory quality evaluation relies on manual tasting, which is costly and subjective. The present application reduces the dependence on manual tasting through an automated prediction model, reduces costs, and improves the objectivity and repeatability of the evaluation.

[0026] As an improvement of this embodiment, the collecting of the intrinsic component data and sensory quality evaluation data of the sample freshly cured tobacco leaves includes: Collect the following intrinsic component data of the sample flue-cured tobacco leaves: protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content, and starch content; The following sensory quality evaluation data of the sample freshly cured tobacco leaves are collected: aroma quality data, aroma intensity data, aroma persistence, taste dryness data, smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data.

[0027] As an improvement of this embodiment, the recursive feature elimination algorithm is used to extract the main feature components related to the sensory quality from the intrinsic component data, including: Determine a component characteristic data set of the intrinsic component data; Use random forest as the base model to build multiple decision trees; Calculating an average impurity reduction for each feature in the component feature data set;

[0028] in, The component characteristic data set includes The average impurity reduction of each feature; The component characteristic data set includes Features is the number of decision trees; is the number of the decision tree; It is decision tree; Represented in the decision tree Using features The amount of impurity reduction when performing the segmentation; According to the first Features Sensory quality evaluation data of the sample first-cured tobacco leaves , a MIC-based nonlinear association test was performed to evaluate the characteristics Sensory quality evaluation data The relationship between them; According to the average impurity reduction of each feature in the component feature data set and the evaluation result of the correlation between each feature and the sensory quality evaluation data, characteristic components related to the sensory quality are extracted from the intrinsic component data.

[0029] Specifically,

[0030] in, Representation characteristics Sensory quality evaluation data The MIC value between To evaluate the characteristics Sensory quality evaluation data The relationship between them; and In the characteristics Sensory quality evaluation data The number of grid divisions in the direction of ; It is a feature Sensory quality evaluation data The mutual information value between ; Indicates that in all possible and Under the combination of The maximum value and The value of .

[0031] Based on the embodiments provided in the present application, the main characteristic components related to sensory quality are extracted from the intrinsic component data by combining the recursive feature elimination algorithm with the MIC nonlinear association test. Thus, the relationship between the features and the sensory quality can be identified and quantified, rather than simply relying on a single statistical indicator. Not only the accuracy of feature selection is improved, but also the depth and breadth of the sensory quality prediction model for the prediction of the sensory quality of the first-cured tobacco leaves are enhanced, thereby improving the prediction ability and practical application value of the model.

[0032] As an improvement of this embodiment, according to the classification of the sensory quality evaluation data of the sample first-cured tobacco leaves, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components, including: Classifying the sensory quality evaluation data of the sample freshly-cured tobacco leaves to obtain a classification result of the sensory quality evaluation data; wherein the classification result of the sensory quality evaluation data includes smoke sensory quality evaluation data, aroma sensory quality evaluation data, and taste sensory quality evaluation data; According to the smoke sensory quality evaluation data, the aroma sensory quality evaluation data and the taste sensory quality evaluation data, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components.

[0033] Based on the embodiments provided in the present application, the characteristic components are divided into three categories: smoke, aroma and taste based on the classification of the sensory quality evaluation data of the sample first-cured tobacco leaves; feature extraction and model construction are performed for different sensory attributes, which not only improves the prediction accuracy of the sensory quality prediction model in specific sensory quality dimensions, but also provides more accurate data support for the personalized customization and quality control of tobacco products, enhancing the practicality and flexibility of the model.

[0034] As an improvement of this embodiment, the sensory quality prediction model of the freshly cured tobacco leaves is obtained based on the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly cured tobacco leaves, and is optimized based on machine learning algorithm training, including: Based on the LAFF algorithm, the first LAFF block is used to perform feature fusion on the smoke sensory characteristics and the smoke sensory quality evaluation data to obtain a first fused feature vector representation; the second LAFF block is used to perform feature fusion on the aroma sensory characteristics and the aroma sensory quality evaluation data to obtain a second fused feature vector representation; the third LAFF block is used to perform feature fusion on the taste sensory characteristics and the taste sensory quality evaluation data to obtain a third fused feature vector representation;

[0035] in, represents the first fused feature vector representation; Indicates the first LAFF block; Indicates the sensory characteristics of smoke; Indicates the smoke sensory quality evaluation data; represents the second fused feature vector representation; Indicates the second LAFF block; Indicates the sensory characteristics of aroma; Indicates aroma sensory quality evaluation data; represents the third fused feature vector representation; Indicates the third LAFF block; Indicates the sensory characteristics of taste; Indicates aroma sensory quality evaluation data; Integrate the first LAFF block, the second LAFF block and the third LAFF block in parallel through a multi-head strategy to obtain an integrated LAFF block; Based on the integrated LAFF block and the deep learning framework, the network structure of the deep belief network is initialized; wherein the network structure of the deep belief network includes an input layer and an RBM layer; the RBM layer includes an original visible layer and a hidden layer; the input layer has a first number of neurons, and the hidden layer has a second number of neurons; For each RBM layer, the following pre-training steps are performed: calculating the probability distribution of the hidden layer; sampling the hidden state from the probability distribution of the hidden layer; reconstructing the original visible layer using the hidden state; updating the connection weights between the reconstructed visible layer and the hidden layer according to the difference between the reconstructed visible layer and the original visible layer; Connect the pre-trained RBM layers in series; add a softmax layer based on the pre-trained deep belief network; The first fused feature vector representation, the second fused feature vector representation and the third fused feature vector representation are used to train and optimize the network parameters of the pre-trained deep belief network until the loss function converges, thereby obtaining a trained sensory quality prediction model for early-cured tobacco leaves.

[0036] Based on the embodiments provided in the present application, the sensory characteristics of three dimensions of smoke, aroma and taste are fused with the corresponding quality evaluation data through the LAFF algorithm, which not only improves the expression ability of the features, but also enhances the ability of the sensory quality prediction model to capture complex sensory quality features through integrated learning; the LAFF algorithm combines the sensory characteristics with the evaluation data, rather than just processing a single sensory feature, to form a more accurate feature vector representation; the output of the LAFF block is used as the input of the deep belief network, thereby realizing deep learning and prediction of the sensory quality of the first-cured tobacco leaves; by pre-training the RBM layer and fine-tuning it, the model can better adapt to the specific data distribution, thereby improving the generalization ability and prediction accuracy of the model; the method of integrating different LAFF blocks in parallel with a multi-head strategy not only improves the processing efficiency of the model, but also enhances the model's understanding and prediction ability of the multi-dimensional characteristics of the sensory quality of tobacco leaves by processing different sensory characteristics in parallel; in the network structure design of the deep belief network, by precisely controlling the number of neurons in the input layer and the hidden layer, and adding a softmax layer on the basis of the pre-trained network, this refined design helps the model to better learn the data features and improve the accuracy of the prediction.

[0037] As an improvement of this embodiment, the selecting of sample primary-cured tobacco leaves according to the administrative division, topography and soil data of the tobacco planting area includes: Analyze the topography of tobacco-growing areas in different administrative regions using geographic information systems and digital elevation model data; wherein the topography includes slope, slope aspect and altitude; Soil samples were collected from tobacco-growing areas in different administrative divisions to determine soil pH, organic matter content, and nutrient levels; Integrate the topographic analysis and soil data of tobacco-growing areas in different administrative divisions; The sample first-cured tobacco leaves are selected according to the integration results.

[0038] The digital elevation model data includes one or more of the following data: Data provided by national or local surveying and mapping units: These units usually have topographic data covering a specific area, including digital elevation model data. Data is obtained through ground surveying, aerial photogrammetry, satellite remote sensing, etc., and processed and updated; Geographic Information System Data: The National Basic Geographic Information System may contain digital elevation model data, which can reflect the geographical spatial distribution characteristics of China, including provincial and county administrative boundaries, river systems, road and railway networks, etc., which also include digital elevation model data; Research data from scientific research institutions and universities: When conducting research related to geographic information systems and remote sensing, scientific research institutions and universities may generate or have digital elevation model data, which can be used for topographic analysis of tobacco growing areas; On-site field measurement data: For specific areas of tobacco planting sites, the three-dimensional coordinates of the measuring points can be directly measured using a total station, rangefinder + electronic tablet to obtain accurate digital elevation model data; Digital topographic map data: Apply existing large-scale topographic map scanning and vectorization to obtain digital elevation model data. This method is suitable for local areas that require accurate topographic data; Satellite remote sensing data: Satellite remote sensing technology, such as the Space Shuttle Radar Topography Mission, can be used to obtain global or regional digital elevation model data, which can be used for topographic analysis of tobacco growing areas; UAV aerial survey data: Using the photogrammetry equipment carried by drones, high-resolution digital elevation model data of small areas can be quickly obtained, which is suitable for local topographic analysis of tobacco planting areas.

[0039] Based on the embodiments provided in this application, by using geographic information system and digital elevation model data to analyze the topography of tobacco planting areas, and combining soil data to select sample primary flue-cured tobacco leaves, it is possible to grasp the impact of tobacco planting environment on sensory quality from a macro level. This method of integrating geographic and soil information provides more comprehensive and in-depth data support for the prediction of tobacco sensory quality, helps to improve the prediction accuracy and applicability of the model, and enhances the environmental adaptability of the model and the accuracy of prediction.

[0040] As an improvement of this embodiment, the free amino acids include the following 21 amino acids: lysine, histidine, arginine, phenylalanine, tryptophan, tyrosine, aspartic acid, glutamic acid, alanine, glycine, isoleucine, leucine, valine, threonine, serine, cystine, glutamine, methionine, 4-aminobutyric acid, asparagine and proline.

[0041] The dichloromethane extract mainly comprises carotenoid degradation products, browning reaction products, aromatic amino acid cracking products, ceramsite degradation products and chlorophyll degradation products.

[0042] The polyphenols mainly include: chlorogenic acid, scopoletin and rutin.

[0043] As an improvement of this embodiment, the smoke sensory quality evaluation data includes smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data; the aroma sensory quality evaluation data includes aroma quality data, aroma intensity data and aroma persistence; the taste sensory quality evaluation data includes taste dryness data.

[0044] Example 3 Application of sensory quality prediction model based on the main internal components of first-cured tobacco leaves in tobacco leaf sensory quality evaluation, including Collecting the internal component data of the target first-cured tobacco leaves; wherein the internal component data include protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content and starch content; Inputting the intrinsic component data into the sensory quality prediction model described in Example 1; Output the sensory quality prediction result of the target freshly-cured tobacco leaves.

[0045] Based on the embodiments provided in this application, the sensory quality prediction model is applied to the sensory quality evaluation of tobacco leaves, and its beneficial effects are: it can provide a fast and accurate tobacco leaf quality evaluation method; the sensory quality prediction model described in Example 1 is applied to the actual tobacco leaf quality control process, which helps the tobacco industry to control quality during production and processing and improve the market competitiveness of products. Through this application, tobacco companies can more effectively manage and optimize tobacco leaf resources, thereby improving overall production efficiency and economic benefits.

[0046] It should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A sensory quality prediction model based on the main internal components of first-cured tobacco leaves, characterized by: include A feature input layer, used to receive the intrinsic component data of the target first-cured tobacco leaves; A feature processing layer, for performing feature extraction and feature conversion on the input intrinsic component data to identify feature patterns related to sensory quality; A prediction output layer, used for outputting a prediction result of the sensory quality of the target freshly-cured tobacco leaves based on the characteristic pattern related to the sensory quality; Among them, the sensory quality prediction model based on the main internal components of freshly cured tobacco leaves is obtained through training and optimization based on a machine learning algorithm.

2. The sensory quality prediction model according to claim 1, characterized in that: The intrinsic component data include protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content and starch content.

3. A method for constructing a sensory quality prediction model based on the main internal components of first-cured tobacco leaves, characterized in that: include Select the sample first-cured tobacco leaves according to the administrative division, topography and soil data of the tobacco growing area; Collecting the intrinsic component data and sensory quality evaluation data of the sample first-cured tobacco leaves; extracting main characteristic components related to sensory quality from the intrinsic component data using a recursive feature elimination algorithm; According to the classification of the sensory quality evaluation data of the sample freshly cured tobacco leaves, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components; According to the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly-cured tobacco leaves, a sensory quality prediction model for freshly-cured tobacco leaves is obtained based on machine learning algorithm training and optimization.

4. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: The collecting of the intrinsic component data and sensory quality evaluation data of the sampled first-cured tobacco leaves includes: Collect the following intrinsic component data of the sample flue-cured tobacco leaves: protein, free amino acids, dichloromethane extract, total organic acid, polyphenols, nicotine, total sugar, reducing sugar, total nitrogen, potassium content, chlorine content, and starch content; The following sensory quality evaluation data of the sample freshly cured tobacco leaves are collected: aroma quality data, aroma intensity data, aroma persistence, taste dryness data, smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data.

5. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: The recursive feature elimination algorithm is used to extract the main characteristic components related to the sensory quality from the intrinsic component data, including: Determine a component characteristic data set of the intrinsic component data; Use random forest as the base model to build multiple decision trees; Calculating an average impurity reduction for each feature in the component feature data set; According to the first Features Sensory quality evaluation data of the sample first-cured tobacco leaves , a MIC-based nonlinear association test was performed to evaluate the characteristics Sensory quality evaluation data The relationship between them; According to the average impurity reduction of each feature in the component feature data set and the evaluation result of the correlation between each feature and the sensory quality evaluation data, characteristic components related to the sensory quality are extracted from the intrinsic component data.

6. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: According to the classification of the sensory quality evaluation data of the sample freshly cured tobacco leaves, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components, including: Classifying the sensory quality evaluation data of the sample freshly-cured tobacco leaves to obtain a classification result of the sensory quality evaluation data; wherein the classification result of the sensory quality evaluation data includes smoke sensory quality evaluation data, aroma sensory quality evaluation data, and taste sensory quality evaluation data; According to the smoke sensory quality evaluation data, the aroma sensory quality evaluation data and the taste sensory quality evaluation data, the extracted characteristic components related to the sensory quality are classified to obtain the smoke sensory characteristic components, the aroma sensory characteristic components and the taste sensory characteristic components.

7. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: The sensory quality prediction model of the freshly cured tobacco leaves is obtained based on the sensory characteristic components of the smoke, the sensory characteristic components of the aroma, the sensory characteristic components of the taste and the sensory quality evaluation data of the sample freshly cured tobacco leaves, and is optimized based on machine learning algorithm training, including: Based on the LAFF algorithm, the first LAFF block is used to perform feature fusion on the smoke sensory characteristics and the smoke sensory quality evaluation data to obtain a first fused feature vector representation; the second LAFF block is used to perform feature fusion on the aroma sensory characteristics and the aroma sensory quality evaluation data to obtain a second fused feature vector representation; the third LAFF block is used to perform feature fusion on the taste sensory characteristics and the taste sensory quality evaluation data to obtain a third fused feature vector representation; Integrate the first LAFF block, the second LAFF block and the third LAFF block in parallel through a multi-head strategy to obtain an integrated LAFF block; Based on the integrated LAFF block and the deep learning framework, the network structure of the deep belief network is initialized; wherein the network structure of the deep belief network includes an input layer and an RBM layer; the RBM layer includes an original visible layer and a hidden layer; the input layer has a first number of neurons, and the hidden layer has a second number of neurons; For each RBM layer, the following pre-training steps are performed: calculating the probability distribution of the hidden layer; sampling the hidden state from the probability distribution of the hidden layer; reconstructing the original visible layer using the hidden state; updating the connection weights between the reconstructed visible layer and the hidden layer according to the difference between the reconstructed visible layer and the original visible layer; Connect the pre-trained RBM layers in series; add a softmax layer based on the pre-trained deep belief network; The first fused feature vector representation, the second fused feature vector representation and the third fused feature vector representation are used to train and optimize the network parameters of the pre-trained deep belief network until the loss function converges, thereby obtaining a trained sensory quality prediction model for early-cured tobacco leaves.

8. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: The sample first-cured tobacco leaves are selected according to the administrative division, topography and soil data of the tobacco planting area, including: Analyze the topography of tobacco-growing areas in different administrative regions using geographic information systems and digital elevation model data; wherein the topography includes slope, slope aspect and altitude; Soil samples were collected from tobacco-growing areas in different administrative divisions to determine soil pH, organic matter content, and nutrient levels; Integrate the topographic analysis and soil data of tobacco-growing areas in different administrative divisions; The sample first-cured tobacco leaves are selected according to the integration results.

9. The method for constructing a sensory quality prediction model according to claim 3, characterized in that: The smoke sensory quality evaluation data includes smoke stimulation data, smoke roundness data, smoke clumping data, smoke fineness data and smoke softness data; the aroma sensory quality evaluation data includes aroma quality data, aroma intensity data and aroma persistence; the taste sensory quality evaluation data includes taste dryness data.

10. Application of the sensory quality prediction model based on the main internal components of freshly cured tobacco leaves as claimed in any one of claims 1 to 9 in the sensory quality evaluation of tobacco leaves.

Images