Tobacco product raw material sensory evaluation method and sensory evaluation model training method

Abstract

The invention relates to a sensory evaluation method of tobacco product raw materials and a training method of a sensory evaluation model, and relates to the technical field of tobacco production. The sensory evaluation method comprises the following steps: obtaining a solid product and a liquid product generated after pyrolysis of a tobacco product raw material; performing Raman spectrum detection on the solid product to obtain first spectrum data associated with chemical characteristics of non-volatile substances in the tobacco product raw materials, and performing fluorescence spectrum detection on the liquid product to obtain second spectrum data associated with chemical characteristics of volatile substances in the tobacco product raw materials; determining first feature data based on the first spectral data, and determining second feature data based on the second spectral data; and based on the first feature data and the second feature data, utilizing a sensory evaluation model corresponding to the target evaluation index to predict the sensory grade of the tobacco product raw material in the aspect of the target evaluation index as a sensory evaluation result of the tobacco product raw material.

Description

Technical Field

[0001] This disclosure relates to the field of tobacco production technology, and in particular to a sensory evaluation method for tobacco product raw materials and a training method for a sensory evaluation model. Background Technology

[0002] Sensory evaluation of tobacco product raw materials is a crucial step in the quality control and product development of tobacco products. It aims to comprehensively and systematically evaluate the characteristics of the smoke produced after combustion of these raw materials through the sensory judgment of professionals, thereby assessing their intrinsic quality. The reliability of sensory evaluation of tobacco product raw materials has a significant impact on the quality of the subsequently manufactured tobacco products.

[0003] In related technologies, the sensory evaluation of tobacco product raw materials is mainly carried out by smokers who use artificial smoke to score the raw materials in various evaluation indicators such as sweetness and strength, so as to determine the sensory level of the raw materials in each evaluation indicator. Summary of the Invention

[0004] The inventors of this disclosure have discovered the following problem in the above-mentioned related technologies: the reliability of sensory evaluation of tobacco product raw materials is low.

[0005] To address the aforementioned problems, the present disclosure provides the following solutions.

[0006] According to some embodiments of this disclosure, a sensory evaluation method for tobacco product raw materials is provided, comprising: acquiring solid and liquid products generated after pyrolysis of the tobacco product raw materials; performing Raman spectroscopy on the solid products to obtain first spectral data associated with the chemical properties of non-volatile substances in the tobacco product raw materials, and performing fluorescence spectroscopy on the liquid products to obtain second spectral data associated with the chemical properties of volatile substances in the tobacco product raw materials; determining first feature data based on the first spectral data, and determining second feature data based on the second spectral data; and predicting the sensory level of the tobacco product raw materials in relation to the target evaluation index using a sensory evaluation model corresponding to the target evaluation index based on the first feature data and the second feature data, as the sensory evaluation result of the tobacco product raw materials.

[0007] According to some embodiments of this disclosure, a training method for a sensory evaluation model is provided. The sensory evaluation model is used to predict the sensory level of tobacco product raw materials in terms of a target evaluation index. The training method includes: acquiring the solid and liquid products of each tobacco product raw material sample after pyrolysis from a plurality of tobacco product raw material samples; performing Raman spectroscopy on the solid products to obtain first sample spectral data associated with the chemical characteristics of non-volatile substances in each tobacco product raw material sample, and performing fluorescence spectroscopy on the liquid products to obtain second sample spectral data associated with the chemical characteristics of volatile substances in each tobacco product raw material sample; determining first sample feature data based on the first sample spectral data, and determining second sample feature data based on the second sample spectral data; using the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index for each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material sample in terms of the target evaluation index as output, training the sensory evaluation model corresponding to the target evaluation index using a supervised training method until the training termination condition is met.

[0008] According to further embodiments of this disclosure, a sensory evaluation device for tobacco product raw materials is provided, comprising: a first acquisition module configured to acquire solid and liquid products generated after pyrolysis of the tobacco product raw materials; a first detection module configured to perform Raman spectroscopy detection on the solid products to obtain first spectral data associated with the chemical properties of non-volatile substances in the tobacco product raw materials, and to perform fluorescence spectroscopy detection on the liquid products to obtain second spectral data associated with the chemical properties of volatile substances in the tobacco product raw materials; a first determination module configured to determine first feature data based on the first spectral data, and to determine second feature data based on the second spectral data; and a prediction module configured to predict the sensory level of the tobacco product raw materials in relation to the target evaluation index using a sensory evaluation model corresponding to the target evaluation index, based on the first feature data and the second feature data, as the sensory evaluation result of the tobacco product raw materials.

[0009] According to further embodiments of this disclosure, a training apparatus for a sensory evaluation model is provided, comprising: a second acquisition module configured to acquire solid and liquid products of each tobacco product raw material sample after pyrolysis from a plurality of tobacco product raw material samples; a second detection module configured to perform Raman spectroscopy detection on the solid products to obtain first sample spectral data associated with the chemical properties of non-volatile substances in each tobacco product raw material sample, and to perform fluorescence spectroscopy detection on the liquid products to obtain second sample spectral data associated with the chemical properties of volatile substances in each tobacco product raw material sample; a second determination module configured to determine first sample feature data based on the first sample spectral data, and to determine second sample feature data based on the second sample spectral data; and a training module configured to use the first sample feature data, the second sample feature data, and a sensory level label corresponding to the target evaluation index of each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material sample in relation to the target evaluation index as output, and to train the sensory evaluation model corresponding to the target evaluation index using a supervised training method until the training termination condition is met.

[0010] According to some embodiments of this disclosure, an electronic device is provided, including: a memory; and a processor coupled to the memory, the processor being configured to execute, based on instructions stored in the memory device, a sensory evaluation method for tobacco product raw materials or a training method for a sensory evaluation model according to any of the above embodiments.

[0011] According to further embodiments of the present disclosure, a computer-readable storage medium is provided, having stored thereon computer instructions that, when executed by a processor, implement the sensory evaluation method for tobacco product raw materials or the training method for a sensory evaluation model in any of the above embodiments.

[0012] According to further embodiments of this disclosure, a computer program product is also provided, including instructions that, when executed by a processor, cause the processor to perform a sensory evaluation method for tobacco product raw materials or a training method for a sensory evaluation model according to any of the foregoing embodiments.

[0013] In the above embodiments, by using Raman spectroscopy and fluorescence spectroscopy in combination, characteristic data reflecting the chemical properties of the pyrolysis products generated after the pyrolysis of tobacco product raw materials are obtained. The characteristic data is then processed using a sensory evaluation model corresponding to the target evaluation index, thereby obtaining accurate and reliable sensory evaluation results for tobacco product raw materials and improving the reliability of sensory evaluation of tobacco product raw materials. Attached Figure Description

[0014] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.

[0015] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein:

[0016] Figure 1 A flowchart illustrating a sensory evaluation method for tobacco product raw materials according to some embodiments of the present disclosure;

[0017] Figure 2 A schematic diagram showing Raman characteristic curves according to some embodiments of the present disclosure;

[0018] Figure 3 A schematic diagram showing fluorescence spectrum curves according to some embodiments of the present disclosure;

[0019] Figure 4 A flowchart illustrating a method for training a sensory evaluation model according to some embodiments of the present disclosure;

[0020] Figure 5 A flowchart illustrating a specific example of a training method for a sensory evaluation model according to some embodiments of the present disclosure;

[0021] Figure 6 A schematic diagram showing Raman characteristic curves according to other embodiments of the present disclosure;

[0022] Figure 7 A schematic diagram showing fluorescence spectral curves of different tobacco product raw material samples according to some embodiments of the present disclosure;

[0023] Figure 8 A schematic diagram illustrating the processing method of five-fold cross-validation in some embodiments of this disclosure;

[0024] Figure 9 A schematic diagram of a confusion matrix according to some embodiments of the present disclosure is shown.

[0025] Figure 10 A block diagram showing a sensory evaluation apparatus for tobacco product raw materials according to some embodiments of the present disclosure;

[0026] Figure 11 A block diagram showing a training apparatus for a sensory evaluation model according to some embodiments of the present disclosure;

[0027] Figure 12 A block diagram of an electronic device according to some embodiments of the present disclosure is shown;

[0028] Figure 13 Block diagrams of electronic devices according to other embodiments of the present disclosure are shown. Detailed Implementation

[0029] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.

[0030] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.

[0031] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.

[0032] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.

[0033] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0034] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.

[0035] As mentioned earlier, related technologies mainly rely on manual sensory evaluation to score tobacco product raw materials across various evaluation indicators, thereby assessing their sensory level. However, this approach has several drawbacks. First, the sensory evaluation results are easily influenced by individual differences and subjective preferences of the evaluators, leading to high subjectivity and low consistency. Second, the manual sensory evaluation process is cumbersome and time-consuming, making it difficult to meet the needs of large-scale production and rapid research and development.

[0036] The inventors of this disclosure have discovered that during the pyrolysis and combustion of tobacco product raw materials, these materials undergo complex physicochemical reactions at high temperatures, such as cracking and oxidation, generating volatile substances (e.g., acids, alcohols, ketones) and non-volatile substances (e.g., coke, ash). These products are the core factors affecting the sensory characteristics of tobacco product raw materials. Therefore, by deeply analyzing the correlation between the chemical properties of these products and the sensory characteristics of tobacco product raw materials, a more scientific basis for sensory evaluation of tobacco product raw materials can be provided.

[0037] In view of this, this disclosure proposes a sensory evaluation method for tobacco product raw materials. By combining Raman spectroscopy and fluorescence spectroscopy, characteristic data reflecting the chemical properties of the pyrolysis products generated after pyrolysis of tobacco product raw materials are obtained. The characteristic data is then processed using a sensory evaluation model corresponding to the target evaluation index, thereby obtaining accurate and reliable sensory evaluation results for tobacco product raw materials and improving the reliability of sensory evaluation of tobacco product raw materials.

[0038] Figure 1 A flowchart illustrating a sensory evaluation method for tobacco product raw materials according to some embodiments of the present disclosure is shown.

[0039] like Figure 1 As shown, in step 110, solid and liquid products generated after the pyrolysis of tobacco product raw materials are obtained.

[0040] In this disclosure, tobacco products can be products made wholly or partially from tobacco leaves as the raw material for tobacco products, intended for smoking, chewing, snorting, or otherwise use. Tobacco products can include, but are not limited to, cigarettes, cigars, pipe tobacco, or hookah. Cigarettes can be distinguished by tobacco leaf formulation and processing methods, including but not limited to flue-cured tobacco, burley tobacco, aromatic tobacco, or sun-cured tobacco. Tobacco products release chemicals such as nicotine through heating (e.g., heated tobacco products) and / or combustion (e.g., cigarettes, cigars, etc.). Tobacco products may or may not have a paper wrapping and may or may not have a filter. In this disclosure, for ease of description, cigarettes are sometimes used as examples of tobacco products. However, it should be understood that the various features or limitations described herein regarding cigarettes also apply to other types of tobacco products.

[0041] Raw materials for tobacco products may include tobacco leaves in various forms, such as tobacco leaves or tobacco shreds, tobacco sheets, tobacco powder or tobacco blocks processed from tobacco leaves.

[0042] In some embodiments, pyrolysis experiments can be performed on tobacco product raw materials to collect the solid and liquid products generated after pyrolysis. For example, the solid products may include coke, and the liquid products may include tar. For example, during the pyrolysis of tobacco product raw materials, the temperature can be set to 500°C, the nitrogen flow rate can be maintained at 400 ml / min, and the pyrolysis time can be 30 minutes to allow the tobacco product raw material sample to react fully and generate stable pyrolysis products.

[0043] It should be understood that pyrolysis is one of the main chemical reactions that occur when tobacco product raw materials are heated further within a temperature range exceeding the pyrolysis threshold temperature in an anaerobic environment. Pyrolysis can occur in the heating-without-combustion stage or in the combustion stage, with the former generally occurring at a lower temperature than the latter.

[0044] In some embodiments, liquid products (e.g., tar) can be collected by a two-stage dry ice condensation method, while solid products (e.g., coke) can be collected after cooling to room temperature, in order to improve the stability of the chemical properties of the pyrolysis products for subsequent analysis.

[0045] In step 120, the solid product is subjected to Raman spectroscopy to obtain first spectral data, and the liquid product is subjected to fluorescence spectroscopy to obtain second spectral data.

[0046] The first spectral data is associated with the chemical properties of non-volatile substances in tobacco product raw materials, while the second spectral data is associated with the chemical properties of volatile substances in tobacco product raw materials.

[0047] In some embodiments, the volatile substances in tobacco product raw materials may include one or more of acids, alcohols, and ketones. The non-volatile substances in tobacco product raw materials may include one or more of coke and ash.

[0048] For example, solid products can include coke. Coke is an important non-volatile organic carbonized residue produced after the pyrolysis of tobacco product raw materials, and is a core component of non-volatile substances in tobacco product raw materials. By performing Raman spectroscopy on coke, first-hand spectral data can be obtained to reveal the chemical characteristics of non-volatile substances in tobacco product raw materials.

[0049] For example, liquid products may include tar. Since aromatic compounds are the core components of tar, and volatile substances (such as ketones, aldehydes, etc.) are precursors or intermediates in the formation of aromatic compounds, fluorescence spectroscopy of tar can provide second spectral data to reveal the chemical characteristics of volatile substances in tobacco product raw materials.

[0050] It should be noted that Raman spectroscopy is a non-destructive molecular vibrational spectroscopy analysis technique based on the Raman scattering effect. The basic principle of Raman spectroscopy is that when monochromatic light (such as laser light) shines on a sample, most photons undergo elastic scattering, and their frequency is the same as that of the incident light. However, a very small number of photons undergo inelastic collisions with molecules in the sample, causing changes in photon energy and a shift in frequency. This phenomenon is called Raman scattering.

[0051] The horizontal axis of a Raman spectroscopy curve represents the change in frequency (also known as the Raman shift, measured in cm⁻¹), reflecting the vibrational modes of the chemical bonds in the molecule, while the vertical axis represents the intensity of scattered light. Because each molecule has a corresponding Raman spectral curve due to its unique chemical structure, much like a "molecular fingerprint," Raman spectroscopy can be used for the identification and analysis of substances.

[0052] Fluorescence spectroscopy is an analytical technique that detects the fluorescence emitted by a substance after absorbing light of a specific wavelength. The basic principle of fluorescence spectroscopy is to obtain molecular dynamics information by recording the relationship between the intensity of fluorescence emitted by a sample after excitation and the wavelength. It is suitable for the component analysis of organic compounds (such as aromatic compounds) with fluorescent properties.

[0053] Depending on the measurement mode, fluorescence spectroscopy can be divided into excitation spectroscopy and emission spectroscopy. The horizontal axis of the fluorescence spectrum curve represents the excitation wavelength or emission wavelength (e.g., in nanometers (nm)), and the vertical axis represents the fluorescence intensity.

[0054] Therefore, the spectral data of pyrolysis products obtained by combining Raman spectroscopy and fluorescence spectroscopy can effectively reflect the chemical characteristics of the pyrolysis products. Thus, using such spectral data as an objective analytical basis for sensory evaluation provides an objective scientific basis for the sensory evaluation of tobacco product raw materials. This helps to more comprehensively and objectively understand the chemical changes of tobacco product raw materials during pyrolysis, thereby improving the reliability of sensory evaluation results.

[0055] In some embodiments, the liquid product may be diluted before fluorescence spectroscopy detection. For example, the liquid product may be diluted to a specified concentration to reduce the occurrence of fluorescence signal oversaturation or weakness due to excessively high or low concentrations, thereby improving the accuracy of the fluorescence signal presented in the fluorescence spectrum and thus improving the accuracy of the obtained second spectral data.

[0056] In step 130, first feature data is determined based on first spectral data, and second feature data is determined based on second spectral data.

[0057] In some embodiments, the first feature data can be used to characterize the structural features and relative content of each chemical structure among the various chemical structures of the solid product.

[0058] It should be understood that the structural features of each chemical structure can be used to distinguish the type of that chemical structure, that is, what kind of structure it is. The relative abundance of each chemical structure can represent the proportion of that chemical structure among the various chemical structures of the solid product.

[0059] For example, taking coke as a solid product, its various chemical structures can include carbon-carbon double bonds, aromatic ring structures, and other chemical structures.

[0060] In some embodiments, the second characteristic data can be used to characterize the relative content of each chemical component among a variety of chemical components of the liquid product, wherein the variety of chemical components of the liquid product includes a variety of aromatic compounds.

[0061] It should be understood that the relative content of each chemical component can represent the proportion of that chemical component in the various chemical components of the liquid product.

[0062] In this way, the first characteristic data determined from the first spectral data obtained by Raman spectroscopy can accurately reflect the microstructure and composition of the solid products generated after the pyrolysis of tobacco product raw materials, and the second characteristic data determined from the second spectral data obtained by fluorescence spectroscopy can accurately reflect the composition ratio of aromatic components in the liquid products generated after the pyrolysis of tobacco product raw materials. This achieves a systematic and multi-dimensional analysis of the chemical structure and composition of the pyrolysis products of tobacco product raw materials, providing objective and reliable data support for the sensory evaluation of tobacco product raw materials and helping to improve the reliability of sensory evaluation results.

[0063] In some embodiments, the Raman spectrum of the tobacco product raw material corresponding to the first spectral data can be subjected to peak fitting processing to extract multiple characteristic peaks from the Raman spectrum of the tobacco product raw material, and then the first characteristic data can be determined based on the multiple characteristic peaks. Each of the multiple characteristic peaks corresponds to each of the multiple chemical structures of the solid product. This will be further explained later.

[0064] In some embodiments, multiple characteristic intervals can be determined from the fluorescence spectrum corresponding to the tobacco product raw material corresponding to the second spectral data, and the second characteristic data can be determined based on the fluorescence characteristic data corresponding to these multiple characteristic intervals. The fluorescence characteristic data includes each characteristic wavelength in the multiple characteristic wavelength intervals and the fluorescence intensity corresponding to each characteristic wavelength. This will be further explained later.

[0065] In step 140, based on the first feature data and the second feature data, the sensory evaluation model corresponding to the target evaluation index is used to predict the sensory level of the tobacco product raw material in terms of the target evaluation index, which is then used as the sensory evaluation result of the tobacco product raw material.

[0066] In some embodiments, the target evaluation index may include at least one of the following: sweetness, strength, aroma, concentration, fineness, smoothness, irritation, and aftertaste.

[0067] In some embodiments, the first feature data and the second feature data can be used together as input to a sensory evaluation model corresponding to the target evaluation index to obtain the sensory level output by the sensory evaluation model corresponding to the target evaluation index. For example, the target evaluation index is sweetness, and the sweetness level includes level 1, level 2, and level 3, with higher levels indicating higher sweetness. In response to the sensory level output by the sensory evaluation model corresponding to sweetness being level 1, the sensory evaluation result of the tobacco product raw material in terms of sweetness can be determined to be level 1.

[0068] In the above embodiments, Raman spectroscopy is performed on the solid products generated after the pyrolysis of tobacco product raw materials to obtain first spectral data associated with the chemical characteristics of non-volatile substances in the tobacco product raw materials. Fluorescence spectroscopy is performed on the liquid products generated after the pyrolysis of tobacco product raw materials to obtain second spectral data associated with the chemical characteristics of volatile substances in the tobacco product raw materials. Then, based on the first and second spectral data, characteristic data of the corresponding pyrolysis products are determined, and this characteristic data is processed using a sensory evaluation model corresponding to the target evaluation index to obtain the sensory evaluation results of the tobacco product raw materials.

[0069] In this approach, the first characteristic data obtained by Raman spectroscopy analysis of the solid products generated after the pyrolysis of tobacco product raw materials can accurately reflect the chemical characteristics of non-volatile substances in the tobacco product raw materials, and the second characteristic data obtained by fluorescence spectroscopy analysis of the liquid products generated after the pyrolysis of tobacco product raw materials can accurately reflect the chemical characteristics of volatile substances in the tobacco product raw materials. This provides objective and reliable data support for the sensory evaluation results of tobacco product raw materials, effectively reduces the subjective errors caused by manual evaluation in the sensory evaluation process, and improves the reliability of the sensory evaluation results of tobacco product raw materials.

[0070] The process of obtaining the first feature data and the second feature data will be illustrated below with reference to some embodiments.

[0071] In some embodiments, the first spectral data includes a plurality of Raman shifts and the intensity of scattered light corresponding to each of the plurality of Raman shifts. For example, the abscissa of the Raman spectral curve formed by the first spectral data can represent the plurality of Raman shifts, and the ordinate can represent the intensity of scattered light corresponding to each of the plurality of Raman shifts.

[0072] In some embodiments, Raman characteristic data corresponding to a specified Raman shift range can be determined from the first spectral data, and then the first characteristic data can be determined based on the first spectral curve formed by the Raman characteristic data.

[0073] For example, Raman characteristic data includes multiple characteristic shifts within a specified Raman shift range and the scattered light intensity corresponding to each of the multiple characteristic shifts. The first spectral curve covers the Raman peaks generated by the vibrational modes within the molecule.

[0074] For example, the specified Raman shift interval (also known as the Raman fingerprint region) can be [800 cm⁻¹, 1800 cm⁻¹]. The Raman feature data corresponding to the specified Raman shift interval can include each feature shift in the Raman shift interval [800 cm⁻¹, 1800 cm⁻¹] and the scattered light intensity corresponding to each feature shift.

[0075] Considering that the Raman peaks covered by the first spectral curve formed by the Raman characteristic data corresponding to the Raman fingerprint region are generated by vibrational modes within the molecule, and these vibrational modes reflect the overall molecular structure, even compounds with very similar structures can have significant differences in the peak position, peak shape, and relative intensity of their Raman fingerprint regions. Therefore, by analyzing the Raman characteristic data corresponding to this Raman fingerprint region, the chemical structure and relative content of different substances can be accurately distinguished, thus providing scientific and reliable data support for subsequent sensory evaluation and helping to improve the reliability of sensory evaluation.

[0076] In some embodiments, the first spectral curve formed by the Raman characteristic data can be subjected to peak fitting to determine multiple Raman characteristic peaks, and then the first characteristic data can be determined based on the multiple Raman characteristic peaks. Each of the multiple Raman characteristic peaks corresponds to each of the multiple chemical structures of the solid product.

[0077] In some embodiments, the ten-peak method can be used to perform peak fitting on the first spectral curve to determine ten Raman characteristic peaks from the first spectral curve.

[0078] For example, when the solid product is coke, the first spectral curve is fitted with peaks using the ten-peak method. The chemical structure corresponding to each of the ten Raman characteristic peaks is shown in Table 1.

[0079] Table 1. Peak position information of the ten-peak method peak fitting

[0080]

[0081] Therefore, peak fitting can decompose overlapping peaks in Raman spectra into multiple independent sub-peaks (i.e., Raman characteristic peaks) with specific Raman shifts and shapes. This helps to identify different chemical structures present in solid products, enabling a comprehensive and systematic analysis of the chemical structure of solid products. This provides scientific and reliable data support for subsequent sensory evaluation and improves the reliability of sensory evaluation.

[0082] In some embodiments, the fitted area of ​​each Raman characteristic curve among multiple Raman characteristic curves corresponding to multiple Raman characteristic peaks is calculated, and the first characteristic data is determined based on the fitted area of ​​each Raman characteristic curve. For example, each Raman characteristic curve covers a corresponding Raman characteristic peak. The fitted area of ​​each Raman characteristic curve can be determined by integrating the scattered light intensity of each Raman characteristic curve within a specified Raman shift interval.

[0083] Figure 2 A schematic diagram of Raman characteristic curves according to some embodiments of the present disclosure is shown.

[0084] Figure 2 A Raman characteristic curve R is schematically shown. The fitted area of ​​this Raman characteristic curve R is the area enclosed by the abscissa corresponding to the Raman characteristic curve R and the specified Raman displacement interval [800 cm⁻¹, 1800 cm⁻¹], as shown below. Figure 2 The shaded area is shown.

[0085] For example, the fitted area of ​​each Raman characteristic curve can be accurately calculated by integrating the scattered light intensity of each Raman characteristic curve over a specified Raman shift interval.

[0086] In some embodiments, the size of the fitted area of ​​each Raman characteristic curve is positively correlated with the relative content of the chemical structure characterized by the Raman characteristic peak corresponding to each Raman characteristic curve.

[0087] For example, the larger the fitted area of ​​a Raman characteristic curve, the greater the proportion of the chemical structure represented by the Raman characteristic peak corresponding to that Raman characteristic curve in the solid product.

[0088] In this approach, considering that Raman characteristic peaks of different chemical structures may overlap in Raman spectral curves, the overlapping Raman characteristic peaks can be decomposed by peak fitting, and the fitting area of ​​the Raman characteristic curve corresponding to each Raman characteristic peak can be accurately calculated.

[0089] Based on this, compared to isolated peak information, since the fitting area of ​​different Raman characteristic curves can reflect the signal intensity of different chemical structures due to their unique vibration modes, the first characteristic data is determined according to the fitting area of ​​different Raman characteristic curves. This allows the first characteristic data to more accurately reflect the structural characteristics and relative content of different chemical structures, thus providing scientific and reliable data support for subsequent sensory evaluation and improving the reliability of sensory evaluation.

[0090] In some embodiments, the second spectral data includes multiple wavelengths and fluorescence intensity corresponding to each of the multiple wavelengths.

[0091] For example, the horizontal axis of the fluorescence spectrum curve formed from the second spectral data can represent multiple wavelengths, and the vertical axis can represent the fluorescence intensity corresponding to each of the multiple wavelengths. For example, depending on the measurement mode, the multiple wavelengths in the second spectral data can be multiple emission wavelengths or multiple excitation wavelengths.

[0092] In some embodiments, multiple characteristic wavelength intervals can be determined from wavelength intervals formed by multiple wavelengths, and then fluorescence characteristic data corresponding to each characteristic wavelength interval can be determined from second spectral data. Then, second characteristic data can be determined based on the fluorescence characteristic data corresponding to each characteristic wavelength interval.

[0093] In some embodiments, each of the plurality of characteristic wavelength ranges corresponds to each of the plurality of chemical components of the liquid product.

[0094] In some embodiments, the fluorescence feature data corresponding to each feature wavelength range includes multiple feature wavelengths in the feature wavelength range and the fluorescence intensity corresponding to each of the multiple feature wavelengths.

[0095] For example, the horizontal axis of the fluorescence characteristic curve formed by the fluorescence characteristic data corresponding to each characteristic wavelength range can represent each characteristic wavelength among the multiple characteristic wavelengths in that characteristic wavelength range, and the vertical axis can represent the fluorescence intensity corresponding to each characteristic wavelength.

[0096] It should be understood that the fluorescence characteristic data corresponding to each characteristic wavelength range can reflect relevant information about the chemical components corresponding to that characteristic wavelength range (such as concentration, content, etc.).

[0097] Taking tar as an example of liquid products, each of the multiple characteristic wavelength ranges can correspond to each of the various aromatic compounds in the tar.

[0098] For example, multiple aromatic compounds may include monocyclic aromatic compounds, bicyclic aromatic compounds, and tricyclic and higher aromatic compounds found in tar, and multiple characteristic wavelength ranges may include three characteristic wavelength ranges corresponding to these three types of aromatic compounds. For example, Table 2 schematically shows the compound information corresponding to the three characteristic wavelength ranges.

[0099] Table 2. Compound information corresponding to different characteristic wavelength ranges

[0100] As shown in Table 2, the fluorescence characteristic data corresponding to the characteristic wavelength range [250, 290) can reflect the relevant information of monocyclic aromatic compounds in tar; the fluorescence characteristic data corresponding to the characteristic wavelength range [290, 320) can reflect the relevant information of bicyclic aromatic compounds in tar; and the fluorescence characteristic data corresponding to the characteristic wavelength range [320, 500] can reflect the relevant information of tricyclic and higher aromatic compounds in tar.

[0101] By analyzing the fluorescence characteristic data corresponding to each characteristic wavelength range, a second characteristic data can be obtained to characterize the relative content of the chemical components corresponding to each characteristic wavelength range.

[0102] In the above embodiments, considering that fluorescence spectroscopy can effectively reflect the distribution of various chemical components in the liquid products (e.g., tar) after the pyrolysis of tobacco raw materials, and that different chemical components have their own unique fluorescence characteristics due to differences in their molecular structures, dividing the second spectral data according to multiple characteristic wavelength ranges corresponding to various chemical components enables simultaneous and rapid quantitative analysis of multiple chemical components in the liquid products. This eliminates the need for multiple separate measurements, improving both the reliability and efficiency of subsequent sensory evaluation.

[0103] In some embodiments, the fitted area of ​​the fluorescence characteristic curve formed by the fluorescence characteristic data corresponding to each characteristic wavelength range can be calculated, and then the second characteristic data can be determined based on the fitted area of ​​the fluorescence characteristic curve formed by the fluorescence characteristic data corresponding to each characteristic wavelength range.

[0104] For example, the fitted area of ​​the fluorescence characteristic curve formed by the fluorescence characteristic data corresponding to each characteristic wavelength range can be determined by integrating the fluorescence intensity of the fluorescence characteristic curve within that characteristic wavelength range.

[0105] It should be understood that multiple characteristic wavelength ranges correspond to multiple fluorescence characteristic data to form multiple fluorescence characteristic curves. Each fluorescence characteristic curve can be a part of the fluorescence spectrum curve (also known as the comprehensive fluorescence spectrum) formed by the second spectral data and corresponds to each characteristic wavelength range.

[0106] Figure 3 A schematic diagram showing fluorescence spectrum curves according to some embodiments of the present disclosure is provided.

[0107] Figure 3 A fluorescence spectrum curve and three characteristic wavelength ranges A, B and C are schematically shown.

[0108] The fluorescence characteristic curve 'a' formed by the fluorescence characteristic data corresponding to the characteristic wavelength range A is a portion of the fluorescence spectrum curve within the characteristic wavelength range A. The fitted area of ​​fluorescence characteristic curve 'a' is the area enclosed by the fluorescence characteristic curve 'a' and the abscissa corresponding to the characteristic wavelength range A, such as... Figure 3 The area of ​​region a1 shown.

[0109] The fluorescence characteristic curve b, formed by the fluorescence characteristic data corresponding to the characteristic wavelength range B, is a portion of the fluorescence spectrum curve within the characteristic wavelength range B. The fitted area of ​​the fluorescence characteristic curve b is the area enclosed by the fluorescence characteristic curve b and the abscissa corresponding to the characteristic wavelength range B, such as... Figure 3 The area of ​​region b1 shown.

[0110] The fluorescence characteristic curve c, formed by the fluorescence characteristic data corresponding to the characteristic wavelength range C, is a portion of the fluorescence spectrum curve within the characteristic wavelength range C. The fitted area of ​​the fluorescence characteristic curve c is the area enclosed by the fluorescence characteristic curve c and the abscissa corresponding to the characteristic wavelength range C, such as... Figure 3 The area of ​​region c1 shown.

[0111] For example, the fitted area of ​​each fluorescence characteristic curve can be accurately calculated by integrating the fluorescence intensity of each fluorescence characteristic curve (e.g., fluorescence characteristic curves a, b, c) within the corresponding characteristic wavelength range.

[0112] In some embodiments, the area fitted to each fluorescence characteristic curve is positively correlated with the relative content of the chemical component represented by the characteristic wavelength range corresponding to each fluorescence characteristic curve. For example, the larger the area fitted to a fluorescence characteristic curve, the greater the proportion of the chemical component represented by the characteristic wavelength range corresponding to that fluorescence characteristic curve in the liquid product.

[0113] In this approach, considering that fluorescence spectra may experience overall wavelength shifts due to instrument fluctuations, temperature changes, and other interferences, the peak fluorescence intensity within each characteristic wavelength range can change drastically due to minute shifts in peak position. In contrast, the fitted area of ​​the fluorescence characteristic curve corresponding to each characteristic wavelength range can integrate all fluorescence intensity data within that range, making it insensitive to the minute fluctuations caused by these interferences. Therefore, the characteristic data obtained based on the fitted area is more stable and reliable. This improves the reliability of the second characteristic data, thereby enhancing the reliability of subsequent sensory evaluation.

[0114] The previous section introduced the implementation of obtaining sensory evaluation results for tobacco product raw materials using a sensory evaluation model corresponding to the target evaluation index. The following section further describes the training method for the sensory evaluation model corresponding to the target evaluation index.

[0115] Figure 4 A flowchart illustrating a training method for a sensory evaluation model according to some embodiments of the present disclosure is shown. Figure 4 As shown, for example, the training method for the sensory evaluation model may include steps 410 to 440. The sensory evaluation model is used to predict the sensory level of tobacco product raw materials in terms of target evaluation indicators.

[0116] In step 410, the solid and liquid products of each tobacco product raw material sample after pyrolysis are obtained from multiple tobacco product raw material samples.

[0117] In some embodiments, to improve the generalization ability of the model training, when selecting tobacco product raw material samples for obtaining training data, multiple tobacco product raw material samples covering various flavors with significant differences can be selected. A pyrolysis experiment is performed on each tobacco product raw material sample to collect the solid and liquid products generated after pyrolysis. For example, the solid product may include coke, and the liquid product may include tar.

[0118] In some embodiments, each tobacco product raw material sample may undergo standardized pretreatment before being subjected to a pyrolysis experiment.

[0119] For example, taking each tobacco product raw material sample as a tobacco leaf sample, the standardized pretreatment process may include: First, drying each tobacco leaf sample at the same temperature to ensure consistent initial moisture content; then, grinding each tobacco leaf sample, filtering the ground tobacco powder through a sieve, and thoroughly mixing to ensure consistent particle size; for each tobacco leaf sample, weighing, for example, 20 mg of tobacco powder as the test sample, and air-drying the sieved test sample under the same conditions to remove moisture; finally, sealing each test sample uniformly and storing it in a constant temperature and humidity environment for subsequent pyrolysis experiments.

[0120] This pretreatment aims to eliminate the interference caused by differences in moisture, particle size, etc., on the pyrolysis process of tobacco product raw material samples, so as to maximize the reliability of the data characterized by the pyrolysis products used for subsequent model training.

[0121] It should be understood that the above preprocessing steps can also be applied to Figure 1 In step 110 of the sensory evaluation method shown, the accuracy of the trained sensory evaluation model in processing characteristic data of pyrolysis products of tobacco product raw materials is improved.

[0122] In step 420, Raman spectroscopy is performed on the solid product of the sample to obtain the first sample spectral data, and fluorescence spectroscopy is performed on the liquid product of the sample to obtain the second sample spectral data.

[0123] Specifically, for each tobacco product raw material sample, the spectral data of the first sample is correlated with the chemical characteristics of the non-volatile substances in the tobacco product raw material sample, and the spectral data of the second sample is correlated with the chemical characteristics of the volatile substances in the tobacco product raw material sample.

[0124] In some embodiments, for each tobacco product raw material sample, the sample liquid product may be diluted before fluorescence spectroscopy detection.

[0125] For example, for each tobacco product raw material sample, its liquid product is diluted to a specified concentration to reduce the occurrence of fluorescence signal oversaturation or weakness due to excessively high or low concentration, thereby improving the accuracy of fluorescence signal in fluorescence spectrum and thus improving the accuracy of the obtained second sample spectral data.

[0126] In step 430, first sample characteristic data are determined based on the first sample spectral data of each tobacco product raw material sample, and second sample characteristic data are determined based on the second sample spectral data of each tobacco product raw material sample.

[0127] It should be understood that the implementation methods of steps 420 and 430 are the same as those described above. Figure 1 The implementation of steps 120 and 130 in the sensory evaluation method shown is similar. For details, please refer to the description in the aforementioned related embodiments, which will not be repeated here.

[0128] In step 440, based on the first sample feature data, the second sample feature data, and the sensory level labels corresponding to the target evaluation index, a supervised training method is used to train the sensory evaluation model corresponding to the target evaluation index.

[0129] For example, the sensory evaluation model corresponding to the target evaluation index is trained using the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index for each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material in terms of the target evaluation index as output, in a supervised training manner until the training termination condition is met.

[0130] In some embodiments, evaluators may be organized to conduct sensory evaluations of each of a plurality of tobacco product raw material samples according to a uniform sensory scoring standard, in order to obtain a sensory score for each tobacco product raw material sample in terms of the target evaluation index. For example, the sensory score may be out of ten, with a higher score indicating a higher sensory quality of the tobacco product raw material sample in terms of the target evaluation index.

[0131] In some embodiments, the sensory scores can be divided into intervals, and a sensory level corresponding to each interval can be defined. That is, the sensory scores given by assessors to each tobacco product raw material sample for the target evaluation indicator are discretized to form a sensory level label corresponding to each tobacco product raw material sample and the target evaluation indicator.

[0132] For example, taking sweetness as the target evaluation index, continuous sweetness scores (such as 4-8 points) can be divided into 1, 2, and 3 levels (corresponding to score intervals [4-6), [6-7), and [7-8 respectively) according to their distribution. This transforms the continuous sensory scores into discrete sensory levels, reducing the complexity of data processing and helping to improve the model's classification ability.

[0133] For example, for a certain tobacco product raw material sample, if the sensory score of the tobacco product raw material sample in terms of sweetness is in the range of 4-6 (e.g., 4.5 points), the sensory grade label of the tobacco product raw material sample in terms of sweetness can be determined as level 1.

[0134] In some embodiments, during the training process, a set of feature data (i.e., the first feature data and the second feature data of the tobacco product raw material sample) and a sensory level label (i.e., the sensory level label of the tobacco product raw material sample corresponding to the target evaluation index) are input as training data into the sensory evaluation model corresponding to the target evaluation index to be trained, until multiple sets of training data corresponding to multiple tobacco product raw material samples are input.

[0135] Through this supervised training method, the sensory evaluation model corresponding to the target evaluation index can fully learn the relationship between the chemical and sensory characteristics of different tobacco product raw material samples, thereby providing a more scientific decision-making tool for the sensory evaluation of tobacco product raw materials.

[0136] In some embodiments, the training termination condition may be, for example, reaching a specified number of training iterations and / or the difference between the predicted sensory grade of the tobacco product raw material sample in terms of the target evaluation indicator and the sensory grade label of the tobacco product raw material sample in terms of the target evaluation indicator being less than a threshold. For example, the difference between the predicted sensory grade and the sensory grade label can be characterized by the loss function value of the sensory evaluation model.

[0137] In some embodiments, the sensory evaluation model corresponding to the target evaluation index may include a random forest model. For example, during training, the random forest model generates multiple decision trees by randomly sampling the training data, and predicts the sensory level of the tobacco product raw material sample based on the output of each tree.

[0138] In the above embodiments, each tobacco product raw material sample from multiple samples is pyrolyzed, and the corresponding solid and liquid products are collected. Raman spectroscopy and fluorescence spectroscopy are used to analyze the solid and liquid products, respectively, to construct a solid product feature dataset (including the first feature data of each tobacco product raw material sample) and a liquid product feature dataset (including the second feature data of each tobacco product raw material sample) for model training. Based on this, combined with the sensory rating labels determined for each tobacco product raw material sample, a supervised learning method is used to train a sensory evaluation model for predicting the sensory evaluation rating of tobacco product raw materials in terms of target evaluation indicators.

[0139] In this way, the sensory evaluation model trained can effectively quantify the intrinsic relationship between the chemical and sensory properties of tobacco product raw materials, thus providing a scientific and accurate decision-making tool for the sensory evaluation of tobacco product raw materials.

[0140] It is understood that multiple sensory evaluation models can be trained in the manner described above for multiple different target evaluation indicators. For example, for each evaluation indicator among sweetness, strength, aroma, intensity, smoothness, irritation, and aftertaste, the model can be trained in the manner described above to obtain a corresponding sensory evaluation model for assessing the sensory level of tobacco product raw materials in that evaluation indicator.

[0141] Figure 5 A flowchart illustrating a specific example of a training method for a sensory evaluation model according to some embodiments of the present disclosure.

[0142] like Figure 5 As shown, for example, the training method for the sensory evaluation model may include steps 510 to 550. This method can be used as... Figure 4 This is a specific implementation of the method.

[0143] In step 510, multiple tobacco product raw material samples are selected to obtain the solid and liquid products generated by each tobacco product raw material sample after pyrolysis.

[0144] In step 520, a sensory grade label dataset corresponding to the target evaluation index is obtained for each tobacco product raw material sample.

[0145] For example, assessors conduct sensory evaluations of the sensory quality of each tobacco product raw material sample among the multiple tobacco raw material samples selected in step 510 in terms of the target evaluation indicators, according to a unified sensory scoring standard. Assessors can conduct systematic and quantitative sensory evaluations of each tobacco product raw material sample based on a standardized sensory scoring table, and convert the scoring results into a standardized sensory grade label dataset.

[0146] In step 530, Raman spectroscopy analysis is performed on the solid products of each tobacco product raw material sample to construct a solid product feature dataset.

[0147] In some embodiments, coke is selected as the solid product of each tobacco product raw material sample for Raman spectroscopy analysis to obtain a Raman spectral curve representing each tobacco product raw material sample (i.e., the spectral curve formed by the first spectral data).

[0148] In some embodiments, for each tobacco product raw material sample, the same Raman shift interval (i.e., the specified Raman shift interval) in its Raman spectral curve is selected as a feature region. Peak fitting processing is performed on the first spectral curve formed by the Raman feature data within this feature region to determine multiple Raman feature peaks. Different Raman feature peaks reflect vibrational modes of different chemical structures in the coke. Then, the fitted area of ​​the Raman feature curve corresponding to each Raman feature peak is calculated to obtain multiple fitted areas corresponding to multiple Raman feature peaks, which serve as one set of solid product feature data (i.e., first feature data) corresponding to that tobacco product raw material sample. Thus, multiple sets of solid product feature data corresponding to multiple tobacco product raw material samples can constitute a solid product feature dataset for model training.

[0149] Figure 6 A schematic diagram showing Raman characteristic curves according to other embodiments of the present disclosure is provided.

[0150] like Figure 6 As shown, Raman spectroscopy analysis was performed on a tobacco product raw material sample using coke as the solid product, with the specified Raman shift range being [800 cm⁻¹, 1800 cm⁻¹]. The “data_modified curve” and the “fit curve” represent the first spectral curves before and after data fitting of the Raman characteristic data within the range of [800 cm⁻¹, 1800 cm⁻¹], respectively.

[0151] The first spectral curve (e.g., the "fit curve") was fitted with peaks using the ten-peak method to determine ten Raman characteristic peaks. The chemical structures corresponding to the ten Raman characteristic peaks are shown in Table 1 above. Figure 6 The diagram illustrates the ten Raman characteristic curves corresponding to these ten Raman characteristic peaks.

[0152] For example, referring to the chemical structures shown in Table 1, the "g920.0 curve" represents the Raman characteristic curve corresponding to the Raman characteristic peak at a Raman shift of 920.0 cm⁻¹, which can reflect the C–C structure of alkanes or cycloalkanes and the C–H structure of aromatic rings in coke; the "g1050.0 curve" represents the Raman characteristic curve corresponding to the Raman characteristic peak at a Raman shift of 1050.0 cm⁻¹, which can reflect the C–H structure of aromatic rings in coke; the "g1180.0 curve" represents the Raman characteristic curve corresponding to the Raman characteristic peak at a Raman shift of 1180.0 cm⁻¹, which can reflect the Caromatic–Calkyl structure and the CH structure on aromatic rings in coke. And so on.

[0153] The fitted area of ​​each Raman characteristic curve can be accurately calculated by integrating the scattered light intensity of each Raman characteristic curve within [800 cm⁻¹, 1800 cm⁻¹]. This yields multiple fitted areas of multiple Raman characteristic curves, which serve as a set of solid product characteristic data (i.e., the first characteristic data) corresponding to the tobacco product raw material sample.

[0154] Raman spectroscopy analysis was performed on the solid products of each tobacco product raw material sample in a similar manner to obtain multiple sets of first feature data corresponding to multiple tobacco product raw material samples to form a solid product feature dataset for model training.

[0155] In step 540, fluorescence spectroscopy analysis is performed on the liquid products of each tobacco product raw material sample to construct a liquid product feature dataset.

[0156] In some embodiments, tar is selected as the liquid product of each tobacco product raw material sample for fluorescence spectroscopy analysis to obtain a fluorescence spectral curve representing each tobacco product raw material sample (i.e., the spectral curve formed by the second spectral data).

[0157] In some embodiments, for each tobacco product raw material sample, the fluorescence spectrum curve of the tobacco product raw material sample is divided into multiple characteristic regions according to multiple different characteristic wavelength ranges, and second characteristic data is determined based on the fluorescence characteristic data in each characteristic region.

[0158] Figure 7 A schematic diagram showing fluorescence spectral curves of different tobacco product raw material samples according to some embodiments of the present disclosure is provided.

[0159] Figure 7 The diagram schematically illustrates three fluorescence spectral curves formed by the second spectral data of three different tobacco product raw material samples obtained by fluorescence spectroscopy analysis using liquid product tar as an example. These curves are denoted by the numbers "120360", "120362", and "120370", respectively.

[0160] For example, combining the compound information corresponding to the three characteristic wavelength ranges shown in Table 2, there are significant differences in fluorescence intensity for each fluorescence spectrum curve within the same characteristic wavelength range. This indicates that fluorescence characteristic data within different characteristic wavelength ranges can reflect the unique fluorescence characteristics of different chemical components. Thus, for each fluorescence spectrum curve, the different chemical components of tar in the sample can be distinguished and quantified (i.e., semi-quantitative characterization) by using fluorescence characteristic data within different characteristic wavelength ranges.

[0161] In some embodiments, for each tobacco product raw material sample, multiple fluorescence feature data corresponding to multiple characteristic wavelength ranges can be obtained. Then, the fitted area of ​​the fluorescence feature curve formed by each fluorescence feature data is calculated to obtain multiple fitted areas of multiple fluorescence feature curves as one liquid product feature data (i.e., second feature data) corresponding to that tobacco product raw material sample. In this way, multiple liquid product feature data corresponding to multiple tobacco product raw material samples can constitute a liquid product feature dataset for model training.

[0162] In step 550, a random forest algorithm is used to train a model based on the sensory level label dataset, solid product feature dataset, and liquid product feature dataset corresponding to the target evaluation index, so as to obtain a sensory evaluation model corresponding to the target evaluation index.

[0163] In some embodiments, the solid product feature datasets and liquid product feature datasets obtained in steps 530 and 440 can be standardized, for example, by z-score standardization.

[0164] In some embodiments, a five-fold cross-validation method can be used to partition the constructed sample dataset, dividing it evenly into five subsets of equal size. In each validation iteration, four subsets are selected for model training, while the remaining subset is used for model testing. This partitioning and validation process is repeated five times. In each validation, a different subset is selected as the test set, and the remaining subsets are used as the training set. The prediction accuracy of the trained sensory evaluation model on each test set is recorded, and the average prediction accuracy across the five validations is calculated as a comprehensive evaluation metric for the model's prediction performance.

[0165] Figure 8 A schematic diagram illustrating the processing method of five-fold cross-validation in some embodiments of this disclosure is shown.

[0166] like Figure 8 As shown, the sample dataset can be a solid product feature dataset or a liquid product feature dataset. The sample dataset is uniformly divided into five subsets of equal size, denoted as D1, D2, D3, D4, and D5, respectively.

[0167] For example, in the first validation process, D2, D3, D4, and D5 are selected as the training set, and D1 is selected as the test set. The prediction accuracy U1 of the trained sensory evaluation model on the test set D1 will be recorded. In the second validation process, D1, D3, D4, and D5 are selected as the training set, and D2 is selected as the test set. The prediction accuracy U2 of the trained sensory evaluation model on the test set D1 will be recorded. This process continues, and the average of the prediction accuracies U1, U2, U3, U4, and U5 from the five validations is calculated as the comprehensive evaluation index of the model.

[0168] Taking sweetness as the target evaluation index as an example, Table 3 shows the prediction results of the sensory evaluation model corresponding to sweetness obtained by training in the above manner.

[0169] Table 3. Prediction results of the sensory evaluation model corresponding to sweetness.

[0170] As shown in Table 3, Table 3 schematically illustrates the sensory rating prediction results for the sweetness of each of the 20 tobacco product raw material samples. The prediction accuracy reached 76.2%, which is significantly higher than the random baseline (33.3%). This indicates that the sensory evaluation model corresponding to sweetness, trained according to the training method proposed in this disclosure, has high accuracy in predicting the sensory rating of sweetness of tobacco product raw materials.

[0171] It should be understood that those skilled in the art will recognize that the random baseline represents the performance benchmark of random forest models in the art, and a higher baseline indicates better prediction performance.

[0172] Figure 9 A schematic diagram of a confusion matrix according to some embodiments of the present disclosure is shown.

[0173] Figure 9 The confusion matrix corresponding to the prediction results shown in Table 3 is illustrated schematically. For example... Figure 9 As shown, the horizontal axis of this confusion matrix represents the sensory rating predicted by the sensory evaluation model, and the vertical axis represents the actual sensory rating (such as sensory rating label values). For example, sweetness ratings include levels 1, 2, and 3, with P1, P2, and P3 representing the predicted sensory ratings of level 1, 2, and 3, respectively, and NO1, NO2, and NO3 representing the actual sensory ratings of level 1, 2, and 3, respectively.

[0174] The value (or color intensity) of each cell represents the number of samples with both the actual and predicted sensory levels. Cells on the diagonal of the matrix represent the number of samples with correct predictions, while cells off the diagonal represent the number of samples with incorrect predictions.

[0175] For example, taking the data 4, 2, 0 in the first row of the confusion matrix as an example, this data indicates that there are a total of 6 samples. Among them, 4 samples have a true sensory level of 1 and a predicted sensory level of 1, that is, the prediction results of 4 samples are correct; 2 samples have a true sensory level of 1 and a predicted sensory level of 2, that is, the prediction results of 2 samples are incorrect and they are misclassified as level 2; 0 samples have a true sensory level of 1 and a predicted sensory level of 3, that is, no samples are misclassified as level 3.

[0176] In the above example, the correspondence between the true label and the prediction result is quantified by the confusion matrix, which comprehensively demonstrates the classification performance of the sensory evaluation model. The diagonal grid reflects the sample cases with accurate prediction, while the off-diagonal grid reflects the sample cases with incorrect prediction. The corresponding results show that the sensory evaluation model corresponding to sweetness trained according to the training method proposed in this disclosure has high accuracy in predicting the sensory level of tobacco product raw materials in terms of sweetness, and the overall prediction effect is good.

[0177] It should be understood that Figure 5 More embodiments of the method shown and its corresponding beneficial effects can be found in the preceding text. Figure 4 The descriptions in the relevant embodiments of the method shown will not be repeated here.

[0178] Figure 10 A block diagram of a sensory evaluation apparatus for tobacco product raw materials according to some embodiments of the present disclosure is shown.

[0179] like Figure 10 As shown, the sensory evaluation device 1000 for tobacco product raw materials includes a first acquisition module 1001, a first detection module 1002, a first determination module 1003, and a prediction module 1004.

[0180] The first acquisition module 1001 is configured to acquire the solid and liquid products generated after the pyrolysis of tobacco product raw materials.

[0181] The first detection module 1002 is configured to perform Raman spectroscopy on the solid product to obtain first spectral data associated with the chemical properties of non-volatile substances in the tobacco product raw materials, and to perform fluorescence spectroscopy on the liquid product to obtain second spectral data associated with the chemical properties of volatile substances in the tobacco product raw materials.

[0182] The first determining module 1003 is configured to determine first feature data based on first spectral data and second feature data based on second spectral data.

[0183] The prediction module 1004 is configured to predict the sensory level of tobacco product raw materials in terms of the target evaluation index based on the first feature data and the second feature data, using a sensory evaluation model corresponding to the target evaluation index as the sensory evaluation result of the tobacco product raw materials.

[0184] In some embodiments, the sensory evaluation device 1000 for tobacco product raw materials may further include the apparatus described above. Figures 1 to 3 Other modules of other operations in the illustrated embodiments.

[0185] Figure 11 A block diagram of a training apparatus for a sensory evaluation model according to some embodiments of the present disclosure is shown.

[0186] like Figure 11 As shown, the training device 1100 for the sensory evaluation model includes a second acquisition module 1101, a second detection module 1102, a second determination module 1103, and a training module 1104.

[0187] The second acquisition module 1101 is configured to acquire the solid and liquid products of each of the multiple tobacco product raw material samples after pyrolysis.

[0188] The second detection module 1002 is configured to perform Raman spectroscopy on the solid product of the sample to obtain first sample spectral data associated with the chemical characteristics of non-volatile substances in each tobacco product raw material sample, and to perform fluorescence spectroscopy on the liquid product of the sample to obtain second sample spectral data associated with the chemical characteristics of volatile substances in each tobacco product raw material sample.

[0189] The second determining module 1103 is configured to determine first sample characteristic data based on the first sample spectral data and to determine second sample characteristic data based on the second sample spectral data.

[0190] The training module 1104 is configured to take the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index as input for each tobacco product raw material sample, and the predicted sensory level of each tobacco product raw material sample in terms of the target evaluation index as output. The sensory evaluation model corresponding to the target evaluation index is trained in a supervised training manner until the training termination condition is met.

[0191] In some embodiments, the training device 1100 for the sensory evaluation model may further include the execution of the foregoing text. Figures 5 to 9 Other modules of other operations in the illustrated embodiments.

[0192] Figure 12 A block diagram of an electronic device according to some embodiments of the present disclosure is shown.

[0193] like Figure 12 As shown, the electronic device 1200 of this embodiment includes a memory 1201 and a processor 1202 coupled to the memory 1201. The processor 1202 is configured to execute the method in any embodiment of this disclosure based on instructions stored in the memory 1201.

[0194] The memory 1201 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, a database, and other programs.

[0195] In some embodiments, the electronic device 1200 can serve as a sensory evaluation device for tobacco product raw materials that perform the operations of any of the above embodiments. In other embodiments, the electronic device 1200 can serve as a training device for a sensory evaluation model that performs the operations of any of the above embodiments.

[0196] Figure 13 Block diagrams of electronic devices according to other embodiments of the present disclosure are shown.

[0197] like Figure 13 As shown, the electronic device 1300 of this embodiment includes: a memory 1301 and a processor 1302 coupled to the memory 1301, the processor 1302 being configured to execute the method of any of the foregoing embodiments based on instructions stored in the memory 1301.

[0198] The memory 1301 may include, for example, system memory, fixed non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, and other programs.

[0199] Electronic device 1300 may also include input / output interface 1303, network interface 1304, storage interface 1305, etc. These interfaces 1303, 1304, 1305, as well as the memory 1301 and processor 1302, can be connected, for example, via bus 1306. Specifically, input / output interface 1303 provides a connection interface for input / output devices such as monitors, mice, keyboards, touchscreens, microphones, and speakers. Network interface 1304 provides a connection interface for various networked devices. Storage interface 1305 provides a connection interface for external storage devices such as SD cards and USB flash drives.

[0200] In some embodiments, the electronic device 1300 can serve as a sensory evaluation device for tobacco product raw materials that perform the operations of any of the above embodiments. In other embodiments, the electronic device 1300 can serve as a training device for a sensory evaluation model that performs the operations of any of the above embodiments.

[0201] This disclosure also provides a computer-readable storage medium including computer program instructions that, when executed by a processor, implement the method of any of the above embodiments.

[0202] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the above embodiments.

[0203] Those skilled in the art will understand that embodiments of this disclosure can be provided as methods, systems, or computer program products. Therefore, this disclosure can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this disclosure can take the form of a computer program product embodied on one or more computer-usable non-transitory storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0204] The sensory evaluation techniques for tobacco product raw materials according to this disclosure have now been described in detail. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the disclosed techniques based on the above description.

[0205] The methods and systems of this disclosure may be implemented in many ways. For example, they may be implemented by software, hardware, firmware, or any combination of software, hardware, and firmware. The above-described order of steps for the methods is for illustrative purposes only, and the steps of the methods of this disclosure are not limited to the specific order described above unless otherwise specifically stated. Furthermore, in some embodiments, this disclosure may also be implemented as a program recorded on a recording medium, the program including machine-readable instructions for implementing the methods according to this disclosure. Thus, this disclosure also covers recording media storing programs for performing the methods according to this disclosure.

[0206] While specific embodiments of this disclosure have been described in detail by way of example, those skilled in the art should understand that the examples are for illustrative purposes only and not intended to limit the scope of this disclosure. Those skilled in the art should understand that modifications can be made to the above embodiments without departing from the scope and spirit of this disclosure. The scope of this disclosure is defined by the appended claims.

Claims

1. A sensory evaluation method for tobacco product raw materials, comprising: To obtain the solid and liquid products generated after the pyrolysis of tobacco product raw materials; Raman spectroscopy was performed on the solid product to obtain first spectral data associated with the chemical properties of non-volatile substances in the tobacco product raw materials, and fluorescence spectroscopy was performed on the liquid product to obtain second spectral data associated with the chemical properties of volatile substances in the tobacco product raw materials. First feature data is determined based on the first spectral data, and second feature data is determined based on the second spectral data; Based on the first feature data and the second feature data, the sensory evaluation model corresponding to the target evaluation index is used to predict the sensory level of the tobacco product raw material in terms of the target evaluation index, which is then used as the sensory evaluation result of the tobacco product raw material.

2. The sensory evaluation method according to claim 1, wherein, The first feature data is used to characterize the structural features and relative content of each chemical structure among the various chemical structures of the solid product. The second feature data is used to characterize the relative content of each chemical component among the various chemical components of the liquid product, the various chemical components including various aromatic compounds.

3. The sensory evaluation method according to claim 1, wherein, The solid product includes coke, and the liquid product includes tar.

4. The sensory evaluation method according to claim 1, wherein, The first spectral data includes multiple Raman shifts and the intensity of scattered light corresponding to each of the multiple Raman shifts. The step of determining the first feature data based on the first spectral data includes: Raman feature data corresponding to a specified Raman shift interval is determined from the first spectral data. The Raman feature data includes multiple feature shifts within the Raman shift interval and the scattered light intensity corresponding to each of the multiple feature shifts. The first characteristic data is determined based on the first spectral curve formed by the Raman characteristic data, and the first spectral curve covers the Raman peaks generated by the vibrational modes inside the molecule.

5. The sensory evaluation method according to claim 4, wherein, The first spectral curve formed based on the Raman feature data, wherein the first feature data includes: The first spectral curve formed by the Raman characteristic data is subjected to peak fitting processing to determine multiple Raman characteristic peaks, each of the multiple Raman characteristic peaks corresponding to each of the multiple chemical structures of the solid product; The first feature data is determined based on the multiple Raman characteristic peaks.

6. The sensory evaluation method according to claim 5, wherein, The determination of the first feature data based on the plurality of Raman feature peaks includes: Calculate the fitted area of ​​each Raman characteristic curve among the multiple Raman characteristic curves corresponding to the multiple Raman characteristic peaks; The first feature data is determined based on the fitted area of ​​each Raman feature curve.

7. The sensory evaluation method according to claim 1, wherein, The second spectral data includes multiple wavelengths and the fluorescence intensity corresponding to each of the multiple wavelengths. The determination of the second feature data based on the second spectral data includes: Multiple characteristic wavelength intervals are determined from the wavelength intervals formed by the multiple wavelengths, and each of the multiple characteristic wavelength intervals corresponds to each of the multiple chemical components of the liquid product; Fluorescence feature data corresponding to each characteristic wavelength range is determined from the second spectral data. The fluorescence feature data includes multiple characteristic wavelengths in each characteristic wavelength range and fluorescence intensity corresponding to each of the multiple characteristic wavelengths. The second feature data is determined based on the fluorescence feature data corresponding to each feature wavelength range.

8. The sensory evaluation method according to claim 7, wherein, The determination of the second feature data based on the fluorescence feature data corresponding to each feature wavelength range includes: Calculate the area under the fluorescence characteristic curve formed by the fluorescence characteristic data corresponding to each characteristic wavelength range; The second feature data is determined based on the fitted area of ​​the fluorescence feature curve.

9. The sensory evaluation method according to any one of claims 1-8, wherein, The sensory evaluation model corresponding to the target evaluation index is trained in the following manner: Obtain the solid and liquid products of each tobacco product raw material sample after pyrolysis from multiple tobacco product raw material samples; Raman spectroscopy was performed on the solid product of the sample to obtain first sample spectral data associated with the chemical characteristics of non-volatile substances in each tobacco product raw material sample, and fluorescence spectroscopy was performed on the liquid product of the sample to obtain second sample spectral data associated with the chemical characteristics of volatile substances in each tobacco product raw material sample. First sample characteristic data are determined based on the first sample spectral data, and second sample characteristic data are determined based on the second sample spectral data; Using the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index for each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material sample in terms of the target evaluation index as output, a supervised training method is used to train the sensory evaluation model corresponding to the target evaluation index until the training termination condition is met.

10. The sensory evaluation method according to any one of claims 1-8, wherein, The sensory evaluation model corresponding to the target evaluation index includes the random forest model.

11. A method for training a sensory evaluation model, wherein the sensory evaluation model is used to predict the sensory level of tobacco product raw materials in terms of target evaluation indicators, the training method comprising: Obtain the solid and liquid products of each tobacco product raw material sample after pyrolysis from multiple tobacco product raw material samples; Raman spectroscopy was performed on the solid product of the sample to obtain first sample spectral data associated with the chemical characteristics of non-volatile substances in each tobacco product raw material sample, and fluorescence spectroscopy was performed on the liquid product of the sample to obtain second sample spectral data associated with the chemical characteristics of volatile substances in each tobacco product raw material sample. First sample characteristic data are determined based on the first sample spectral data, and second sample characteristic data are determined based on the second sample spectral data; Using the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index for each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material sample in terms of the target evaluation index as output, a supervised training method is used to train the sensory evaluation model corresponding to the target evaluation index until the training termination condition is met.

12. A sensory evaluation device for tobacco product raw materials, comprising: The first acquisition module is configured to acquire the solid and liquid products generated after the pyrolysis of tobacco product raw materials; The first detection module is configured to perform Raman spectroscopy on the solid product to obtain first spectral data associated with the chemical properties of non-volatile substances in the tobacco product raw material, and to perform fluorescence spectroscopy on the liquid product to obtain second spectral data associated with the chemical properties of volatile substances in the tobacco product raw material. The first determining module is configured to determine first feature data based on the first spectral data and to determine second feature data based on the second spectral data. The prediction module is configured to predict the sensory level of the tobacco product raw material in terms of the target evaluation index based on the first feature data and the second feature data, using a sensory evaluation model corresponding to the target evaluation index as the sensory evaluation result of the tobacco product raw material.

13. A training device for a sensory evaluation model, comprising: The second acquisition module is configured to acquire the solid and liquid products of each tobacco product raw material sample after pyrolysis from a plurality of tobacco product raw material samples. The second detection module is configured to perform Raman spectroscopy on the solid product of the sample to obtain first sample spectral data associated with the chemical characteristics of non-volatile substances in each tobacco product raw material sample, and to perform fluorescence spectroscopy on the liquid product of the sample to obtain second sample spectral data associated with the chemical characteristics of volatile substances in each tobacco product raw material sample. The second determining module is configured to determine first sample characteristic data based on the first sample spectral data, and to determine second sample characteristic data based on the second sample spectral data; The training module is configured to take the first sample feature data, the second sample feature data, and the sensory level label corresponding to the target evaluation index of each tobacco product raw material sample as input, and the predicted sensory level of each tobacco product raw material sample in terms of the target evaluation index as output, and train the sensory evaluation model corresponding to the target evaluation index in a supervised training manner until the training termination condition is met.

14. An electronic device comprising: Memory; and A processor coupled to the memory is configured to execute, based on instructions stored in the memory, the sensory evaluation method for tobacco product raw materials according to any one of claims 1-10 or the training method for the sensory evaluation model according to claim 11.

15. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the sensory evaluation method for tobacco product raw materials according to any one of claims 1-10 or the training method for the sensory evaluation model according to claim 11.

16. A computer program product comprising instructions that, when executed by a processor, cause the processor to perform a sensory evaluation method for tobacco product raw materials according to any one of claims 1-10 or a training method for a sensory evaluation model according to claim 11.

Images