A method for measuring similarity of cigarette flavor characteristics

Abstract

The present application provides a cigarette flavor feature similarity measurement method, which comprises the following steps: selecting representative cigarettes of different brands in a set brand cigarette, organizing panelists to evaluate the cigarettes, and recording the scores of each flavor index; calculating the mean and standard deviation of the scores of each flavor index in different brands of cigarettes, and constructing a multi-peak normal distribution model; constructing a brand flavor feature generation probability model according to the multi-peak normal distribution model; setting the weight values of each flavor index, and adding each weight value to the brand flavor feature generation probability model to construct a similarity calculation formula; organizing panelists to evaluate test cigarettes, recording the scores of each flavor index of the test cigarettes, and substituting them into the similarity calculation formula to calculate the similarity score; and evaluating the similarity of the test cigarette and the set brand cigarette according to the similarity score. The present application can effectively improve the accuracy of the measurement results of the similarity of the flavor features of cigarettes.

Description

Technical Field

[0001] This invention belongs to the field of cigarette evaluation technology, specifically relating to a method for measuring the similarity of cigarette aroma characteristics. Background Technology

[0002] In the cigarette production process, in order to effectively examine the differences between different batches of cigarette products, assess the differences between cigarette products processed in different locations, examine the similarity between different brands, and evaluate the similarity between new products and the aroma characteristics of the enterprise brand, a similarity measurement method suitable for evaluating the aroma characteristics of cigarettes is needed.

[0003] Compared with existing research results, the invention patent CN105335734A, "A Star-Shaped Recognition Method for Cigarette Aroma Characteristics," uses a star-shaped diagram to compare the similarity of aroma characteristics; however, it cannot provide a specific numerical score for the similarity. The invention patent CN106770971A, "A Method for Evaluating Cigarette Style Characteristics," uses a contour diagram to analyze the evaluation results, but it also cannot provide a specific numerical score for the similarity.

[0004] Current research on similarity measurement of cigarette aroma characteristics mainly focuses on distance measurement. The *China Tobacco Science Journal*, in its article "Comprehensive Evaluation of Cigarette Quality Stability—Based on Multi-Feature Similarity Analysis and Principal Component Analysis," utilizes principal component analysis to reduce the dimensionality of cigarette feature data and employs cosine similarity and Euclidean distance as similarity measures. The *Jiangxi Agricultural Journal*, in its article "Application of Similarity Distance Analysis in Cigarette Sensory Evaluation Analysis," summarizes the application of Euclidean distance, Mahalanobis distance, and Minkowski distance in cigarette sensory evaluation methods. The *Tobacco Science and Technology Journal*, in its article "Establishment and Application of a Tobacco Sensory Evaluation Analysis Method," proposes a sensory deviation evaluation index, the D-value, based on Euclidean distance. However, the similarity measurement method using Euclidean distance (i.e., total squared difference or root mean squared error) has potential problems. When using Euclidean distance for similarity measurement, a lower total squared difference between the test sample and the brand's core aroma characteristics across all aroma indicators indicates that the product better matches the brand's characteristics. However, there are significant differences among different aroma indicators. Some aroma indicators have large variances, resulting in highly discrete data and large fluctuations in distances. Other aroma indicators have large values, often leading to larger distances. Some aroma indicators better highlight brand characteristics but are not reflected in traditional similarity measurements, thus requiring improvement in the accuracy of the measurement results. Therefore, how to effectively improve the accuracy of cigarette aroma feature similarity measurement results has become a pressing technical problem for those skilled in the art. Summary of the Invention

[0005] The purpose of this invention is to provide a method for measuring the similarity of cigarette aroma features, so as to solve the above-mentioned technical problems in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for measuring the similarity of cigarette aroma features includes the following steps:

[0008] Step S1: Select representative brands of cigarettes from the designated brand, organize tasters to taste them, and record the scores of each aroma index.

[0009] Step S2: Calculate the mean and standard deviation of the scores of each aroma index in different brands of cigarettes, and construct a multimodal normal distribution model;

[0010] Step S3: Based on the multimodal normal distribution model, construct a probability model for generating brand fragrance characteristics;

[0011] Step S4: Set the weight values ​​of each fragrance index and add each weight value to the brand fragrance feature generation probability model to construct a similarity calculation formula;

[0012] Step S5: Organize the taste testers to taste the test cigarettes, record the scores of each aroma index of the test cigarettes, and substitute them into the similarity calculation formula to calculate the similarity score.

[0013] Step S6: Evaluate the similarity between the test cigarette and the set brand cigarette based on the similarity score.

[0014] Preferably, in step S3, the brand fragrance feature generation probability model is as follows:

[0015]

[0016] Where: A = (a1, a2, a3, ..., a k ), where a1, a2, a3, ..., a are vector representations of the aroma characteristics of the tested cigarettes; k Θ is the vector representation of the various aroma indices of the tested cigarettes; Θ = (θ1, θ2, θ3, ..., θ k ), where θ1, θ2, θ3, ..., θ are vector representations of the aroma characteristics of a brand of cigarettes. k , which is a vector representation of the various aroma indicators of a brand of cigarettes.

[0017] Preferably, in step S4, the specific steps for setting the weight values ​​of each aroma index are as follows:

[0018] Step S41: Select other representative leading cigarette brands in the industry as industry samples representing the average level of the cigarette industry, organize taste testers to taste them, and record the scores of each aroma index.

[0019] Step S42: Obtain the test values ​​of each aroma index of the set brand of cigarettes relative to each aroma index of the industry samples through hypothesis testing, and use the test value as the weight value of the aroma index.

[0020] Preferably, in step S42, a one-tailed t-test is used as the hypothesis test, and the weight value of the aroma index is equal to the logarithm of the t-value of the aroma index of the set brand cigarettes divided by the sum of the logarithms of the t-values ​​of all aroma indices of the industry samples.

[0021] Preferably, in step S4, for each aroma index whose score distribution during the evaluation conforms to a normal distribution, the similarity calculation formula is as follows:

[0022]

[0023] Where S is the similarity; t j The t-value is the aroma index of the industry sample; t i To set the t-value for the aroma index of branded cigarettes, t i The value can be:

[0024]

[0025] Preferably, for each aroma index whose score distribution during the smoking evaluation does not conform to a normal distribution, the lower limit of S is set to 0. The baseline similarity when each aroma index of the test cigarette is equal to the average value of the designated brand of cigarettes is calculated and denoted as S'. This S' is then added as a subtraction factor to the similarity calculation formula, i.e.:

[0026]

[0027] Preferably, it further includes step S7: performing thermogravimetric analysis on the test cigarette and on different brands of cigarettes that are representative of the set brand, and calculating the pyrolysis difference.

[0028] Preferably, the method for obtaining the pyrolysis difference is as follows:

[0029] Cigarettes from various brands were used as analytical samples. A certain amount of tobacco shreds was pulverized using a high-speed pulverizer with a 60-80 mesh sieve. 10-20 mg of each sample was weighed and placed in an alumina plate crucible. Using a thermal analyzer under N2 purging at a flow rate of 100 mL / min, the samples were heated from room temperature to 873 K at a rate of 10 K / min. Data from thermogravimetric analysis and differential scanning calorimetry were collected simultaneously, and the pyrolysis difference was calculated using the following formula.

[0030]

[0031] Where: NRMSE represents the pyrolysis difference; N represents the number of recording points during thermogravimetric analysis in the range of 100–600℃; dm represents the mass percentage of the sample analyzed at time T; and T represents time. To set the quality loss rate for the l-th record point of the brand; The mass loss rate at the l-th recording point of the test cigarette.

[0032] The beneficial effects of this invention are as follows:

[0033] The present invention provides a method for measuring the similarity of cigarette aroma features. It constructs a brand aroma feature generation probability model and sets weight values ​​for each aroma indicator, thereby establishing a similarity calculation formula. The scores of each aroma indicator of the test cigarette during tasting are then substituted into the similarity calculation formula to calculate the similarity score. Based on the similarity score, the similarity of the aroma features of the test cigarette and the designated brand cigarette can be easily determined. Therefore, the similarity calculation formula of the present invention not only utilizes the brand aroma feature generation probability model but also sets weight values ​​for each aroma indicator, effectively avoiding the problems caused by using Euclidean distance for similarity measurement in existing technologies. This significantly improves the accuracy of cigarette aroma feature similarity measurement results. Attached Figure Description

[0034] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly described below, and the specific embodiments of the present invention will be further described in detail with reference to the drawings, wherein...

[0035] Figure 1 A flowchart of a method for measuring the similarity of cigarette aroma features provided in an embodiment of the present invention. Detailed Implementation

[0036] To enable those skilled in the art to better understand the technical solution of the present invention, the present solution will be further described in detail below with reference to specific embodiments.

[0037] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for measuring the similarity of cigarette aroma features, which includes the following steps:

[0038] Step S1: Select representative brands of cigarettes from the designated brand, organize tasters to taste them, and record the scores of each aroma index.

[0039] Step S2: Calculate the mean and standard deviation of the scores of each aroma index in different brands of cigarettes, and construct a multimodal normal distribution model;

[0040] Step S3: Based on the multimodal normal distribution model, construct a probability model for generating brand fragrance characteristics;

[0041] Step S4: Set the weight values ​​of each fragrance index and add each weight value to the brand fragrance feature generation probability model to construct a similarity calculation formula;

[0042] Step S5: Organize the taste testers to taste the test cigarettes, record the scores of each aroma index of the test cigarettes, and substitute them into the similarity calculation formula to calculate the similarity score.

[0043] Step S6: Evaluate the similarity between the test cigarette and the set brand cigarette based on the similarity score.

[0044] The cigarette aroma feature similarity measurement method provided in this invention constructs a brand aroma feature generation probability model and sets weight values ​​for each aroma indicator. Based on this, a similarity calculation formula is built. Then, the scores of each aroma indicator of the test cigarette during tasting are substituted into the similarity calculation formula to calculate the similarity score. Based on the similarity score, the similarity of the aroma features of the test cigarette and the designated brand cigarette can be easily determined. It is evident that the similarity calculation formula of this invention not only utilizes the brand aroma feature generation probability model but also sets weight values ​​for each aroma indicator, effectively avoiding the problems caused by using Euclidean distance for similarity measurement in existing technologies. Therefore, it can effectively improve the accuracy of cigarette aroma feature similarity measurement results.

[0045] Specifically, the sensory evaluation method for Chinese-style cigarette style can be used to evaluate and score the aroma indicators.

[0046] Furthermore, in step S3, the brand fragrance feature generation probability model is as follows:

[0047]

[0048] Where: A = (a1, a2, a3, ..., a k ), where a1, a2, a3, ..., a are vector representations of the aroma characteristics of the tested cigarettes; k Θ is the vector representation of the various aroma indices of the tested cigarettes; Θ = (θ1, θ2, θ3, ..., θ k ), where θ1, θ2, θ3, ..., θ are vector representations of the aroma characteristics of a brand of cigarettes. k , which is a vector representation of the various aroma indicators of a brand of cigarettes.

[0049] It should be noted that the original probability model for generating brand fragrance features could be:

[0050]

[0051] However, for ease of calculation and to avoid numerical underflow during algorithm implementation, the negative logarithmic function is used to transform the generation probability model, resulting in the above formula (1-1).

[0052] Understandably, the smaller the negative logarithm of the generation probability, the more it tends to resemble the aroma characteristics of the given brand. However, if only P(A|Θ) is considered, the following situation may affect the judgment: For example, given two test cigarettes A1 and A2, P(A1|Θ) is slightly larger than P(A2|Θ), but test cigarette A2 performs similarly to the brand's more prominent aroma indicators. In this case, comparing only P(A|Θ) might lead to the conclusion that test cigarette A1 better matches the brand's aroma characteristics. However, considering both P(A|Θ) and its performance on the brand's more prominent aroma indicators, the true conclusion might be that test cigarette A2 is closer to the brand's aroma characteristics. To avoid the above problems, it is necessary to set weight values ​​for each aroma indicator.

[0053] In step S3, when constructing the brand aroma feature generation probability model based on the multimodal normal distribution model, since the probability density function of the normal distribution is continuous, this embodiment can divide the value interval of the normal distribution (μ-3*σ, μ+3*σ) into multiple small intervals with a unit length of 0.01. 1,000,000 random data points are generated using the Box-Muller algorithm, and the probability of falling within each interval is calculated as a method for discretizing the continuous function. Here, μ is the mean, and σ is the standard deviation.

[0054] To ensure that the similarity metric highlights the aroma characteristics of the brand that receive high attention, it is necessary to assess the degree of difference between the brand and the overall industry level in each aroma indicator, and use this assessment as the weight value for each aroma indicator in the similarity metric. Taking flue-cured cigarettes as an example, since there are hundreds of different brands of flue-cured cigarettes on the market, it can be assumed that the score distribution of flue-cured cigarettes in each aroma indicator follows a normal distribution. The normal distribution of aroma indicators provides a prerequisite for hypothesis testing. The test values ​​obtained through hypothesis testing reflect the difference between the brand's aroma indicators and other brands in the industry, and can be used to indicate the degree to which each aroma indicator highlights the brand's aroma characteristics. Therefore, the specific steps for setting the weight values ​​of each aroma indicator in step S4 are as follows:

[0055] Step S41: Select other representative leading cigarette brands in the industry as industry samples representing the average level of the cigarette industry, organize taste testers to taste them, and record the scores of each aroma index.

[0056] Step S42: Obtain the test values ​​of each aroma index of the set brand of cigarettes relative to each aroma index of the industry samples through hypothesis testing, and use the test value as the weight value of the aroma index.

[0057] Furthermore, in step S42, a one-tailed t-test is used as a hypothesis test, and the weight value of the aroma index is equal to the logarithm of the t-value of the aroma index of the set brand cigarettes divided by the sum of the logarithms of the t-values ​​of all aroma indices of the industry samples.

[0058] Specifically, in step S4, for each aroma index whose score distribution during the evaluation conforms to a normal distribution, the similarity calculation formula is as follows:

[0059]

[0060] Where S is the similarity; t j Let t be the t-value of the j-th aroma index of the industry sample; t i To set the t-value, or test value, for the i-th aroma index of a branded cigarette, t i The value can be:

[0061]

[0062] It should be noted that, due to t i The absolute value of t may be less than 1, therefore it is necessary to adjust t. i Perform the appropriate processing: when t i When the absolute value is less than or equal to 1, it indicates that the brand of cigarettes is almost indistinguishable from other brands in the industry in terms of aroma. In this case, the numerator of its weighting term is set to 0, even if t i =1, indicating that this fragrance index contributes negligibly to the brand's fragrance characteristics; when t i When the absolute value of t is greater than 1, i The value of t is not processed. i A larger absolute value indicates that the brand of cigarettes is significantly different from other brands in the industry in terms of aroma. In this case, the weight is larger, and its generation probability will contribute much more to the similarity S than other aroma indicators.

[0063] When the score distribution of certain aroma indicators does not conform to a normal distribution, such as the absence of "sun-cured tobacco aroma" in most flue-cured cigarettes, the "green aroma" mainly found in "Nanjing" cigarettes, and the "herbal aroma" mainly found in "Jinsheng" cigarettes, a correction term needs to be added. When the generation probability P(A|Θ) is large, S is smaller; and when S is smaller, the test cigarette A is closer to the aroma characteristics of the designated brand of cigarette. Therefore, for ease of representation, the lower limit of S is set to 0. The baseline similarity is calculated when the average score of each aroma indicator of the test cigarette is equal to that of each aroma indicator of the designated brand of cigarette, denoted as S'. This is then added as a subtraction factor to the similarity calculation formula, i.e.:

[0064]

[0065] It is understandable that for aroma indicators whose score distribution during the smoking evaluation conforms to a normal distribution, formula (2-1) is used to calculate the similarity. For aroma indicators whose score distribution during the smoking evaluation does not conform to a normal distribution, formula (2-2) is used to calculate the similarity.

[0066] In other embodiments, when there is sufficient data, a one-tailed z-test can also be used as a hypothesis test, which will not be elaborated here.

[0067] Furthermore, the method for measuring the similarity of cigarette aroma features also includes step S7: performing thermogravimetric analysis on the test cigarette and different brands of cigarettes that are representative of the set brand, and calculating the pyrolysis difference to verify the effectiveness and feasibility of the measurement method.

[0068] Specifically, the method for obtaining the pyrolysis difference is as follows:

[0069] Cigarettes from various brands were used as analytical samples. A certain amount of tobacco shreds was pulverized using a high-speed pulverizer with a 60-80 mesh sieve. 10-20 mg of each sample was weighed and placed in an alumina plate crucible. Using a thermal analyzer under N2 purging at a flow rate of 100 mL / min, the samples were heated from room temperature to 873 K (600 °C) at a rate of 10 K / min. Data from thermogravimetric analysis and differential scanning calorimetry were collected simultaneously, and the pyrolysis difference was calculated using the following formula.

[0070]

[0071] Where: NRMSE represents the pyrolysis difference; N represents the number of recording points during thermogravimetric analysis in the range of 100–600℃; dm represents the mass percentage of the sample analyzed at time T, in %; T represents time, in min. To set the quality loss rate for the l-th recording point of the brand, the unit is: % / min; The mass loss rate at the l-th recording point of the test cigarette is expressed in % / min.

[0072] It should be noted that cigarette products under the same brand often share similar aroma characteristics. For example, "Jin Sheng" cigarettes have a prominent herbal aroma, "Nanjing" cigarettes have a prominent fresh and savory aroma, and "Golden Leaf" cigarettes have a prominent flue-cured tobacco aroma. Driven by market competition and corporate innovation, each cigarette company's products have gradually developed distinctive brand aroma characteristics. Therefore, measuring the similarity between sample aroma characteristics and brand characteristics is of great significance for product design, product development, and market analysis.

[0073] This invention selects representative cigarette brands from the same brand to form a reference sample. The generation probability of the test sample under the reference sample generation model can be used to measure the similarity between the test sample and the reference sample. During the production process, fluctuations in tobacco quality, flavoring, additives, and blending, as well as variations in production, transportation, and storage environments, can cause changes in the aroma characteristics of cigarettes within a certain range. Therefore, for a single cigarette in the reference sample, its aroma index score distribution can be considered to follow a normal distribution, while the distribution of the entire reference sample follows a multimodal normal distribution. Based on the aroma characteristic distribution of the reference sample cigarettes, a generation probability model for the test sample can be established, using the generation probability as a metric.

[0074] Comparative experiment:

[0075] Table 1 shows the similarity measurements performed using four leading cigarette brands and four different brands of "Golden Leaf" cigarettes as test samples, employing Euclidean distance, cosine similarity, and the similarity measurement method of "Golden Leaf" brand aroma characteristics (i.e., selecting "Golden Leaf" as the brand of cigarettes mentioned above).

[0076] Table 1. Similarity measurement results and pyrolysis differences using different measurement methods.

[0077]

[0078] As can be seen from Table 1:

[0079] The Euclidean distance between the "Golden Leaf" brand cigarette test samples and the average level of "Golden Leaf" cigarettes is relatively low, indicating that the "Golden Leaf" brand cigarette test samples are closer to the average level of "Golden Leaf". However, the scores of "Core Brand Sample 2" and "Golden Leaf Sample 3" are at the same level, indicating that using Euclidean distance as a similarity measure failed to take into account the distribution of various aroma indicators, thus resulting in inaccurate measurement results.

[0080] The cosine similarity results show that the scores of the "Golden Leaf" brand test samples are generally higher than those of other leading brands in the industry. This indicates that the vector formed by the smoking evaluation results of the "Golden Leaf" brand cigarette test samples has a smaller angle with the vector formed by the average level of "Golden Leaf" cigarettes, indicating a higher similarity. However, the cosine similarity results are not significant; differences can only be seen when four decimal places are retained, and the differences are very small and inaccurate.

[0081] The similarity score S shows that the "Golden Leaf" brand cigarette test sample scored significantly lower than other leading brands in the industry. Based on the principle that the lower the score, the higher the similarity, it can be concluded that the "Golden Leaf" brand test sample is quite similar to the aroma characteristics of the "Golden Leaf" brand, which is in line with the experimental expectations.

[0082] The pyrolysis difference values ​​show that the pyrolysis difference between the "Golden Leaf" brand and the "Golden Leaf" test sample is relatively small, which verifies the feasibility of the similarity measurement method of the present invention.

[0083] The cigarette aroma feature similarity measurement method proposed in this invention overcomes the shortcomings of existing technologies, eliminating problems such as overlapping classification boundaries and low recognition. It effectively improves product personnel's ability to recognize product maintenance results and provides guidance for new products. Furthermore, to verify the effectiveness and feasibility of using a probabilistic generation model as a measurement method for cigarette similarity, this invention employs thermogravimetric analysis. A specific brand of cigarettes and other leading brands representing the overall industry level are used as analysis samples. The same thermogravimetric analysis method is used for thermogravimetric experiments, and the standard root error is used to evaluate the pyrolysis differences between the samples, thus verifying the effectiveness and feasibility of the above measurement method. This invention proposes a cigarette aroma feature similarity measurement method combined with thermogravimetric analysis, which can be used to measure the similarity or difference between the aroma features of a test sample and a single brand of cigarettes.

[0084] The above are merely preferred embodiments of the present invention. It should be noted that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. Moreover, after reading the contents of the present invention, those skilled in the art can make various modifications or alterations to the present invention, and these equivalent forms also fall within the scope defined by the appended claims.

Claims

1. A method for measuring the similarity of aroma characteristics in cigarettes, characterized in that, It includes the following steps: Step S1: Select representative brands of cigarettes from the designated brand, organize tasters to taste them, and record the scores of each aroma index. Step S2: Calculate the mean and standard deviation of the scores of each aroma index in different brands of cigarettes, and construct a multimodal normal distribution model; Step S3: Based on the multimodal normal distribution model, construct a probability model for generating brand fragrance characteristics; Step S4: Set the weight values ​​of each fragrance index and add each weight value to the brand fragrance feature generation probability model to construct a similarity calculation formula; Step S5: Organize the taste testers to taste the test cigarettes, record the scores of each aroma index of the test cigarettes, and substitute them into the similarity calculation formula to calculate the similarity score. Step S6: Evaluate the similarity between the test cigarette and the set brand cigarette based on the similarity score; In step S3, the brand fragrance feature generation probability model is as follows: ， Where: A = (a1,a2,a3,…,a…) k ), where a1, a2, a3, ..., a are vector representations of the aroma characteristics of the tested cigarettes; k Let Θ be the vector representation of each aroma index of the tested cigarette; Θ=(θ1, θ2, θ3,…, θ k Let θ1, θ2, θ3, ..., θ be the vector representation of the aroma characteristics of the branded cigarettes. k , is the vector representation of each aroma index of the branded cigarette; k is the total number of aroma indices; In step S4, the specific steps for setting the weight values ​​of each aroma index are as follows: Step S41: Select other representative leading cigarette brands in the industry as industry samples representing the average level of the cigarette industry, organize taste testers to taste them, and record the scores of each aroma index. Step S42: Obtain the test values ​​of each aroma index of the set brand of cigarettes relative to each aroma index of the industry samples through hypothesis testing, and use the test value as the weight value of the aroma index.

2. The method for measuring the similarity of cigarette aroma features according to claim 1, characterized in that, In step S42, a one-tailed t-test is used as a hypothesis test. The weight value of the aroma index is equal to the logarithm of the t-value of the aroma index of the set brand cigarettes divided by the sum of the logarithms of the t-values ​​of all aroma indices of the industry samples.

3. The method for measuring the similarity of cigarette aroma features according to claim 2, characterized in that, In step S4, for each aroma index whose score distribution during the evaluation conforms to a normal distribution, the similarity calculation formula is as follows: ， Where S is the similarity; t j The t-value is the aroma index of the industry sample; t i To set the t-value for the aroma index of branded cigarettes, t i The value can be: 。 4. The method for measuring the similarity of cigarette aroma features according to claim 2, characterized in that, For aroma indicators whose score distribution during the smoking evaluation does not conform to a normal distribution, the lower limit of S is set to 0. The baseline similarity is calculated when the average score of each aroma indicator of the test cigarette is equal to that of each aroma indicator of the designated brand of cigarettes, denoted as S'. This baseline similarity is then added as a subtraction factor to the similarity calculation formula, i.e.: 。 5. The method for measuring the similarity of cigarette aroma characteristics according to any one of claims 1 to 4, characterized in that, It also includes step S7: performing thermogravimetric analysis on the test cigarette and on different brands of cigarettes that are representative of the set brand, and calculating the pyrolysis difference.

6. The method for measuring the similarity of cigarette aroma features according to claim 5, characterized in that, The method for obtaining the pyrolysis difference is as follows: Cigarettes from various brands and representative varieties were used as analytical samples. A certain amount of tobacco shreds was pulverized using a high-speed pulverizer with a 60-80 mesh sieve. 10-20 mg of each sample was weighed and placed in an alumina plate crucible. Using a thermal analyzer under N2 purging at a flow rate of 100 mL / min, the samples were heated from room temperature to 873 K at a rate of 10 K / min. Data from thermogravimetric analysis and differential scanning calorimetry were collected simultaneously, and the pyrolysis difference was calculated using the following formula. NRMSE=100% × , Where: NRMSE represents the pyrolysis difference; N represents the number of recording points during thermogravimetric analysis in the range of 100–600℃; dm represents the mass percentage of the sample analyzed at time T; and T represents time. To set the brand's first Quality loss rate per record point; For the first test of cigarettes Quality loss rate per record point.

Images