阅读说明：本技术 一种烤烟烟丝的产地及等级识别方法 (Method for identifying producing area and grade of flue-cured tobacco shreds ) 是由 毕一鸣 廖付 张立立 何文苗 李永生 帖金鑫 李石头 田雨农 郝贤伟 赵振杰 许 于 2019-10-16 设计创作，主要内容包括：本发明提供一种烤烟烟丝产地及等级识别方法,该方法包括如下步骤：步骤1)获取多个烟叶(烟丝)样本的表征指标,所述的表征指标包括化学指标、香型指标与部位特征；步骤2)获取步骤1)中多个烟叶(烟丝)样本的近红外光谱,利用偏最小二乘方法分别对步骤1)获得的表征指标、步骤2)获得的近红外光谱进行关联,建立模型；步骤3)将全国各产地、各等级烟叶样本的历史近红外光谱应用到步骤2)所述的模型中,得到全国各产地、各等级烟叶样本的模型计算值；统计各产地、等级的各模型阈值范围,建立匹配矩阵；步骤4)获取未知样本的近红外光谱,通过步骤2)模型预测出表征指标后,结合步骤3)的匹配矩阵,预测其产地及等级。(The invention provides a method for identifying the producing areas and the grades of flue-cured tobacco shreds, which comprises the following steps of 1) obtaining characterization indexes of a plurality of tobacco leaf (shred) samples, wherein the characterization indexes comprise chemical indexes, odor indexes and position characteristics, 2) obtaining near infrared spectrums of the tobacco leaf (shred) samples in the step 1), respectively correlating the characterization indexes obtained in the step 1) and the near infrared spectrums obtained in the step 2) by using a partial least square method, establishing a model, 3) applying historical near infrared spectrums of tobacco leaf samples in all producing areas and all grades in the country to the model in the step 2) to obtain model calculation values of the tobacco leaf samples in all producing areas and all grades in the country, counting threshold value ranges of the models in all producing areas and grades, establishing a matching matrix, 4) obtaining the near infrared spectrums of unknown samples, and predicting the representing indexes through the model in the step 2) and then combining the matching matrix in the step 3) to predict the producing areas and the grades.)

1, kinds of flue-cured tobacco shred producing area and grade identification method, which is characterized in that the method comprises the following steps:

step 1) obtaining characterization indexes of a plurality of tobacco leaf (cut tobacco) samples, wherein the characterization indexes comprise chemical indexes, odor indexes and part characteristics; the chemical indexes at least comprise total sugar content, nicotine content, reducing sugar content, chlorine content, potassium content and total nitrogen content; the odor index at least comprises a faint scent index, a middle scent index and a strong scent index; the site characteristic is selected from the upper, middle or lower lobe;

step 2) acquiring near infrared spectrums of a plurality of tobacco leaf (tobacco shred) samples in the step 1), and respectively correlating the characterization indexes obtained in the step 1) and the near infrared spectrums obtained in the step 2) by using a partial least square method to establish a model;

step 3) applying the historical near infrared spectrums of the tobacco leaf samples of all nationwide producing areas and all levels to the model in the step 2) to obtain model calculation values of the tobacco leaf samples of all nationwide producing areas and all levels; counting threshold ranges of the models of all producing areas and levels, and establishing a matching matrix;

and 4) acquiring a near infrared spectrum of an unknown sample, predicting a characterization index through the model in the step 2), and predicting the producing area and the grade of the unknown sample by combining the matching matrix in the step 3).

2. The method as claimed in claim 1, wherein the modeling samples in step 1) cover the national main flue-cured tobacco producing areas and grades, and the number of the samples is not less than 500, preferably the flue-cured tobacco producing areas and grades cover the national 14 flue-cured tobacco producing areas, namely Anhui, Chongqing, Fujian, West, Guizhou, Hunan, Hubei, Henan, Sichuan, Shandong, Yunnan, Jiangxi, Heilongjiang, Liaoning, and 9 main grades, namely B3F, B2F, B1F, C1F, C2F, C3F, C4F, X2F and X3F.

3. The method according to claim 1, wherein in step 1) the scent index, fen scent index [ 100 ], medium scent index [ 010 ], Luzhou scent index [ 001 ]; site features in step 1), upper lobe: 1; middle leaf: 2; lower lobe: 3.

4. the method according to claim 1, characterized in that the collected near infrared spectrum is subjected to a spectral pre-treatment in step 2); preferably, the preprocessing method comprises smoothing, 1 st derivative, 2 nd derivative or standard normal correction; more preferably, 1 st derivative + standard normal correction is used.

5. The method of claim 1, wherein in step 2) the chemical indicators and site characteristics are modeled using PLS1, each indicator being modeled separately, and the flavor indicators are modeled using PLS2 and three indicators .

6. The method of claim 1, wherein in step 3), the mean value and standard deviation of each model (chemical and odor type characterization index) are calculated according to the statistics of the producing area and grade, and the standard deviation of the mean value + - times is used as the threshold range

Wherein m is_i，s_iRespectively representing the mean value and the standard deviation of the ith index of a certain grade of a certain production area; k is a radical of_iIs a counting value, if the counting value is 1 within the threshold condition range, otherwise, the counting value is 0;

counting the producing area and the grade in the historical tobacco sample data, generating a matching matrix, wherein the row number and the column number of the matrix are the producing area number and the grade number in the historical data respectively, and each elements in the matrix are calculated by a formula (1).

7. The method according to claim 6, wherein in step 4), the characterization index of the unknown sample prediction is compared with the matching matrix obtained in step 3),

if the three of the chemistry, the odor type and the part are in accordance with the producing area and the grade, the producing area and the grade to which the unknown sample belongs are determined;

if one or more of items exceed the threshold, the unknown sample is not in the existing modeled pay zone and grade, and a voting algorithm is implemented, wherein the algorithm specifically comprises the following steps:

k represents the comprehensive matching degree of 10 indexes; the maximum value of K is 10, which represents that all 10 indexes of a certain sample fall within a threshold range of a certain grade of a certain producing area; the minimum value of K is 0, which represents that all 10 indexes of a certain sample do not fall within a threshold range of a certain grade of a certain producing area;

applying the prediction index of the unknown sample to K values of all the producing areas and grades in the calculation historical database, and identifying the producing area and grade represented by the maximum value of the K values as the producing area and grade of the unknown sample;

and calculating K values of all the producing areas and grades in the historical database, and identifying the producing area and grade represented by the maximum value of the K values as the producing area and grade of the unknown sample.

Technical Field

The invention belongs to the field of flue-cured tobacco shred attribute identification and production place identification, and particularly relates to flue-cured tobacco production place and grade identification methods based on near infrared spectrum.

Background

As agricultural products, no powerful analytical means exists at present, the grade, the origin and the like of tobacco leaves can be rapidly identified, only limited judgment can be carried out by means of sensory evaluation, and an objective identification method is lacked in the actual requirements of identifying illegal tobacco leaves and the like.

The near infrared spectrum mainly comprises frequency doubling and frequency combining absorption of hydrogen-containing group vibration, contains composition information of most types of organic compounds, has rich information related to chemical components of tobacco leaves, and is favorable for reflecting tobacco leaf information and tracing through spectra.

A method for quickly judging the producing area of fresh tea leaves by near-infrared spectrums includes such steps as scanning by near-infrared spectrometer to obtain the near-infrared spectrums of fresh tea leaves in different producing areas, analyzing the main components of the spectrum of fresh tea leaves, creating artificial neural network prediction models for the producing areas of fresh tea leaves in different information transfer modes, and judging if the tea leaves are Yulu tea.

Cai Heiwuang et al disclose near infrared spectrum method for fast detecting radix tetrastigme producing area and identify five producing areas.

The Shifengcheng and the like use a PLS-DA algorithm based on near infrared spectrum to judge the tobacco production area, respectively establish production area judging models for single cigarettes in 4 production areas of Sichuan, Yunnan, Chongqing and Fujian, and the prediction precision of verification set samples in each production area is more than 93 percent. Preprocessing full-waveband spectral characteristic information of flue-cured tobacco by Wang-Ding and the like, and establishing a flue-cured tobacco flavor type PLS-DA qualitative discrimination model with the recognition accuracy rate of 100%;

y.zhang, Duan J, x.liu et al, respectively, propose methods for modeling tobacco chemical components using near infrared spectroscopy.

Ni, Hana, Shao, Tan, et al, respectively, propose methods for attribute or brand classification of tobacco leaves or cigarettes using near infrared spectroscopy.

The method only provides identification of 3 and 4 producing areas, and relates to dozens of tobacco planting provinces and a plurality of tobacco leaf grades in actual requirements.

Reference documents:

shifengcheng, Lidong Liliang, von Lin, etc. the PLS-DA algorithm based on near infrared spectra discriminates the flue-cured tobacco leaf producing area [ J ] tobacco technology, 2013(4).

Wang ding, zhao mingjingju, bubo, et al, method for identifying flue-cured tobacco of different flavor styles using visible-near infrared spectroscopy [ J ] chinese tobacco science 2015(6).

Application number of near infrared spectrum methods for rapidly determining fresh tea leaf producing area, such as Rongshengceng, is 201610930724.5

Method for rapidly detecting radix tetrastigme producing area by near infrared spectra of Cai Heiwuang et al, application number 201710371389.4

Y.Zhang，Q.Cong，Y.Xie，J.Yang，B.Zhao，Quantitative analysis of routinechemical constituents in tobacco by near-infrared spectroscopy and supportvector machine，Spectrochim. Acta A 71(2008)1408-1413.

Duan J，Huang Y，LiZ，et al.Determination of 27 chemical constituents inChinese southwest tobacco by FT-NIR spectroscopy[J].Industrial Crops andProducts，2012，40(none)：21-26.

X.Liu，H.-C.Chen，T.-A.Liu，Y.-L.Li，Z.-R.Lu，W.-C.Lu，Application of PCA-SVR to NIR prediction model for tobacco chemical composition，Spectrosc.Spectral Anal.27(2007)2460-2463.

L.-J.Ni，L.-G.Zhang，J.Xie，J.-Q.Luo，Pattern recognition of Chineseflue-cured tobaccos by an improved and simplified K-nearest neighborsclassification algorithm on near infrared spectra. Anal.Chim.Acta 633(2009)43-50.

M.Hana，W.F.McClure，T.B.Whitaker，M.W.White，D.R.Bahler，Applyingartificial neural networks：Part II.Using near infrared data to classifytobacco types and identify native grown tobacco，J.Near Infrared Spectrosc.5(1997)19-25.

Y.Shao，Y.He，Y.Wang，A new approach to discriminate varietiesoftobacco using vis/near infrared spectra，Eur.Food Res.Technol.224(2007)591-596.

C.Tan，M.Li，X.Qin，Study of the feasibility of distinguishingcigarettes of different brands using an Adaboost algorithm and near-infraredspectroscopy，Anal.Bioanal.Chem.389(2007) 667-674.

Disclosure of Invention

Aiming at the problems, the invention provides flue-cured tobacco shred producing areas and grade identification methods based on near infrared spectrum, which comprises the following steps of 1) obtaining characterization indexes of a plurality of tobacco shred (tobacco shred) samples, wherein the characterization indexes comprise chemical indexes, odor indexes and position characteristics, the chemical indexes at least comprise total sugar content, nicotine content, reducing sugar content, chlorine content, potassium content and total nitrogen content, the odor indexes at least comprise faint scent indexes, middle scent indexes and strong scent indexes, and the position characteristics are selected from upper leaves, middle leaves or lower leaves;

step 2) acquiring near infrared spectrums of a plurality of tobacco leaf (tobacco shred) samples in the step 1), and respectively correlating the characterization indexes obtained in the step 1) and the near infrared spectrums obtained in the step 2) by using a partial least square method to establish a model;

step 3) applying the historical near infrared spectrums of the tobacco leaf samples of all nationwide producing areas and all levels to the model in the step 2) to obtain model calculation values of the tobacco leaf samples of all nationwide producing areas and all levels; counting threshold ranges of the models of all producing areas and levels, and establishing a matching matrix;

and 4) acquiring a near infrared spectrum of an unknown sample, predicting a characterization index through the model in the step 2), and predicting the producing area and the grade of the unknown sample by combining the matching matrix in the step 3).

Preferably, the modeling samples in the step 1) cover the main flue-cured tobacco producing areas and grades in the country, and the number of the samples is not less than 500, preferably, the flue-cured tobacco producing areas and grades cover 14 flue-cured tobacco producing areas in the country, namely Anhui, Chongqing, Fujian, West, Guizhou, Hunan, Hubei, Henan, Sichuan, Shandong, Yunnan, Jiangxi, Heilongjiang and Liaoning, and 9 main grades are B3F, B2F, B1F, C1F, C2F, C3F, C4F, X2F and X3F.

Preferably, the fragrance model related to the invention is modeled by only using samples of Yunnan, Guizhou, Hubei and Hunan, wherein the Yunnan is a faint scent type and is marked as [ 100 ], the Guizhou and the Hubei are marked as middle scent type and are marked as [ 010 ], and the Hunan is a strong scent type and is marked as [ 001 ]; specifically, in the step 1), the odor type index is a faint odor type mark [ 100 ], the middle odor type mark [ 010 ], and the strong odor type mark [ 001 ]; site features in step 1), upper lobe: 1; middle leaf: 2; lower lobe: 3.

preferably, in the chemical model in step (1), the reference values of total sugar, nicotine, reducing sugar, chlorine, potassium and total nitrogen are measured by a flow analyzer.

Preferably, the collected near infrared spectrum is subjected to spectrum pretreatment in the step 2); preferably, the pre-processing method comprises smoothing, 1 st derivative, 2 nd derivative or standard normal correction; more preferably, 1 st derivative + standard normal correction is used.

Preferably, the chemical indexes and the site characteristics in the step 2) are modeled by PLS1, each index is modeled independently, and the odor type indexes are modeled by PLS2 and three indexes .

Preferably, in step 3), the mean value and standard deviation of the characterization indexes of each model (chemistry, odor type, etc.) are calculated according to the statistics of the producing area and the grade, and the standard deviation of the mean value +/- times is taken as the threshold range, namely

Wherein m is_i，s_iRespectively representing the mean value and the standard deviation of the ith index of a certain grade of a certain production area; y is_iIs the ith index model prediction value, k, of the sample to be measured_iIs a counting value, if the counting value is 1 within the threshold condition range, otherwise, the counting value is 0;

counting the origin and the grade in the historical tobacco sample data to generate a matching matrix, wherein the row number and the column number of the matrix are the production area number and the grade number in the historical data respectively, and each elements in the matrix are calculated by a formula (1).

Preferably, in the step 4), the predicted characterization indexes of the unknown samples are compared with the matching matrix obtained in the step 3), if the producing area and the grade of the unknown samples are consistent with the chemical, the odor type and the part, the producing area and the grade of the unknown samples are determined, if one or more than items exceed the threshold value, the unknown samples are not in the producing area and the grade of the existing modeling, and a voting algorithm is implemented, wherein the algorithm specifically comprises the following steps:

k represents the comprehensive matching degree of 10 indexes; the maximum value of K is 10, which represents that all 10 indexes of a certain sample fall within a threshold range of a certain grade of a certain producing area; the minimum value of K is 0, which represents that all 10 indexes of a certain sample do not fall within a threshold range of a certain grade of a certain producing area;

applying the prediction index of the unknown sample to K values of all the producing areas and grades in the calculation historical database, and identifying the producing area and grade represented by the maximum K value as the producing area and grade of the unknown sample;

and calculating K values of all the producing areas and grades in the historical database, and identifying the producing area and grade represented by the maximum value of the K values as the producing area and grade of the unknown sample.

The method provided by the invention predicts the unknown sample in three attributes of chemistry, flavor type and part by using the model in the step (2), namely the raw tobacco profile, counts the producing area and the grade in historical data, generates a matching matrix, wherein the row number and the column number of the matrix are respectively the producing area number and the grade number in the historical data, each elements in the matrix are calculated by a formula (1), the profile result is voted by combining the matching matrix, wherein the chemical calculation is that 6 indexes accord with the producing area and the grade which are most produced by a threshold value, the chemistry, the flavor type and the part, and the producing area and the grade which the unknown sample belongs to are determined, and if one or more indexes exceed the threshold value, the unknown sample is not in the producing area and the grade which are modeled currently.

Compared with the prior art, the invention has the following advantages:

1. the method can judge the tobacco shred producing area and grade without appearance and sensory evaluation.

2. Based on near infrared spectrum, the producing area grade of an unknown sample is judged by combining multiple indexes of chemistry, parts and odor, and the existing identification method is judged by using a single index.

3. The method can simultaneously judge the producing area and the grade.

4. The method can judge the sample attributes of 9 grades of 14 provinces and cities, and is more comprehensive and specific than other reporting methods.

5. The method for identifying the origin and the registration is stable and reliable.

Drawings

FIG. 1 shows training data and modeling results for a part model;

fig. 2 is a flowchart of an implementation of the identification method provided in the present invention.

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below, and the scope of the present invention is not limited by the embodiments, and is determined by the claims. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without any inventive step, are within the scope of the present invention.