Near-infrared qualitative analysis model based on PCA (Principal Component Analysis) algorithm and application
Through the near-infrared qualitative analysis model based on PCA algorithm, the production characteristics of tobacco leaves are analyzed using near-infrared spectroscopy technology, the problem of relying on empirical methods for the analysis of tobacco leaves in the existing technology is solved, and the rapid and accurate distinction of tobacco leaves is achieved, and scientific basis is provided to replace tobacco leaves in scarce production areas.
Patent Information
- Application Number
- CN202411643101.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-02-25
AI Technical Summary
The existing technology has not yet applied near-infrared spectroscopy technology to the analysis and identification of tobacco leaf production areas, resulting in the distinction and judgment of tobacco leaf production areas relying on empirical methods and lacking objective and accurate analysis methods.
A near-infrared qualitative analysis model based on PCA algorithm is used to scan the tobacco leaves from different origins through a near-infrared spectrometer, establish a data matrix, and conduct principal component analysis to calculate the mean and standard deviation of the principal component to determine the origin recognition line and achieve rapid and accurate distinction of the tobacco leaves.
It realizes rapid and precise distinction between the production areas of tobacco leaves, and provides a scientific basis for the replacement of tobacco leaves in scarce production areas. It is simple to operate, environmentally friendly, pollution-free, low cost and high efficiency.
Smart Images
Figure CN119516385A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of near-infrared nondestructive testing, and in particular relates to a near-infrared qualitative analysis model based on a PCA algorithm, which is used for the classification and identification of tobacco leaf producing areas. Background Art
[0002] Near Infrared Spectrum (NIRS) is an electromagnetic radiation wave between visible light (Vis) and mid-infrared (MIR), and is the first non-visible light region discovered in the absorption spectrum. In organic molecules, the atoms that make up chemical bonds or functional groups are in a state of constant vibration, and their vibration frequency is equivalent to the vibration frequency of infrared light. Therefore, when infrared light is irradiated on organic molecules, the chemical bonds or functional groups in the molecules can undergo vibration absorption. Different chemical bonds or functional groups have different absorption frequencies and will be in different positions on the infrared spectrum, so that information about what kind of chemical bonds or functional groups are contained in the molecules can be obtained. The near-infrared spectral region is consistent with the absorption region of the combined frequency and various frequency doublings of the vibration of hydrogen-containing groups in organic molecules. By scanning the near-infrared spectrum of the sample, the characteristic information of the hydrogen-containing groups in the organic molecules in the sample can be obtained. In addition, the use of near-infrared spectroscopy technology to analyze samples has the advantages of convenience, rapidity, efficiency, accuracy, low cost, no damage to samples, no consumption of chemical reagents, and no pollution to the environment. Therefore, this technology is favored by more and more people. NIR spectroscopy technology has the modern analytical characteristics of efficient, green and environmentally friendly analysis process, and has become one of the fastest-growing and most eye-catching spectral analysis technologies in recent years. As many as 80% to 90% of the chemical components in tobacco leaves can be studied and detected using near-infrared technology. The near-infrared spectrum contains a wealth of information on the correlation of tobacco chemical components, and there is a reliable material basis for clustering analysis and pattern recognition of tobacco leaves based on near-infrared information. In recent years, this technology has gradually been applied by the domestic tobacco industry to the quality control of raw materials and auxiliary materials and the quality detection of cigarettes, but there has been no report on its use in the analysis of tobacco leaf origin.
[0003] The origin characteristics of tobacco leaves are an important basis for the design of flue-cured tobacco cigarettes and the selection and use of tobacco leaves. Tobacco leaves from different regions have different aroma characteristics, but due to the complexity of tobacco leaves and smoke components, the distinction and judgment of tobacco leaf origin types can only rely on the empirical method of sensory evaluation. Near-infrared spectroscopy can capture data information closely related to regional origin characteristics to describe the origin characteristics of different tobacco leaves, achieve objective description and identification of tobacco leaf origin characteristics, establish a PCA differentiation model for tobacco leaf origins in the region, and use the model to analyze the origin similarity of tobacco leaves from different producing areas, providing a scientific basis for the replacement of tobacco leaves in scarce producing areas.
[0004] The present invention provides a near-infrared qualitative analysis model based on a PCA algorithm and its application. A near-infrared spectrometer is used to scan the spectra of tobacco leaves from different origins, and all the spectra are arranged in rows to obtain a data matrix. The PCA algorithm is used for analysis, and the mean of the principal components is calculated. The classification lines of tobacco leaves from different origins are determined, thereby establishing a PCA analysis model, and providing a scientific basis for the replacement of tobacco leaves in product formulas. Summary of the invention
[0005] The present invention provides a near-infrared qualitative analysis model based on a PCA algorithm. Model establishment comprises the following steps: using a near-infrared spectrometer to scan the spectra of tobacco leaves from the same origin, arranging all the near-infrared spectra from the same origin in rows to obtain a data matrix, performing principal component analysis on the matrix, determining an origin identification line by calculating relevant data, and establishing a qualitative model.
[0006] The near infrared qualitative analysis model of the present invention can quickly and accurately distinguish the origin of tobacco leaves. The present invention is simple and easy to operate, environmentally friendly and pollution-free, and saves reagents.
[0007] Preferably, model building includes the following steps:
[0008] (1) Spectral scanning: Scan the near-infrared spectrum of the sample to be tested and extract the characteristics of the spectrum;
[0009] (2) Establishing a data matrix: Arrange the near-infrared spectra of n tobacco leaves from the same origin in rows to obtain a data matrix D;
[0010] (3) Principal component analysis: Perform principal component analysis on D, calculate the principal components through the eigenvectors, and then retain the first principal component P1 and the third principal component P3;
[0011] (4) Calculate the mean of P1 and P3 and use this mean as the model differentiation center of tobacco leaves from the origin of “a”;
[0012] (5) Calculate the distance d from all tobacco leaves from the origin "X" to this center i (i=1, 2, ..., n); calculate all distances d i The mean and standard deviation s, with It is the identification line of tobacco leaves originating from “certain” place.
[0013] More preferably, the method comprises the following steps:
[0014] (1) Spectral scanning: The tobacco leaf samples were scanned using a near-infrared spectrometer to obtain their near-infrared spectra. The operating parameters were: spectral scanning range 12,000 to 4,000 cm-1, spectral resolution 8 cm-1, and scanning times 64 times (about 30 seconds). Spectral data were collected in a transmission mode and processed into the first-order differential of the absorption spectrum. Each sample was scanned 10 times to obtain the average spectrum. Each sample was scanned to obtain more than 2 average spectra.
[0015] (2) Spectral preprocessing: The standard normal variable transformation method is used to eliminate the differences caused by the particle size, surface scattering and optical path change of the solid sample. The spectrum is smoothed and noise-filtered by wavelet transform. The original spectrum is processed by second-order derivative to eliminate baseline drift, improve spectral resolution, and extract characteristic spectra.
[0016] (3) Exporting spectral data: exporting data points in the spectral range of 12 000 to 4 000 cm-1;
[0017] (4) Establishing a data matrix: Arranging the near-infrared spectra of tobacco leaves in the same area by rows to obtain a data matrix;
[0018] (5) Principal component analysis: Calculate the principal components through the eigenvectors, analyze and retain them;
[0019] (6) Calculate the mean of the retained principal components and use this mean as the center of the differentiation model of tobacco leaf producing areas in the region;
[0020] (7) Calculate the distance from all tobacco leaves in the area to this center, and calculate the mean and standard deviation of all distances. The mean value plus 1.2 times the standard deviation is used as the classification line for the tobacco leaf producing area in the area.
[0021] (8) Establishing PCA model: The model is established using the spectrum collected under normal conditions. The formula is as follows:
[0022]
[0023] Where: X(m×n) represents the spectral data matrix of m spectra and n wavelength points, T i is the score vector, P i is the loading vector. A is the number of factors, E is the error term, and T in the above table represents the transpose of the matrix.
[0024] The invention also provides an application method of the model.
[0025] When applying this model to identify the origin of unknown tobacco leaf samples, follow the steps below:
[0026] (1) Spectral scanning: Scan the near-infrared spectrum of the sample to be tested and extract the characteristic vector of the spectrum;
[0027] (2) Establishing a data matrix: Arrange the near-infrared spectra of n tobacco leaves from the same origin in rows to obtain a data matrix D;
[0028] (3) Principal component analysis: Perform principal component analysis on D, calculate the principal components through the eigenvectors, and then retain the first principal component P1 and the third principal component P3;
[0029] (4) Calculate the mean of P1 and P3 and use this mean as the model differentiation center of tobacco leaves from the origin of “a”;
[0030] (5) Calculate the distance di (i=1, 2, ..., n) from all tobacco leaves of "certain" origin to this center. Calculate the mean and standard deviation s of all distances di as the identification line of tobacco leaves of "certain" origin.
[0031] When applying this model to identify unknown brand products, follow the steps below:
[0032] (1) scanning and obtaining a near infrared spectrum of the sample;
[0033] (2) calculating the principal component of the tobacco leaf sample according to the characteristic vector of the spectrum;
[0034] (3) Calculate the distance from the principal component to the center of the constructed model, and judge the similarity of the origin based on the distance. If the distance falls within the classification identification line, it can be determined that the origin of the tobacco leaf is similar to that of the constructed model; otherwise, it is determined that they are of different origins.
[0035] The preferred steps are:
[0036] (1) scanning and obtaining a near infrared spectrum of the sample;
[0037] (2) Calculate the first principal component PC1 and the third principal component PC3 of the product through the feature vector in the modeling step 2;
[0038] (3) Calculate the distance d unknown from PC1 and PC3 to the center of tobacco leaves from the origin of "a", and determine the origin based on the distance. If d unknown < , then it can be determined that the unknown product belongs to the origin; otherwise, it is determined that it does not belong to the origin.
[0039] The beneficial effects of the present invention are as follows:
[0040] 1. The present invention uses the PCA algorithm to perform qualitative analysis on near-infrared spectra and establishes a similarity classification and recognition model for tobacco origins in different regions, which can quickly and accurately distinguish tobacco origins.
[0041] 2. The method of the present invention is simple and quick to operate, non-destructive to samples and non-polluting to the environment.
[0042] 3. The present invention has the advantages of low cost and high efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 This is the original spectrum of near infrared scanning of tobacco leaves;
[0044] Figure 2 It is a recognition model established by near infrared spectrum of tobacco leaves from GZ production area; DETAILED DESCRIPTION
[0045] The specific implementation modes of the present invention are further described in detail below with reference to the accompanying drawings.
[0046] Example 1
[0047] 1. Experimental instruments
[0048] MPA Fourier near-infrared spectrometer and ZM200 precision pulverizer produced by BRUKER (Germany).
[0049] 2. Sample collection
[0050] In order to make the established classification model have wide applicability, this embodiment selected 180 samples of first-cured tobacco samples of the same grade from the YN production area from 2021 to 2023 to establish the tobacco leaf origin model of the YN production area, and selected 20 unknown samples for external verification of the model.
[0051] 3. Sample preparation
[0052] The tobacco leaves were dried in an oven at 40°C to keep the moisture content of the samples basically consistent, and then fully crushed with a ZM200 precision grinder and passed through a 60-mesh sieve.
[0053] 4. Spectral scanning and data processing
[0054] The spectrum of tobacco leaf samples was scanned using the MPA Fourier near-infrared spectrometer (with near-infrared quantitative analysis diffuse reflection gold-plated large integrating sphere and sample rotator sampling accessories) produced by BRUKER (Germany), and the spectrum was processed using the qualitative analysis software QUANT6.5 in Bruker OPUS. The specific operation is as follows: put tobacco powder into the sample cup, the height in the cup is about 3 cm, press the weight on the sample for 10 seconds and then take it out, wipe the quartz glass at the bottom of the cup with gauze, and then place the sample cup on a rotating platform for NIR scanning. The operating parameters are: spectral scanning range 12 000~4 000cm -1 , spectral resolution 8cm -1 , scanning times 64 times (about 30S). Spectral data were collected in a transmission mode and processed into the first-order differential of the absorption spectrum. The original scan of the cigarette tobacco is shown in Figure 1In the modeling process, in order to eliminate the influence of noise and baseline, the standard normal variable transformation method was used to eliminate the differences caused by the particle size, surface scattering and optical path change of solid samples, the spectrum was smoothed and filtered by wavelet transform, and the second-order derivative was used for preprocessing. After the sample was scanned, the spectral data was processed using statistical software.
[0055] 5. Model building
[0056] The steps to build the model are as follows:
[0057] (1) Spectral scanning: The tobacco leaf samples were scanned using a near-infrared spectrometer to obtain their near-infrared spectra. The operating parameters were: spectral scanning range 12,000 to 4,000 cm-1, spectral resolution 8 cm-1, and scanning times 64 times (about 30 seconds). Spectral data were collected in a transmission mode and processed into the first-order differential of the absorption spectrum. Each sample was scanned 10 times to obtain the average spectrum. Each sample was scanned to obtain more than 3 average spectra.
[0058] (2) Spectral preprocessing: The standard normal variable transformation method is used to eliminate the differences caused by the particle size, surface scattering and optical path change of the solid sample. The spectrum is smoothed and noise-filtered by wavelet transform. The original spectrum is processed by second-order derivative to eliminate baseline drift, improve spectral resolution, and extract characteristic spectra.
[0059] (3) By analyzing the spectra of the scanned tobacco leaves in the producing areas, a PCA model is established: the model is established using the spectra collected under normal conditions, and the formula is as follows:
[0060]
[0061] (4) Establishing a data matrix: Arrange the near-infrared spectra of n tobacco leaves from the same origin in rows to obtain a data matrix D;
[0062] (5) Principal component analysis: Perform principal component analysis on D, calculate the principal components through the eigenvectors, and then retain the first principal component P1 and the third principal component P3;
[0063] (6) Calculate the mean of P1 and P3, and use this mean as the model differentiation center of tobacco leaves from the "certain" origin;
[0064] (7) Calculate the distance di (i=1, 2, ..., n) from all tobacco leaves of "certain" origin to this center. Calculate the mean and standard deviation s of all distances di as the identification line of tobacco leaves of "certain" origin.
[0065] When applying this model to identify the origin, follow the steps below:
[0066] 1) Scanning and obtaining the near infrared spectrum of the sample;
[0067] 2) Calculate the first principal component PC1 and the third principal component PC3 of the product through the feature vector in the modeling step 2;
[0068] 3) Calculate the distance d unknown from PC1 and PC3 to the center of tobacco leaves from the "certain" origin, and determine the origin based on the distance. If d unknown <, then it can be determined that the unknown product belongs to the origin, otherwise it is determined that it does not belong to the origin.
[0069] The recognition model established by the near infrared spectrum of tobacco leaves from GX origin is shown in Figure 2 .
[0070] 6. Model prediction results
[0071] In order to test the accuracy of the prediction of the above model, 20 tobacco leaf samples from different cities in the GZ production area in 2023 (all within the range of production predicted by the model) were randomly selected from the Guangxi Tobacco Raw Material Supply Department, and the identification ability of the model was externally tested. After the samples were processed, the spectra were collected using near-infrared, and finally the origin identification model was used to predict the origin of the spectra. The results are shown in Table 1.
[0072] Table 1 Recognition results of tobacco leaf feature model from “GZ” origin
[0073]
[0074] As can be seen from the above table, the model predicted the correct origins for 19 out of 20 samples, and the accuracy of origin identification reached 95%, indicating that the model's origin prediction accuracy is very high.
[0075] Example 2 Establishment of a near-infrared tobacco leaf origin qualitative model
[0076] 1. Experimental instruments
[0077] MPA Fourier near-infrared spectrometer and ZM200 automatic precision pulverizer produced by BRUKER (Germany).
[0078] 2. Sample collection
[0079] In this embodiment, a total of 306 samples of first-cured tobacco of different grades from the GX and YN production areas close to the modeled GZ production area from 2021 to 2023 were selected for the similarity analysis of tobacco leaf production areas of different origins in the constructed model.
[0080] 3. Model application
[0081] The sample preparation and spectral scanning model establishment methods in this embodiment are the same as those in Example 1. A total of 8 samples with similar origins to the model tobacco leaves were screened out from 306 tobacco leaves of different origins and grades. The specific analysis results are shown in Table 2. The selected similar samples were compared with the samples to be replaced by sensory evaluation to confirm whether they can be replaced (Table 2).
[0082] Table 2 Similarity recognition results of tobacco origin
[0083] Sample No. Sample name Similarity (%) Can sensory evaluation replace 1 <![CDATA[GX-BS(C3F)-2023]]> 97.2 yes 2 <![CDATA[GX-HZ(C2F)-2022]]> 93.6 yes 3 <![CDATA[GX-LL(C2F)-2023]]> 94.1 yes 4 <![CDATA[YN-QJ(C3F)-2022]]> 96.3 yes 5 <![CDATA[YN-DL(C2F)-2021]]> 97.0 yes 6 <![CDATA[YN-BS(C3F)-2023]]> 92.6 yes 7 <![CDATA[YN-QJ(C3F)-2022]]> 93.5 yes 8 <![CDATA[YN-BJ(C3F)-2021]]> 96.3 yes
[0084] From the above table, we can see that the model predicted 8 samples with similar origins from 306 samples of neighboring origins. Through smoking evaluation and verification, the identified samples can all be used for substitution, indicating that the model has a high prediction accuracy for the similarity of tobacco leaf origins.
[0085] It can be seen from the above embodiments that the model established by the present invention using near-infrared spectroscopy technology combined with the PCA method has a prediction accuracy of more than 92.6% for tobacco leaf origin, and a high recognition rate for the similarity of tobacco leaf origin. It can be seen that the application of near-infrared analysis technology can be well used for tobacco leaf origin similarity analysis, which is an effective and feasible method with strong practicality and realistic significance.
[0086] The above embodiments of the present invention are merely examples to clearly illustrate the present invention, and are not limitations on the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to enumerate the implementation methods used here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.
Claims
1. A near infrared qualitative analysis model based on PCA algorithm, characterized in that: The model establishment includes the following steps: using a near-infrared spectrometer to scan the spectra of tobacco leaves from the same origin, arranging all near-infrared spectra from the same origin in rows to obtain a data matrix, performing principal component analysis on the matrix, determining the origin identification line by calculating relevant data, and establishing a qualitative model.
2. A near infrared qualitative analysis model based on PCA algorithm according to claim 1, characterized in that: The model building includes the following steps: (1) Spectral scanning: Scan the near-infrared spectrum of the sample to be tested and extract the characteristics of the spectrum; (2) Establishing a data matrix: Arrange the near-infrared spectra of n tobacco leaves from the same origin in rows to obtain a data matrix D; (3) Principal component analysis: Perform principal component analysis on D, calculate the principal components through the eigenvectors, and then retain the first principal component P1 and the third principal component P3; (4) Calculate the mean of P1 and P3 and use this mean as the model differentiation center of tobacco leaves from "a" origin; (5) Calculate the distance d from all tobacco leaves from the origin "X" to this center i (i=1, 2, ..., n); calculate all distances d i The mean and standard deviation s, with It is the identification line of tobacco leaves originating from "certain" place.
3. The near infrared qualitative analysis model based on PCA algorithm according to claim 1, characterized in that: The model establishment includes the following steps: (1) Spectral scanning: The tobacco leaf samples were scanned using a near-infrared spectrometer to obtain their near-infrared spectra. The operating parameters were: spectral scanning range 12 000-4 000 cm-1, spectral resolution 8 cm-1, and scanning times 64 times; spectral data were collected in a transmission mode and processed into the first-order differential of the absorption spectrum; each sample was scanned 10 times to obtain the average spectrum, and more than 2 average spectra were obtained for each sample scan; (2) Spectral preprocessing: The standard normal variable transformation method is used to eliminate the differences caused by the particle size, surface scattering and optical path change of the solid sample. The spectrum is smoothed and noise-filtered by wavelet transform. The original spectrum is processed by second-order derivative to eliminate baseline drift, improve spectral resolution, and extract characteristic spectra. (3) Exporting spectral data: exporting data points in the spectral range of 12 000 to 4 000 cm-1; (4) Establishing a data matrix: Arranging the near-infrared spectra of tobacco leaves in the same area by rows to obtain a data matrix; (5) Principal component analysis: Calculate the principal components through the eigenvectors, analyze and retain them; (6) Calculate the mean of the retained principal components and use this mean as the center of the differentiation model of tobacco leaf producing areas in the region; (7) Calculate the distance from all tobacco leaves in the region to this center, and calculate the mean and standard deviation of all distances, and use the mean plus 1.2 times the standard deviation as the classification line for the tobacco leaf producing area in the region; (8) Establishing PCA model: The model is established using the spectrum collected under normal conditions. The formula is as follows: Where: X(m×n) represents the spectral data matrix of m spectra and n wavelength points, T i is the score vector, P i is the loading vector, A is the number of factors, E is the error term, and T in the above table represents the transpose of the matrix.
4. The application of a near infrared qualitative analysis model based on a PCA algorithm according to any one of claims 1 to 3, characterized in that: When applying this model to identify unknown samples, the analysis steps are as follows: (1) scanning and obtaining a near infrared spectrum of the sample; (2) Calculate the first principal component PC1 and the third principal component PC3 of the product through the feature vector in the modeling step 2; (3) Calculate the distance d between PC1 and PC3 and the center of tobacco leaves from the origin "a" 未知 , determine the territory based on the distance, if It can be determined that the unknown product belongs to the place of origin, otherwise it can be determined that it does not belong to the place of origin.
5. The use according to claim 4, characterized in that: The method comprises the following steps before scanning: crushing the sample into 40-80 meshes, and the sample is shredded tobacco, tobacco stems, or tobacco dust.
6. The application of a near infrared qualitative analysis model based on a PCA algorithm according to any one of claims 1 to 3, characterized in that: When applying this model to identify the origin of unknown tobacco leaf samples, follow the steps below: (1) scanning and obtaining a near infrared spectrum of the sample; (2) calculating the principal component of the tobacco leaf sample according to the characteristic vector of the spectrum; (3) Calculate the distance from the principal component to the center of the constructed model, and judge the similarity of the origin based on the distance. If the distance falls within the classification identification line, it can be determined that the origin of the tobacco leaf is similar to that of the constructed model; otherwise, it is determined that they are of different origins.
7. The application of a near infrared qualitative analysis model based on a PCA algorithm according to any one of claims 1 to 3, characterized in that: When applying this model to identify the origin of unknown tobacco leaf samples, follow the steps below: (1) Spectral scanning: Scan the near-infrared spectrum of the sample to be tested and extract the characteristic vector of the spectrum; (2) Establishing a data matrix: Arrange the near-infrared spectra of n tobacco leaves from the same origin in rows to obtain a data matrix D; (3) Principal component analysis: Perform principal component analysis on D, calculate the principal components through the eigenvectors, and then retain the first principal component P1 and the third principal component P3; (4) Calculate the mean of P1 and P3 and use this mean as the model differentiation center of tobacco leaves from "a" origin; (5) Calculate the distance di (i=1, 2, ..., n) from all tobacco leaves of "certain" origin to this center, and calculate the mean and standard deviation s of all distances di to serve as the identification line of tobacco leaves of "certain" origin.