CN105181650B - A method of quickly differentiating local tea variety using near-infrared spectrum technique - Google Patents

Abstract

The present invention is a kind of method quickly differentiating local tea variety using near-infrared spectrum technique, the near-infrared diffusing reflection spectrum of tealeaves is acquired near infrared spectrometer first, dimension-reduction treatment is carried out to the higher-dimension near infrared spectrum of tealeaves with principal component analysis (PCA) again, the extraction of the variety classification information of tealeaves spectroscopic data is carried out with linear discriminant analysis (LDA), and the discriminatory analysis of local tea variety is finally carried out using a kind of new broad sense noise clustering method.The present invention has detection speed fast, differentiates that accuracy rate is high, it is environmentally protective, it can be achieved that local tea variety accurate discriminating.

Description

A method for rapid identification of tea varieties using near-infrared spectroscopy

technical field

The invention relates to the technical field of a method for identifying tea varieties, in particular to a method for quickly identifying tea varieties using near-infrared spectroscopy.

Background technique

Tea is one of the three major beverages in the world. It contains organic substances such as tea polyphenols, protein and amino acids, as well as inorganic substances such as potassium, calcium and magnesium. Healthy body. Leshan Zhuyeqing is a unique tea brand in Leshan. However, there is a phenomenon of shoddy tea in the tea market, and ordinary consumers cannot distinguish between high-quality famous tea and low-quality tea, and are often deceived. In addition, shoddy and inferior tea leaves have damaged the brand reputation of famous and high-quality teas, violated the rights and interests of consumers, and brought troubles to the marketing of famous and high-quality teas. Therefore, it is very necessary to study a method for identifying tea varieties that is simple, easy to operate, and fast in detection speed.

Near-infrared spectroscopy detection technology, as a rapid non-destructive detection technology, has been used in the detection and analysis of tea quality in recent years. Zhang Long et al. used near-infrared spectroscopy, principal component analysis and canonical discriminant analysis to classify non-fermented tea, semi-fermented tea and fermented tea. Ning Jingming et al. used near-infrared spectroscopy and neural networks to distinguish three types of Pu-erh tea with different degrees of fermentation. Huang et al. used near-infrared spectroscopy and an ant colony optimization model to detect the total anthocyanin content of scented tea. Ren et al. used near-infrared spectroscopy to detect the chemical composition of black tea and identify the traceability of tea. He et al. used near-infrared spectroscopy, partial least squares discriminant analysis and Euclidean distance method to detect the traceability of tea. Xiong et al. used near-infrared spectroscopy and multi-spectral imaging system to detect the total polyphenol content of Tieguanyin tea.

Fuzzy C-means clustering (FCM) is a well-known fuzzy clustering algorithm, which is widely used, but FCM is sensitive to noise data. Noise clustering is a fuzzy clustering algorithm, which is suitable for cluster analysis of noise-containing data. Noise clustering treats noise data as a category, but noise clustering has a dependence on parameters. At the same time, noise clustering The objective functions of the class are all based on the square of the Euclidean distance from the sample to the class center vector, and their accuracy is often not ideal when clustering data with a complex topology.

The near-infrared diffuse reflectance spectrum data of tea collected by near-infrared spectrometer is a kind of high-dimensional data. After dimensionality compression and feature extraction, the cluster topology of the data is relatively complex. If the Euclidean distance used by clustering is used to measure data, the clustering effect is not ideal.

Contents of the invention

Aiming at the defects and insufficiencies of the noise clustering method in the prior art, the present invention proposes a fast detection speed, high identification accuracy, green and environmental protection, and can realize accurate identification of tea varieties using near-infrared spectrum technology. A method for identifying tea varieties; thereby solving the problem that the noise clustering method can only cluster data with a simple topological structure, and improving the accuracy of the noise clustering.

The object of the present invention is achieved by the following technical means: a method for quickly identifying tea varieties using near-infrared spectroscopy, characterized in that it comprises the following steps:

Step 1. Collection of near-infrared spectra of tea samples: collecting tea samples of different varieties with a near-infrared spectrometer to obtain near-infrared diffuse reflectance spectra of tea samples;

Step 2, performing dimensionality reduction processing on the near-infrared spectrum of the tea sample: using principal component analysis (PCA) to convert the near-infrared spectrum of the tea sample from high-dimensional data to low-dimensional data;

Step 3, extracting the identification information of the near-infrared spectrum of the tea sample: using linear discriminant analysis (LDA) to extract the identification information of the near-infrared spectrum of the tea sample;

Step 4, run fuzzy C-means clustering to obtain initial cluster centers;

Step 5. Use a generalized noise clustering method to identify tea varieties: run the generalized noise clustering method based on the initial clustering center in step 4 to obtain fuzzy membership degrees, and tea varieties can be identified according to the fuzzy membership degrees.

For the near-infrared diffuse reflectance spectra described in steps 1, 2, and 3, because the near-infrared diffuse reflectance spectra of different tea samples contain different internal quality information of tea leaves, the internal quality of tea leaves with different varieties is different, and the corresponding near-infrared reflectance spectra The infrared diffuse reflectance spectrum is also different, which is the principle of the present invention.

The generalized noise clustering method in the step 5 adopts the generalized noise clustering based on the p power of Euclidean distance to classify the tea varieties, as follows:

(1).Initialization

Set the number of tea near-infrared spectrum samples n (+∞>n>1), the number of sample categories c (n>c>1), the weight index m (+∞>m>1) and p (+∞>p>1) , the initial number of iterations r=1, the maximum number of iterations r _max , the upper limit of error ε, the initialization class center v _i,0 ;

(2). Calculation parameter α _ik :

Here σ ² is the variance of the sample; α _ik is the parameter of the kth (k=1, 2, ..., n) sample of the i (i=1, 2, ..., c) category; D _ik,r ＝||x _k -v _{i, r-1} || is the Euclidean distance of x _k -v _{i, r-1} , x _k is the kth sample, v _{i, r-1} is the r-1th iteration The class center vector of the i-th class; D _jk,r =||x _k -v _j,r-1 || is the Euclidean distance of x _k -v _j,r-1 , ν _j,r-1 is the r-th The class center vector of the jth class at 1 iteration; is the overall sample mean, x _j is the jth sample;

(3). Calculate the fuzzy membership value u _ik,r during the rth iteration;

Here the membership value u _ik,r represents the fuzzy membership value of the k-th sample belonging to the i-th class during the r-th iterative calculation; D _ik,r =||x _k -v _i,r-1 ||, v _{i, r-1} is the class center vector of the i-th class at the r-1th iteration;

(4). Calculate the class center v _i,r at the rth iteration;

When max _i ||v _{i, r} -v _{i, r-1} ||<ε or r=r _max , the iteration terminates; otherwise, r=r+1, return to step (2) to continue the iterative calculation.

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

1. The present invention adopts the generalized noise clustering based on the pth power of Euclidean distance to classify tea varieties; thereby solving the problem that the noise clustering method can only cluster data with simple topology, and improving the accuracy of the noise clustering. 2, the inventive method collects the near-infrared diffuse reflectance spectrum of tealeaves with near-infrared spectrometer, then uses principal component analysis (PCA) to carry out dimensionality reduction process to the high-dimensional near-infrared spectrum of tealeaves, carries out tealeaves spectral data with linear discriminant analysis (LDA) Finally, a new generalized noise clustering method is used to identify and analyze tea varieties. 3. The present invention has the advantages of fast detection speed, high identification accuracy, environmental protection, and accurate identification of tea varieties.

Description of drawings

Fig. 1 is a schematic flow sheet of the present invention;

Fig. 2 is the diffuse reflectance near-infrared spectrogram of tea sample among the present invention;

Fig. 3 is the two-dimensional data diagram that obtains after linear discriminant analysis feature extraction among the present invention;

Fig. 4 is the fuzzy degree of membership figure of the inventive method;

Fig. 5 is a diagram of the clustering accuracy rate of tea variety identification realized by the method of the present invention.

Detailed ways

Below in conjunction with accompanying drawing description and specific embodiment, the present invention is described in further detail: a kind of near-infrared spectrum tea variety identification method of generalized noise clustering of the present invention is applicable to the identification analysis of tea variety, and the implementation process of the present invention is as shown in Figure 1 shown.

Example

Step 1. Collection of near-infrared spectra of tea samples: collecting tea samples of different varieties with a near-infrared spectrometer to obtain near-infrared diffuse reflectance spectra of tea samples.

Three kinds of tea were collected: high-quality Leshan Zhuyeqing, inferior Leshan Zhuyeqing and Emeishan Maofeng. The number of samples for each tea was 32, totaling 96 samples. All the tea samples were ground and then filtered through a 40-mesh sieve. 0.5 g of each sample was uniformly mixed with potassium bromide at a ratio of 1:100, and 1 g of the mixture was taken for film-pressing treatment. When collecting near-infrared spectra, the laboratory temperature is about 25°C, the relative humidity is about 50%, and the FTIR-7600 Fourier transform near-infrared spectroscopy analyzer is turned on and preheated for 1 hour. The spectrum analyzer scans each tea sample 32 times. The wavenumber range of spectral scanning is 4001.569～401.1211cm ^-1 , and the scanning interval is 1.9285cm ^-1 . The near-infrared spectrum of each tea sample is 1868-dimensional high-dimensional data. Each sample was sampled 3 times, and the average value was taken as the experimental data for subsequent model establishment. The NIR spectra of the tea samples are shown in Figure 2.

Step 2. Dimensionality reduction processing of the near-infrared spectrum of the tea sample: the near-infrared spectrum of the tea sample is transformed from high-dimensional data to low-dimensional data by using principal component analysis (PCA).

The near-infrared spectral data of 96 samples were compressed into 20-dimensional data by principal component analysis.

Step 3, extracting the identification information of the near-infrared spectrum of the tea sample: using linear discriminant analysis (LDA) to extract the identification information of the near-infrared spectrum of the tea sample.

Select 13 samples from each tea sample to form a training set of tea samples, and the total number of samples in the training set is 39, and the remaining samples form a test set of tea samples, and the total number of samples in the test set is 57. Calculate the discriminant vector of the 20-dimensional training set sample by running LDA, and take the first two discriminant vectors, project the 20-dimensional test set sample onto these two discriminative vectors, and the LDA score map of the test sample is shown in Figure 3 .

Step 4. Run fuzzy C-means clustering to obtain initial cluster centers.

Set the weight index m=2.0 of fuzzy C-means clustering (FCM), the maximum number of iterations _rmax =100, the upper limit of error ε=0.00001, the initial class center vector of FCM is the first 3 data of the test data in Fig. 3 . The calculated class center vector of FCM is:

v _1,0 = [-0.097 0.0026]

v _2,0 = [0.0198 -0.0910]

v _3,0 = [0.0660 0.0472]

Step 5: Use the generalized noise clustering method to identify tea varieties: run the generalized noise clustering method based on the initial clustering center in step 4 to obtain the fuzzy membership degree, and the tea variety identification can be realized according to the fuzzy membership degree.

The generalized noise clustering method in the step five is as follows:

(1).Initialization

Set the number of tea near-infrared spectrum samples n=57, the number of sample categories c=3, the weight index m=2 and p (+∞>p>1), the initial iteration number r=1, the maximum iteration number r _max =100, the error Upper limit ε=0.00001, initialize class center v _i,0 (i=1,2,3);

(2). Calculation parameter α _ik :

Here σ ² is the variance of the sample; α _ik is the parameter of the kth (k=1, 2, ..., n) sample of the i (i=1, 2, ..., c) category; D _ik,r ＝||x _k -v _{i, r-1} || is the Euclidean distance of x _k -v _{i, r-1} , x _k is the kth sample, v _{i, r-1} is the r-1th iteration The class center vector of the i-th class; D _jk,r =||x _k -v _j,r-1 || is the Euclidean distance of x _k -v _j,r-1 , ν _j,r-1 is the r-th The class center vector of the jth class at 1 iteration; is the overall sample mean, and x _j is the jth sample.

(3). Calculate the fuzzy membership value u _ik,r during the rth iteration;

Here the membership value u _ik,r represents the fuzzy membership value of the k-th sample belonging to the i-th class during the r-th iterative calculation; D _ik,r =||x _k -v _i,r-1 ||, v _{i, r-1} is the class center vector of the i-th class at the r-1th iteration.

Experimental results: when iterating 7 times (r=7), the iteration terminates, and the value of the fuzzy membership value u _ik,7 at this time is shown in Figure 4, which corresponds to the maximum value of u _ik,7 in the kth sample The i value of , that is, to determine that the kth sample belongs to the i-th class. When the weight index p is 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, the clustering accuracy can be obtained according to the value of fuzzy membership, as shown in Figure 5.

(4). Calculate the class center v _i,r at the rth iteration;

When max _i ||v _{i, r} -v _{i, r-1} ||<ε or r=r _max , the iteration terminates; otherwise, r=r+1, return to step (2) to continue the iterative calculation.

Experimental results: when the iteration terminates r=7, v _i,7 is:

a Determine which category the tea with v _0,7 as the center of the category belongs to:

so, The value of is the smallest, then it is determined that the tea with v _0,7 as the center of the class belongs to high-quality Leshan Zhuyeqing.

b Determine which category the tea with v _1,7 as the center of the category belongs to:

so, The value of is the smallest, then it is determined that the tea with v _1,7 as the center of the class belongs to Maofeng of Mount Emei.

c Determine which category the tea with v _2,7 as the center of the category belongs to:

so, The value of is the smallest, then it is determined that the tea with v _2,7 as the center of the class belongs to inferior Leshan Zhuyeqing.

The above is only a part of the specific implementation methods of the present invention, and the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present invention. All should be covered within the protection scope of the present invention.

Claims (1)

1. A method for quickly identifying tea varieties using near-infrared spectroscopy, characterized in that it may further comprise the steps: Step 1. Collection of near-infrared spectra of tea samples: collecting tea samples of different varieties with a near-infrared spectrometer to obtain near-infrared diffuse reflectance spectra of tea samples; Step 2: Perform dimensionality reduction processing on the near-infrared spectrum of the tea sample: use principal component analysis to convert the near-infrared spectrum of the tea sample from high-dimensional data to low-dimensional data; Step 3, extracting the identification information of the near-infrared spectrum of the tea sample: using linear discriminant analysis to extract the identification information of the near-infrared spectrum of the tea sample; Step 4, run fuzzy C-means clustering to obtain initial cluster centers; Step five, carry out the discrimination of tea variety with a kind of generalized noise clustering method: run generalized noise clustering method according to the initial clustering center of step four to obtain fuzzy membership degree, can realize the discrimination of tea variety according to fuzzy membership degree; The generalized noise clustering method in the step 5 adopts the generalized noise clustering based on the p power of Euclidean distance to classify the tea varieties, as follows: (1).Initialization Set the number of tea near-infrared spectrum samples n (+∞>n>1), the number of sample categories c (n>c>1), the weight index m (+∞>m>1) and p (+∞>p>1) , the initial number of iterations r=1, the maximum number of iterations r _max , the upper limit of error ε, the initialization class center v _i,0 ; (2). Calculation parameter α _ik : Here σ ² is the variance of the sample; α _ik is the parameter of the kth (k=1, 2, ..., n) sample of the i (i=1, 2, ..., c) category; D _ik,r ＝||x _k -v _{i, r-1} || is the Euclidean distance of x _k -v _{i, r-1} , x _k is the kth sample, v _{i, r-1} is the r-1th iteration The class center vector of the i-th class; D _jk,r =||x _k -v _j,r-1 || is the Euclidean distance of x _k -v _j,r-1 , ν _j,r-1 is the r-th The class center vector of the jth class at 1 iteration; is the overall sample mean, x _j is the jth sample; (3). Calculate the fuzzy membership value u _ik,r during the rth iteration; Here the membership value u _ik,r represents the fuzzy membership value of the k-th sample belonging to the i-th class during the r-th iterative calculation; D _ik,r =||x _k -v _i,r-1 ||, v _{i, r-1} is the class center vector of the i-th class at the r-1th iteration; (4). Calculate the class center v _i,r at the rth iteration; When max _i ||v _{i, r} -v _{i, r-1} ||<ε or r=r _max , the iteration terminates; otherwise, r=r+1, return to step (2) to continue the iterative calculation.