CN105181650A - Method for quickly identifying tea varieties through near-infrared spectroscopy technology - Google Patents

Abstract

The invention provides a method for quickly identifying tea varieties through a near-infrared spectroscopy technology. The method comprises the steps that firstly, near-infrared diffuse reflection spectrum of tea is collected through a near-infrared spectroscopy instrument, dimension reduction treatment is conducted on the high-dimension near-infrared spectrum of the tea through principal component analysis (PCA), the variety classification information of tea spectrum data is extracted through linear discriminant analysis (LDA), and finally, the tea varieties are identified and analyzed through a new generalized noise clustering method. The method is high in detection speed, high in identification accuracy, environmentally friendly and capable of accurately identifying the tea varieties.

Description

A kind of method using near-infrared spectrum technique to differentiate local tea variety fast

Technical field

The present invention relates to a kind of technical field of local tea variety discrimination method, be specifically related to a kind of method using near-infrared spectrum technique to differentiate local tea variety fast.

Background technology

Tealeaves is one of large beverage in the world three, and it contains the organic substances such as Tea Polyphenols, protein and amino acid, also containing dead matter such as potassium, calcium and magnesium, has and calms the nerves, the effect such as improving eyesight and heat-clearing, often drink tea and be of value to the healthy of people.Leshan green bamboo snake is the distinctive tealeaves brand in Leshan, but there is phenomenon of adulterating in tea market, and ordinary consumer is beyond recognition high-quality well-known tea and tealeaves inferior, is often deceived.In addition, shoddy tealeaves inferior compromises the brand prestige of Famous High-quality Tea, has encroached on consumers' rights and interests, brings puzzlement to the marketing of Famous High-quality Tea.So study a kind of method simple, be easy to operate, the discrimination method of local tea variety that detection speed is fast is very important.

Near Infrared Spectroscopy Detection Technology, as a kind of Fast nondestructive evaluation technology, is applied in the detection analysis of tea leaf quality in recent years.The near-infrared spectrum technique such as Zhang Long, principal component analysis (PCA) and Dian Ze discriminatory analysis are to non-fermented tea, and semi-fermented tea and fermented tea carry out sort research.Near-infrared spectrum technique and the neural network such as Ning Jingming distinguish the Pu'er tea of three kinds of Various Fermenting Degree.Total anthocyanidin content of near-infrared spectrum technique and ant group optimization model inspection jasmine tea such as Huang.Ren etc. with near-infrared spectrum technique detect black tea chemical composition and identify tealeaves trace to the source ground.The near-infrared spectrum technique such as He, what partial least squares discriminant analysis and Euclidean distance method detected tealeaves traces to the source ground.The Determination of Polyphenols of near-infrared spectrum technique and multispectral image systems axiol-ogy extra-strong tea such as Xiong.

Fuzzy C-Means Clustering (FCM) is famous fuzzy clustering algorithm, and it is applied widely, but FCM is responsive to noise data.Noise cluster is a kind of fuzzy clustering algorithm, it is applicable to the cluster analysis processing Noise Data, noise data is regarded as a classification and is processed by noise cluster, but noise cluster has dependence to parameter, simultaneously, the objective function of noise cluster is all be based upon sample on square basis of the Euclidean distance of class center vector, and their accuracys rate when the data of cluster topological structure more complicated are not often very desirable.

The tealeaves near-infrared diffuse reflection spectrum data collected with near infrared spectrometer is a kind of high dimensional data, bunch topological structure more complicated of data after dimension polynomiol and feature extraction, when carrying out data clusters according to noise cluster, the Euclidean distance adopted due to noise cluster carrys out metric data, then Clustering Effect is undesirable.

Summary of the invention

The present invention is directed to the defect of noise clustering method in prior art and not enough problem, propose a kind of detection speed fast, discriminating accuracy rate is high, environmental protection, can realize a kind of method using near-infrared spectrum technique to differentiate local tea variety fast of the accurate discriminating of local tea variety; Thus solve noise clustering method can only the simple data problem of cluster topological structure, improve the accuracy rate of noise cluster.

The object of the invention is to be realized by following technological means: a kind of method using near-infrared spectrum technique to differentiate local tea variety fast, is characterized in that comprising the following steps:

The collection of step one, tealeaves sample near infrared spectrum: the tealeaves sample gathering different cultivars with near infrared spectrometer, obtains the near-infrared diffuse reflection spectrum of tealeaves sample;

Step 2, dimension-reduction treatment is carried out to tealeaves sample near infrared spectrum: adopt principal component analytical method (PCA) that tealeaves sample near infrared spectrum is transformed to low-dimensional data from high dimensional data;

The authentication information of step 3, extraction tealeaves sample near infrared spectrum: adopt linear discriminant analysis (LDA) to extract the authentication information of tealeaves sample near infrared spectrum;

Step 4, operation Fuzzy C-Means Clustering are to obtain initial cluster center;

Step 5, carry out the discriminating of local tea variety with a kind of broad sense noise clustering method: run broad sense noise clustering method according to the initial cluster center of step 4 and obtain fuzzy membership, the discriminating of local tea variety can be realized according to fuzzy membership.

Step one, the near-infrared diffuse reflection spectrum described in two, three, near-infrared diffuse reflection spectrum because of different tealeaves samples contains the different inside quality information of tealeaves, its inside quality of tealeaves that kind is different is different, corresponding near-infrared diffuse reflection spectrum is not identical yet, and this is principle of the present invention.

Broad sense noise clustering method in described step 5 adopts the broad sense noise cluster based on the p power of Euclidean distance to carry out the classification of local tea variety, specific as follows:

(1). initialization

Tealeaves near infrared spectrum number of samples n (+∞ > n > 1) is set, sample class number c (n > c > 1), weighted index m (+∞ > m > 1) and p (+∞ > p > 1), primary iteration number of times r=1, greatest iteration number r _max, error higher limit ε, initialization class center v _{i, 0};

(2). calculating parameter α _ik:

$a_{ik}=\frac{{\mathrm{\σ}}^2}{m^{2}c}{\[\underset{j=1}{\overset{c}{\Σ}}{\left(\frac{D_{ik,r}}{D_{jk,r}}\right)}^{\frac{2}{m-1}}\]}^{-1},{\mathrm{\σ}}^2=\frac{1}{n}{\Σ}_{k=1}^n\left|\right|x_k-\stackrel{\‾}{x}|{|}^2,\stackrel{\‾}{x}=\frac{1}{n}{\Σ}_{j=1}^n{x}_j$

Here σ ²it is the variance of sample; α _ikbe i-th (i=1,2 ..., c) classification kth (k=1,2 ..., the n) parameter of individual sample; D _{ik, r}=|| x _k-v _{i, r-1}|| be x _k-v _{i, r-1}euclidean distance, x _kfor a kth sample, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration; D _{jk, r}=|| x _k-v _{j, r-1}|| be x _k-v _{j, r-1}euclidean distance, ν _{j, r-1}the class center vector of jth class when being the r-1 time iteration; for population sample average, x _jfor a jth sample;

(3). calculate fuzzy membership angle value u during the r time iteration _{ik, r};

$u_{ik,r}=\frac{{\left({\mathrm{\α}}_{ik}{D}_{ik,r}^{-p}\right)}^{\frac{1}{m-1}}}{\underset{i=1}{\overset{c}{\Σ}}{\left({\mathrm{\α}}_{ik}{D}_{ik,r}^{-p}\right)}^{\frac{1}{m-1}}+1}$

Here angle value u is subordinate to _{ik, r}when representing the r time iterative computation, a kth sample is under the jurisdiction of the fuzzy membership angle value of the i-th class; D _{ik, r}=|| x _k-v _{i, r-1}||, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration;

(4). calculate the r time iteration Shi Lei center v _i,r;

$v_{i,r}=\frac{\underset{k=1}{\overset{n}{\Σ}}{u}_{ik,r}^m{x}_k}{\underset{k=1}{\overset{n}{\Σ}}{u}_{ik,r}^m},\∀i$

Work as max _i|| v _i,r-v _{i, r-1}|| < ε or r=r _maxtime, iteration ends; Otherwise r=r+1, returns step (2) and continues iterative computation.

Compared with prior art the present invention has following obvious advantage:

1, the present invention adopts the broad sense noise cluster based on the p power of Euclidean distance to carry out the classification of local tea variety; Thus solve noise clustering method can only the simple data problem of cluster topological structure, improve the accuracy rate of noise cluster.2, the inventive method gathers the near-infrared diffuse reflection spectrum of tealeaves with near infrared spectrometer, the higher-dimension near infrared spectrum of principal component analysis (PCA) (PCA) to tealeaves is used to carry out dimension-reduction treatment again, carry out the extraction of the variety classification information of tealeaves spectroscopic data with linear discriminant analysis (LDA), finally utilize a kind of new broad sense noise clustering method to carry out the discriminatory analysis of local tea variety.3, the present invention has detection speed soon, and differentiate that accuracy rate is high, environmental protection, can realize the accurate discriminating of local tea variety.

Accompanying drawing explanation

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

Fig. 2 is the diffuse reflection near infrared light spectrogram of tealeaves sample in the present invention;

Fig. 3 is the 2-D data figure obtained after neutral line discriminatory analysis feature extraction of the present invention;

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

Fig. 5 is the cluster accuracy rate figure that the inventive method realizes local tea variety discriminating.

Embodiment

Illustrate that the present invention is described in further detail with embodiment below in conjunction with accompanying drawing: the near infrared spectrum local tea variety discrimination method of a kind of broad sense noise cluster of the present invention is applicable to the discriminatory analysis of local tea variety, and implementing procedure of the present invention as shown in Figure 1.

Embodiment

The collection of step one, tealeaves sample near infrared spectrum: the tealeaves sample gathering different cultivars with near infrared spectrometer, obtains the near-infrared diffuse reflection spectrum of tealeaves sample.

Gather high-quality Leshan green bamboo snake, Leshan inferior green bamboo snake and Mount Emei Mao Fengsan kind tealeaves, the sample number of often kind of tealeaves is 32, adds up to 96 samples.Through 40 mesh screen after all tealeaves samples are polished and shatter, each sample is got 0.5g and is carried out press mold process with potassium bromide by getting potpourri 1g after 1:100 Homogeneous phase mixing respectively.The laboratory temperature about 25 DEG C when carrying out collection near infrared spectrum, relative humidity was about 50%, FTIR-7600 type Fourier transform near infrared analyser start preheating 1 hour.Spectroanalysis instrument scans each tealeaves sample 32 times, and the wave-number range of spectral scan is 4001.569 ~ 401.1211cm ^-1, sweep spacing is 1.9285cm ^-1, the near infrared spectrum of each tealeaves sample is the high dimensional data of 1868 dimensions.Each specimen sample 3 times, gets the experimental data that its mean value is set up as following model.The near infrared light spectrogram of tealeaves sample as shown in Figure 2.

Step 2, dimension-reduction treatment is carried out to tealeaves sample near infrared spectrum: adopt principal component analytical method (PCA) that tealeaves sample near infrared spectrum is transformed to low-dimensional data from high dimensional data.

Adopt the data that the near infrared spectrum data boil down to 20 of 96 samples is tieed up by principal component analysis (PCA).

The authentication information of step 3, extraction tealeaves sample near infrared spectrum: adopt linear discriminant analysis (LDA) to extract the authentication information of tealeaves sample near infrared spectrum.

From often kind of tealeaves sample, choose 13 sample composition tealeaves sample training collection, then training set total sample number is 39, and remaining sample composition tealeaves test sample collection, then test set total sample number is 57.Calculate the discriminant vectors of the training set sample of 20 dimensions by running LDA, and get front 2 discriminant vectorses, projected on these 2 discriminant vectorses by the test set sample that 20 tie up, the LDA shot chart of its test sample book as shown in Figure 3.

Step 4, operation Fuzzy C-Means Clustering are to obtain initial cluster center.

The weighted index m=2.0 of Fuzzy C-Means Clustering (FCM) is set, greatest iteration number r _max=100, error higher limit ε=0.00001, the initial classes center vector of FCM is front 3 data of the test data of Fig. 3.The class center vector calculating the FCM of gained is:

v _1,0＝[-0.0970.0026]

v _2,0＝[0.0198-0.0910]

v _3,0＝[0.06600.0472]

Step 5, carry out the discriminating of local tea variety with broad sense noise clustering method: run broad sense noise clustering method according to the initial cluster center of step 4 and obtain fuzzy membership, the discriminating of local tea variety can be realized according to fuzzy membership.

Broad sense noise clustering method in described step 5 is as follows:

(1). initialization

Tealeaves near infrared spectrum number of samples n=57 is set, sample class number c=3, weighted index m=2 and p (+∞ > p > 1), primary iteration number of times r=1, greatest iteration number r _max=100, error higher limit ε=0.00001, initialization class center v _{i, 0}(i=1,2,3);

(2). calculating parameter α _ik:

$a_{ik}=\frac{{\mathrm{\&sigma;}}^2}{m^{2}c}{\&lsqb;\underset{j=1}{\overset{c}{\&Sigma;}}{\left(\frac{D_{ik,r}}{D_{jk,r}}\right)}^{\frac{2}{m-1}}\&rsqb;}^{-1},{\mathrm{\&sigma;}}^2=\frac{1}{n}{\&Sigma;}_{k=1}^n\left|\right|x_k-\stackrel{\&OverBar;}{x}|{|}^2,\stackrel{\&OverBar;}{x}=\frac{1}{n}{\&Sigma;}_{j=1}^n{x}_j$

Here σ ²it is the variance of sample; α _ikbe i-th (i=1,2 ..., c) classification kth (k=1,2 ..., the n) parameter of individual sample; D _{ik, r}=|| x _k-v _{i, r-1}|| be x _k-v _{i, r-1}euclidean distance, x _kfor a kth sample, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration; D _{jk, r}=|| x _k-v _{j, r-1}|| be x _k-v _{j, r-1}euclidean distance, ν _{j, r-1}the class center vector of jth class when being the r-1 time iteration; for population sample average, x _jfor a jth sample.

(3). calculate fuzzy membership angle value u during the r time iteration _{ik, r};

$u_{ik,r}=\frac{{\left({\mathrm{\&alpha;}}_{ik}{D}_{ik,k}^{-p}\right)}^{\frac{1}{m-1}}}{\underset{i=1}{\overset{c}{\&Sigma;}}{\left({\mathrm{\&alpha;}}_{ik}{D}_{ik,k}^{-p}\right)}^{\frac{1}{m-1}}+1}$

Here angle value u is subordinate to _{ik, r}when representing the r time iterative computation, a kth sample is under the jurisdiction of the fuzzy membership angle value of the i-th class; D _{ik, r}=|| x _k-v _{i, r-1}||, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration.

Experimental result: the iteration ends as iteration 7 times (r=7), fuzzy membership angle value u now _{ik, 7}numerical value as shown in Figure 4, get u in a kth sample _{ik, 7}the i value corresponding to maximal value, namely judge that a kth sample belongs to the i-th class.When weighted index p is respectively 2,3,4,5,6,7,8,9,10,11,12, when 13, the value according to fuzzy membership can obtain cluster accuracy rate as shown in Figure 5.

(4). calculate the r time iteration Shi Lei center v _i,r;

$v_{i,r}=\frac{\underset{k=1}{\overset{n}{\&Sigma;}}{u}_{ik,r}^m{x}_k}{\underset{k=1}{\overset{n}{\&Sigma;}}{u}_{ik,r}^m},\&ForAll;i$

Work as max _i|| v _i,r-v _{i, r-1}|| < ε or r=r _maxtime, iteration ends; Otherwise r=r+1, returns step (2) and continues iterative computation.

Experimental result: r=7, v during iteration ends _{i, 7}for:

$\left[\begin{array}{c}{v}_{0,7}\\ v_{1,7}\\ v_{2,7}\end{array}\right]=\left[\begin{array}{cc}-0.094145& 0.02393\\ 0.020499& -0.095007\\ 0.063514& 0.046156\end{array}\right]$

The tealeaves of test sample book can be divided into three classifications by the fuzzy membership angle value according to Fig. 4, and they are respectively with v _0,7, v _1,7and v _2,7for three data acquisitions at class center.Training sample is known three kind tealeaves (i.e. high-quality Leshan green bamboo snake, Leshan inferior green bamboo snake and Mount Emei Mao Feng), and the mean value calculating the training sample of often kind of tealeaves is: Mount Emei's hair peak-to-average is ${\stackrel{\&OverBar;}{x}}_{em}=\left[\begin{array}{cc}0.021737& -0.080187\end{array}\right];$ High-quality Leshan green bamboo snake mean value is ${\stackrel{\&OverBar;}{x}}_{yz}=\left[\begin{array}{cc}-0.10105& 0.035855\end{array}\right],$ Leshan inferior green bamboo snake mean value is ${\stackrel{\&OverBar;}{x}}_{lz}=\left[\begin{array}{cc}0.060952& 0.084049\end{array}\right].$ (note: training sample here and test sample book to refer in step 3 the data sample that obtains after LDA calculates) judges that the method which kind of three classifications of the tealeaves of test sample book belong to respectively is: the Euclidean distance calculating certain cluster centre of test sample book and the mean value of training sample three class tealeaves respectively, certain cluster centre is minimum from the Euclidean distance of which kind of training local tea variety, judges that local tea variety belonging to this cluster centre and this training local tea variety are same breed.Concrete computation and analysis is as follows:

A judges with v _0,7for which kind of the tealeaves at class center belongs to:

$\left|\right|v_{0,7}-{\stackrel{\&OverBar;}{x}}_{em}\left|\right|=0.1558$

$\left|\right|v_{0,7}-{\stackrel{\&OverBar;}{x}}_{yz}\left|\right|=0.0138$

$\left|\right|v_{0,7}-{\stackrel{\&OverBar;}{x}}_{lz}\left|\right|=0.1663$

So, value minimum, then judge with v _0,7tealeaves for class center belongs to high-quality Leshan green bamboo snake.

B judges with v _1,7for which kind of the tealeaves at class center belongs to:

$\left|\right|v_{1,7}-{\stackrel{\&OverBar;}{x}}_{em}\left|\right|=0.0149$

$\left|\right|v_{1,7}-{\stackrel{\&OverBar;}{x}}_{yz}\left|\right|=0.1786$

$\left|\right|v_{1,7}-{\stackrel{\&OverBar;}{x}}_{lz}\left|\right|=0.1836$

So, value minimum, then judge with v _1,7tealeaves for class center belongs to Mount Emei Mao Feng.

C judges with v _2,7for which kind of the tealeaves at class center belongs to:

$\left|\right|v_{2,7}-{\stackrel{\&OverBar;}{x}}_{em}\left|\right|=0.1331$

$\left|\right|v_{2,7}-{\stackrel{\&OverBar;}{x}}_{yz}\left|\right|=0.1649$

$\left|\right|v_{2,7}-{\stackrel{\&OverBar;}{x}}_{lz}\left|\right|=0.0380$

So, value minimum, then judge with v _2,7tealeaves for class center belongs to Leshan inferior green bamboo snake.

The above; be only a part of embodiment of the present invention, protection scope of the present invention is not limited thereto, and is anyly familiar with those skilled in the art in the technical scope that the present invention discloses; the change that can expect easily or replacement, all should be encompassed within protection scope of the present invention.

Claims (3)

1. use near-infrared spectrum technique to differentiate a method for local tea variety fast, it is characterized in that comprising the following steps:

The collection of step one, tealeaves sample near infrared spectrum: the tealeaves sample gathering different cultivars with near infrared spectrometer, obtains the near-infrared diffuse reflection spectrum of tealeaves sample;

Step 2, dimension-reduction treatment is carried out to tealeaves sample near infrared spectrum: adopt principal component analytical method (PCA) that tealeaves sample near infrared spectrum is transformed to low-dimensional data from high dimensional data;

The authentication information of step 3, extraction tealeaves sample near infrared spectrum: adopt linear discriminant analysis (LDA) to extract the authentication information of tealeaves sample near infrared spectrum;

Step 4, operation Fuzzy C-Means Clustering are to obtain initial cluster center;

Step 5, carry out the discriminating of local tea variety with a kind of broad sense noise clustering method: run broad sense noise clustering method according to the initial cluster center of step 4 and obtain fuzzy membership, the discriminating of local tea variety can be realized according to fuzzy membership.

2. a kind of method using near-infrared spectrum technique to differentiate local tea variety fast according to claim 1, it is characterized in that: step one, the near-infrared diffuse reflection spectrum described in two, three, near-infrared diffuse reflection spectrum because of different tealeaves samples contains the different inside quality information of tealeaves, its inside quality of tealeaves that kind is different is different, corresponding near-infrared diffuse reflection spectrum is not identical yet, and this is principle of the present invention.

3. a kind of method using near-infrared spectrum technique to differentiate local tea variety fast according to claim 1, it is characterized in that, broad sense noise clustering method in described step 5 adopts the broad sense noise cluster based on the p power of Euclidean distance to carry out the classification of local tea variety, specific as follows:

(1). initialization

Tealeaves near infrared spectrum number of samples n (+∞ > n > 1) is set, sample class number c (n > c > 1), weighted index m (+∞ > m > 1) and p (+∞ > p > 1), primary iteration number of times r=1, greatest iteration number r _max, error higher limit ε, initialization class center v _{i, 0};

(2). calculating parameter α _ik:

${\mathrm{\&alpha;}}_{ik}=\frac{{\mathrm{\&sigma;}}^2}{m^{2}c}{\&lsqb;\underset{j=1}{\overset{c}{\&Sigma;}}{\left(\frac{D_{ik,r}}{D_{jk,r}}\right)}^{\frac{2}{m-1}}\&rsqb;}^{-1},{\mathrm{\&sigma;}}^2=\frac{1}{n}{\mathrm{\&Sigma;}}_{k=1}^n\left|\right|x_k-\stackrel{\&OverBar;}{x}|{|}^2,\stackrel{\&OverBar;}{x}=\frac{1}{n}{\mathrm{\&Sigma;}}_{j=1}^n{x}_j$

Here σ ²it is the variance of sample; α _ikbe i-th (i=1,2 ..., c) classification kth (k=1,2 ..., the n) parameter of individual sample; D _{ik, r}=|| x _k-v _{i, r-1}|| be x _k-v _{i, r-1}euclidean distance, x _kfor a kth sample, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration; D _{jk, r}=|| x _k-v _{j, r-1}|| be x _k-v _{j, r-1}euclidean distance, ν _{j, r-1}the class center vector of jth class when being the r-1 time iteration; for population sample average, x _jfor a jth sample;

(3). calculate fuzzy membership angle value u during the r time iteration _{ik, r};

$u_{ik,r}=\frac{{\left({\mathrm{\&alpha;}}_{ik}{D}_{ik,r}^{-p}\right)}^{\frac{1}{m-1}}}{\underset{i=1}{\overset{c}{\&Sigma;}}{\left({\mathrm{\&alpha;}}_{ik}{D}_{ik,r}^{-p}\right)}^{\frac{1}{m-1}}+1}$

Here angle value u is subordinate to _{ik, r}when representing the r time iterative computation, a kth sample is under the jurisdiction of the fuzzy membership angle value of the i-th class; D _{ik, r}=|| x _k-v _{i, r-1}||, v _{i, r-1}the class center vector of the i-th class when being the r-1 time iteration;

(4). calculate the r time iteration Shi Lei center v _i,r;

$v_{i,r}=\frac{\underset{k=1}{\overset{n}{\&Sigma;}}{u}_{ik,r}^m{x}_k}{\underset{k=1}{\overset{n}{\&Sigma;}}{u}_{ik,r}^m},\&ForAll;i$

Work as max _i|| v _i,r-v _{i, r-1}|| < ε or r=r _maxtime, iteration ends; Otherwise r=r+1, returns step (2) and continues iterative computation.