CN104897729A - Sorting method of storage time of melon slice tea by using electronic nose - Google Patents

Abstract

The invention discloses a sorting method of the storage time of melon slice tea by using an electronic nose. According to the sorting method, an electronic nose sensor is used for simulating the functions and features of human sensory evaluation, and in combination with an SMO algorithm, sorting of the storage time can be performed, so that corresponding predication model and method can be further obtained. The sorting method has the beneficial effects that the SMO algorithm is applied to the sorting of the actual storage time of the melon slice tea, and compared with other methods, the tea data can be effectively, quickly and accurately converted into valuable information, thus having great significance for further improvement of the research and development of tea science and technology in China.

Description

The storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea

Technical field

The present invention relates to a kind of sorting technique, especially the storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea.

Background technology

Utilize intelligent sensory analytical technology to simulate function and the feature of people's sensory evaluation, extract valuable, novel, refining, intelligible information in intelligent sensory detection in conjunction with many algorithms, and then obtain corresponding forecast model and method.Electronic Nose is the novel intelligent sense organ instrument risen from the nineties in 20th century, is a kind of instrument of simulated animal olfactory organ.The fields such as food, beverage, environmental monitoring and processing of farm products are widely used at present.Compare traditional determination method, Electronic Nose Technology is simple to operate, highly sensitive, and measurement result is more objective, reliable.

Along with the development of Agricultural informatics, at the Researching and practicing of long-term tea science technology, produced and accumulate a large amount of having important practical significance and the tealeaves data message of scientific value by Electronic Nose Technology.How these data collected fast and effeciently being carried out analyzing, process and refining, is an important topic of tea science technical research.

There is a large amount of classification problems in tealeaves data research, existing sorting algorithm has a variety of, and such as classics has the methods such as decision tree, artificial neural network, Bayes, K nearest neighbor algorithm, support vector machine.Sequential minimal optimization algorithm (Sequential Minimal Optimization, SMO) is proposed in 1998 by Platt, is the highly effective method of current SVM process large data sets.It is the special case of steady job collection in decomposition algorithm, and the scale of working set is reduced to minimum---two samples.In addition, this algorithm also has not to be needed to store kernel matrix, simple, does not need the features such as matrix operation, makes this algorithm usually show overall Fast Convergent characteristic.

Summary of the invention

The technical problem to be solved in the present invention is to provide a kind of storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea,

The present invention is achieved through the following technical solutions.

A storage time sorting technique for Electronic Nose green tea produced in Anhui Province tea, step comprises:

(1) green tea produced in Anhui Province Tea Samples packing and storing;

(2) use the Electronic Nose containing 10 different metal oxide sensors to detect green tea produced in Anhui Province Tea Samples, the different pieces of information detected according to lower 10 sensors of different time builds data set;

(3) sample set is cut into 10 equal mutually disjoint subsamples, in turn will wherein 9 increments this build model as training sample set, the model built is verified in 1 remaining subsample, above step repeats the average after 10 times as the estimation to arithmetic accuracy, adopts SMO algorithm to classify to the green tea produced in Anhui Province tea storage time:

A () selects Two Variables, if be α ₁, α ₂, its dependent variable α _i(i=3,4 ..., N) be fixing, the subproblem of optimization problem can be write as:

$W=\frac{1}{2}{K}_{11}{{\mathrm{\α}}_1}^2+\frac{1}{2}{K}_{22}{{\mathrm{\α}}_2}^2+y_1{y}_2{K}_{12}{\mathrm{\α}}_1{\mathrm{\α}}_2-({\mathrm{\α}}_1+{\mathrm{\α}}_2)+y_1{\mathrm{\α}}_1\underset{i=3}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\α}}_i{K}_{i1}+y_2{\mathrm{\α}}_2\underset{i=3}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\α}}_i{K}_{i2}$

Constraint condition becomes:

${\mathrm{\α}}_1{y}_1+{\mathrm{\α}}_2{y}_2=-\underset{i=3}{\overset{N}{\mathrm{\Σ}}}{y}_i{\mathrm{\α}}_i=\mathrm{\ζ}$

0≤α _i≤C(i＝1,2)

Wherein: K _ij=K (x _i, x _j), i, j=1,2 ..., N, ζ are constants.Simple in order to describe, note

$g\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_i{y}_{i}K(x_i,x)+b$

Can obtain through derivation:

${\mathrm{\α}}_2^{\mathrm{new}}={\mathrm{\α}}_2^{\mathrm{old}}+\frac{y_2(E_1-E_2)}{\mathrm{\η}}$

Wherein:

$E_i=\underset{j=1}{\overset{N}{\mathrm{\Σ}}}{\mathrm{\α}}_j{y}_{j}K(x_j,x_i)+b-y_i,i=\mathrm{1,2}$

η＝K ₁₁+K ₂₂-2K ₁₂

α after pruning ₂

${\mathrm{\α}}_2^{\mathrm{new},\mathrm{clip}}=\left\{\begin{array}{cc}H,& {\mathrm{\α}}_2^{\mathrm{new}}\≥H\\ {\mathrm{\α}}_2^{\mathrm{new}},& L\≤{\mathrm{\α}}_2^{\mathrm{new}}\≤H\\ L,& {\mathrm{\α}}_2^{\mathrm{new}}\≤L\end{array}\right.$

Wherein:

If y ₁=y ₂, $L=\max (0,{\mathrm{\α}}_2^{\mathrm{old}}-{\mathrm{\α}}_1^{\mathrm{old}}),H=\min (C,C+{\mathrm{\α}}_2^{\mathrm{old}}-{\mathrm{\α}}_1^{\mathrm{old}})$

If y ₁≠ y ₂, $L=\max (0,{\mathrm{\α}}_2^{\mathrm{old}}+{\mathrm{\α}}_1^{\mathrm{old}}-C),H=\min (C,{\mathrm{\α}}_2^{\mathrm{old}}+{\mathrm{\α}}_1^{\mathrm{old}})$

According to can obtain ${\mathrm{\α}}_1^{\mathrm{new}}={\mathrm{\α}}_1^{\mathrm{old}}+y_1{y}_2({\mathrm{\α}}_2^{\mathrm{old}}-{\mathrm{\α}}_2^{\mathrm{new},\mathrm{clip}});$

B () adopts heuristic to carry out variables choice, algorithm specifically comprises two-layer circulation: outer loop and interior loop, the selection of first variable is at outer loop, the selection of second variable is in interior loop, outer loop selects the sample point violating KKT condition in all samples, all samples are checked whether to meet KKT condition, namely

$y_{i}g\left(x_i\right)=\left\{\begin{array}{c}\&GreaterEqual;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=0\}\\ =1,\left\{x_i\right|0<{\mathrm{\&alpha;}}_i<C\}\\ \&le;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=C\}\end{array}\right.$

Interior loop mainly seeks to make | E ₁-E ₂| the variable of the maximum correspondence of value as second variable, if the optimal value of objective function can not be made to have certain reduction by said method, then travel through whole training sample set, otherwise enter outer loop and again find first variable.After often completing a suboptimization, the E reset threshold b and upgrade corresponding to each variable _ivalue.

Further, green tea produced in Anhui Province tea tea sample is carried out pouch independent packaging with tin paper bag, every bag of 50g by (1) respectively, and sack seals, and encloses hygroscopic agent simultaneously and make moist to prevent tealeaves in bag; Then put into refrigerator and cooled and hide (4 DEG C) preservation.Carried out a test every 15 days, carry out 6 tests altogether.The bag number needed is taken out in each test, and all the other are motionless.

Further, (2) data set comprises 2112 samples, 10 sample attributes, 6 classifications.

Beneficial effect of the present invention:

The present invention is by being applied to the storage time classification of actual green tea produced in Anhui Province tea by SMO sorting algorithm, compared with other method, more effectively these tea data are converted into valuable information rapidly and accurately, this is significant to improving further of the research and development of China's tea science technology.

Accompanying drawing explanation

Fig. 1 is the schematic flow sheet of the storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea of the present invention;

Fig. 2 is the storage time point visualization result figure by mistake of the SMO algorithm that the present invention is based on Polynomial kernel function;

Fig. 3 is the storage time point visualization result figure by mistake of the SMO algorithm that the present invention is based on gaussian kernel function.

Embodiment

According to drawings and embodiments the present invention is described in further detail below.

1. the packing of Tea Samples and storage

Study the change with the storage time, the change of the fragrance component Electronic Nose response signal at the bottom of tealeaves, tea, tea, by tea sealing, lucifuge, refrigeration.Specific implementation method is as follows:

Green tea produced in Anhui Province tea tea sample tea processing factory business provided carries out pouch independent packaging with tin paper bag respectively.Every bag of 50g, sack seals, and encloses hygroscopic agent simultaneously and make moist to prevent tealeaves in bag; Then put into refrigerator and cooled and hide (4 DEG C) preservation.Carried out a test every 15 days, carry out 6 tests altogether.The bag number needed is taken out in each test, and all the other are motionless.

During each test, the tealeaves of each grade prepares 70 repeat samples respectively, and each repeat samples quality is 5g, and bilayer film is sealed in the beaker of 500ml, leaves standstill 45min, room temperature 25 ± DEG C.

2. detection by electronic nose method

The PEN3 type Electronic Nose that Electronic Nose adopts German Airsense company to produce, containing 10 different metal oxide sensors, obtains 10 sensor characteristic values.The principle of Electronic Nose utilizes specific metal oxide and biological membrane, the power that the change causing film potential small judges whether scent of and smell is contacted according to volatile substance molecules, the comprehensive information of Quick for sample is carried out with specific sensor and pattern recognition system, the hidden feature of prompting sample, this gas sensor has that reliability is high, sensitivity good and the feature such as repeatability is strong.

The standard transducer array of Electronic Nose PEN3 is in table 1.

The standard transducer array of table 1 PEN3

By detection by electronic nose to the details situation of the partial data of green tea produced in Anhui Province tea tea data in table 2.Wherein, 1 to 10 sensor record, under different time, produce different resistance values, and the storage time is the storage number of days of tealeaves.This data set comprises 2112 samples, 10 sample attributes, 6 classifications.

Table 2 green tea produced in Anhui Province tea data cases

3. tealeaves storage time modeling method

The modeling of tealeaves storage time utilizes ten folding cross-validation methods, 10 equal mutually disjoint subsamples are cut into by sample set, in turn will wherein 9 increments this build model as training sample set, the model built is verified in 1 remaining subsample, and above step repeats the average after 10 times as the estimation to arithmetic accuracy.

4. as Fig. 1, the present invention adopts SMO algorithm to classify to the green tea produced in Anhui Province tea storage time, and the concrete steps of this algorithm comprise as follows:

Step one: the analytic method solving the quadratic programming of Two Variables;

Select Two Variables, if be α ₁, α ₂, its dependent variable α _i(i=3,4 ..., N) be fixing, the subproblem of optimization problem can be write as:

$W=\frac{1}{2}{K}_{11}{{\mathrm{\&alpha;}}_1}^2+\frac{1}{2}{K}_{22}{{\mathrm{\&alpha;}}_2}^2+y_1{y}_2{K}_{12}{\mathrm{\&alpha;}}_1{\mathrm{\&alpha;}}_2-({\mathrm{\&alpha;}}_1+{\mathrm{\&alpha;}}_2)+y_1{\mathrm{\&alpha;}}_1\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i{K}_{i1}+y_2{\mathrm{\&alpha;}}_2\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i{K}_{i2}---\left(1\right)$

Constraint condition becomes:

${\mathrm{\&alpha;}}_1{y}_1+{\mathrm{\&alpha;}}_2{y}_2=-\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i=\mathrm{\&zeta;}$

0≤α _i≤C(i＝1,2) (2)

Wherein: K _ij=K (x _i, x _j), i, j=1,2 ..., N, ζ are constants.Simple in order to describe, note

$g\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i{y}_{i}K(x_i,x)+b---\left(3\right)$

Can obtain through derivation:

${\mathrm{\&alpha;}}_2^{\mathrm{new}}={\mathrm{\&alpha;}}_2^{\mathrm{old}}+\frac{y_2(E_1-E_2)}{\mathrm{\&eta;}}---\left(4\right)$

Wherein:

$E_i=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_j{y}_{j}K(x_j,x_i)+b-y_i,i=\mathrm{1,2}---\left(5\right)$

η＝K ₁₁+K ₂₂-2K ₁₂(6)

α after pruning ₂

${\mathrm{\&alpha;}}_2^{\mathrm{new},\mathrm{clip}}=\left\{\begin{array}{cc}H,& {\mathrm{\&alpha;}}_2^{\mathrm{new}}\&GreaterEqual;H\\ {\mathrm{\&alpha;}}_2^{\mathrm{new}},& L\&le;{\mathrm{\&alpha;}}_2^{\mathrm{new}}\&le;H\\ L,& {\mathrm{\&alpha;}}_2^{\mathrm{new}}\&le;L\end{array}\right.---\left(7\right)$

Wherein:

If y ₁=y ₂, $L=\max (0,{\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_1^{\mathrm{old}}),H=\min (C,C+{\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_1^{\mathrm{old}})$

If y ₁≠ y ₂, $L=\max (0,{\mathrm{\&alpha;}}_2^{\mathrm{old}}+{\mathrm{\&alpha;}}_1^{\mathrm{old}}-C),H=\min (C,{\mathrm{\&alpha;}}_2^{\mathrm{old}}+{\mathrm{\&alpha;}}_1^{\mathrm{old}})$

According to can obtain ${\mathrm{\&alpha;}}_1^{\mathrm{new}}={\mathrm{\&alpha;}}_1^{\mathrm{old}}+y_1{y}_2({\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_2^{\mathrm{new},\mathrm{clip}})$

Step 2: the heuristic of choice variable.

SMO adopts heuristic to carry out variables choice, and algorithm specifically comprises two-layer circulation: outer loop and interior loop.The selection of first variable is at outer loop, and the selection of second variable is in interior loop.Outer loop selects the sample point violating KKT condition in all samples, particularly, checks all samples whether to meet KKT condition, namely

$y_{i}g\left(x_i\right)=\left\{\begin{array}{c}\&GreaterEqual;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=0\}\\ =1,\left\{x_i\right|0<{\mathrm{\&alpha;}}_i<C\}\\ \&le;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=C\}\end{array}\right.---\left(8\right)$

Interior loop mainly seeks to make | E ₁-E ₂| the variable of the maximum correspondence of value as second variable, if the optimal value of objective function can not be made to have certain reduction by said method, then travel through whole training sample set, otherwise enter outer loop and again find first variable.After often completing a suboptimization, the E reset threshold b and upgrade corresponding to each variable _ivalue.

5. according to said method, this example selects gaussian kernel function and Polynomial kernel function, carries out training and testing, be analyzed the performance of kernel function the sorter of SMO algorithm.Concrete outcome is as shown in table 3.

Table 3 kernel function performance comparison

After table 3 shows and adopts gaussian kernel function and Polynomial kernel function, the classification performance comparing result of SMO algorithm, as can be seen from the table, on the modeling time, the SMO algorithm based on Polynomial kernel function has modeling speed relatively faster, has only used 0.48s; On classification accuracy, the classification accuracy based on the SMO algorithm of Polynomial kernel function is relatively much high, compared with gaussian kernel function, high by sixties percent.Result shows, the SMO algorithm based on Polynomial kernel function has the classification accuracy of training speed and Geng Gao faster.The classifying quality of the disaggregated model adopting Polynomial kernel function to obtain is better, especially when in the face of large data, and more remarkable effect.

Tealeaves data set comprises 6 category attributes: " 0 ", " 15 ", " 30 ", " 45 ", " 60 ", " 75 ", represents the number of days that tealeaves stores respectively.Fig. 2 is the storage time point visualization result figure by mistake based on the SMO algorithm of Polynomial kernel function.Fig. 3 is the storage time point visualization result figure by mistake based on the SMO algorithm of gaussian kernel function.In figure, X-axis represents concrete class, and Y-axis represents prediction classification, the sample of square expression mis-classification in figure, and fork represents the sample of correct classification.

Above-described embodiment, only for technical conceive of the present invention and feature are described, its object is to allow the personage being familiar with this art can understand content of the present invention and be implemented, can not limit the scope of the invention with this.All equivalences done according to Spirit Essence of the present invention change or modify, and all should be encompassed in protection scope of the present invention.

Claims (3)

1. a storage time sorting technique for Electronic Nose green tea produced in Anhui Province tea, it is characterized in that, step comprises:

(1) green tea produced in Anhui Province Tea Samples packing and storing;

(2) use the Electronic Nose containing 10 different metal oxide sensors to detect green tea produced in Anhui Province Tea Samples, the different pieces of information detected according to lower 10 sensors of different time builds data set;

(3) sample set is cut into 10 equal mutually disjoint subsamples, in turn will wherein 9 increments this build model as training sample set, the model built is verified in 1 remaining subsample, above step repeats the average after 10 times as the estimation to arithmetic accuracy, adopts SMO algorithm to classify to the green tea produced in Anhui Province tea storage time:

A () selects Two Variables, if be α ₁, α ₂, its dependent variable α _i(i=3,4 ..., N) and be fixing, the subproblem of optimization problem can be write as:

$W=\frac{1}{2}{K}_{11}{{\mathrm{\&alpha;}}_1}^2+\frac{1}{2}{K}_{22}{{\mathrm{\&alpha;}}_2}^2+y_1{y}_2{K}_{12}{\mathrm{\&alpha;}}_1{\mathrm{\&alpha;}}_2-({\mathrm{\&alpha;}}_1+{\mathrm{\&alpha;}}_2)+y_1{\mathrm{\&alpha;}}_1\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i{K}_{i1}+y_2{\mathrm{\&alpha;}}_2\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i{K}_{i2}$

Constraint condition becomes:

${\mathrm{\&alpha;}}_1{y}_1+{\mathrm{\&alpha;}}_2{y}_2=-\underset{i=3}{\overset{N}{\mathrm{\&Sigma;}}}{y}_i{\mathrm{\&alpha;}}_i=\mathrm{\&zeta;}$

0≤α _i≤C(i＝1,2)

Wherein: K _ij=K (x _i, x _j), i, j=1,2 ..., N, ζ are constants.Simple in order to describe, note

$g\left(x\right)=\underset{i=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_i{y}_{i}K(x_i,x)+b$

Can obtain through derivation:

${\mathrm{\&alpha;}}_2^{\mathrm{new}}={\mathrm{\&alpha;}}_2^{\mathrm{old}}+\frac{y_2(E_1-E_2)}{\mathrm{\&eta;}}$

Wherein:

$E_i=\underset{j=1}{\overset{N}{\mathrm{\&Sigma;}}}{\mathrm{\&alpha;}}_j{y}_{j}K(x_j,x_i)+b-y_i,i=\mathrm{1,2}$

η＝K ₁₁+K ₂₂-2K ₁₂

α after pruning ₂

${\mathrm{\&alpha;}}_2^{\mathrm{new},\mathrm{clip}}=\left\{\begin{array}{cc}H,& {\mathrm{\&alpha;}}_2^{\mathrm{new}}\&GreaterEqual;H\\ {\mathrm{\&alpha;}}_2^{\mathrm{new}},& L\&le;{\mathrm{\&alpha;}}_2^{\mathrm{new}}\&le;H\\ L,& {\mathrm{\&alpha;}}_2^{\mathrm{new}}\&le;L\end{array}\right.$

Wherein:

If y ₁=y ₂, $L=\max (0,{\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_1^{\mathrm{old}}),H=\min (C,C+{\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_1^{\mathrm{old}})$

If y ₁≠ y ₂, $L=\max (0,{\mathrm{\&alpha;}}_2^{\mathrm{old}}+{\mathrm{\&alpha;}}_1^{\mathrm{old}}-C),H=\min (C,{\mathrm{\&alpha;}}_2^{\mathrm{old}}+{\mathrm{\&alpha;}}_1^{\mathrm{old}})$

According to can obtain ${\mathrm{\&alpha;}}_1^{\mathrm{new}}={\mathrm{\&alpha;}}_1^{\mathrm{old}}+y_1{y}_2({\mathrm{\&alpha;}}_2^{\mathrm{old}}-{\mathrm{\&alpha;}}_2^{\mathrm{new},\mathrm{clip}});$

B () adopts heuristic to carry out variables choice, algorithm specifically comprises two-layer circulation: outer loop and interior loop, the selection of first variable is at outer loop, the selection of second variable is in interior loop, outer loop selects the sample point violating KKT condition in all samples, all samples are checked whether to meet KKT condition, namely

$y_{i}g\left(x_i\right)=\left\{\begin{array}{c}\&GreaterEqual;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=0\}\\ =1,\left\{x_i\right|0<{\mathrm{\&alpha;}}_i<C\}\\ \&le;1,\left\{x_i\right|{\mathrm{\&alpha;}}_i=C\}\end{array}\right.$

Interior loop mainly seeks to make | E ₁-E ₂| the variable of the maximum correspondence of value as second variable, if the optimal value of objective function can not be made to have certain reduction by said method, then travel through whole training sample set, otherwise enter outer loop and again find first variable.After often completing a suboptimization, the E reset threshold b and upgrade corresponding to each variable _ivalue.

2. the storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea according to claim 1, it is characterized in that, green tea produced in Anhui Province tea tea sample is carried out pouch independent packaging with tin paper bag, every bag of 50g by (1) respectively, sack seals, and encloses hygroscopic agent simultaneously and make moist to prevent tealeaves in bag; Then put into refrigerator and cooled and hide (4 DEG C) preservation.Carried out a test every 15 days, carry out 6 tests altogether.The bag number needed is taken out in each test, and all the other are motionless.

3. the storage time sorting technique of Electronic Nose green tea produced in Anhui Province tea according to claim 1, is characterized in that, (2) data set comprises 2112 samples, 10 sample attributes, 6 classifications.