CN110186871A - A kind of method of discrimination in the fresh tea leaves place of production - Google Patents
A kind of method of discrimination in the fresh tea leaves place of production Download PDFInfo
- Publication number
- CN110186871A CN110186871A CN201910555468.XA CN201910555468A CN110186871A CN 110186871 A CN110186871 A CN 110186871A CN 201910555468 A CN201910555468 A CN 201910555468A CN 110186871 A CN110186871 A CN 110186871A
- Authority
- CN
- China
- Prior art keywords
- sample
- tea leaves
- production
- place
- fresh tea
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 82
- 238000000034 method Methods 0.000 title claims abstract description 49
- 241001122767 Theaceae Species 0.000 title claims abstract 25
- 238000001228 spectrum Methods 0.000 claims abstract description 49
- 230000003595 spectral effect Effects 0.000 claims abstract description 33
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 25
- 238000002329 infrared spectrum Methods 0.000 claims abstract description 23
- 238000004611 spectroscopical analysis Methods 0.000 claims abstract description 16
- 230000002068 genetic effect Effects 0.000 claims abstract description 15
- 238000012216 screening Methods 0.000 claims abstract description 15
- 238000012360 testing method Methods 0.000 claims abstract description 14
- 230000006870 function Effects 0.000 claims description 36
- 210000002569 neuron Anatomy 0.000 claims description 23
- 230000005284 excitation Effects 0.000 claims description 22
- 238000004364 calculation method Methods 0.000 claims description 13
- 230000000694 effects Effects 0.000 claims description 11
- 238000007781 pre-processing Methods 0.000 claims description 6
- 230000004069 differentiation Effects 0.000 claims description 5
- 239000000126 substance Substances 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 229910000530 Gallium indium arsenide Inorganic materials 0.000 claims description 3
- 239000012141 concentrate Substances 0.000 claims description 3
- 230000003287 optical effect Effects 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 abstract description 4
- 244000269722 Thea sinensis Species 0.000 description 47
- 235000013616 tea Nutrition 0.000 description 46
- 230000005540 biological transmission Effects 0.000 description 5
- 238000010606 normalization Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000002708 enhancing effect Effects 0.000 description 2
- 238000012417 linear regression Methods 0.000 description 2
- 210000005036 nerve Anatomy 0.000 description 2
- 238000004497 NIR spectroscopy Methods 0.000 description 1
- 230000003466 anti-cipated effect Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 235000009569 green tea Nutrition 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000012821 model calculation Methods 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 239000002994 raw material Substances 0.000 description 1
- 238000004904 shortening Methods 0.000 description 1
- 238000007619 statistical method Methods 0.000 description 1
- 238000010025 steaming Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 238000010408 sweeping Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/359—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light
Landscapes
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
A kind of method of discrimination in the fresh tea leaves place of production, is related to rapid test paper identification technology field.The method is that the prediction model in the fresh tea leaves place of production is established using the near infrared spectrum of fresh tea leaves, is then determined according to the place of production of the prediction model to unknown fresh tea leaves；It is characterized by: the method for building up of the prediction model in the fresh tea leaves place of production are as follows: scanning obtains the near infrared spectrum of different sources fresh tea leaves, then after carrying out pretreatment cancelling noise information to sample spectra, using the applying frequency of genetic algorithm screening spectroscopic data point, the spectral information data point of modeling is then filtered out using Partial Least Squares；The extreme learning machine model i.e. prediction model in the fresh tea leaves place of production is established with the spectral information data point filtered out again.The present invention realizes quick, accurate, the objective prediction to the fresh tea leaves place of production, and model structure is simple, modeling rate is fast, prediction accuracy is high, Model Practical is strong.
Description
Technical field
The present invention relates to rapid test paper identification technology fields, more specifically to a kind of to apply near-infrared spectrum technique
Differentiate the method in the fresh tea leaves place of production.
Background technique
Beautiful dew of bestowing favour is the famous steaming green tea in China and national geography sign protection product, it is desirable that the tea of processing is fresh
Leaf must pick up from its scope of conservation area, and protection zone is mainly Enshi City Baiyangping township, military outpost township and sun river township.Due to grace
The huge market influence of Shi Yulu brand, the tea grower in the area Zhou Biancha are driven by interests, often pick in non-protection area
Fresh tea leaves pretend to be the fresh leaf of protection zone, and are sold to higher price the Yu Lucha processing factory that bestows favour and earn additional interests, and tea
Leaf purchases personnel when purchasing fresh leaf, and the place of production of fresh leaf, but this side are often differentiated with the feeling of itself and working experience
Method is subjective, also vulnerable to the influence of external environment, often judges incorrectly, and brings in this way to subsequent tea processing
Negative consequence also affects greatly the brand reputation for beautiful dew of bestowing favour, due to a lack of the means that effectively can accurately differentiate the place of production,
If things go on like this, the beautiful dew brand that can make to bestow favour loses the market competitiveness, becomes the synonym of public tea.It therefore, is effectively maintenance grace
The brand reputation of Shi Yulu is badly in need of establishing a kind of method that is accurate, objectively differentiating the fresh tea leaves place of production.
And near-infrared spectrum technique has quick, lossless, the objective advantage for differentiating the sample place of production.Cai Hailan etc. is reviewed closely
Progress of the infrared spectrum technology in the control of fresh tea leaf raw material, process control and made tea is examined, and to the skill
Application prospect of the art in this field is looked forward to；Zhao Jiewen etc. is realized using near-infrared spectrum technique to the made tea place of production
Quick nondestructive differentiates, but the problems such as the uniqueness due to fresh tea leaves and sample homogeneity control difficulty, not by near infrared light
The place of production that spectral technology is applied to fresh tea leaves differentiates；Application No. is 201610930724.5 Chinese patent (publication No. CN
A kind of method that near infrared spectrum quickly determines the fresh tea leaves place of production 106568741A) is disclosed, this method tentatively realizes difference
The quick discrimination of place of production fresh leaf, but there is also following shortcomings for this method: do not screened when modeling fresh leaf characteristic spectrum section,
Non- cancelling noise information, this is easy to bring over-fitting, is unfavorable for the steady of model, moreover, existing between sample spectra a large amount of
Interference information and group frequency and frequency multiplication information, inevitably reduce forecast result of model, and it is longer to model the time.
Summary of the invention:
In view of the above-mentioned problems of the prior art, the object of the present invention is to provide it is a kind of it is quick, lossless, accurately differentiate fresh tea leaves
The method in the place of production realizes simplified model structure, improves modeling rate, improves fresh leaf sample place of production prediction accuracy and enhancing model
The purpose of practicability.
To achieve the above object, the present invention adopts the following technical scheme:
A kind of method of discrimination in the fresh tea leaves place of production, the method are to establish the fresh tea leaves place of production using the near infrared spectrum of fresh tea leaves
Then prediction model determines according to the place of production of the prediction model to unknown fresh tea leaves；It is characterized by: the fresh tea leaves produce
The method for building up of the prediction model on ground are as follows: scanning obtain different sources fresh tea leaves near infrared spectrum, then to sample spectra into
After row pretreatment cancelling noise information, using the applying frequency of genetic algorithm screening spectroscopic data point, then using minimum two partially
Multiplication filters out the spectral information data point of modeling；Extreme learning machine model is established i.e. with the spectral information data point filtered out again
The prediction model in the fresh tea leaves place of production.
A kind of method of discrimination in the fresh tea leaves place of production, it is characterised in that: the method for building up of the prediction model, specifically
The following steps are included:
Step 1: the acquisition of fresh leaf sample and classification
Fresh leaf sample in the fresh leaf sample and non-protection area in protection zone is acquired respectively, according to place of production difference, by fresh leaf sample
It is divided into 2 set of calibration set and verifying collection, is respectively used to establish calibration set near infrared prediction model and calibration set is predicted
Model robustness is tested；Different chemical scores is assigned respectively to the fresh tea leaves sample of different sources；
Step 2: spectral scan
The near infrared spectrum for scanning fresh tea leaves sample respectively using Fourier transformation type near infrared spectrometer, obtains spectral information；
Step 3: spectral noise information pre-processing
Applied Chemometrics software carries out denoising using vector method for normalizing near infrared spectrum obtained in step 2
Pretreatment；After spectrum denoising, then convert sample spectra to pairs of data point；
Step 4: screening optimal spectrum Information Number strong point
Using the applying frequency of genetic algorithm screening spectral information data point, modeling then is filtered out using Partial Least Squares
Optimal spectrum Information Number strong point；
Step 5: extreme learning machine model foundation
Using the optimal spectrum Information Number strong point filtered out in step 4 as input value, using fresh leaf sample different sources as output valve,
The prediction model in the fresh tea leaves place of production is established using the extreme learning machine program bag in Matlab2017b software, excitation function includes 2
Kind: sigmoid function and logistic function；The neuron that hidden layer contains is respectively 5,10,15 and 20；Comparison model
Coefficient R c and validation-cross root mean square variance RMSECV size, obtain optimal near infrared prediction model, remember simultaneously
The time required to record modeling；Wherein, Rc is bigger, RMSECV is smaller, indicates that forecast result of model is better；
Wherein, RMSECV calculation formula are as follows:,
Rc calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets,For sample
Concentrate the average value of the measured value of i-th of sample, i≤n in formula；
Step 6: model robustness is examined
Application verification collection sample tests to the fresh leaf sample extreme learning machine prediction model effect of different sources, acquired results
It indicates that wherein Rp is bigger, RMSEP is smaller with coefficient R p, verifying mean square deviation RMSEP and differentiation rate, indicates model robustness
It is better, it can accurately predict the place of production of fresh leaf sample；
Wherein RMSEP calculation formula are as follows:,
Rp calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets, i in formula≤
n。
The method of discrimination in a kind of fresh tea leaves place of production, it is characterised in that: the acquisition of fresh leaf sample is protected in the step 1
Protecting sample in area is 60,60, sample in non-protection area；Fresh leaf sample standard of plucking is respectively as follows: bud, the first leaf, the second leaf,
Third leaf, three leaf of one leaf of a bud, two leaves and a bud and a bud.
A kind of method of discrimination in fresh tea leaves place of production, it is characterised in that: the Fourier transformation type near infrared spectrum
Instrument is the silent winged generation that II type Fourier Transform Near Infrared instrument of Antaris of U.S.'s match, selects integrating sphere diffusing reflection optics flat
Platform；Spectral scanning range 4000-10000cm-1；Resolution ratio 8cm-1, detector InGaAs；Each sample acquires 3 spectrum,
Scanning 64 times every time, the spectrum acquired to 3 times is averaged, using averaged spectrum as the final spectrum of the fresh leaf sample；It is sweeping
Before retouching fresh leaf sample spectra, which is preheated into 1h, after keeping room temperature and humidity almost the same, then by sample
It is fitted into rotating cup matched with instrument and carries out spectral scan, the dress sample thickness of each sample is consistent, and guarantees near infrared light
Sample can not be penetrated.
The method of discrimination in a kind of fresh tea leaves place of production, it is characterised in that: fresh leaf sample size is 120 in step 1
Part, wherein 90, calibration set sample, verifying collect 30, sample.
A kind of method of discrimination in fresh tea leaves place of production is differentiating the application bestowed favour on the beautiful dew fresh tea leaves place of production.
Compared with prior art, after first rejecting sample noise information the invention has the following beneficial effects: (1) present invention,
Pairs of data point is converted by sample spectra to save in excel, is then screened using genetic algorithm and Partial Least Squares
The spectral information data point in the optimal reflection fresh leaf place of production out；On this basis, limits of application learning machine algorithm establishes fresh tea leaves
The near-infrared model in the place of production realizes quick, accurate, the objective prediction to the fresh tea leaves place of production, plays simplified model structure, improves
Rate is modeled, fresh leaf sample place of production prediction accuracy is improved and enhances the purpose of Model Practical；Modeling data point only accounts for whole
The best fresh leaf place of production extreme learning machine model RMSECV of the 2.4% of spectroscopic data point, foundation is 0.1002, Rc 0.990, is built
The mould time is only 3 seconds.(2) present invention is combined using genetic algorithm and partial least squares algorithm, precisely screening reflection fresh leaf sample
The spectral information data point in the product place of production；As input data, by the neuron number for constantly optimizing extreme learning machine repeatedly
With excitation function, it is finally reached the accurate purpose for differentiating the fresh leaf sample place of production.(3)) by genetic algorithm, partial least squares algorithm
It is combined with extreme learning machine algorithm, perfection is realized to bud, the first leaf, the second leaf, third leaf, one leaf of a bud, two leaves and a bud
With the accurate prediction in the standard of plucking fresh leaf sample place of production such as three leaf of a bud, wherein motivating letter using 15 neurons and sigmoid
Predictablity rate is counted up to 100%；Deviation < 0.15.
Detailed description of the invention
Fig. 1 is the near infrared spectrum of fresh tea leaves sample；
Fig. 2 is the spectroscopic data point applying frequency of genetic algorithm screening；
Fig. 3 relationship between RMSECV and modeling spectroscopic data point；
Fig. 4 is extreme learning machine internal structure.
Specific embodiment
Below in conjunction with the drawings and specific embodiments, the present invention is described in further detail.
A kind of method of discrimination in the fresh tea leaves place of production, scanning obtain the near infrared spectrum of different sources fresh tea leaves sample, then
Sample spectra is carried out after pre-processing cancelling noise information, then convert pairs of data point in excel table for sample spectra
It saves；Using the applying frequency of genetic algorithm screening spectroscopic data point, modeling then is filtered out most using Partial Least Squares
Good spectral information data point；Finally optimal spectral information data point is input in extreme learning machine algorithm, by constantly anti-
The neuron number and excitation function for optimizing extreme learning machine again, establish the near-infrared spectroscopy in the fresh leaf sample place of production, are used for
Predict the place of production of fresh leaf sample.Specifically includes the following steps:
Step 1: the acquisition of fresh leaf sample and classification
Acquisition is bestowed favour in beautiful dew protection zone fresh leaf sample in fresh leaf sample and non-protection area, and totally 120.It is different according to the place of production,
Fresh leaf sample is divided into 2 set of calibration set and verifying collection, is respectively used to establish calibration set near infrared prediction model and right
Calibration set prediction model robustness is tested.Different chemical scores is assigned respectively to the fresh tea leaves sample of different sources, is protected
Fresh leaf sample place of production chemical score is respectively set as 1.00 in area, and fresh leaf place of production chemical score is set as 2.00 in non-protection area.
Wherein 60, sample in protection zone, 60, sample in non-protection area.Fresh leaf sample standard of plucking is respectively as follows: bud, the
(sample includes in protection zone and non-protection area for one leaf, the second leaf, third leaf, one leaf of a bud, two leaves and a bud and three leaf of a bud
Bud, the differing maturities fresh tea leaves samples such as the first leaf, the second leaf, third leaf, three leaf of one leaf of a bud, two leaves and a bud and a bud).Its
In, one leaf of a bud is made of simple bud, the first leaf and longer stalk, and two leaves and a bud is made of simple bud, the first leaf, the second leaf and long stalk,
One bud, three leaf is made of simple bud, the first leaf, the second leaf, third leaf and longer stalk.
Step 2: spectral scan
Using the silent winged generation that II type Fourier Transform Near Infrared instrument (FT-NIR) of Antaris of U.S.'s match, integrating sphere is selected
Diffusing reflection optical platform；Spectral scanning range 4000-10000cm-1；Resolution ratio 8cm-1, detector InGaAs.Each sample
3 spectrum are acquired, every time scanning 64 times, the spectrum acquired to 3 times is averaged, most using averaged spectrum as the fresh leaf sample
Whole spectrum.Before scanning fresh leaf sample spectra, which is preheated into 1h, keeps room temperature and humidity almost the same
Afterwards, then sample is fitted into rotating cup matched with instrument and carries out spectral scan, the dress sample thickness of each sample is consistent, and is protected
Card near infrared light can not penetrate sample.
The near infrared spectrum of fresh tea leaves sample is referring to Fig. 1.
Step 3: spectral noise information pre-processing
During spectra collection, it will usually which generating high-frequency noise and baseline drift etc. influences the noise letter of forecast result of model
Breath, if not pre-processed to spectral noise, being directly used in and establish prediction model it will cause the prediction effect of model is poor,
And model is also unstable, therefore needs to carry out noise suppression preprocessing to spectral information before modeling.Applied chemistry meter in this step
Amount learns software TQ Analyst 9.4.45 software and 7.0 software of OPUS to the near infrared spectrum of whole different sources fresh leaf samples
Smooth, first derivative is carried out respectively, and second dervative, multiplicative scatter correction and vector normalization pretreatment improve the noise of spectrum
Than to be conducive to establish steady prediction model；By comparing, obtain optimal spectrum preprocess method for vector normalization,
Its influence that can deduct the linear translation in sample spectra, and every spectrum is individually corrected, there is stronger information
Processing capacity.After spectrum denoising, then by sample spectra pairs of data point (X-Y one-to-one correspondence) is converted, is stored in excel
In table, prediction model is established for subsequent.
Step 4: screening optimal spectrum Information Number strong point
1) spectroscopic data point is screened using genetic algorithm
Near infrared spectrum contains all information of sample, such as the place of production, plucking time, kind and component content information, therefore,
In order to improve the prediction effect of model, need to screen the spectral information in the reflection fresh leaf sample place of production, the removal light useless with modeling
Spectrum information.Model prediction accuracy not only can be improved in this, can also greatly simplify the structure of model, reduces the operation of model
Amount, reduces the operation time of modeling, reduces modeling cost.Therefore, present invention application genetic algorithm screens useful spectroscopic data
Point calculates the applying frequency of spectroscopic data point, establishes Partial Least Squares model (Fig. 2) for subsequent.
Figure it is seen that most spectroscopic data points are 2 times for frequency, maximum application frequency is 7 times, can
See, the same information that spectrally different data point represents is different, and the importance of modeling is also different.Therefore, it attempts to apply this
A little data establish Partial Least Squares model.
2) Partial Least Squares model foundation
Partial Least Squares is a kind of statistical method, and predictive variable and observational variable are projected to one newly respectively by projecting
Space, to find a linear regression model (LRM), it has very strong stability, and the models fitting effect of foundation is good, has relatively strong
Practical application.Acquired results often use validation-cross root mean square variance (RMSECV) and related coefficient (Rc) to indicate.Wherein, Rc
It is bigger, RMSECV is smaller, indicate forecast result of model it is better.
Wherein, RMSECV calculation formula are as follows:,
Rc calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets,For sample
Product concentrate the average value of the measured value of i-th of sample, i≤n in formula.
Therefore, in order to which the place of production of fresh leaf sample is better anticipated, present invention application Partial Least Squares establishes prediction model,
Preliminary screening goes out the optimal spectrum Information Number strong point modeled, reaches the mesh of precisely screening reflection fresh leaf sample place of production spectral information
's.Meanwhile the Partial Least Squares also selects to be which kind of optimal spectrum preprocess method in verification step three in turn.
Establish the prediction model of different data point respectively using Partial Least Squares, acquired results are shown in Fig. 3:
From figure 3, it can be seen that the RMSECV of Partial Least Squares model is minimum at this time when modeling spectroscopic data point is 37,
It is 0.1836；Coefficient R c is 0.865.At this point, the modeling optimal spectrum data point of screening is respectively as follows: 4065.21,
4335.19, 4389.19, 4551.18, 5654.26, 5820.11, 5839.40, 5908.82, 5947.39,
5997.53, 6032.24, 6055.39, 6741.92, 6954.05, 6981.05, 7409.17 7482.45,
7679.15, 7729.29, 8072.56, 8103.42, 8176.70, 8427.40, 8616.39, 8631.82,
8670.39, 8693.53, 8724.38, 8789.95, 8932.66, 8994.37, 9052.22, 9102.36,
9144.79, 9191.07, 9503.48 and 9958.60 cm-1.When establishing best model, the spectroscopic data of application is located in advance
Reason method is vector method for normalizing.
Step 5: extreme learning machine model foundation
Extreme learning machine (Extreme learning machine, ELM) is a kind of Single hidden layer feedforward neural networks study calculation
Method, advantage are: will not fall into local optimum, without iteration, can rapid solving, no setting is required complicated parameter, pass through intersection
Verifying optimizes excitation function and node in hidden layer repeatedly, and then obtains optimum prediction model.
The present invention is on the basis of step 4, although tentatively having obtained the spectral information data point in the reflection fresh leaf sample place of production,
But due to the presence of group frequency and frequency multiplication information between spectral information, it is likely between each data point there is also non-linear relation, because
This, is for the place of production of more accurate prediction fresh leaf sample, the further precisely pre- test sample of limits of application learning machine algorithm of the present invention
The place of production of product, extreme learning machine internal structure are shown in Fig. 4.
The 37 optimal spectrum data points screened in applying step four of the present invention further establish the fresh tea leaves place of production limit
Habit machine model.Using optimal spectrum Information Number strong point as input value, using fresh leaf sample different sources as output valve, application
Extreme learning machine program bag in Matlab2017b software establishes the discrimination model in the fresh tea leaves place of production, and excitation function includes 2 kinds:
Sigmoid function and logistic function；The number of nodes that hidden layer contains has 4 kinds, respectively 5,10,15 and 20.In order to reach
To optimal prediction effect, lot of experimental data is needed to verify repeatedly to 8 kinds of obtained extreme learning machine models, further
The combination of optimal neuron number and excitation function is obtained, optimal prediction effect, comparison model phase relation can be reached
Number (correlation coefficient of calibration, Rc) and validation-cross root mean square variance (root mean
Square error of calibration, RMSECV) size, obtain best near infrared prediction model, wherein Rc is got over
Greatly, RMSECV is smaller, indicates that forecast result of model is better.Best calibration set model is obtained after comparison；Modeling institute is recorded simultaneously
It takes time.8 kinds of extreme learning machine model results are shown in Table 1, as it can be seen from table 1 the best fresh leaf place of production extreme learning machine established
Model RMSECV is 0.1002, Rc 0.990, and the modeling time is only 3 seconds, at this point, neuron number used in modeling is 15,
Excitation function is sigmoid function.
Step 6: model robustness is examined
To avoid the occurrence of over-fitting, a steady fresh leaf sample place of production prediction model is established, the mesh of practical application is reached
, therefore, test using collection sample is all verified to the fresh leaf sample extreme learning machine prediction model effect of different sources,
Acquired results related coefficient (correlation coefficient of prediction, Rp), verifying mean square deviation (root
Mean square error of prediction, RMSEP) and differentiation rate indicate that wherein Rp is bigger, the smaller then table of RMSEP
Representation model robustness is better, can accurately predict the place of production of fresh leaf sample.
Wherein RMSEP calculation formula are as follows:,
Rp calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets, i in formula≤
n。
Fresh leaf sample size is 120 parts in the present invention, and fresh leaf sample is calibration set and verifying according to the ratio cut partition of 3:1
Collection, wherein 90, calibration set sample, verifying collect 30, sample.30 parts of samples of application verification collection examine calibration set model at this time
It tests, acquired results indicate that concrete outcome is referring to table 1 with coefficient R p and verifying collection mean square deviation RMSEP.
As it can be seen from table 1 in different sources fresh leaf sample extreme learning machine model, when neuron is 5, excitation function
When for sigmoid, the modeling time is 8 seconds, and calibration set model Rc is 0.947, RMSECV 0.2802, when with all 30 verifyings
Collection sample is when testing to calibration set model robustness, be verified collection model Rp be 0.932, RMSEP 0.3025.Work as mind
When through member be 10, transmission function is logistic, the modeling time is 12 seconds, and calibration set model Rc is 0.914, RMSECV is
0.3853, when with all 30 verifying collection samples test to calibration set model robustness, being verified collection model Rp is
0.904,0.3526 RMSEP.When neuron is 15, transmission function is sigmoid, the modeling time is 3 seconds, calibration set mould
Type Rc is 0.990, RMSECV 0.1002, is tested when collecting samples with all 30 verifyings to calibration set model robustness
When, be verified collection model Rp be 0.986, RMSEP 0.1084.When neuron is 20, transmission function is logistic,
Modeling the time is 5 seconds, and calibration set model Rc is 0.965, RMSECV 0.2003, when collecting samples to correction with all 30 verifyings
Collection model robustness is when testing, be verified collection model Rp be 0.954, RMSEP 0.2118.When neuron is 5, swashs
When to encourage function be logistic, the modeling time is 6 seconds, and calibration set model Rc is 0.956, RMSECV 0.2115, when with whole
When 30 verifyings collection samples test to calibration set model robustness, it is verified that collection model Rp is 0.951, RMSEP is
0.2715.When neuron is 10, excitation function is sigmoid, modeling the time be 6 seconds, calibration set model Rc be 0.960,
RMSECV is 0.2104, when with all 30 verifying collection samples test to calibration set model robustness, is verified collection
Model Rp is 0.927, RMSEP 0.3103.When neuron is 15, excitation function is logistic, the modeling time is 9
Second, calibration set model Rc is 0.965, RMSECV 0.2007, steady to calibration set model when collecting samples with all 30 verifyings
Property when testing, be verified collection model Rp be 0.941, RMSEP 0.2848.When neuron is 20, excitation function is
When sigmoid, the modeling time is 8 seconds, and calibration set model Rc is 0.936, RMSECV 0.2985, is collected when with all 30 verifyings
When sample tests to calibration set model robustness, be verified collection model Rp be 0.912, RMSEP 0.3958.
As it can be seen that being built in the case where limits of application learning machine method but internal different neuron numbers and different excitation functions
In vertical prediction model, with the fresh leaf sample different sources pole established when being sigmoid with 15 neurons and excitation function
Limit learning machine model prediction result is best, and forecast result of model is best, most short the time required to modeling；Secondly for 20 nerves
The fresh leaf sample different sources extreme learning machine prediction model that member and transmission function are established when being logistic, modeling time are 5
Second；The worst fresh leaf sample different sources limit study established when be with 15 neurons and transmission function being logistic
Machine prediction model, modeling time are 9 seconds.It follows that same extreme learning machine modeling method, but intrinsic nerve member number with
The difference of excitation function can generate large effect to the prediction result for establishing model, therefore, in limits of application learning machine method
When establishing model, neuron number and excitation function are reasonably selected, can just reach optimal prediction effect.
Optimal limit learning machine model using 15 neurons and excitation function to establish when sigmoid verifies 30
Collection fresh leaf sample the place of production predicted, the table 2 that prediction result is seen below.From table 2 it can be seen that the true value in the fresh leaf sample place of production
All < 0.15 with the absolute value of the difference of predicted value (| deviation |), show that model predicts all samples correct, differentiation rate is
100%.As it can be seen that the different sources fresh leaf sample limit study established when 15 neurons of application and excitation function are sigmoid
Quick, Accurate Prediction of the machine model realization to the fresh leaf sample place of production.
In conclusion the present invention provides a kind of application near-infrared spectrum technique combination genetic algorithm, partial least squares algorithm
With extreme learning machine algorithm for accurately predicting the place of production of fresh leaf sample, fresh leaf sample noise information is first rejected, is obtained best
Preprocessing procedures are vector normalization；Then the applying frequency of spectroscopic data point is obtained using genetic algorithm, and application is inclined
The optimal spectrum data point of least square method screening modeling is 37, accounts for the 2.4% of whole spectroscopic data points；Again with preferred light
Modal data point is that input value establishes the extreme learning machine prediction model in the fresh leaf place of production, by repeatedly constantly preferably neuron number and
Excitation function, the extreme learning machine model prediction effect established when being finally sigmoid with 15 neurons of application and excitation function
Fruit is best (the modeling time is 3 seconds, Rp=0.986, RMSEP=0.1084), determines prediction knot to the place of production of verifying collection fresh leaf sample
Fruit is all correct, is 100%.Therefore, the present invention is by genetic algorithm, partial least squares algorithm and extreme learning machine algorithm (15 minds
Through member and sigmoid excitation function) it combines, perfection is realized to bud, the first leaf, the second leaf, third leaf, one leaf of a bud, one
The standard of plucking fresh leaf sample such as three leaf of two leaf of bud and the bud place of production accurate prediction (| deviation | all < 0.15, predictablity rate is
100%), the prediction model of foundation does not only reach the model calculation amount that substantially reduces (modeling data point accounts for whole spectroscopic data points
2.4%), the purpose of simplified model and shortening modeling time (the modeling time is only 3 seconds), while also acting as the prediction for improving model
The purpose of accuracy and enhancing Model Practical.
When the fresh tea leaves to the unknown place of production differentiate, the near infrared spectrum of its fresh tea leaves is first scanned, through vector normalizing
After the pretreatment of change method, call in above-mentioned established model and quick predict carried out to unknown spectrum place of production value, when output valve 1 ±
The fresh tea leaves are determined when in 0.15 range from beautiful dew protection zone of bestowing favour, determining when output valve is in 2 ± 0.15 ranges should
Fresh tea leaves are from non-beautiful dew protection zone of bestowing favour.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those of ordinary skill in the art to which the present invention belongs, exist
Under the premise of not departing from present inventive concept, a number of simple deductions or replacements can also be made, and above structure all shall be regarded as belonging to
Protection scope of the present invention.
Claims (7)
1. a kind of method of discrimination in the fresh tea leaves place of production, the method is to establish the fresh tea leaves place of production using the near infrared spectrum of fresh tea leaves
Prediction model, then determined according to the place of production of the prediction model to unknown fresh tea leaves；It is characterized by: the fresh tea leaves
The method for building up of the prediction model in the place of production are as follows: scanning obtains the near infrared spectrum of different sources fresh tea leaves, then to sample spectra
After carrying out pretreatment cancelling noise information, using the applying frequency of genetic algorithm screening spectroscopic data point, then using partially minimum
Square law filters out the spectral information data point of modeling；Extreme learning machine model is established with the spectral information data point filtered out again
That is the prediction model in the fresh tea leaves place of production.
2. a kind of method of discrimination in fresh tea leaves place of production according to claim 1, it is characterised in that: the prediction model is built
Cube method, specifically includes the following steps:
Step 1: the acquisition of fresh leaf sample and classification
Fresh leaf sample in the fresh leaf sample and non-protection area in protection zone is acquired respectively, according to place of production difference, by fresh leaf sample
It is divided into 2 set of calibration set and verifying collection, is respectively used to establish calibration set near infrared prediction model and calibration set is predicted
Model robustness is tested；Different chemical scores is assigned respectively to the fresh tea leaves sample of different sources；
Step 2: spectral scan
The near infrared spectrum for scanning fresh tea leaves sample respectively using Fourier transformation type near infrared spectrometer, obtains spectral information；
Step 3: spectral noise information pre-processing
Applied Chemometrics software carries out denoising using vector method for normalizing near infrared spectrum obtained in step 2
Pretreatment；After spectrum denoising, then convert sample spectra to pairs of data point；
Step 4: screening optimal spectrum Information Number strong point
Using the applying frequency of genetic algorithm screening spectral information data point, modeling then is filtered out using Partial Least Squares
Optimal spectrum Information Number strong point；
Step 5: extreme learning machine model foundation
Using the optimal spectrum Information Number strong point filtered out in step 4 as input value, using fresh leaf sample different sources as output valve,
The prediction model in the fresh tea leaves place of production is established using the extreme learning machine program bag in Matlab2017b software, excitation function includes 2
Kind: sigmoid function and logistic function；The neuron that hidden layer contains is respectively 5,10,15 and 20；Comparison model
Coefficient R c and validation-cross root mean square variance RMSECV size, obtain optimal near infrared prediction model, remember simultaneously
The time required to record modeling；Wherein, Rc is bigger, RMSECV is smaller, indicates that forecast result of model is better；
Wherein, RMSECV calculation formula are as follows:,
Rc calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets,For sample
Concentrate the average value of the measured value of i-th of sample, i≤n in formula；
Step 6: model robustness is examined
Application verification collection sample tests to the fresh leaf sample extreme learning machine prediction model effect of different sources, acquired results
It indicates that wherein Rp is bigger, RMSEP is smaller with coefficient R p, verifying mean square deviation RMSEP and differentiation rate, indicates model robustness
It is better, it can accurately predict the place of production of fresh leaf sample；
Wherein RMSEP calculation formula are as follows:,
Rp calculation formula are as follows:,
In formula, n indicate sample number, yi andy i ’ The measured value and predicted value of i-th of sample respectively in sample sets, i in formula≤
n。
3. a kind of method of discrimination in fresh tea leaves place of production according to claim 2, it is characterised in that: fresh leaf in the step 1
It is 60 that sample, which acquires sample in protection zone, 60, sample in non-protection area；Fresh leaf sample standard of plucking is respectively as follows: bud, and first
Leaf, the second leaf, third leaf, three leaf of one leaf of a bud, two leaves and a bud and a bud.
4. a kind of method of discrimination in fresh tea leaves place of production according to claim 2, it is characterised in that: in the step 2 in Fu
Leaf transformation type near infrared spectrometer is the silent winged generation that II type Fourier Transform Near Infrared instrument of Antaris of U.S.'s match, is selected
Integrating sphere diffusing reflection optical platform；Spectral scanning range 4000-10000cm-1；Resolution ratio 8cm-1, detector InGaAs；Often
A sample acquires 3 spectrum, every time scanning 64 times, and the spectrum acquired to 3 times is averaged, using averaged spectrum as the fresh leaf sample
The final spectrum of product；Before scanning fresh leaf sample spectra, which is preheated into 1h, keeps room temperature and humidity base
After this is consistent, then sample is fitted into rotating cup matched with instrument and carries out spectral scan, the dress sample thickness of each sample is kept
Unanimously, guarantee that near infrared light can not penetrate sample.
5. a kind of method of discrimination in fresh tea leaves place of production according to claim 2, it is characterised in that: fresh leaf sample in step 1
Quantity is 120 parts, wherein 90, calibration set sample, verifying collection 30, sample.
6. a kind of method of discrimination in fresh tea leaves place of production according to claim 2, it is characterised in that: in step 5 prediction model
Neuron number uses 15, and excitation function is sigmoid function.
7. a kind of method of discrimination in the fresh tea leaves place of production described in any one of -6 is bestowed favour Yu Lucha in differentiation according to claim 1
Application on the fresh leaf place of production.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910555468.XA CN110186871A (en) | 2019-06-25 | 2019-06-25 | A kind of method of discrimination in the fresh tea leaves place of production |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910555468.XA CN110186871A (en) | 2019-06-25 | 2019-06-25 | A kind of method of discrimination in the fresh tea leaves place of production |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110186871A true CN110186871A (en) | 2019-08-30 |
Family
ID=67723410
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910555468.XA Pending CN110186871A (en) | 2019-06-25 | 2019-06-25 | A kind of method of discrimination in the fresh tea leaves place of production |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110186871A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111488851A (en) * | 2020-04-17 | 2020-08-04 | 成都曙光光纤网络有限责任公司 | Traceability detection method, device, equipment and medium for fruit production place |
| CN113033075A (en) * | 2021-02-26 | 2021-06-25 | 柳州融安金园食品有限公司 | Storage and preservation method and device for kumquat fruits |
| CN113484249A (en) * | 2020-07-18 | 2021-10-08 | 北京立鼎智行科技有限公司 | Asphalt brand identification system based on extreme learning machine |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106560700A (en) * | 2016-10-20 | 2017-04-12 | 中国计量大学 | Machine learning method for identifying origin of Wuyi rock tea automatically |
| CN109001147A (en) * | 2018-08-24 | 2018-12-14 | 湖北省农业科学院果树茶叶研究所 | A method of fresh tea leaves geography information is differentiated using near infrared spectrum |
-
2019
- 2019-06-25 CN CN201910555468.XA patent/CN110186871A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106560700A (en) * | 2016-10-20 | 2017-04-12 | 中国计量大学 | Machine learning method for identifying origin of Wuyi rock tea automatically |
| CN109001147A (en) * | 2018-08-24 | 2018-12-14 | 湖北省农业科学院果树茶叶研究所 | A method of fresh tea leaves geography information is differentiated using near infrared spectrum |
Non-Patent Citations (3)
| Title |
|---|
| 张海东等: "近红外光谱结合极限学习机和GA_PLS算法检测普洱茶茶多酚含量", 《激光与光电子学进展》 * |
| 王毅等: "优化茶多酚近红外光谱定量分析模型", 《计算机与应用化学》 * |
| 聂钰洪等: "近红外分析技术在茶叶检测方面的研究进展", 《现代农业科技》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111488851A (en) * | 2020-04-17 | 2020-08-04 | 成都曙光光纤网络有限责任公司 | Traceability detection method, device, equipment and medium for fruit production place |
| CN113484249A (en) * | 2020-07-18 | 2021-10-08 | 北京立鼎智行科技有限公司 | Asphalt brand identification system based on extreme learning machine |
| CN113033075A (en) * | 2021-02-26 | 2021-06-25 | 柳州融安金园食品有限公司 | Storage and preservation method and device for kumquat fruits |
| CN113033075B (en) * | 2021-02-26 | 2022-09-09 | 柳州融安金园食品有限公司 | Storage and preservation method and device for kumquat fruits |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105928901B (en) | A kind of near-infrared quantitative model construction method that qualitative, quantitative combines | |
| CN105630743B (en) | A kind of system of selection of spectrum wave number | |
| CN106568738A (en) | Method of using near infrared spectroscopy to rapidly determine fresh leaves of tea in different quality grades | |
| CN102564993B (en) | Method for identifying rice varieties by using Fourier transform infrared spectrum and application of method | |
| CN109253985B (en) | Method for identifying wood grade for koto panel by near infrared spectrum based on neural network | |
| CN110186871A (en) | A kind of method of discrimination in the fresh tea leaves place of production | |
| CN108872132A (en) | A method of fresh tea leaves kind is differentiated using near infrared spectrum | |
| CN105628708A (en) | Quick nondestructive testing method for multi-parameter quality of south Xinjiang red dates | |
| CN105717066A (en) | Near-infrared spectrum recognition model based on weighting association coefficients | |
| CN110398473A (en) | A kind of rapid test paper detection method and system | |
| Cui et al. | Identification of maize seed varieties based on near infrared reflectance spectroscopy and chemometrics | |
| CN109115692A (en) | A kind of spectral data analysis method and device | |
| CN110186870A (en) | A kind of method that extreme learning machine spectral model differentiates the beautiful dew fresh tea leaves place of production of bestowing favour | |
| CN110308111A (en) | A method of using the near-infrared spectrum technique quick predict Yuanan yellow tea bored yellow time | |
| CN108663334A (en) | The method for finding soil nutrient spectral signature wavelength based on multiple Classifiers Combination | |
| CN113030011A (en) | Rapid nondestructive testing method and system for sugar content of fruits | |
| CN106442399B (en) | A kind of method that near infrared spectrum differentiates the different same kind fresh tea leaves of planting environment | |
| CN106568740A (en) | Method for rapid judging of varieties of fresh tea leaves by near infrared spectroscopy | |
| CN110084227A (en) | Mode identification method based on near-infrared spectrum technique | |
| CN106442400B (en) | A kind of method that near infrared spectrum quickly determines soil type fresh tea leaves | |
| CN106568741B (en) | A kind of method that near infrared spectrum quickly determines the fresh tea leaves place of production | |
| JP7203252B2 (en) | A Sensory Evaluation Method for Spectral Data of Mainstream Cigarette Smoke | |
| CN105866042B (en) | Based on pixel index unbiased esti-mator method biology index of quality spatial distribution detection method | |
| CN110334714A (en) | A kind of machine based on artificial neural network technology adopts mee tea vehicle tinctorial pattern product grade prediction technique | |
| CN110320173A (en) | The method for rapidly judging of machine fresh tea picking mee tea vehicle tinctorial pattern product grade based on particle swarm optimization algorithm |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190830 |
|
| RJ01 | Rejection of invention patent application after publication |