CN107192689A - A kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra - Google Patents
A kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra Download PDFInfo
- Publication number
- CN107192689A CN107192689A CN201710299188.8A CN201710299188A CN107192689A CN 107192689 A CN107192689 A CN 107192689A CN 201710299188 A CN201710299188 A CN 201710299188A CN 107192689 A CN107192689 A CN 107192689A
- Authority
- CN
- China
- Prior art keywords
- prediction
- high density
- milk powder
- density wavelet
- tera
- 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.)
- Granted
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 52
- 239000000843 powder Substances 0.000 title claims abstract description 48
- 235000013336 milk Nutrition 0.000 title claims abstract description 45
- 239000008267 milk Substances 0.000 title claims abstract description 45
- 210000004080 milk Anatomy 0.000 title claims abstract description 45
- 238000001514 detection method Methods 0.000 title claims abstract description 42
- 238000012856 packing Methods 0.000 title claims abstract description 24
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 22
- 235000016709 nutrition Nutrition 0.000 claims abstract description 22
- 239000004615 ingredient Substances 0.000 claims abstract description 15
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 238000002790 cross-validation Methods 0.000 claims description 22
- 238000000034 method Methods 0.000 claims description 21
- 230000004927 fusion Effects 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 11
- 230000035764 nutrition Effects 0.000 claims description 10
- 230000000694 effects Effects 0.000 claims description 7
- 239000000203 mixture Substances 0.000 claims description 6
- 238000005267 amalgamation Methods 0.000 claims description 3
- 230000001419 dependent effect Effects 0.000 claims description 3
- 238000004806 packaging method and process Methods 0.000 description 12
- 239000013078 crystal Substances 0.000 description 10
- 230000003595 spectral effect Effects 0.000 description 9
- HCHKCACWOHOZIP-UHFFFAOYSA-N Zinc Chemical compound [Zn] HCHKCACWOHOZIP-UHFFFAOYSA-N 0.000 description 8
- 238000005070 sampling Methods 0.000 description 8
- 239000011701 zinc Substances 0.000 description 8
- 229910052725 zinc Inorganic materials 0.000 description 8
- IJGRMHOSHXDMSA-UHFFFAOYSA-N Atomic nitrogen Chemical compound N#N IJGRMHOSHXDMSA-UHFFFAOYSA-N 0.000 description 6
- XUIMIQQOPSSXEZ-UHFFFAOYSA-N Silicon Chemical compound [Si] XUIMIQQOPSSXEZ-UHFFFAOYSA-N 0.000 description 5
- 238000004458 analytical method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 5
- 229910052710 silicon Inorganic materials 0.000 description 5
- 239000010703 silicon Substances 0.000 description 5
- JBRZTFJDHDCESZ-UHFFFAOYSA-N AsGa Chemical compound [As]#[Ga] JBRZTFJDHDCESZ-UHFFFAOYSA-N 0.000 description 4
- 229910001218 Gallium arsenide Inorganic materials 0.000 description 4
- 238000000862 absorption spectrum Methods 0.000 description 4
- 230000007547 defect Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 239000000463 material Substances 0.000 description 4
- 238000001328 terahertz time-domain spectroscopy Methods 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 230000006378 damage Effects 0.000 description 3
- 229910052757 nitrogen Inorganic materials 0.000 description 3
- 235000015097 nutrients Nutrition 0.000 description 3
- 235000008476 powdered milk Nutrition 0.000 description 3
- 102000004169 proteins and genes Human genes 0.000 description 3
- 108090000623 proteins and genes Proteins 0.000 description 3
- 239000000126 substance Substances 0.000 description 3
- 238000010521 absorption reaction Methods 0.000 description 2
- -1 carton Substances 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 230000005684 electric field Effects 0.000 description 2
- 235000021393 food security Nutrition 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000003340 mental effect Effects 0.000 description 2
- 238000009659 non-destructive testing Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 229920003023 plastic Polymers 0.000 description 2
- 239000004033 plastic Substances 0.000 description 2
- 230000010287 polarization Effects 0.000 description 2
- 238000002203 pretreatment Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 230000004044 response Effects 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- GUBGYTABKSRVRQ-XLOQQCSPSA-N Alpha-Lactose Chemical compound O[C@@H]1[C@@H](O)[C@@H](O)[C@@H](CO)O[C@H]1O[C@@H]1[C@@H](CO)O[C@H](O)[C@H](O)[C@H]1O GUBGYTABKSRVRQ-XLOQQCSPSA-N 0.000 description 1
- GYHNNYVSQQEPJS-UHFFFAOYSA-N Gallium Chemical compound [Ga] GYHNNYVSQQEPJS-UHFFFAOYSA-N 0.000 description 1
- GUBGYTABKSRVRQ-QKKXKWKRSA-N Lactose Natural products OC[C@H]1O[C@@H](O[C@H]2[C@H](O)[C@@H](O)C(O)O[C@@H]2CO)[C@H](O)[C@@H](O)[C@H]1O GUBGYTABKSRVRQ-QKKXKWKRSA-N 0.000 description 1
- 229920000877 Melamine resin Polymers 0.000 description 1
- 229910052782 aluminium Inorganic materials 0.000 description 1
- XAGFODPZIPBFFR-UHFFFAOYSA-N aluminium Chemical compound [Al] XAGFODPZIPBFFR-UHFFFAOYSA-N 0.000 description 1
- 230000001174 ascending effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 150000001720 carbohydrates Chemical class 0.000 description 1
- 235000014633 carbohydrates Nutrition 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 235000013365 dairy product Nutrition 0.000 description 1
- 238000005034 decoration Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000004744 fabric Substances 0.000 description 1
- 235000019197 fats Nutrition 0.000 description 1
- 230000004907 flux Effects 0.000 description 1
- 239000006260 foam Substances 0.000 description 1
- 239000011888 foil Substances 0.000 description 1
- 229910052733 gallium Inorganic materials 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000002427 irreversible effect Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 239000008101 lactose Substances 0.000 description 1
- 239000010985 leather Substances 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- JDSHMPZPIAZGSV-UHFFFAOYSA-N melamine Chemical compound NC1=NC(N)=NC(N)=N1 JDSHMPZPIAZGSV-UHFFFAOYSA-N 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 235000008935 nutritious Nutrition 0.000 description 1
- 230000035699 permeability Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000005180 public health Effects 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 238000000411 transmission spectrum Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Chemical compound O XLYOFNOQVPJJNP-UHFFFAOYSA-N 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
- G01N21/3581—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation
- G01N21/3586—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using far infrared light; using Terahertz radiation by Terahertz time domain spectroscopy [THz-TDS]
Landscapes
- Physics & Mathematics (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Health & Medical Sciences (AREA)
- Toxicology (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
The present invention relates to a kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra, including：Gathered by tera-hertz spectra signal acquiring system known to nutritional ingredient actual value and qualified original packing milk powder tera-hertz spectra.High density wavelet transformation is carried out, optimal high density wavelet basis is chosen.The otherness of original spectrum signal different sample room signals under different decomposition yardstick is contrasted, optimal decomposition scale is established in the way of crossing prediction error is minimum.Multiple dimensioned high density wavelet decomposition is carried out to terahertz light spectrum signal, each floor height density wavelet coefficient is obtained.To acquired each floor height density wavelet coefficient, prediction submodel is set up.For the high density wavelet coefficient of different levels, corresponding prediction submodel is set up respectively, constitutes prediction submodel group.By suitable convergence strategy, all prediction submodels are merged.
Description
Technical field
A kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra
Background technology
Milk powder has the advantages that nutritious, composition proportion rationally, is easily absorbed by the body, and is loved by people, for a long time with
Come one of the Major Nutrient source of always infant and middle-aged and old Major Nutrient supplies.However, in recent years in great number profit
Under the driving of profit, emerged in an endless stream by milk powder safety problems such as dominating, " melamine " " leather milk " of nutrition fraud,
As one of principal contradiction under China's politics, economy, the multiple field of public health.However, traditional milk powder detection method is big
The instrument and equipment that high level easily consumes is all relied on, has that testing cost is high, sample pre-treatments are cumbersome, need to tear packaging detection etc. open and ask
Topic, wastes time and energy and costly, seriously hinders popularization of the traditional instrument detection method in vast basic unit's Site Detection, more can not
Meet the severe situation of the safe examination of China's present milk powder.In addition, current detection method is required for tearing milk powder packaging open
Envelope processing, can cause irreversible damage to commodity packaging, and cause food security waste and spending unnecessary in enforcing the law, difficult
To adapt to the milk powder safety situation of China's current rigorous.Therefore, in the urgent need to developing the new and effective lossless inspection of milk powder original packing
Survey technology, with the nutrition and quality information of quick obtaining milk powder.
Spectrum detection technique is currently a kind of good Fast Detection Technique, but traditional spectrum modeling method is typically used
Multiple linear regression correction method, such as PLS, ridge regression, but because spectral signal has serious baseline drift mostly
Move and the interference such as background, matrix is, it is necessary to carry out denoising to it before spectrum carries out multiple regression correction, smooth etc. pre-process.But
In preprocessing process, different pretreatments algorithm overcomes the weighting point of spectra1 interfer- to differ greatly, and easily causes losing for effective information
Lose so that the forecast result of model and robustness built out are poor.In the terahertz light time spectrum in face of complex system, this kind of algorithm
Defect is especially pronounced, and its reason is that the spectral peak of tera-hertz spectra is wider, and simple Pretreated spectra is difficult to accurately extract to be measured
The tera-hertz spectra spectrum information of material.Therefore, the efficient tera-hertz spectra analytic technique of Development of Novel is extremely urgent.
The content of the invention
Based on the defect of transmission spectra modeling method, to meet the demand of powder quality high flux examination, the present invention is provided
A kind of original packing milk powder lossless detection method.Technical scheme is as follows：
A kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra, comprises the following steps：
S1, by tera-hertz spectra signal acquiring system gather nutritional ingredient actual value known to and qualified original packing milk powder
Tera-hertz spectra, as the original spectrum of calibration set.
S2, progress high density wavelet transformation, according to the feature of original packing milk powder terahertz light spectrum signal, choose optimal highly dense
Spend wavelet basis.
S3, using optimal high density wavelet basis selected in step S2, contrast original spectrum signal is in different decomposition chi
The otherness of different sample room signals under degree, optimal decomposition scale is established in the way of crossing prediction error is minimum.
S4, using the optimal high density wavelet decomposition parameter established in step S2, S3, including optimal high density wavelet basis and
Optimal Decomposition yardstick, multiple dimensioned high density wavelet decomposition is carried out to terahertz light spectrum signal, obtains each floor height density wavelet coefficient.
S5, to each floor height density wavelet coefficient acquired in step S4, PLS algorithm is respectively adopted and builds
Vertical prediction submodel, process is as described in step S6, S7.
A certain layer in S6, each floor height density wavelet coefficient obtained using in step S4 is independent variable, correspondence nutritional ingredient
Content be dependent variable, carry out the offset minimum binary Monte Carlo Cross-Validation of 1000-5000 time, tested according to Monte Carlo intersection
The optimal number of principal components that root-mean-square error minimum principle determines prediction submodel is demonstrate,proved, and it is square to record corresponding minimum cross validation
Root error.
S7, according to the optimal number of principal components established in step S6, correspondence is set up using nonlinear iterative partial least square method
Prediction submodel.
S8, the high density wavelet coefficient for different levels, corresponding prediction is set up by repeat step S6 and S7 respectively
Submodel, constitutes prediction submodel group.
S9, by suitable convergence strategy, all prediction submodels are merged：Gained in step S5 can be used
Each submodel Monte Carlo Cross-Validation root-mean-square error is fusion foundation, with Monte Carlo Cross-Validation root-mean-square error square
Inverse be submodel weight, and final Multiscale Fusion model is obtained by weights amalgamation mode.
S10, further collection checking sample, verify the prediction effect of set up Multiscale Fusion model, such as model
Predicated error can reach detection demand, then retain the model, otherwise recalculate and correct the model, until the model reaches inspection
Survey demand.
S11, for Different Nutrition composition, repeat step S1-S10, complete the instruction of Different Nutrition ingredient prediction model
Practice.
The present invention is due to taking above technical scheme, and it has advantages below：
The Non-Destructive Testing of original packing milk powder is realized using tera-hertz spectra first, without sample pretreatment during detection, kept away
The destruction to nonmetallic milk powder outer packing during traditional detection is exempted from, and internal powder quality will not have been impacted；Secondly
Use high density wavelet transformation be obviously improved in the way of over-sampling tera-hertz spectra when/the multiple dimensioned resolution capability of frequency, have
Beneficial to the accurate Terahertz spectrum information for extracting test substance in milk powder system；Analysing terminal is real by the way of multi-scale Modeling
Existing operating in a key, possesses good ease for use；The present invention also has that detection speed is fast, system response time (is only needed tens of soon
Second can complete one-time detection), easy to operate, precision of prediction high achievable Site Detection the advantages of.In addition, this method compares
In conventional milk powder detection method, nutritional ingredient detection and the Quality Identification of milk powder can be achieved without destroying milk powder packaging, greatly
Improve the detection efficiency and performance of milk powder, the extensive clearance detection of the customs that is particularly suitable for use in, the detection of food security on-site law-enforcing
And the field such as terminal client Site Detection, the efficiency and economy of milk powder safety examination are lifted, to improve dairy products security control
Efficiency provides strong technical support, is had broad application prospects in milk powder field of non destructive testing.
Brief description of the drawings
Fig. 1 is the key structure schematic diagram of spectroscopic acquisition system.
Label declaration in figure：1 femto-second laser；2 be half-wave plate；3 be beam splitter；4 deferred mounts；5 choppers；6 lens；
7 high pass photoconductive antenna gallium arsenides；8 off axis paraboloidal mirrors；9 be silicon lens；10 be zinc telluridse crystal；11 be lens；12
For polarizer；13 be quarter-wave plate；14 be Wollaston prism；15 be balance diode；16 be lock-in amplifier；17 are
Computer；18 be speculum；19 be the casing filled with nitrogen；20 be the original packing powdered milk sample of nonmetallic packaging.
Fig. 2 is the algorithm flow block diagram of the present invention.
Fig. 3 is milk powder terahertz absorption spectra.
Fig. 4 is high density wavelet conversion coefficient sequence.
Fig. 5 is Monte Carlo Cross-Validation schematic diagram.
Embodiment
Tera-hertz spectra is the new spectral measurement methodses based on femtosecond laser technology, with penetrability is strong, energy is relatively low,
It is safe, the advantages of, the present invention is relatively low and most of nonmetallic materials and apolar substance are worn using THz wave energy
The extremely strong characteristic of permeability, for the milk powder product of nonmetallic packaging, penetrates its outer packing using THz wave and directly obtains inside
The terahertz light spectrum information of milk powder, is handled by carrying out the modeling of high density multi-scale wavelet, obtain critical nutrients in sample into
Divide information.Detection process neither destroys milk powder original packing and powder quality is not caused to damage again.
The present invention is made up of signal acquisition terminal and signal analysis terminal two parts.Each several part is described in detail as follows：
1st, signal acquisition terminal：
Signal acquisition terminal is main by terahertz time-domain spectroscopy signal acquiring system (as shown in Figure 1), and control module
Constitute.
(1) terahertz time-domain spectroscopy signal acquiring system is main by Terahertz light source, sample cell, the portion such as Terahertz detector
Divide and constitute, its main task is collected to produce the sample in terahertz pulse ripple and transmission sample pond by terahertz detector
Terahertz signal.
(2) control module is made up of high performance computation chip and control circuit, main to be responsible for coordination, control, monitoring terahertz
The hereby gatherer process of time-domain spectroscopy signal, it is ensured that the correctness of spectral signal, and be responsible for sending the spectral signal of collection to letter
Number analysing terminal is handled.
(3) signal acquisition terminal works process is summarized as follows：
Femto-second laser 1 produces laser pulse in S1, terahertz time-domain spectroscopy analysis system, passes through half-wave plate 2, beam splitter
3, laser is divided into pump light and detection light.
S2, pump light pass through deferred mount 4, chopper 5, lens 6, high pass photoconductive antenna gallium arsenide 7, by off-axis
Collimation is mapped to the original packing powdered milk sample 20 of nonmetallic packaging after paraboloidal mirror 8 and silicon lens 9 are focused on, through being passed through again after sample 20
Focused on again on zinc telluridse crystal 10 after silicon lens 9 and off axis paraboloidal mirror 8 are conllinear with detection light.
S3, detection light are focused on zinc telluridse crystal 10 again through lens 11 after polarizer 12 is conllinear with pump light.
After S4, terahertz pulse are by zinc telluridse crystal 10, pass through quarter-wave plate 13, Wollaston prism 14, warp
Balance diode 15 is detected, and input computer 17 is handled after signal feeding lock-in amplifier 16 is amplified.
Change optical path direction using speculum 18 in S5, the system.
S6, lens 6, high pass photoconductive antenna gallium arsenide 7, vertical shaft paraboloidal mirror 8, silicon lens 9, nonmetallic packaging
Original packing powdered milk sample 20, zinc telluridse crystal 10, lens 11, polarizer 12, quarter-wave plate 13, the He of Wollaston prism 14
Balance diode 15 is sealed in the casing 19 filled with nitrogen.
2nd, signal analysis and processing terminal：
The terminal is typically made up of the high-performance workstation with stronger operational capability, and main responsible collect is adopted from signal
Collect terminal terahertz light spectrum signal and use high density wavelet transformation be obviously improved in the way of over-sampling at that time/frequently it is multiple dimensioned
Resolution capability, and avoid milk powder quantitative information from losing by data prediction and the integrated computing of Multivariate Correction, and then complete
The multi-scale Modeling process of tera-hertz spectra, the nutritional ingredient and quality information of milk powder are obtained with this.
, it is necessary to carry out multi-resolution decomposition to spectral signal, its isolation is by scaling algorithm in multi-scale Modeling algorithm
Determine, therefore, scaling algorithm is the core of multi-scale Modeling algorithm, and its performance will directly affect the effect finally modeled.However,
Traditional wavelet transform, can only realize roughly signal when/frequency multi-resolution decomposition, and have that resolution ratio is poor, time frequency analysis
Not enough fine, signal easily many defects such as distortion, are not suitable for the parsing of tera-hertz spectra.High density wavelet transformation has resolution ratio
The advantages of height, compact schemes, over-sampling, can effectively be lifted in the way of over-sampling original spectrum when/the multiple dimensioned resolution ratio of frequency,
And then dimensional informations more more than wavelet transform are provided, the material for being obviously improved complex system tera-hertz spectra differentiates energy
Power.Therefore, high density wavelet transformation can effectively overcome the current defect that terahertz light spectral resolution is relatively low, frequency band is narrower.
Characteristic differences of the invention according to different material THz wave bands of a spectrum, with the THz wave time domain absorption spectra of sample
For spectral signal, multiple dimensioned parsing is carried out to terahertz time-domain absorption spectrum using high density multi-scale wavelet modeling algorithm, had
The resolution ratio of effect lifting original spectrum, response of the prominent features signal on frequency domain.For each yardstick high density wavelet systems number sequence
Row, are respectively adopted partial least squares algorithm and set up prediction submodel, and submodel is merged using weight fusion strategy, enter
And final high density multi-scale wavelet Fusion Model is obtained, and containing with this Different Nutrition composition for measuring milk powder in original packing
Measure information.On this basis, the accurate judgement of powder quality is realized.
Multi-scale Modeling algorithm steps are as follows：
S1, pass through tera-hertz spectra signal acquiring system and gather representational original packing milk powder spectrum, its nutritional ingredient
Actual value can be obtained by traditional detection method, and as the original spectrum of calibration set.
S2, the feature according to original packing milk powder terahertz light spectrum signal, choose optimal high density wavelet basis, in practice may be used
Different optimal high density wavelet basis are determined according to the absorption peak character of Different Nutrition component substances in milk powder, such as 2vm, 4vm,
Bi4 etc..
S3, using optimal high density wavelet basis selected in step S2, contrast original spectrum signal is in different decomposition chi
The otherness of different sample room signals under degree, optimal decomposition scale is established in the way of crossing prediction error is minimum.
S4, using the optimal high density wavelet decomposition parameter established in step S2, S3 to terahertz light spectrum signal carry out it is many
Yardstick high density wavelet decomposition, obtains each floor height density wavelet coefficient.
S5, to each floor height density wavelet coefficient acquired in step S4, PLS algorithm is respectively adopted and builds
Vertical prediction submodel, its detailed process is as described in step S6, S7.
A certain layer in S6, each floor height density wavelet coefficient obtained using in step S4 is independent variable, correspondence nutritional ingredient
Content be dependent variable, carry out the offset minimum binary Monte Carlo Cross-Validation of 1000-5000 times, and intersect according to Monte Carlo
Verify that root-mean-square error minimum principle determines the optimal number of principal components of prediction submodel, and it is equal to record corresponding minimum cross validation
Square error.
Meng Tekate cross validation root-mean-square errors described in step S6 are specific as follows：
In formula：MCRMSECV is Monte Carlo Cross-Validation root-mean-square error, and t is Monte-Carlo step number of times, and n is each
The sample number of sampling prediction, PRESSiFor the Prediction sum squares of ith Monte-Carlo step, CirFor ith Monte Carlo
The sample actual value of sampling, CipFor the sample predicted value of ith Monte-Carlo step.
S7, according to the optimal number of principal components established in step S6, correspondence is set up using nonlinear iterative partial least square method
Prediction submodel.
S8, the wavelet coefficient for different levels, corresponding prediction submodel is set up by repeat step S6 and S7 respectively,
Constitute prediction submodel group.
S9, by suitable convergence strategy, all prediction submodels are merged.In practice, step S5 can be used
Each submodel Monte Carlo Cross-Validation root-mean-square error of middle gained is fusion foundation.For example, with Monte Carlo Cross-Validation
The inverse of root-mean-square error square is the weight of submodel, and obtains final Multiscale Fusion model by weights amalgamation mode.
The specific formula of convergence strategy based on Meng Tekate cross validation root-mean-square errors is as follows：
In formula：MfFor Fusion Model, MsiFor i-th of submodel, m is submodel sum, WiIt is the power of i-th of submodel
Weight, MCRMSECViFor the Meng Tekate cross validation root-mean-square errors of i-th of submodel.
S10, further collect checking sample, the prediction effect of Multiscale Fusion model is set up in checking, such as model it is pre-
Detection demand can be reached by surveying error, then retains the model, otherwise recalculate the model, until the model reaches detection demand.
S11, for Different Nutrition composition, repeat step S1-S10, complete the instruction of Different Nutrition ingredient prediction model
Practice.
Illustrated with reference to embodiment.
The optimum working mode for the spectroscopic acquisition system that the present invention is used is as described below：
Femto-second laser 1 sends the femto-second laser pulse that centre wavelength is 800nm, and pulse width is 100fs, repetition rate
For 80MHz, power output is 720mW.Into after terahertz time-domain spectroscopy analysis system, light beam is after half-wave plate 2 and beam splitter 3
It is divided into stronger pump light and weaker detection light beam.Pump light is cut by delayer 4 through frequency for 1.1kHz chopper 5
Ripple, is incided on high pass photoconductive antenna GaAs GaAs7 crystal after being focused on through lens 6, and frequency is produced by optical rectification effect
The terahertz pulse that scope is about 0.2~3THz.Terahertz pulse is incided after being focused on through off axis paraboloidal mirror 8 and silicon lens 9
On sample 20 with nonmetallic packaging.Light is detected by lens 11, is total to after polarizer 12 with the pump light that is projected through sample
Focus on and incide on detecting element zinc telluridse ZnTe10 crystal again after line.At this moment the electric field of terahertz pulse passes through electrooptic effect
The index ellipsoid of electro-optic crystal zinc telluridse crystal 10 is modulated, the polarization state of direct impulse is changed.Pulse passes through four points
One of wave plate 13, after Wollaston rib 14, the change of the polarization state through balancing the detection light of diode 15, you can obtain being loaded with sample
The size and variable signal of the terahertz pulse electric field of information, signal feeding lock-in amplifier 16 is amplified, and by changing
The method for becoming delay length detects the whole time domain waveform of terahertz signal, is finally handled signal feeding computer 17.
To prevent influence of the water vapor in air to terahertz signal, from gallium arsenide 7, the testing sample 20 for producing terahertz signal
This section of light path to crystal detection zinc telluridse ZnTe10 is sealed in the casing 19 filled with nitrogen.It is relatively wet in casing 19
Degree is less than 2%, and temperature is 294K21 DEG C.In detection process, system signal noise ratio is 1000, and spectral resolution is better than 40GHz.
Because THz wave can be completely through for the nonmetallic packaging such as plastics, carton, paper bag, cloth bag and foam,
And transmitance is far above the electromagnetic wave of other wavelength；But for mental package and aluminum foil composite, Terahertz can not pass through, therefore
And inventive samples are the original packing milk powder (20) of the nonmetallic packaging such as carton, paper bag, plastics, mental package milk powder is not herein
Row.
Prediction process is as follows：
S1, the milk powder standard sample 300 for collecting existing nutritional ingredient actual value, the model that its every nutritional ingredient is covered
Enclose as wide as possible, be distributed as uniform as possible.By taking protein as an example, the protein content coverage in sample is 10g/
100g-21g/100g, sample concentration is at intervals of 0.1g/100g.
S2,300 standard samples to being collected into obtain its former using terahertz signal collection system progress spectra collection
Beginning absorption spectrum, as the original spectrum of calibration set sample, as shown in Figure 3.Accompanying drawing 3 absorbs for the Terahertz of milk powder in packaging
Spectrum example.
S3, the characteristic according to protein absorption peak in milk powder, choose optimal high density wavelet basis：“4vm”.
S4, by contrasting under different decomposition yardstick the otherness between different sample signals, protein content can be established most
Excellent decomposition scale is 4 layers.And obtain each layer wavelet coefficient of the calibration set sample under the decomposition scale.In this example, signal will be divided
Solve as 9 floor height density wavelet coefficient sequences, as shown in Figure 4.Accompanying drawing 4 is the small echo after each level wavelet coefficient is arranged by ascending order
Coefficient schematic diagram.
S5, Monte Carlo Cross-Validation is respectively adopted to each layer wavelet coefficient determines optimal number of principal components, frequency in sampling is
1000 times, sampling proportion is 70%.Wherein the Monte Carlo Cross-Validation result of first layer wavelet coefficient is as shown in figure 5, by covering
Special Caro cross validation root-mean-square error minimum principle, it may be determined that number of principal components is 10, and record minimum Monte Carlo and intersect and test
Demonstrate,prove error.
S6, the optimal number of principal components according to obtained by step S5, prediction is set up using nonlinear iterative partial least square algorithm
Model.
S7, the high density wavelet coefficient for different levels, repeat step S5, S6 set up all prediction submodels.
S8, the Monte Carlo Cross-Validation root-mean-square error according to each submodel, calculate the weight of each submodel, and carry out
Weight fusion, obtains final Multiscale Fusion model.
S9, complete model set up after, newly collect 100 samples as checking collection be predicted, and with actual value carry out
Contrast verification.As a result show, for protein, detection error is no more than 5%, detection demand is reached enough.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, for example, by the technology application
In the detection of other nutritional ingredients (such as fat, carbohydrate, linoleic acid, leukotrienes, lactose composition), these improve and moistened
Decorations also should be regarded as protection scope of the present invention.
Claims (1)
1. a kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra, comprises the following steps：
S1, gathered known to nutritional ingredient actual value by tera-hertz spectra signal acquiring system and qualified original packing milk powder too
Hertz spectrum, as the original spectrum of calibration set.
S2, progress high density wavelet transformation, according to the feature of original packing milk powder terahertz light spectrum signal, choose optimal high density small
Ripple base；
S3, using optimal high density wavelet basis selected in step S2, contrast original spectrum signal is under different decomposition yardstick
The otherness of different sample room signals, optimal decomposition scale is established in the way of crossing prediction error is minimum；
S4, using the optimal high density wavelet decomposition parameter established in step S2, S3, including optimal high density wavelet basis and optimal
Decomposition scale, multiple dimensioned high density wavelet decomposition is carried out to terahertz light spectrum signal, obtains each floor height density wavelet coefficient；
S5, to each floor height density wavelet coefficient acquired in step S4, PLS algorithm is respectively adopted and sets up pre-
Submodel is surveyed, process is as described in step S6, S7；
A certain layer in S6, each floor height density wavelet coefficient obtained using in step S4 corresponds to containing for nutritional ingredient as independent variable
Measure as dependent variable, the offset minimum binary Monte Carlo Cross-Validation that progress is 1000-5000 times is equal according to Monte Carlo Cross-Validation
Square error minimum principle determines the optimal number of principal components of prediction submodel, and records corresponding minimum cross validation root mean square mistake
Difference；
S7, according to the optimal number of principal components established in step S6, set up corresponding pre- using nonlinear iterative partial least square method
Survey submodel；
S8, the high density wavelet coefficient for different levels, corresponding prediction submodule is set up by repeat step S6 and S7 respectively
Type, constitutes prediction submodel group；
S9, by suitable convergence strategy, all prediction submodels are merged：Each son of gained in step S5 can be used
Model Monte Carlo Cross-Validation root-mean-square error is fusion foundation, with falling for Monte Carlo Cross-Validation root-mean-square error square
Number is the weight of submodel, and obtains final Multiscale Fusion model by weights amalgamation mode；
S10, further collection checking sample, verify the prediction effect of set up Multiscale Fusion model, the prediction of such as model
Error can reach detection demand, then retain the model, otherwise recalculate and correct the model, until the model reaches that detection is needed
Ask；
S11, for Different Nutrition composition, repeat step S1-S10, complete the training of Different Nutrition ingredient prediction model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710299188.8A CN107192689B (en) | 2017-04-28 | 2017-04-28 | Original packaged milk powder nondestructive testing method based on multi-scale terahertz spectrum |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710299188.8A CN107192689B (en) | 2017-04-28 | 2017-04-28 | Original packaged milk powder nondestructive testing method based on multi-scale terahertz spectrum |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107192689A true CN107192689A (en) | 2017-09-22 |
| CN107192689B CN107192689B (en) | 2020-06-26 |
Family
ID=59872625
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710299188.8A Active CN107192689B (en) | 2017-04-28 | 2017-04-28 | Original packaged milk powder nondestructive testing method based on multi-scale terahertz spectrum |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107192689B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019223372A1 (en) * | 2018-05-22 | 2019-11-28 | 深圳市太赫兹科技创新研究院 | Method for detecting melamine |
| CN111721754A (en) * | 2020-06-24 | 2020-09-29 | 天津大学 | Method for detecting mineral elements in liquid milk based on laser-induced breakdown spectroscopy |
| CN112862902A (en) * | 2021-02-24 | 2021-05-28 | 中国资源卫星应用中心 | Relative radiation correction method of space linear array camera |
| CN117710379A (en) * | 2024-02-06 | 2024-03-15 | 杭州灵西机器人智能科技有限公司 | Nondestructive testing model construction method, nondestructive testing device and medium |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106124435A (en) * | 2016-07-04 | 2016-11-16 | 江苏大学 | Based on visible ray, near-infrared, the rice new-old quality inspection device of Terahertz fusion spectral technique and detection method |
-
2017
- 2017-04-28 CN CN201710299188.8A patent/CN107192689B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106124435A (en) * | 2016-07-04 | 2016-11-16 | 江苏大学 | Based on visible ray, near-infrared, the rice new-old quality inspection device of Terahertz fusion spectral technique and detection method |
Non-Patent Citations (5)
| Title |
|---|
| CHOONWOO RYU等: "Atmospheric degradation correction of terahertz beams using multiscale signal restoration", 《APPLIED OPTICS》 * |
| 刘云曼等: "近红外光谱测量中的多尺度建模新方法", 《纳米技术与精密工程》 * |
| 李军等: "基于离散小波变换的高光谱特征提取中分解尺度的确定方法", 《自然科学进展》 * |
| 滕学明等: "太赫兹技术对营养品中蛋白质含量的研究", 《现代科学仪器》 * |
| 陈达等: "多尺度建模在近红外光谱模型传递中的应用", 《纳米技术与精密工程》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019223372A1 (en) * | 2018-05-22 | 2019-11-28 | 深圳市太赫兹科技创新研究院 | Method for detecting melamine |
| CN111721754A (en) * | 2020-06-24 | 2020-09-29 | 天津大学 | Method for detecting mineral elements in liquid milk based on laser-induced breakdown spectroscopy |
| CN112862902A (en) * | 2021-02-24 | 2021-05-28 | 中国资源卫星应用中心 | Relative radiation correction method of space linear array camera |
| CN112862902B (en) * | 2021-02-24 | 2024-05-07 | 中国资源卫星应用中心 | Relative radiation correction method of space linear array camera |
| CN117710379A (en) * | 2024-02-06 | 2024-03-15 | 杭州灵西机器人智能科技有限公司 | Nondestructive testing model construction method, nondestructive testing device and medium |
| CN117710379B (en) * | 2024-02-06 | 2024-05-10 | 杭州灵西机器人智能科技有限公司 | Nondestructive testing model construction method, nondestructive testing device and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107192689B (en) | 2020-06-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107192689A (en) | A kind of original packing milk powder lossless detection method based on multiple dimensioned tera-hertz spectra | |
| CN101975764B (en) | Near infrared spectrum technology-based multiband soil nitrogen detection device and method | |
| CN103134767B (en) | Method for liquor quality identification through infrared spectrum revision | |
| Hua et al. | Quantitative determination of cyfluthrin in n-hexane by terahertz time-domain spectroscopy with chemometrics methods | |
| CN108982405B (en) | Oil water content measuring method and instrument based on deep learning | |
| CN105388120B (en) | Calibration Wavelength modulation spectroscopy gas detection method is exempted from based on WMRF models | |
| CN101769866B (en) | Device for detecting milk components and method thereof | |
| CN105548075A (en) | Device and method for detecting oxygen content in glass medicine bottle | |
| CN108846203A (en) | The method and device of fruit non-destructive testing | |
| CN101424636A (en) | A kind of device and method of rapidly and nondestructively detecting content of green tea composition | |
| CN102175638A (en) | Device for rapidly and nondestructively detecting component content of yellow rice wine | |
| CN101975759A (en) | Transmission-type nondestructive measuring device and method of water content of plant leaves | |
| CN106596499A (en) | Real-time Raman spectrum calibration method | |
| CN105784672A (en) | Drug detector standardization method based on dual-tree complex wavelet algorithm | |
| CN105044024A (en) | Method for nondestructive testing of grape berries based on near infrared spectrum technology | |
| Huang et al. | Assessment of tomato maturity in different layers by spatially resolved spectroscopy | |
| CN101968443A (en) | Nondestructive detection device and method of water content of reflective near infrared plant leaf | |
| CN106383088A (en) | A seed purity rapid nondestructive testing method based on a multispectral imaging technique | |
| CN211347925U (en) | Gas concentration measuring device | |
| CN105527236A (en) | Method for determination of main nutritional components of agricultural product by use of spectroscopy method | |
| CN205844176U (en) | A kind of near infrared spectrum analysis device | |
| Liu et al. | Real-time measurement of moisture content of paddy rice based on microstrip microwave sensor assisted by machine learning strategies | |
| Lan et al. | Multi-harmonic measurements of line shape under low absorption conditions | |
| CN103149180B (en) | Detection method of soil spectral reflectivity and specific conductance | |
| Gao et al. | Research on the seed respiration CO2 detection system based on TDLAS technology |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20181206 Address after: 300000 Jinzhong Highway 3699, Dongli District, Tianjin City, 14-314, G District, Hardware City, North China Applicant after: TIANJIN ZHIQIAO DATA TECHNOLOGY CO., LTD. Address before: Room 139, No. 11, Zhongshan Branch Road, Jiaojiang District, Taizhou City, Zhejiang Province Applicant before: Zhejiang Da Da Technology Co., Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |