CN106680238B - Method based on infrared spectrum analysis material component content - Google Patents
Method based on infrared spectrum analysis material component content Download PDFInfo
- Publication number
- CN106680238B CN106680238B CN201710009518.5A CN201710009518A CN106680238B CN 106680238 B CN106680238 B CN 106680238B CN 201710009518 A CN201710009518 A CN 201710009518A CN 106680238 B CN106680238 B CN 106680238B
- Authority
- CN
- China
- Prior art keywords
- data
- aiming field
- source domain
- component content
- model
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000004458 analytical method Methods 0.000 title claims abstract description 48
- 239000000463 material Substances 0.000 title claims abstract description 48
- 238000002329 infrared spectrum Methods 0.000 title claims abstract description 41
- 230000003595 spectral effect Effects 0.000 claims abstract description 62
- 238000012546 transfer Methods 0.000 claims abstract description 34
- 206010027476 Metastases Diseases 0.000 claims abstract description 19
- 230000009401 metastasis Effects 0.000 claims abstract description 19
- 238000001228 spectrum Methods 0.000 claims description 52
- 238000012360 testing method Methods 0.000 claims description 28
- 238000004566 IR spectroscopy Methods 0.000 claims description 21
- 238000000605 extraction Methods 0.000 claims description 18
- 239000011159 matrix material Substances 0.000 claims description 16
- 238000012545 processing Methods 0.000 claims description 14
- 239000000126 substance Substances 0.000 claims description 13
- 230000008859 change Effects 0.000 claims description 11
- 238000010238 partial least squares regression Methods 0.000 claims description 11
- 230000017105 transposition Effects 0.000 claims 1
- 238000004422 calculation algorithm Methods 0.000 abstract description 102
- 230000008569 process Effects 0.000 abstract description 12
- 238000013526 transfer learning Methods 0.000 abstract description 4
- 230000009466 transformation Effects 0.000 abstract description 3
- 238000004364 calculation method Methods 0.000 abstract description 2
- 240000008042 Zea mays Species 0.000 description 37
- 235000005824 Zea mays ssp. parviglumis Nutrition 0.000 description 37
- 235000002017 Zea mays subsp mays Nutrition 0.000 description 37
- 235000005822 corn Nutrition 0.000 description 37
- 238000013508 migration Methods 0.000 description 33
- 230000005012 migration Effects 0.000 description 33
- 238000012549 training Methods 0.000 description 17
- 238000010586 diagram Methods 0.000 description 16
- 230000000694 effects Effects 0.000 description 14
- 239000000470 constituent Substances 0.000 description 13
- 238000002790 cross-validation Methods 0.000 description 11
- 230000006872 improvement Effects 0.000 description 11
- 229920002472 Starch Polymers 0.000 description 9
- 235000019698 starch Nutrition 0.000 description 9
- 239000008107 starch Substances 0.000 description 9
- 239000004615 ingredient Substances 0.000 description 6
- 239000004576 sand Substances 0.000 description 6
- 238000012512 characterization method Methods 0.000 description 5
- 238000002474 experimental method Methods 0.000 description 5
- 238000010987 Kennard-Stone algorithm Methods 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 102000004169 proteins and genes Human genes 0.000 description 4
- 108090000623 proteins and genes Proteins 0.000 description 4
- 239000004480 active ingredient Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000004590 computer program Methods 0.000 description 2
- 239000012141 concentrate Substances 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 230000007423 decrease Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 230000001373 regressive effect Effects 0.000 description 2
- 238000004611 spectroscopical analysis Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 1
- 238000009415 formwork Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005086 pumping Methods 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
- 210000004885 white matter Anatomy 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
- G01N21/3563—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light for analysing solids; Preparation of samples therefor
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
The present invention relates to a kind of methods based on infrared spectrum analysis material component content, including establishing the first regression model according to source domain ir data and source domain material component content corresponding with the source domain ir data, the parameter in first regression model is sought；Aiming field ir data is obtained, the metastasis model between aiming field ir data and source domain ir data is established, seeks the parameter in the metastasis model；According to the aiming field ir data, the metastasis model, aiming field material component content corresponding with the aiming field ir data is obtained using first regression model.Feature moving method and partial least squares algorithm in analysis method combination transfer learning of the invention, realize the transformation of aiming field to source domain Spectral feature scale, redundancy can be not only removed, improves the accuracy of transfer, and can largely reduce the calculation amount of transfer process.
Description
Technical field
The present invention relates to infrared spectrum analysis fields, are based on infrared spectrum analysis material composition in particular to one kind
The method of content.
Background technique
It would know that material component content by infrared spectrum analysis.By measurement infrared spectroscopy, analyze it, thus
Know material component content, not only can be with qualitative analysis, it can also quantitative analysis.But in existing infrared spectrometry process
In, the change of measuring instrument or measuring condition all will lead to original peg model failure, and re-establishing model will waste largely
Time and cost cause analysis result inaccuracy, the low situation of analysis efficiency.
Summary of the invention
The present invention is low in order to solve the problems, such as existing modeling efficiency again, proposes a kind of based on infrared spectrum analysis object
The method of matter component content, comprising the following steps:
S1, according to source domain ir data and source domain material component content corresponding with the source domain ir data
The first regression model is established, the parameter in first regression model is sought；
S2 obtains aiming field ir data, establish aiming field ir data and source domain ir data it
Between metastasis model, seek the parameter in the metastasis model；
S3, according to the aiming field ir data, the metastasis model, using first regression model obtain with
The corresponding aiming field material component content of the aiming field ir data.
Further, first regression model is Partial Least-Squares Regression Model, and the step S1 includes, to the source
Domain ir data carries out feature extraction and obtains the first spectral signature, according to first spectral signature and source domain material composition
Content establishes the Partial Least-Squares Regression Model, finds out regression coefficient.
Further, the aiming field ir data includes that aiming field infrared spectroscopy normal data and aiming field are infrared
Spectrum test data, the step S2 include carrying out feature extraction according to the aiming field infrared spectroscopy normal data to obtain second
Standard spectrum feature；The metastasis model is established according to first spectral signature and the second standard spectrum feature, is found out
Transfer matrix.
Further, the step S3 includes, according to the aiming field examination of infrared spectrum data acquisition third Spectral Properties
Sign, is brought into the minimum inclined two for the third spectral signature and the metastasis model and multiplies and obtain the target in regression model
Domain material component content.
Further, described that the step of feature extraction obtains the first spectral signature is carried out to the source domain ir data
Including centralization processing being carried out to the source domain ir data and source domain material component content, after handling according to centralization
Source domain ir data and source domain material component content establish least square regression model and obtain first spectral signature.
Further, also obtaining includes aiming field standard substance component content, described according to the aiming field infrared spectroscopy
The step of normal data progress feature extraction obtains the second standard spectrum feature includes: to the aiming field infrared spectroscopy criterion numeral
Centralization processing is carried out according to the aiming field standard substance component content, according to centralization treated aiming field infrared spectroscopy
Normal data and aiming field standard substance component content establish Partial Least-Squares Regression Model and obtain the second standard spectrum feature.
Further, while the step S2 obtains the second standard spectrum feature, the second standard projection number is also obtained
According to the second normal loading data；According to the aiming field examination of infrared spectrum data acquisition third Spectral Properties in the step S3
The step of sign, includes, using the mean value of the aiming field infrared spectroscopy normal data to the aiming field examination of infrared spectrum data
Centralization processing is carried out, successively recursion obtains the according to the following formula using centralization treated aiming field examination of infrared spectrum data
Three spectral signatures:Wherein, i is more than or equal to 1 and is less than or equal to k, TT_testFor
Third spectral signature, k are the number of third spectral signature,For i-th of component of the second standard projection data,Centered on change i-th of residual error item of treated aiming field examination of infrared spectrum data,For the second standard
I-th of component of load data.
Further, by solving the optimization problem of following formula,Wherein,
B indicates that the coefficient based on source domain feature regression model, M indicate transfer matrix of the target domain characterization to source domain feature, WSAnd WTRespectively
Indicate the projection matrix of source domain and aiming field；Pass through TS=XS*WSThe first spectral signature is solved, wherein the first spectral signature isI is more than or equal to 1 and is less than or equal to k, and k is the number of the first spectral signature；Pass throughCalculate regression coefficient ΒT=[b1,b2,...,bk], y indicates source domain material component content.
Further, the second standard spectrum feature, T are sought by following formulaT=XT*WT, wherein the second standard spectrum feature isI is more than or equal to 1 and is less than or equal to k, and k is the number of the second spectral signature.
Further, the second standard spectrum feature is utilizedWith the first spectral signaturePass through following formulaObtain transfer matrix Μ=[m1,m2,...,
mk], i is more than or equal to 1 and is less than or equal to k, and k is the number of the second standard spectrum feature, whereinFromMiddle selection.
Technical solution through the foregoing embodiment, the method for the invention based on infrared spectrum analysis material component content are built
Transfer relationship between vertical source domain and aiming field sample characteristics, on the one hand can remove redundancy, obtain more accurate simple
Transfer relationship, and then preferable prediction effect can be obtained, on the other hand can very great Cheng for higher-dimension Small Sample Database collection
Operand is reduced on degree.
Detailed description of the invention
The features and advantages of the present invention will be more clearly understood by referring to the accompanying drawings, and attached drawing is schematically without that should manage
Solution is carries out any restrictions to the present invention, in the accompanying drawings:
Fig. 1 is the flow diagram of method of the embodiment of the present invention based on infrared spectrum analysis material component content；
Fig. 2 is the flow diagram of method of the embodiment of the present invention based on infrared spectrum analysis material component content；
Fig. 3 is the principal and subordinate's spectrum and deviation spectrum for the corn data set verified to analysis method of the invention；
Fig. 4 is the principal and subordinate's spectrum and deviation spectrum for the tablet data set that analysis method of the invention is verified；
Fig. 5 is that the number of principal components of the PLS model for the corn that analysis method of the invention is verified chooses process；
Fig. 6 is the window size selection course of the PDS model for the corn that analysis method of the invention is verified；
Fig. 7 is moisture content true value and predicted value under each model in the corn verified of analysis method of the invention
Comparison schematic diagram；
Fig. 8 is oil content true value and predicted value under each model in the corn verified of analysis method of the invention
Comparison schematic diagram；
Fig. 9 be in the corn verified to analysis method of the invention protein content under each model true value with
The comparison schematic diagram of predicted value；
Figure 10 be in the corn verified to analysis method of the invention content of starch under each model true value with
The comparison schematic diagram of predicted value；
Figure 11 is prediction of the water content before and after calibration migration in the corn verified to analysis method of the invention
The comparison schematic diagram of value and true value；
Figure 12 is prediction of the oil content content before and after calibration migration in the corn verified to analysis method of the invention
The comparison schematic diagram of value and true value；
Figure 13 is that protein content is pre- before and after calibration migration in the corn verified to analysis method of the invention
The comparison schematic diagram of measured value and true value；
Figure 14 is prediction of the content of starch before and after calibration migration in the corn verified to analysis method of the invention
The comparison schematic diagram of value and true value；
Figure 15 is that the number of principal components of the PLS model for the corn verified to analysis method of the invention chooses process signal
Figure；
Figure 16 is that the window size selection course of the PDS model for the corn verified to analysis method of the invention is illustrated
Figure；
Figure 17 be in the tablet verified to analysis method of the invention the first active constituent under different models
The comparison schematic diagram of predicted value and true value；
Figure 18 be in the tablet verified to analysis method of the invention second active ingredient under different models
The comparison schematic diagram of predicted value and true value；
Figure 19 be in the tablet verified to analysis method of the invention the third active constituent under different models
The comparison schematic diagram of predicted value and true value；
Figure 20 be in the tablet verified to analysis method of the invention 1 content of active constituent calibration migration before and after
The comparison schematic diagram of predicted value and true value；
Figure 21 be in the tablet verified to analysis method of the invention 2 content of active constituent calibration migration before and after
The comparison schematic diagram of predicted value and true value；
Figure 22 be in the tablet verified to analysis method of the invention 3 content of active constituent calibration migration before and after
The comparison schematic diagram of predicted value and true value.
Specific embodiment
To better understand the objects, features and advantages of the present invention, with reference to the accompanying drawing and specific real
Applying mode, the present invention is further described in detail.It should be noted that in the absence of conflict, the implementation of the application
Feature in example and embodiment can be combined with each other.
In the following description, numerous specific details are set forth in order to facilitate a full understanding of the present invention, still, the present invention may be used also
To be implemented using other than the one described here other modes, therefore, protection scope of the present invention is not by described below
Specific embodiment limitation.
Embodiment one
As shown in Figure 1, the present invention provides a kind of method based on infrared spectrum analysis material component content, including it is following
Step:
S101 contains according to source domain ir data and source domain material composition corresponding with the source domain ir data
Amount establishes the first regression model, seeks the parameter in first regression model；First regression model is, for example, partially minimum
Two multiply regression model, carry out feature extraction to the source domain ir data and obtain the first spectral signature, according to described first
Spectral signature and source domain material component content establish the Partial Least-Squares Regression Model, find out regression coefficient；Specifically, described
Carrying out the step of feature extraction obtains the first spectral signature to the source domain ir data includes, to the source domain infrared light
Modal data and source domain material component content carry out centralization processing, according to centralization treated source domain ir data and source
Domain material component content establishes least square regression model and obtains first spectral signature.The operation of centralization processing is to use
Source domain ir data subtracts the mean value of source domain ir data, subtracts source domain material composition with source domain material component content
The mean value of content reduces influence of the deviation to model is established.
Specifically, by solving the optimization problem of following formula,Wherein, B
Indicate that the coefficient based on source domain feature regression model, M indicate transfer matrix of the target domain characterization to source domain feature, WSAnd WTRespectively
Indicate the projection matrix of source domain and aiming field.Pass through TS=XS*WSThe first spectral signature is solved, wherein the first spectral signature isI is more than or equal to 1 and is less than or equal to k, and k is the number of the first spectral signature；Pass throughCalculate regression coefficient ΒT=[b1,b2,...,bk], y indicates source domain material component content.
S102 obtains aiming field ir data, establishes aiming field ir data and source domain ir data
Between metastasis model, seek the parameter in the metastasis model；The aiming field ir data includes that aiming field is infrared
Spectrum normal data and aiming field examination of infrared spectrum data carry out feature according to the aiming field infrared spectroscopy normal data and mention
It takes and obtains the second standard spectrum feature；The transfer is established according to first spectral signature and the second standard spectrum feature
Model finds out transfer matrix, in order to improve accuracy, can from first spectral signature selected part spectral signature with it is described
Second standard spectrum feature establishes metastasis model, and when selection is corresponding according to the corresponding selection of material concentration, can such as take, source domain object
Matter component content data set identical with aiming field standard substance concentration carries out operation.
Specifically, the second standard spectrum feature, T are sought by following formulaT=XT*WT, wherein the second standard spectrum feature isI is more than or equal to 1 and is less than or equal to k, and k is the number of the second spectral signature.
Utilize the second standard spectrum featureWith the first spectral signature
Pass through following formulaObtain transfer matrix Μ=[m1,m2,...,mk], i is more than or equal to 1 and small
In being equal to k, k is the number of the second standard spectrum feature, whereinFromMiddle choosing
It takes.
S103 is obtained according to the aiming field ir data, the metastasis model using first regression model
Aiming field material component content corresponding with the aiming field ir data；According to the aiming field examination of infrared spectrum number
According to third spectral signature is obtained, the third spectral signature and the metastasis model are brought into the minimum inclined two and multiply recurrence mould
The aiming field material component content is obtained in type.
It is described according to the aiming field infrared light the embodiment of the invention also includes acquisition aiming field standard substance component content
Composing normal data and carrying out the step of feature extraction obtains the second standard spectrum feature includes: to the aiming field infrared spectroscopy standard
Data and the aiming field standard substance component content carry out centralization processing, according to centralization treated aiming field infrared light
Spectrum normal data and aiming field standard substance component content establish Partial Least-Squares Regression Model and obtain the second standard spectrum feature.
The step of centralization processing, is with above-mentioned to similar derived from the processing step of ir data.
While step S102 described in the embodiment of the present invention obtains the second standard spectrum feature, the second standard is also obtained
Data for projection and the second normal loading data；According to the aiming field examination of infrared spectrum data acquisition in the step S103
The step of three spectral signatures, includes, using the mean value of the aiming field infrared spectroscopy normal data to the aiming field infrared spectroscopy
Test data carries out centralization processing, and using centralization, treated that aiming field examination of infrared spectrum data are successively passed according to the following formula
It pushes away and obtains third spectral signature:Wherein, i is more than or equal to 1 and is less than or equal to
K, TT_testFor third spectral signature, k is the number of third spectral signature,It is i-th point of the second standard projection data
Amount,Centered on change i-th of residual error item of treated aiming field examination of infrared spectrum data,For the second mark
I-th of component of quasi- load data.
Method based on infrared spectrum analysis material component content of the invention establish source domain and aiming field sample characteristics it
Between transfer relationship, on the one hand can remove redundancy, obtain more accurate simple transfer relationship, thus can obtain compared with
Good prediction effect, on the other hand can largely reduce operand for higher-dimension Small Sample Database collection.In addition, only inclined
One parameter of latent variable of least-squares algorithm (PLS algorithm) needs to be arranged, and realizes process very simple.It should be noted that
" infrared spectroscopy " word is used in the present invention, it will be appreciated that at including near infrared spectrum, may also comprise middle infrared spectrum, remote red
External spectrum.
Embodiment two
Method combination transfer learning and PLS algorithm based on infrared spectrum analysis material component content of the invention is formed
A kind of migration calibration algorithm (CT_pls algorithm), basic thought source is in the transfer learning method based on feature, by target
Characteristic of field maps to source domain feature space, and then the model that can use source domain handles the data of aiming field.This method
Feature extraction is carried out to source domain sample and target sample first with PLS algorithm, then establishes the polynary mark based on source domain feature
Linear transition model between cover half type and source domain and target domain characterization, finally in an identical manner to unknown aiming field
After being shifted after sample progress feature extraction, the feature after transfer is predicted using source domain peg model.
Assuming that being respectively present source domain data set { XS, y } and aiming field data set { XT, y }, wherein XSAnd XTRespectively by key light
It spectrometer and is measured from spectrometer, establishes the calibration migration models between source domain and aiming field, actually solution formula (3.1)
Optimization problem.
In formula (3.1), B indicates that the coefficient based on source domain feature regression model, M indicate target domain characterization to source domain spy
The transfer matrix of sign, WSAnd WTRespectively indicate the projection matrix of source domain and aiming field.Select partial least squares algorithm as master herein
Body algorithm, WSAnd WTRespectively by establishing { XS, y } and { XS, y } PLS model acquire, the feature T of source domainSWith the feature of aiming field
TTIt is acquired by formula (3.2).
Obtaining source domain feature TSAfterwards, source domain characteristic { T is utilizedS,ySPolynary peg model is established, whereinCalculate regression coefficient ΒT=[b1,b2,...,bk], k indicates the main Characteristic Number extracted.
In order to realize that target numeric field data is effectively predicted in source domain model, need to carry out spectral space progress using standard set
Transformation, formula (3.4) (3.5) show that spectral signature transforms to the implementation method of source domain from aiming field.
Τ'S←ΤTΜ (3.4)
Wherein,Τ'SAnd TTIt is the feature of source domain and aiming field sample set, Τ ' respectivelySFrom
Middle acquisition TS, for calculating transfer matrix Μ=[m1,m2,...,mk]。
After metastasis model between the peg model and source domain and aiming field for establishing source domain, it can be realized to aiming field
Sample is effectively predicted, as shown in formula (3.6).
yT=TT*M*B (3.6)
Specifically, as shown in Fig. 2, the method for the invention based on infrared spectrum analysis material component content includes acquisition source
Domain training set, i.e. acquisition source domain ir data and source domain material component content；Aiming field standard set is obtained, i.e. acquisition target
Domain infrared spectroscopy normal data and aiming field standard substance component content；Aiming field test set is obtained, i.e. acquisition aiming field is infrared
Spectrum test data and aiming field test substances component content；To source domain data carry out centralization processing, to target numeric field data into
Row centralization processing；The first Spectra feature extraction is carried out using pls model to source domain data, forms assemblage characteristic data set, from
It is middle to extract feature (i.e. material component content is corresponding) corresponding with standard set, it is established using assemblage characteristic data set and pls algorithm
First regression model, aiming field standard set carry out feature extraction using pls and obtain the second standard spectrum feature, pass through pls model
The transfer matrix between the first spectral signature after selecting and the second standard spectrum feature is sought, aiming field test data is utilized
Pls model seeks third spectral signature, and third spectral signature and transfer matrix are brought into the first regression model, to obtain
Material component content corresponding with aiming field test data.Specific implementation process, including data prediction, feature extraction, build
Vertical source domain peg model calculates transfer relationship, predict to unknown object numeric field data.
Specifically, it can be realized by being loaded with the processor circuit of computer program, computer program process is as follows:
The method based on infrared spectrum analysis material component content of the embodiment of the present invention uses Partial Least Squares Regression
Analysis, partial least-squares regressive analysis (PLS) provide a kind of method of multipair Multilinear Regression modeling, especially work as two groups of variables
Very much, and all there is multiple correlation, and when the quantity (sample size) for observing data has less, use partial least-squares regressive analysis
The model of foundation has the advantages that the methods of traditional classical regression analysis is unexistent.When two groups of measurement samples of same article
When from different meter device or measuring state, two groups of samples are not identical related, it is possible to move the sample from new space
It moves to reference to space, and then directly new samples can be predicted using the model in reference space.Original mould is re-used
Type reduces modeling cost.
1. establishing the PLS regression model based on spectral signature
Partial Least-Squares Regression Model is established to ir data and its corresponding constituent concentration first, for obtaining light
The number of spectrum signature, spectral signature is chosen by cross validation method.Then dense to spectral signature and its corresponding ingredient
Degree re-establishes PLS model, and for the regression coefficient of computation model, main feature (spectral signature) number at this time still passes through friendship
Fork verification method is selected.It establishes PLS model twice to ir data and once directly establishes PLS model in prediction essence
It is not influenced substantially on degree, the spectral signature progress that the regression coefficient calculated using spectral signature can directly to aiming field after transfer
Prediction.
2. realizing the transfer learning between spectral signature
Different spectrometers measure the conditional probability of ir data or marginal probability distribution may be different, so that original
Polynary peg model can not the ir data to aiming field accurately predicted that it is inclined often to there is very big prediction
Difference, since modeling cost is very high again, it is therefore desirable to the spectral signature of aiming field be migrated to source domain, and then reduce source domain and mesh
It marks domain and is being distributed upper difference.Feature extraction is carried out to the standard spectrum sample in source domain and aiming field first, then establishes feature
To the PLS model of feature, transfer matrix is calculated.So that target domain characterization is multiplied with transfer matrix, the migration of feature can be realized.
3. a pair aiming field spectroscopic data is predicted
The feature of aiming field is migrated to the feature space of source domain, recurrence mould of the source domain based on feature can be directly utilized
Type predicts the feature of aiming field.Model is re-established so as to avoid to aiming field sample, is greatly reduced
Modeling cost.
Corn and tablet data are analyzed respectively for the analysis method in the present invention, specific as follows:
1. corn data set
Corn data set has 80 samples, corresponds to the content of moisture, oil, four kinds of protein, starch substances, Ke Yicong
(http://www.eigenvector.com/Data/Data_sets.html) is obtained.For ir data collection respectively by
Tri- kinds of different instruments of m5, mp5, mp6 are measured in 1100-2498nm of wave-length coverage by interval of 2nm, totally 700 channels.This reality
The middle spectrum for measuring m5 is tested as principal spectrum, spectroscopic data is as source domain data set XS, since the mp6 spectrum measured and m5 are surveyed
The difference obtained more greatly, is chosen as from spectrum, corresponding data set is as aiming field data set XT.Spectrogram is as shown in figure 3, wherein
Subgraph (A), (B), (C) respectively indicate key light spectrogram, from spectrogram and principal spectrum and from the SPECTRAL DIVERSITY figure between spectrum.
In experiment, data set is divided using Kennard-Stone (KS) algorithm, first from source domain and aiming field number
According to the data for extracting 20% respectively are concentrated, as test sample, respectively 16, wherein the test sample of aiming field is for testing
Demarcate migration models.Remaining 80% sample is as training sample, respectively 64, and wherein the training sample of source domain is for establishing
Reference model, can the migration sample to aiming field predict, aiming field for establishing the master pattern of aiming field, in order to
Compare the performance of other migration models.Several samples are extracted by KS algorithm respectively from the training sample of source domain and aiming field again
Transfer relationship as master sample collection, for establishing between source domain sample and aiming field sample.The quantity of master sample to turn
Shifting relationship affect is larger, and master sample quantity is very little, can not obtain sufficient sample information, and quantity is too many, is readily incorporated redundancy
Information, both of which can not obtain accurate transfer relationship.Both in order to balance, this experiment using KS algorithm from source domain with
50% sample is extracted in the training sample of aiming field respectively as master sample collection, respectively 32.
2. tablet data set
2002, issued on international diffusing reflection meeting (IDRC) " Shootout " data set include by two spectrometers
Respectively in the ir data of the wave-length coverage 600-1898nm tablet sample measured with the interval 2nm, respectively as source domain number
According to target numeric field data, include 650 variables, for analyzing the content of three kinds of active constituents in tablet.These samples difference
It is divided into source domain calibration sample set and aiming field calibration sample set, respectively includes 155 samples, source domain test set and aiming field are surveyed
Examination collection respectively includes 460 samples.The sample conduct of extraction 50% respectively is concentrated from the calibration of source domain and aiming field by KS algorithm
Standard set, respectively 78.The infrared spectrogram of tablet is presented in Fig. 4, and wherein Fig. 4 (A) indicates that principal spectrum, Fig. 4 (B) indicate
From spectrum, Fig. 4 (C) indicates principal spectrum and from the SPECTRAL DIVERSITY figure between spectrum.From in Fig. 4 (C) it can be seen that in wave numberWithRange, principal spectrum and from spectrum there is difference and the difference of front end there is
Biggish fluctuation, and in other wave-number ranges, existing difference is smaller.Illustrate to be easier to introduce noise at the both ends of spectrum.By
In principal spectrum and from the difference between spectrum and little, therefore it can guess, before and after model migration, the performance of prediction will not
There is too big transformation.
Detailed process is as follows:
1. data preprocessing method
Before training pattern, the method for selecting centralization pre-processes data, can to avoid due to numerical value difference compared with
Deviation caused by big.
2. parameter selection
The parameter selection of model can produce very big influence to model performance, selects an optimal parameter, can make
It obtains model and obtains optimal performance.For example, selecting optimal number of principal components for PLS algorithm, model can be made to obtain best
Prediction effect.In present invention experiment, SBC (slope and deviation correction method), (segmentation is direct by MSC (multiplicative scatter correction), PDS
Standard), CT_pls is all made of the polynary peg model that PLS algorithm establishes key light modal data, therefore is determining master sample quantity
Later, SBC and CT_pls algorithm only has one parameter needs of number of principal components and is set, and PDS algorithm also needs in addition to number of principal components
Window size is configured.In the present invention, the method for 10 folding cross validations is selected to carry out the principal component number of PLS algorithm
Selection is arranged number of principal components from 1 to 5, is divided into 1, calculates separately its corresponding cross validation error (RMSECV), chooses minimum
The corresponding number of principal components of RMSECV be best number of principal components.For PDS algorithm, since standard data set sample number is less,
When establishing PLS submodel to each window, using 5 folding cross validations, window size is set from 3 to 20, is divided into 2, window is big
The small odd number that should be not less than 3, calculates the corresponding RMSECV of each window size, and the pumping mouth of corresponding minimum RMSECV is best window
Mouthful.Model evaluation
In present invention experiment, using root-mean-square error (RMSE) as parameter selection and the index of model evaluation.The meter of RMSE
Calculation method such as formula (3.11).
Wherein,For predicted value,For reference value (true value or fiducial value),For test sample number.
RMSEC indicates that the training error of calibration collection, RMSEP indicate the prediction error of test set, and RMSECV is indicated to intersect and be tested
Demonstrate,prove error.For the cross validation error of PLS algorithm,Indicate true value.For PDS algorithm, the friendship of selected window size
Validation error is pitched,Indicate the predicted value of principal spectrum standard set.
For more intuitively CT_pls model more proposed by the present invention and other classical models and PLS benchmark model
The difference degree in estimated performance, use formula (3.12) calculate CT_pls algorithm with respect to other algorithm performances improvement rate
OrRate of descent.
In formula (3.12), RMSEPCT_plsIndicate the prediction error of CT_pls algorithm,Indicate that other are right
Than the error of the prediction of algorithm.
In addition, the present invention is examined between CT_pls method and other algorithms using rank sum test method with the presence or absence of significant
Sex differernce directly calculates the p value between predicted value using the wilcoxon function in python in scipy packet, if p > 0.05,
Illustrate that there is no significant differences between two kinds of algorithms, otherwise there are significant differences for explanation.
The present invention selects corn data set, tablet data set to test.SBC, PDS, CT_pls algorithm are all made of
PLS algorithm uses source domain data to establish polynary peg model as reference model, for the target to migration as main body algorithm
Domain forecast sample is predicted.Meanwhile using PLS algorithm, the polynary peg model of aiming field training sample is established, for comparing
The estimated performance of migration models is demarcated, more comprehensively, is accurately assessed convenient for being made to SBC, PDS, CT_pls calibration moving method.
Experimental result mainly includes following components:
(1) number of principal components of PLS algorithm chooses process and the result of RMSEC, RMSEP, RMSECV are shown.
(2) selection course of PDS algorithm window size.
(3) under different master sample numbers, the situation of change of the RMSEP of tri- kinds of migration algorithms of SBC, PDS, CT_pls.
(4) the fixed master sample number of setting, the comparison of tri- kinds of migration algorithm predictive abilities of SBC, PDS, CT_pls.
(5) calibration migration front and back, the comparison and parameter setting of model prediction ability.
It is tested using corn data set.Table 3.1 is illustrated directly to be established using the aiming field training set of corn and be corresponded to
Moisture content, oil, protein, the training error of the PLS model of content of starch, cross validation error, prediction error and principal component
Number.
The error and parameter of the aiming field data set PLS model of 3.1 corn of table
From RMSEC, RMSECV, the RMSEP that can be seen that every kind of ingredient in corn in table 3.1 without very big difference, say
It is bright not occur over-fitting, and RMSEP is smaller, illustrates also do not occur poor fitting phenomenon, and then can illustrate that number of principal components selects
What is taken is reasonable.The present invention chooses the principal component of PLS algorithm using 10 folding cross validation methods, Fig. 5 (A) (B) (C) (D)
Be set forth about moisture content in corn, oil content, protein, content of starch PLS model RMSECV with number of principal components change
Change process obtains the minimum value of RMSECV, therefore contain about component each in corn respectively when number of principal components is 5,5,5,5
The best number of principal components of the PLS model of amount is respectively 5,5,5,5.Although it is 5 that maximum number of principal components, which is arranged, corn data set is each
Component RMSECV is not restrained with the variation of number of principal components, can not obtain global minimum value, but if number of principal components selects
Take it is excessive will appear over-fitting, and will increase the complexity of PLS model, analyzed by many experiments, choose it is maximum it is main at
More satisfied effect can be obtained by being divided into 5.
It for PDS algorithm, needs to reasonably select window size, the present invention passes through the method pair of 5 folding cross validations
Window size is selected, and Fig. 6 (A) (B) (C) (D) is set forth about moisture content, oil content, protein, content of starch in corn
PDS model window size selection course, choose the corresponding window size of minimum RMSECV for PDS model best window.
For the PDS model of water content from Fig. 6, optimum window size 13, and the PDS model of other three kinds of ingredients, best window
Mouth size is 3.
For SBC, PDS, CT_pls algorithm, estimated performance is influenced by master sample quantity.Because of the number of master sample
Amount affects transfer relationship, and transfer relationship directly affects precision of prediction again, moves so the quantity of master sample affects calibration
The estimated performance of shifting formwork type.Table 3.2- table 3.5 is illustrated in the case where master sample number is different, moisture, oil, egg in corn
The prediction error of four kinds of white matter, starch content of material under different models, wherein the N of the first row indicates master sample number.Herein
PLS model indicate the benchmark model directly established using aiming field training data, therefore carried out to aiming field test sample
It when prediction, does not need to migrate sample, so prediction error is unrelated with master sample number.
The prediction error of moisture content in 3.2 corn of table
From table 3.2 as can be seen that for SBC algorithm, the smallest prediction error is 0.3081, the prediction with PLS method
Error is 0.1916, and the two difference is larger.In the case where being only applicable to systematization error due to SBC, illustrate for the pre- of moisture
It surveys, SBC method is not appropriate for.For PDS algorithm, the smallest prediction error obtains at N=45, and RMSECP=0.1767 is right
In CT_pls algorithm, the smallest prediction error obtains at N=13, and the lesser prediction error of RMSEP=0.1678, the two are equal
It is obtained at N=32, respectively 0.1860,0.1831.It can be seen that master sample number it is excessive or it is very few cannot all obtain it is best
Transfer relationship.
As can be seen that SBC method obtains the smallest prediction error 0.0668 in N=52 from table 3.3, but removing N=
The prediction error under other master sample numbers outside 26 is all proximate to it, and all close to the prediction error 0.0624 of PLS, explanation
SBC method is suitble to the prediction of oil.The minimum prediction error of PDS algorithm obtains at N=52, RMSEP=0.0787, CT_pls
The minimum prediction error of algorithm obtains at N=45, and RMSEP=0.0723, smaller value obtains all at N=32, respectively
0.0832 and 0.0740, and from after N=32, the RMSEP variation of PDS and CT_pls are all little.
The prediction error of oil content in 3.3 corn of table
The prediction error of protein content in 3.4 corn of table
From table 3.4 as can be seen that SBC method obtains minimum N=39 at predicts error, RMSEP=0.2552, and whole
During a master sample number variation, the variation of RMSEP is simultaneously little.PDS algorithm obtains minimum value 0.2296 at N=45,
And from after N=32, RMSEP is relatively stable.CT_pls algorithm obtains minimum prediction error 0.2093 at N=45, and from N=
After 26, RMSEP is relatively stable.
The prediction error of content of starch in 3.5 corn of table
From table 3.5 as can be seen that the minimum prediction error of SBC algorithm obtains at N=39, RMSEP=0.5775, and
RMSEP is relatively stable.PDS algorithm obtains at N=32, RMSEP=0.4964, and smaller value obtains at N=26 and N=39,
Respectively 0.5101,0.5270.CT_pls algorithm obtained at N=52 prediction error minimum value 0.4592, and N=(26,
32,39,45) place obtains smaller value.
By analyzing table 3.2- table 3.5, it can be deduced that draw a conclusion: first, the variation of master sample number is to SBC
The predictive ability of algorithm is simultaneously little, and the predictive ability of SBC algorithm and unstable.For example, being obtained for the prediction of oil content fine
Effect, be slightly better than PDS and CT_pls algorithm, and close to PLS algorithm, but very poor for the effect of the prediction of moisture content, far
Not as good as PDS and CT_pls algorithm, and differ larger with the prediction error of PLS.Second, the prediction for PDS and CT_pls algorithm
Error is affected by master sample number, and generally, in N < 32, prediction error is larger, and with the increase of sample number,
RMSEP can decline, and minimum value or smaller value are obtained at N=32, hereafter, with sample number increase RMSEP variation less or
Decline, therefore select 32 master samples (i.e. the 50% of training sample) that preferable migration effect can be obtained.Third, comprehensive ratio
Compared with the estimated performance of SBC, PDS, CT_pls algorithm, the estimated performance of CT_pls is best, followed by PDS algorithm, is that SBC is calculated again
Method.
In order to it is more fair, intuitively compare the prediction effect of calibration migration algorithm, the present invention selects 32 standard samples
This establishes the transfer relationship between source domain and aiming field, and Fig. 7-Figure 10 gives each algorithm corresponding to components various in corn
Predicted value figure compared with true value, predicted value closer to true value, corresponding mark point then closer to this straight line of y=x, because
This can put the intensity near straight line y=x according to the corresponding mark of every kind of algorithm, to judge the estimated performance of algorithm,
Their prediction effect can be more intuitively observed in turn.
It, can not be according to the concentration of mark point by Fig. 7-Figure 10 since the prediction error difference of PDS, CT_pls model is little
Degree contrasts two kinds of algorithm superiority and inferiority, therefore the prediction error corresponding to Fig. 7-Figure 10 predicted value is illustrated in table 3.6.Simultaneously
Table 3.7- table 3.10 gives CT_pls between the prediction error improvement rate or rate of descent of PLS, SBC, PDS algorithm and they
Carry out the p value of rank sum test.
Prediction error of the 3.6 each constituent concentration of corn data set of table under different models
Water content CT_pls algorithm is to the improvement rate of other algorithms and the p value of rank sum test in 3.7 corn of table
Oil content content CT_pls algorithm is to the improvement rate of other algorithms and the p value of rank sum test in 3.8 corn of table
Protein content CT_pls algorithm is to the improvement rate of other algorithms and the p value of rank sum test in 3.9 corn of table
Content of starch CT_pls algorithm is to the improvement rate of other algorithms and the p value of rank sum test in 3.10 corn of table
From table 3.6- table 3.10, further illustrate in tri- kinds of migration algorithms of SBC, PDS, CT_pls, CT_pls algorithm
Estimated performance is optimal, and PDS algorithm takes second place, and SBC algorithm is worst.Also, it since the p value in table 3.7-3.10 is all larger than 0.05, says
Significant difference is not present between bright CT_pls algorithm and other algorithms.
Finally, using directly being predicted using source domain model the domain test sample that do not carry out diverting the aim, and with use
CT_pls algorithm carry out prediction be compared, and then can the transfer ability intuitively to CT_pls model assess.Figure
11- Figure 14 is given the predicted value of model for not carrying out calibration migration and the comparison figure of true value and is carried out using CT_pls algorithm
Demarcate the predicted value of migration and the comparison figure of true value.
In Figure 11-Figure 14, when dot expression does not carry out calibration migration, aiming field test sample true value and predicted value it
Between relationship point, five-pointed star indicates using CT_pls algorithm to carry out the aiming field predicted value after calibration migration and between true value
Relationship point.It can be seen that dark dot all substantial deviation straight line y=x from Figure 11-Figure 14, and five-pointed star all concentrates on straight line y
Near=x, illustrate that directly using source domain model to carry out prediction to target numeric field data will appear very big deviation, this deviation is not by
Same measuring instrument introduces, and after carrying out calibration migration using CT_pls algorithm, it can largely reduce source domain number
According to the deviation between target numeric field data, and then directly the target after transfer can be carried out with data using source domain model pre-
It surveys, and obtains and good prediction effect.
It is tested using tablet data set.Table 11, which illustrates, directly establishes corresponding three using the aiming field training set of tablet
Training error, cross validation error, prediction error and the number of principal components of the PLS model of kind active component content.
The error and parameter of the aiming field data set PLS model of 3.11 tablet of table
From can be seen that RMSEC, RMSECV, RMSEP of every kind of ingredient in tablet in table 3.11 all in the identical order of magnitude
On, illustrate do not occur over-fitting, and RMSEP is smaller, illustrates also do not occur poor fitting phenomenon, and then can illustrate principal component
Number is chosen reasonable.The embodiment of the present invention chooses the principal component of PLS algorithm using 10 folding cross validation methods, Figure 15
(A) RMSECV about the PLS model of three kinds of active component contents in tablet is set forth with the change of number of principal components in (B) (C)
Change process obtains the minimum value of RMSECV, therefore about constituent content each in tablet respectively when number of principal components is 3,2,5
The best number of principal components of PLS model be respectively 3,2,5.
For PDS algorithm, the embodiment of the present invention selects window size by the method for 5 folding cross validations, figure
The window size selection course of the PDS model about three kinds of active component contents of tablet is set forth in 16 (A) (B) (C).From figure
As can be seen that corresponding to the PDS model of the first active constituent in 16, optimum window size 19, and its two kinds of active components
PDS model, optimum window size are respectively 3 and 13.
For SBC, PDS, CT_pls algorithm, table 12- table 14 is illustrated in the case where master sample number is different, in tablet
Prediction error of three kinds of active component contents under different models, wherein the N of the first row indicates that master sample number, PLS model are
The model that aiming field training data is established.
The prediction error of the first active component content in 3.12 tablet of table
The prediction error of second active ingredient content in 3.13 tablet of table
The prediction error of the third active component content in 3.14 tablet of table
From table 3.12- table 3.14 as can be seen that in the change procedure of master sample number, CT_pls algorithm can be predicted
Error is substantially all the prediction error of slightly less than PDS algorithm, and the prediction error of SBC algorithm is often higher than the prediction of PDS algorithm
Error.Illustrate the estimated performance of CT_pls algorithm better than PDS algorithm, the estimated performance of PDS algorithm is better than SBC algorithm, and CT_
The prediction error of pls and PDS algorithm illustrates that the two has preferable calibration transfer ability all close to the prediction error of PLS algorithm.
Furthermore SBC algorithm in the prediction error to second active ingredient also close to the prediction error of PLS algorithm, but to the first activity
The prediction error of ingredient and differing for PLS algorithm are larger, further illustrate the not popularity of SBC algorithm application.
Figure 17, Figure 18, Figure 19 respectively show in N=78 (i.e. 50% sample of training set), correspond to three kinds of work
The true value of tetra- kinds of models of PLS, SBC, PDS, CT_pls of property ingredient figure compared with predicted value.
Two kinds of algorithm superiority and inferiority clearly very can not be contrasted according to the intensity of mark point by Figure 17, Figure 18, Figure 19,
Therefore the prediction error corresponding to Figure 17, Figure 18, Figure 19 predicted value is illustrated in table 3.14.Table 3.16, table 3.17, table simultaneously
3.18, which give CT_pls, carries out sum of ranks between the prediction error improvement rate or rate of descent of PLS, SBC, PDS algorithm and they
The p value of inspection.
Prediction error of the 3.15 each active component content of tablet data set of table under different models
The CT_pls model of 1 content of active constituent is to the improvement rate of other models and the p of rank sum test in 3.16 tablet of table
Value
The CT_pls model of 2 content of active constituent is to the improvement rate of other models and the p of rank sum test in 3.17 tablet of table
Value
The CT_pls model of 3 content of active constituent is to the improvement rate of other models and the p of rank sum test in 3.18 tablet of table
Value
It can be seen that the prediction for tablet data active component content from table 3.15- table 3.17, CT_pls algorithm
Estimated performance reaches most preferably, the PLS model even better than directly established using target numeric field data, the estimated performance ten of PDS algorithm
Nearly PLS model is tapped, the estimated performance of SBC model is worst.And every group of p value is both less than 0.05, illustrates CT_pls algorithm and its
There is significant differences between his algorithm.
Figure 20, Figure 21, Figure 22 give the predicted value of model for not carrying out calibration migration and the comparison figure of true value and make
The predicted value of calibration migration and the comparison figure of true value are carried out with CT_pls algorithm.
From Figure 20, Figure 21, Figure 22 as can be seen that five-pointed star type mark point than ring-dot type mark point be more nearly and
It concentrates near straight line y=x, illustrates after carrying out calibration migration using CT_pls algorithm, obtain better prediction effect.So
And compared with corn data set, the migration effect of tablet data set is not obvious, this is because the principal spectrum of tablet data set and
Not too much big from SPECTRAL DIVERSITY, this point is as can be seen from Figure 4.
Analysis method of the invention uses aiming field training sample to establish PLS model as benchmark model, for comparing
The transfer ability of tri- kinds of SBC, PDS, CT_pls calibration migration models.The experimental results showed that the prediction of PDS and CT_pls model misses
Difference illustrates that the two all has preferable transfer ability, and the prediction error of CT_pls model is small all close to the prediction error of PLS
In the prediction error of PDS model.And SBC model is not the prediction effect that total energy has obtained, and illustrates its stability and predictive ability
Far away from PDS and CT_pls model.Therefore, in general, in three kinds of migration models, CT_pls model has optimal predictability
Can, PDS takes second place, and SBC is worst.
In the present invention, term " first ", " second ", " third " are used for description purposes only, and should not be understood as instruction or
Imply relative importance.Term " multiple " refers to two or more, unless otherwise restricted clearly.
The foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, for the skill of this field
For art personnel, the invention may be variously modified and varied.All within the spirits and principles of the present invention, made any to repair
Change, equivalent replacement, improvement etc., should all be included in the protection scope of the present invention.
Claims (5)
1. a kind of method based on infrared spectrum analysis material component content, which comprises the following steps:
S1 is established according to source domain ir data and source domain material component content corresponding with the source domain ir data
First regression model seeks the parameter in first regression model；
S2 obtains aiming field ir data, establishes between aiming field ir data and source domain ir data
Metastasis model seeks the parameter in the metastasis model；
S3, according to the aiming field ir data, the metastasis model, using first regression model obtain with it is described
The corresponding aiming field material component content of aiming field ir data；
Wherein, first regression model is Partial Least-Squares Regression Model, and the step S1 includes, to the source domain infrared light
Modal data carries out feature extraction and obtains the first spectral signature, is established according to first spectral signature and source domain material component content
The Partial Least-Squares Regression Model, finds out regression coefficient；
The aiming field ir data includes aiming field infrared spectroscopy normal data and aiming field examination of infrared spectrum data,
The step S2 includes carrying out feature extraction according to the aiming field infrared spectroscopy normal data to obtain the second standard spectrum feature；
The metastasis model is established according to first spectral signature and the second standard spectrum feature, finds out transfer matrix.
2. the method according to claim 1 based on infrared spectrum analysis material component content, which is characterized in that the step
Rapid S3 includes, according to the aiming field examination of infrared spectrum data acquisition third spectral signature, by the third spectral signature and
The metastasis model, which is brought into the minimum inclined two and multiplies, obtains the aiming field material component content in regression model.
3. the method according to claim 1 based on infrared spectrum analysis material component content, which is characterized in that described right
The step of source domain ir data progress feature extraction obtains the first spectral signature includes, to the source domain infrared spectroscopy
Data and source domain material component content carry out centralization processing, according to centralization treated source domain ir data and source domain
Material component content establishes least square regression model and obtains first spectral signature.
4. the method according to claim 1 based on infrared spectrum analysis material component content, which is characterized in that also obtain
It is described that feature extraction acquisition is carried out according to the aiming field infrared spectroscopy normal data including aiming field standard substance component content
The step of second standard spectrum feature include: to the aiming field infrared spectroscopy normal data and the aiming field standard substance at
Point content carries out centralization processing, according to centralization treated aiming field infrared spectroscopy normal data and aiming field standard substance
Component content establishes Partial Least-Squares Regression Model and obtains the second standard spectrum feature.
5. the method according to claim 2 based on infrared spectrum analysis material component content, which is characterized in that the step
In rapid S2, while obtaining the second standard spectrum feature, the second standard projection data and the second normal loading data are also obtained；
Include according to the step of aiming field examination of infrared spectrum data acquisition third spectral signature in the step S3, using described
The mean value of aiming field infrared spectroscopy normal data carries out centralization processing to the aiming field examination of infrared spectrum data, in utilization
Successively recursion obtains third spectral signature to the heartization treated aiming field examination of infrared spectrum data according to the following formula:Wherein, i is more than or equal to 1 and is less than or equal to k, TT_testFor third Spectral Properties
Sign, k are the number of third spectral signature,For i-th of component of the second standard projection data,Centered on change
I-th of residual error of aiming field examination of infrared spectrum data that treated,For i-th of component of the second normal loading data
Transposition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710009518.5A CN106680238B (en) | 2017-01-06 | 2017-01-06 | Method based on infrared spectrum analysis material component content |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710009518.5A CN106680238B (en) | 2017-01-06 | 2017-01-06 | Method based on infrared spectrum analysis material component content |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106680238A CN106680238A (en) | 2017-05-17 |
| CN106680238B true CN106680238B (en) | 2019-09-06 |
Family
ID=58849170
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710009518.5A Active CN106680238B (en) | 2017-01-06 | 2017-01-06 | Method based on infrared spectrum analysis material component content |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106680238B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107918718B (en) * | 2017-11-03 | 2020-05-22 | 东北大学秦皇岛分校 | Sample component content determination method based on online sequential extreme learning machine |
| CN108152239A (en) * | 2017-12-13 | 2018-06-12 | 东北大学秦皇岛分校 | The sample composition content assaying method of feature based migration |
| CN108509998A (en) * | 2018-03-30 | 2018-09-07 | 中国科学院半导体研究所 | A kind of transfer learning method differentiated on different devices for target |
| CN110632024B (en) * | 2019-10-29 | 2022-06-24 | 五邑大学 | Quantitative analysis method, device and equipment based on infrared spectrum and storage medium |
| CN111220566A (en) * | 2020-01-16 | 2020-06-02 | 东北大学秦皇岛分校 | Infrared spectrum measuring instrument calibration migration method based on OPLS and PDS |
| CN111220565B (en) * | 2020-01-16 | 2022-07-29 | 东北大学秦皇岛分校 | CPLS-based infrared spectrum measuring instrument calibration migration method |
| CN111563436B (en) * | 2020-04-28 | 2022-04-08 | 东北大学秦皇岛分校 | Infrared spectrum measuring instrument calibration migration method based on CT-CDD |
| CN113343804B (en) * | 2021-05-26 | 2022-04-29 | 武汉大学 | Integrated migration learning classification method and system for single-view fully-polarized SAR data |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| GB2311390B (en) * | 1996-03-18 | 2000-03-22 | Ibm | Initial program load in data processing network |
| CN100561194C (en) * | 2007-09-04 | 2009-11-18 | 厦门中药厂有限公司 | A kind of AOTF near infrared spectrometer that utilizes detects method of microorganism in the Chinese medicine |
-
2017
- 2017-01-06 CN CN201710009518.5A patent/CN106680238B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN106680238A (en) | 2017-05-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106680238B (en) | Method based on infrared spectrum analysis material component content | |
| Demattê et al. | Soil analytical quality control by traditional and spectroscopy techniques: Constructing the future of a hybrid laboratory for low environmental impact | |
| CN104062257B (en) | A kind of based on the method for general flavone content near infrared ray solution | |
| CN106596450B (en) | Incremental method based on infrared spectrum analysis material component content | |
| Wang et al. | Soil pH value, organic matter and macronutrients contents prediction using optical diffuse reflectance spectroscopy | |
| Ni et al. | Screening wavelengths with consistent and stable signals to realize calibration model transfer of near infrared spectra | |
| CN110687072B (en) | Calibration set and verification set selection and modeling method based on spectral similarity | |
| EP3066437B1 (en) | Texture analysis of a coated surface using kepler's planetary motion laws | |
| WO2016000088A1 (en) | Hyperspectral waveband extraction method based on optimal index factor-correlation coefficient method | |
| CN110455722A (en) | Rubber tree blade phosphorus content EO-1 hyperion inversion method and system | |
| CN107505346B (en) | The method for predicting to be especially the chemical displacement value of NMR spin system in biological fluid sample in class of fluids sample | |
| CN108152239A (en) | The sample composition content assaying method of feature based migration | |
| CN109444066A (en) | Model transfer method based on spectroscopic data | |
| CN109799207A (en) | The quantitative detecting method of talcum powder is mixed in the root of Dahurain angelica based on near-infrared spectrum analysis | |
| CN106770058A (en) | The quick special purpose device and its application method of the soil nitrate-N based on infrared spectrum | |
| CN108680515A (en) | A kind of simple grain amylose in rice Quantitative Analysis Model structure and its detection method | |
| CN108956583A (en) | Characteristic spectral line automatic selecting method for laser induced breakdown spectroscopy analysis | |
| CN106092962A (en) | A kind of near infrared spectroscopy quickly detects the method for millet crude protein content | |
| CN103592255A (en) | Soft method for measuring total protein content of donkey-hide gelatin skin solution on basis of near infrared spectrum technology | |
| Liu et al. | Prediction of soil organic carbon with different parent materials development using visible-near infrared spectroscopy | |
| Mermet et al. | A logical way through the limits of quantitation in inductively coupled plasma spectrochemistry | |
| CN106770005B (en) | A kind of division methods of calibration set for near-infrared spectrum analysis and verification collection | |
| CN104297201A (en) | Method for quickly, accurately and quantitatively detecting ratio of various oil components in blend oil | |
| CN103398971A (en) | Chemometrics method for determining cetane number of diesel oil | |
| CN104849234A (en) | Assay method for analyzing contents of principal components of imidacloprid based on near-infrared spectrum |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |