CN104502305B - Near infrared spectrum useful information distinguishing method based on wavelet transform - Google Patents
Near infrared spectrum useful information distinguishing method based on wavelet transform Download PDFInfo
- Publication number
- CN104502305B CN104502305B CN201410744881.8A CN201410744881A CN104502305B CN 104502305 B CN104502305 B CN 104502305B CN 201410744881 A CN201410744881 A CN 201410744881A CN 104502305 B CN104502305 B CN 104502305B
- Authority
- CN
- China
- Prior art keywords
- matrix
- useful information
- near infrared
- wavelet transform
- signal
- 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.)
- Expired - Fee Related
Links
Landscapes
- Spectrometry And Color Measurement (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention provides a near infrared spectrum useful information distinguishing method based on wavelet transform. The near infrared spectrum useful information distinguishing method comprises the following steps: firstly, obtaining a near infrared spectrum detection signal of a sample; performing the wavelet transform to obtain a maximum wavelet transform result of a wavelength signal of each point, wherein a scale factor corresponding to the maximum wavelet transform result is used as an optimal scale factor of the wavelength signal of the point; replacing a primary detection signal by using an optimal wavelet transform result of the wavelength signals of all the points, wherein the replaced signal is used as a new sample detection signal; selecting the new signal by using a UVE-PLS method, and establishing a prediction model; and detecting the predictive ability of the model by using a minimum root mean square error (RMSE). According to the distinguishing method, useful information in an original wavelength detection signal is maximally extracted by virtue of the continuous wavelet transform, and the prediction model is established by virtue of the useful information, so that the quality of the prediction model is effectively improved.
Description
Technical field
The invention belongs to chemometric techniques field, it is related to a kind of near infrared spectrum useful information based on wavelet transformation
Resolving method.
Background technology
Due to quickly, no undermining the features such as need not pre-process, near infrared spectrum (near infrared
Spectroscopy, NIR) analytical technology is widely used in analyzing the complex sample in many fields, such as agricultural, food
Sample analysis with various fields such as medicine.The multivariate calibration method of Chemical Measurement (chemometrics), such as partially
Little least square method (partial least squares, PLS), is requisite analysis in Near-Infrared Spectra for Quantitative Analysis
Means.PLS analysis method can be expressed as follows：
Wherein,XIt is the near infrared spectrum detection information matrix of sample,YIt is the corresponding measured value of each sample,BIt is PLS
Coefficient vector.
In Near-Infrared Spectra for Quantitative Analysis, the steady key of multivariate calibration PLS model and the accuracy predicting the outcome are past
Past and unsatisfactory, this is because the bands of a spectrum of near infrared spectrum are weak, wide and overlapping serious.Using all near-infrareds detecting
Spectral signal, as the modeling information of PLS model, inherently brings excessively useless detection information into, serious reduction forecast model
Steady key and accuracy.
The Signal Pre-Processing Method in Chemical Measurement field can effectively improve steady key and the accuracy of model, such as continuously little
Wave conversion method (continuous wavelet transform, CWT).The Continuous Wavelet Transform definition of detection signal
As follows
Wherein, W (a, b) is the CWT of detection signal f (x), a（a>0；a∈R）It is scale factor, b（b∈R）It is
The window factor,ψ * It is small echo.
In the calculating process of wavelet transformation, convolution algorithm makes wavelet transformation can obtain one in each yardstick a
The maximum wavelet conversion signal of specific wavelength signal.This means that if a value is different, gained is little for point wavelength signals
Wave conversion result is different.If a value choose improper it is easy to many useful informations in loss detection signal.
Wavelet transformation has good time-frequency separation characteristic, and information processing capability is strong, and wavelet transformation is located in advance for near-infrared
Reason, extractable useful information, eliminate ambient interferences, improve near infrared analysis precision and model robustness.But often study people
Member simply arbitrarily specifies a less a value when choosing a value（Because if a value larger it is impossible to reflection detection signal details
Information）, such as a=5 or a=10.This means that for whole detection signal, and when being analyzed with wavelet transformation, a value is fixing
Constant it is easy to lose the useful information of point wavelength signals.
Content of the invention
It is an object of the invention to provide a kind of near infrared spectrum useful information resolving method based on wavelet transformation, by even
Continuous wavelet transformation extracts the useful information in former wavelength detecting signal to greatest extent, sets up forecast model by these information,
Effectively improve forecast model quality.
For achieving the above object, the technical solution adopted in the present invention is：A kind of near infrared spectrum based on wavelet transformation
Useful information resolving method, extracts the useful information in former wavelength detecting signal to greatest extent by continuous wavelet transform, then
Set up forecast model by these information, effectively improve forecast model quality it is characterised in that this resolving method is specifically by following
Step is carried out：
Step 1：Obtain the near infrared spectrum detection signal of sample；
Step 2：Wavelet transformation is carried out to the near infrared spectrum detection signal of sample, obtains each point wavelength signals
Big wavelet transform result, using corresponding for maximum wavelet transformation results scale factor as this wavelength signals optimal scale because
Son；
Step 3：With there being the wavelet transform result in the respective optimal scale factor for a wavelength signals to replace initial inspection
Survey signal, and using the signal after replacing as new sample detection signal, form a new detection signal matrix；Then, give
New detection signal matrix merging one is identical with this matrix size but amplitude is 10-15Stochastic variable matrix；
Step 4：One and original matrix size identical amplitude 10 are expanded in former matrix of variables by UVE-PLS method-15
Stochastic variable matrix, form a new matrix, and produce the regression coefficient matrix of new matrix by leave one cross validation,
The maximum absolute stability value of this stochastic variable matrix as cuts off threshold value, and stability is less than the variable of this cut-out threshold value from model
Middle deletion, thus selecting useful information, the useful information selected is built prediction mould as the input information of PLS model
Type；.
Step 5：Carry out detection model using lowest mean square root error (root mean square error, RMSE)
Predictive ability；RMSE value is calculated by following formula：
In above formula,y i Withŷ i It is the corresponding measured value of each sample and predicted value respectively,nIt is number of samples.
Resolving method of the present invention, when near infrared spectrum detection signal is carried out with Wavelet transformation analysis, is not to specify a tool
The a of body（Scale factor）Value, and when being selected at some point signal, wavelet transform result reaches a value conduct during maximum
The a value of this signal, so that each point signal has the optimum a value controlled oneself.Combine near infrared light spectrum signal continuous simultaneously
The peak information of wavelet transformation and paddy information, utilize no information variable removing method (uninformative variable
Elimination-PLS, UVE-PLS) delete united information in garbage, recycle remaining important information data build
Vertical forecast model；It is not only able to effectively reflect the useful information near infrared light spectrum signal, and model matter can be effectively improved
Amount.
Brief description
Fig. 1 is the near infrared spectrum detection signal figure that resolving method of the present invention obtains.
Fig. 2 is the optimal wavelet transformation results figure being had a wavelength signals in resolving method of the present invention, i.e. each point letter
Wavelet transform result number when its a value is for optimal value.
Fig. 3 is that UVE-PLS selects variable schematic diagram.
Specific embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
The invention provides a kind of near infrared spectrum useful information resolving method based on wavelet transformation, specifically press following step
Suddenly carry out：
Step 1：Obtain the near infrared spectrum detection signal of sample using Near-Infrared Spectroscopy Instruments, shown in Fig. 1 be
700 wavelength points of 80 samples；
Step 2：Wavelet transformation is carried out to the near infrared spectrum detection signal of sample, obtains the little of each point wavelength signals
The maximum of wave conversion result, by corresponding for this maximum scale factor（A value）Optimum a value as this wavelength signals；
Step 3：With there being the wavelet transform result in each optimum a value for a wavelength signals to replace initial detection letter
Number, and using the signal after replacing as new sample detection signal, form a new detection signal matrix（As Fig. 2 left side
Shown in point）；Then, to new detection signal matrix merge one identical with this matrix size but very by a small margin（10-15）With
Machine matrix of variables（The stochastic variable matrix merging is as shown in the right half part of Fig. 2）；
Step 4：Using UVE-PLS method, new signal is selected：
No information variable removing method (UVE-PLS) is one kind effectively useless variable（Wavelength）Removing method, by
The information variable that has being successfully applied to continuous detection signal extracts.The method carrys out each change in evaluation model using variable stability
The importance of amount, the variable that stability is less than cut-out threshold value is considered as asemantic variable.In order to obtain cutting off threshold value, UVE-
PLS expand in former matrix of variables one with original matrix size identical very by a small margin (10-15) stochastic variable matrix,
Thus forming a new matrix, and produce the regression coefficient matrix of new matrix by leave one cross validation.Leaving-one method intersects
Checking be every time leave out a line from matrix successively after set up the method that PLS regression coefficient asked for by model.Variable stability
(variable stability, VS) is defined as the mean value of all coefficients and the standard deviation (standard of this variable
Deviation, STD) ratio,
The stability (VS) of variable can be calculated by equation below：
（3）In formula,mean(β j ) andSTD(β j ) it is respectivelyjIndividual variable is (altogetherpIndividual variable) regression coefficient
(β j ) mean value and standard deviation.
Cut-out threshold value is the maximum absolute stability value of the stochastic variable matrix of amplification（As shown in dotted line horizontal in Fig. 3）, stable
Property will delete from model less than the variable of this cut-out threshold value, that is, the hash between two dotted lines will be deleted, by select
Useful information as PLS model input information to build forecast model；
Step 5：Carry out detection model using lowest mean square root error (root mean square error, RMSE)
Predictive ability；RMSE value is calculated by following formula：
（4）In formula,y i Withŷ i It is the corresponding measured value of each sample and predicted value respectively,nIt is number of samples.
The present invention extracts the useful information in former wavelength detecting signal to greatest extent by continuous wavelet transform, passes through
These information set up forecast model, can effectively improve forecast model quality.Small echo is being carried out near infrared spectrum detection signal
During transform analysis, it is not to specify a specific a value, and when being selected at some point signal, wavelet transform result reaches
A value during big value is as a value of this signal, so that each point signal has the optimum a value controlled oneself.For continuous wavelet
Transformation results, it is on the occasion of the useful information representing near infrared spectrum peak-to-peak signal, and negative value represents near infrared light spectral valley signal
Useful information, for complicated near infrared light spectrum signal, the information of peak and valley is all very important, but paddy information is ground
Study carefully personnel to ignore.The peak information of present invention joint near infrared light spectrum signal continuous wavelet transform and paddy information, are become using no information
Amount removing method (uninformative variable elimination-PLS, UVE-PLS) is deleted in united information
Garbage, then sets up forecast model using remaining important information data.The method is not only able to effectively reflect near-infrared
Useful information in spectral signal, and model quality can be effectively improved.
Claims (1)
1. a kind of near infrared spectrum useful information resolving method based on wavelet transformation, obtains the near infrared spectrum detection letter of sample
Number, wavelet transformation is carried out to the near infrared spectrum detection signal of sample, obtains the maximum wavelet conversion of each point wavelength signals
As a result, using corresponding for maximum wavelet transformation results scale factor as this wavelength signals the optimal scale factor, its feature exists
In this resolving method is specifically carried out according to the following steps：
Step 1：Useful information in former wavelength detecting signal is extracted to greatest extent by continuous wavelet transform, then passes through these
Information sets up forecast model, effectively improves forecast model quality,
Step 2：With there being the wavelet transform result in the respective optimal scale factor for a wavelength signals to replace initial detection letter
Number, and using the signal after replacing as new sample detection signal, form a new detection signal matrix；Then, to new
Detection signal matrix merging one is identical with this matrix size but amplitude is 10-15Stochastic variable matrix；
Step 3：One and original matrix size identical amplitude 10 are expanded in former matrix of variables by UVE-PLS method-15With
Machine matrix of variables, forms a new matrix, and produces the regression coefficient matrix of new matrix by leave one cross validation, this with
The maximum absolute stability value of machine matrix of variables as cuts off threshold value, and the variable that stability is less than this cut-out threshold value is deleted from model
Removing, thus selecting useful information, the useful information selected being built forecast model as the input information of PLS model；
Step 4：Using lowest mean square root error come the predictive ability of detection model；RMSE value is calculated by following formula：
In above formula,y i Withŷ i It is the corresponding measured value of each sample and predicted value respectively,nIt is number of samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410744881.8A CN104502305B (en) | 2014-12-09 | 2014-12-09 | Near infrared spectrum useful information distinguishing method based on wavelet transform |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410744881.8A CN104502305B (en) | 2014-12-09 | 2014-12-09 | Near infrared spectrum useful information distinguishing method based on wavelet transform |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104502305A CN104502305A (en) | 2015-04-08 |
| CN104502305B true CN104502305B (en) | 2017-02-22 |
Family
ID=52943723
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410744881.8A Expired - Fee Related CN104502305B (en) | 2014-12-09 | 2014-12-09 | Near infrared spectrum useful information distinguishing method based on wavelet transform |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104502305B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002090299A (en) * | 2000-09-11 | 2002-03-27 | Opt Giken Kk | Method for distinguishing grade of high-molecular material |
| CN101520412A (en) * | 2009-03-23 | 2009-09-02 | 中国计量学院 | Near infrared spectrum analyzing method based on isolated component analysis and genetic neural network |
| CN101055248B (en) * | 2007-04-28 | 2010-12-15 | 吉林燃料乙醇有限责任公司 | Method for analyzing high moisture corn and freezing corn moisture using near infrared spectrum technology |
| CN102507495A (en) * | 2011-11-23 | 2012-06-20 | 浙江大学 | Method for rapidly and nondestructively detecting green tea water content based on wavelet transformation |
| CN103854305A (en) * | 2014-03-19 | 2014-06-11 | 天津大学 | Module transfer method based on multiscale modeling |
| CN103913432A (en) * | 2014-03-25 | 2014-07-09 | 西安交通大学 | Near infrared spectrum wavelength selecting method based on particle swarm optimization |
| CN104020135A (en) * | 2014-06-18 | 2014-09-03 | 中国科学院重庆绿色智能技术研究院 | Calibration model establishing method based on near infrared spectrum |
| CN104165861A (en) * | 2014-08-22 | 2014-11-26 | 云南中烟工业有限责任公司 | Near infrared spectrum quantitative model simplification method based on principal component analysis |
-
2014
- 2014-12-09 CN CN201410744881.8A patent/CN104502305B/en not_active Expired - Fee Related
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002090299A (en) * | 2000-09-11 | 2002-03-27 | Opt Giken Kk | Method for distinguishing grade of high-molecular material |
| CN101055248B (en) * | 2007-04-28 | 2010-12-15 | 吉林燃料乙醇有限责任公司 | Method for analyzing high moisture corn and freezing corn moisture using near infrared spectrum technology |
| CN101520412A (en) * | 2009-03-23 | 2009-09-02 | 中国计量学院 | Near infrared spectrum analyzing method based on isolated component analysis and genetic neural network |
| CN102507495A (en) * | 2011-11-23 | 2012-06-20 | 浙江大学 | Method for rapidly and nondestructively detecting green tea water content based on wavelet transformation |
| CN103854305A (en) * | 2014-03-19 | 2014-06-11 | 天津大学 | Module transfer method based on multiscale modeling |
| CN103913432A (en) * | 2014-03-25 | 2014-07-09 | 西安交通大学 | Near infrared spectrum wavelength selecting method based on particle swarm optimization |
| CN104020135A (en) * | 2014-06-18 | 2014-09-03 | 中国科学院重庆绿色智能技术研究院 | Calibration model establishing method based on near infrared spectrum |
| CN104165861A (en) * | 2014-08-22 | 2014-11-26 | 云南中烟工业有限责任公司 | Near infrared spectrum quantitative model simplification method based on principal component analysis |
Non-Patent Citations (2)
| Title |
|---|
| 小波变换在近红外光谱分析中的应用进展;田高友等;《光谱学与光谱分析》;20031231;第23卷(第6期);第1112-1114页 * |
| 小波变换用于近红外光谱性质分析;田高友等;《分析化学》;20040930;第32卷(第9期);第1125-1130页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104502305A (en) | 2015-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105300923B (en) | Without measuring point model of temperature compensation modification method during a kind of near-infrared spectrometers application on site | |
| Rodriguez-Saona et al. | Rapid determination of Swiss cheese composition by Fourier transform infrared/attenuated total reflectance spectroscopy | |
| KR101939887B1 (en) | Texture analysis of a coated surface using pivot-normalization | |
| CN107179310B (en) | Raman spectrum characteristic peak recognition methods based on robust noise variance evaluation | |
| Ranzan et al. | Wheat flour characterization using NIR and spectral filter based on Ant Colony Optimization | |
| CN102608061B (en) | Improved method for extracting Fourier transformation infrared spectrum characteristic variable of multi-component gas by aid of TR (Tikhonov regularization) | |
| CN105158200B (en) | A kind of modeling method for improving the Qualitative Analysis of Near Infrared Spectroscopy degree of accuracy | |
| CN109253985B (en) | Method for identifying wood grade for koto panel by near infrared spectrum based on neural network | |
| CN109115692B (en) | Spectral data analysis method and device | |
| CN105334185A (en) | Spectrum projection discrimination-based near infrared model maintenance method | |
| CN109283153B (en) | Method for establishing quantitative analysis model of soy sauce | |
| CN104502306B (en) | Near-infrared spectrum wavelength system of selection based on variable importance | |
| CN107976417B (en) | Crude oil type identification method based on infrared spectrum | |
| CN102854151B (en) | Chemometrics method for classifying sample sets in spectrum analysis | |
| CN107271389B (en) | A kind of spectral signature variable fast matching method based on index extreme value | |
| CN104502305B (en) | Near infrared spectrum useful information distinguishing method based on wavelet transform | |
| Xia et al. | Non-destructive analysis the dating of paper based on convolutional neural network | |
| CN106970042B (en) | Method for detecting impurity and moisture content of carrageenin | |
| Xie et al. | Quantitative analysis of steel samples by laser-induced-breakdown spectroscopy with wavelet-packet-based relevance vector machines | |
| CN110632024B (en) | Quantitative analysis method, device and equipment based on infrared spectrum and storage medium | |
| CN113702328A (en) | Method, device, equipment and storage medium for analyzing properties of product oil | |
| Li et al. | An improved ensemble model for the quantitative analysis of infrared spectra | |
| Tan et al. | A multi-model fusion strategy for multivariate calibration using near and mid-infrared spectra of samples from brewing industry | |
| Zhang et al. | On-line monitoring of pharmaceutical production processes using hidden Markov model | |
| CN113674814B (en) | Method and device for constructing spectrum quantitative analysis model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170222 Termination date: 20201209 |