CN105628670A - Two-dimensional correlation spectroscopy multi-scale modeling method for olive oil impurity identification - Google Patents

Abstract

The invention discloses a two-dimensional correlation spectroscopy multi-scale modeling method for olive oil impurity identification. The method comprises selecting the optimal two-dimensional wavelet basis, carrying out multi-scale decomposition and all layer respective reconstitution through the optimal two-dimensional wavelet basis, carrying out modeling and predicting on each layer of the reconstituted correlation spectrum through NPLS, acquiring a root-mean-square error subjected to cross validation, carrying out submodel fusion through the calculated weight, and evaluating multiscale-two dimensional correlation spectrum model result and performances through the predicted root-mean-square error and correlation coefficient. Compared with the routine model, the modeling method substantially improves a precision and reliability of a conventional Raman spectroscopy model, finds novel symptom information in a sample spectrum, prevents information loss, realizes simple and reliable Raman spectrum analysis and can be widely used in complex system spectral analysis.

Description

A kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification

Technical field

The present invention relates to a kind of modeling method, be specifically related to a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification.

Background technology

Olive oil originates from the coastal all states in Mediterranean, so far there is the history of several thousand, it is described as " vegetable oil queen " by west, it can not only blood circulation promoting, improve hormonal system, not only there is abundant nutrition, also have certain beauty functions, therefore won the favor of more and more consumer. The remote super other kinds of vegetable oil of value of olive oil, illegal retailer, in order to pursue juice, blends vegetable oil cheap in a large number in olive oil and carries out squeezing extraction together, such as Oleum Helianthi, soybean oil, hazelnut oil. These adulterated oils inferior once come into the market and health may being produced serious threat, so need a kind of can the method for fast and convenient accurate detection olive oil doping.

Although each olive oil composition is all extremely complex system, but each crude oil all has relatively-stationary compositional system, this system has respective characteristic just as the fingerprint of people, and the diversity making full use of these " fingerprints " is expected to realize the doping identification of olive oil. Raman spectra yet with olive oil is complicated and overlapping serious, each quasi-grease raman spectra characteristic close especially adulterated, and causes that traditional Raman spectrum is poor to olive oil doping recognition resolution. Therefore, we introduce two-dimensional correlation spectra technology, put forth effort to improve the apparent resolution of Raman spectrum. Two-dimensional correlation spectra is analyzed by testing sample applies outside perturbation, the corresponding dynamic spectrum of sample under record perturbation state, then a series of dynamic spectrum is carried out correlation analysis, again result is showed with the form of two dimension equal pitch contour or 3-D graphic, possessed the feature that the one-dimensional spectrum of many routines does not possess. This technology can be effectively improved the resolution of spectrum, and disclosing in molecule, intermolecular interaction and judge to have played in the research of the sequencing of each functional group reactions in molecule important effect, olive oil doping identification is significant.

Multi-scale Modeling passes through multi-resolution decomposition and data fusion, the ingenious multiple dimensioned characteristic of time-frequency that make use of signal, accurately extract the characteristic information of spectrum, and effectively achieve the integrated computing of data prediction and Multivariate Correction to extract the information in the two-dimensional correlation spectra of olive oil, use NPLS that the information obtained is modeled, to work in coordination with the multiple dimensioned characteristic of time-frequency utilizing signal accurately to extract spectral signature information, and avoid information dropout.

Summary of the invention

For solving the problems referred to above, the invention provides a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification, in conjunction with multi-scale Modeling method, the two-dimensional correlation raman spectral signal of olive oil is processed, for adulterated oil lipid species in examination olive oil, best information in adaptive extraction two-dimensional correlation spectra can set up for the qualitative and quantitative model of alloy, and then be obviously improved degree of accuracy and the prediction effect of model.

For achieving the above object, the technical scheme that the present invention takes is:

A kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification, comprises the steps:

S1, gather the original spectrum of different olive oil sample;

S2, generation step S1 gather the two-dimensional correlation spectra of original spectrum;

S3, characteristic in conjunction with two-dimensional wavelet transformation and two-dimensional correlation spectra, select best wavelet that two-dimensional correlation spectra is carried out 2-d wavelet multi-resolution decomposition, obtain 2-d wavelet coefficient;

S4,2-d wavelet coefficient to step S3 gained carry out image reconstruction;

S5, spectrum picture to each layer of reconstruct carry out multidimensional offset minimum binary modeling respectively, obtain submodel, and corresponding group doping content predictive value, and the doping content root-mean-square error of each layer of reconstruct image modeling;

S6, use weights that the submodel of step S5 gained carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model prediction effect.

The original spectrum gathered in described step S1 is the spectrum that same sample gathers same instrument under condition of different temperatures, and the temperature conditions of change therein needs guarantee identical for different samples.

Two-dimensional correlation spectra in described step S2 is generated by below equation:

$\mathrm{\Φ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^{T}y\left(v_2\right)$

$\mathrm{\Ψ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^T\mathrm{Ny}\left(v_2\right)$

In formula: y (v) is input spectrum matrix, �� (v1, v2) is the relevant spectrum picture matrix of the synchronization generated, and �� (v1, v2) is the asynchronous correlation spectrum image array generated.

In described step S4, reconstruct refers to and each layer of wavelet coefficient after the decomposition of the two-dimensional correlation spectra of same sample is reconstructed respectively.

Root-mean-square error in described step S5 is RMSECV, and formula is as follows:

$\mathrm{RMSECV}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{\mathrm{sum}{C}_{\mathrm{REF}}^2}}$

In formula: C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

RMSEP in described step S6 is predicted root mean square error, below equation obtain:

$\mathrm{RMSEP}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{n}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

Correlation coefficient in described step S6 is R, below equation obtain:

$R=\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}{C}_{\mathrm{REF}}-\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}\mathrm{\Σ}{C}_{\mathrm{PEF}}}{n}}{\sqrt{(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left(\mathrm{\Σ}{C}_{\mathrm{NIR}}\right)}^2}{n})(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left({\mathrm{\ΣC}}_{\mathrm{REF}}\right)}^2}{n})}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

Model Fusion in described step S6 refers to that each layer of two-dimensional wavelet transformation coefficient is reconstructed image carries out NPLS modeling, is predicted the outcome and predicted root mean square error.

Weights in described step S6 are obtained by below equation:

$W_i=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\mathrm{\Σ}}_i^m{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}$

Wherein, RMSECV_iIt it is the predicted root mean square error after i-th submodel cross validation. Submodel is merged by described step S6 by below equation:

$C=\underset{i}{\overset{m}{\mathrm{\Σ}}}{W}_i\·C_{\mathrm{iREF}}$

In formula: C_iREFBeing predicting the outcome of submodel, m is the yardstick decomposed, and C is predicting the outcome after Model Fusion, namely final model prediction final result.

Wherein, the selection best wavelet described in step S3, is that wavelet basis mathematical characteristic is analyzed, and obtains the wavelet basis function with symmetry, compact sup-port, orthogonality and high-order vanishing moment, has Daubechies, Symlets, Coiflets etc.; Multidimensional partial least squares algorithm (Multi-waypartialleastsquare described in step S5, N-PLS), it is based on the Multidimensional Data Model algorithm on offset minimum binary basis, the load vectors directly related with each dimension can be obtained, be conducive to each dimension of model is made independent explanation.

The method have the advantages that

The present invention first selects best 2-d wavelet base that two-dimensional correlation spectra is carried out multi-resolution decomposition and each layer reconstructs respectively; Secondly each layer of correlation spectrum of reconstruct is modeled prediction and obtains the root-mean-square error of cross validation by application NPLS; Then pass through the weights calculated and carry out submodel fusion; Finally by predicted root mean square error and correlation coefficient, result and the performance of multiple dimensioned-two-dimensional correlation spectra model are evaluated. This method is compared to conventional model, it is obviously improved precision and the reliability of normal Raman spectrum analysis model, not only carry and excavated characterization information new in sample spectra, and avoid the loss of information, make Raman spectrum analysis simpler, reliably, it is expected to be widely used in complex system spectrum analysis.

Detailed description of the invention

In order to make objects and advantages of the present invention clearly understand, below in conjunction with embodiment, the present invention is further elaborated. Should be appreciated that specific embodiment described herein is only in order to explain the present invention, is not intended to limit the present invention.

Embodiments provide a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification, it is characterised in that comprise the steps:

S1, gather the original spectrum of different olive oil sample; The original spectrum gathered is the spectrum that same sample gathers same instrument under condition of different temperatures, and the temperature conditions of change therein needs guarantee identical for different samples;

S2, generation step S1 gather the two-dimensional correlation spectra of original spectrum; Described two-dimensional correlation spectra is generated by below equation:

$\mathrm{\Φ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^{T}y\left(v_2\right)$

$\mathrm{\Ψ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^T\mathrm{Ny}\left(v_2\right)$

In formula: y (v) is input spectrum matrix, �� (v1, v2) is the relevant spectrum picture matrix of the synchronization generated, and �� (v1, v2) is the asynchronous correlation spectrum image array generated.

S3, characteristic in conjunction with two-dimensional wavelet transformation and two-dimensional correlation spectra, select best wavelet that two-dimensional correlation spectra is carried out 2-d wavelet multi-resolution decomposition, obtain 2-d wavelet coefficient; Described selection best wavelet, is that wavelet basis mathematical characteristic is analyzed, and obtains the wavelet basis function with symmetry, compact sup-port, orthogonality and high-order vanishing moment, has Daubechies, Symlets, Coiflets etc.;

S4,2-d wavelet coefficient to step S3 gained carry out image reconstruction; Described reconstruct refers to and each layer of wavelet coefficient after the decomposition of the two-dimensional correlation spectra of same sample is reconstructed respectively.

S5, spectrum picture to each layer of reconstruct carry out multidimensional offset minimum binary modeling respectively, obtain submodel, and corresponding group doping content predictive value, and the doping content root-mean-square error of each layer of reconstruct image modeling; Described multidimensional partial least squares algorithm (Multi-waypartialleastsquare, N-PLS), it is based on the Multidimensional Data Model algorithm on offset minimum binary basis, it is possible to obtain the load vectors directly related with each dimension, is conducive to each dimension of model is made independent explanation; Described root-mean-square error is RMSECV, and formula is as follows:

$\mathrm{RMSECV}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{\mathrm{sum}{C}_{\mathrm{REF}}^2}}$

In formula: C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped;

S6, use weights that the submodel of step S5 gained carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model prediction effect; Described RMSEP is predicted root mean square error, below equation obtain:

$\mathrm{RMSEP}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{n}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped;

Described correlation coefficient is R, below equation obtain:

$R=\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}{C}_{\mathrm{REF}}-\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}\mathrm{\Σ}{C}_{\mathrm{PEF}}}{n}}{\sqrt{(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left(\mathrm{\Σ}{C}_{\mathrm{NIR}}\right)}^2}{n})(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left({\mathrm{\ΣC}}_{\mathrm{REF}}\right)}^2}{n})}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

Described Model Fusion refers to that each layer of two-dimensional wavelet transformation coefficient is reconstructed image carries out NPLS modeling, is predicted the outcome and predicted root mean square error.

Weights in described step S6 are obtained by below equation:

$W_i=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\mathrm{\Σ}}_i^m{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}$

Wherein, RMSECV_iIt it is the predicted root mean square error after i-th submodel cross validation. Submodel is merged by described step S6 by below equation:

$C=\underset{i}{\overset{m}{\mathrm{\Σ}}}{W}_i\·C_{\mathrm{iREF}}$

In formula: C_iREFBeing predicting the outcome of submodel, m is the yardstick decomposed, and C is predicting the outcome after Model Fusion, namely final model prediction final result.

Embodiment

Sample and instrument

Sample: the olive oil of four kinds of different brands and soybean oil, Oleum Helianthi, Petiolus Trachycarpi oil, Oleum Arachidis hypogaeae semen.

Instrument: Portable Raman spectrometer

Concrete steps include:

(1) the doping sample of different brands olive oil is prepared: experiment arranges 4 groups altogether, often group 40 doping sample. Wherein by soybean oil according to volume fraction 0.5% to 3%, increment is 0.5%, 3% to 10%, and increment is 1%, 10% to 40%, and increment is 10%, mixes in olive oil. Oleum Helianthi, Petiolus Trachycarpi oil, Oleum Arachidis hypogaeae semen are also adopted by identical mode and are adulterated.

(2) sample is placed in water bath with thermostatic control, sample is separately heated to 40,50,60,70,80,90 degrees Celsius, after temperature constant, adopt Portable Raman spectrometer collected specimens spectrum.

(3) the data application matlab2011b collected is read, use two-dimensional correlation spectra formula algorithm to generate the two-dimensional correlation spectra of sample;

(4) in conjunction with the characteristic of discrete two-dimensional wavelet transformation and two-dimensional correlation spectra, select best wavelet db5 and the best wavelet decomposition number of plies 3, two-dimensional correlation spectra is carried out 2-d wavelet multi-resolution decomposition;

(5) carry out image reconstruction by decomposing the 2-d wavelet coefficient obtained, respectively obtain the image after 10 groups of 2-d wavelet reconstruct.

(6) spectrum picture reconstructed each group carries out multidimensional offset minimum binary modeling respectively, and obtains other 40 the doping content predictive values of respective sets, and the doping content root-mean-square error of each layer of reconstruct image modeling;

(7) distribute weights according to the doping content root-mean-square error in step (6), use weights that 10 submodels carry out Model Fusion, and calculate final RMSEP value and correlation coefficient carrys out evaluation model prediction effect.

In sum, originally it is embodied as first selecting best 2-d wavelet base that two-dimensional correlation spectra is carried out multi-resolution decomposition and each layer reconstructs respectively; Secondly each layer of correlation spectrum of reconstruct is modeled prediction and obtains the root-mean-square error of cross validation by application NPLS; Then pass through the weights calculated and carry out submodel fusion; Finally by predicted root mean square error and correlation coefficient, result and the performance of multiple dimensioned-two-dimensional correlation spectra model are evaluated. This method is compared to conventional model, it is obviously improved precision and the reliability of normal Raman spectrum analysis model, not only carry and excavated characterization information new in sample spectra, and avoid the loss of information, make Raman spectrum analysis simpler, reliably, it is expected to be widely used in complex system spectrum analysis.

The above is only the preferred embodiment of the present invention; it should be pointed out that, for those skilled in the art, under the premise without departing from the principles of the invention; can also making some improvements and modifications, these improvements and modifications also should be regarded as protection scope of the present invention.

Claims (10)

1. one kind is suitable to the two-dimensional correlation spectra multi-scale Modeling method that olive oil doping identifies, it is characterised in that comprise the steps:

S1, gather the original spectrum of different olive oil sample;

S2, generation step S1 gather the two-dimensional correlation spectra of original spectrum;

S3, characteristic in conjunction with two-dimensional wavelet transformation and two-dimensional correlation spectra, select best wavelet that two-dimensional correlation spectra is carried out 2-d wavelet multi-resolution decomposition, obtain 2-d wavelet coefficient;

S4,2-d wavelet coefficient to step S3 gained carry out image reconstruction;

S5, spectrum picture to each layer of reconstruct carry out multidimensional offset minimum binary modeling respectively, obtain submodel, and corresponding group doping content predictive value, and the doping content root-mean-square error of each layer of reconstruct image modeling;

S6, use weights that the submodel of step S5 gained carries out Model Fusion, and calculate RMSEP value and correlation coefficient carrys out evaluation model prediction effect.

2. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterized in that, the original spectrum gathered in described step S1 is the spectrum that same sample gathers same instrument under condition of different temperatures, and the temperature conditions of change therein needs guarantee identical for different samples.

3. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that the two-dimensional correlation spectra in described step S2 is generated by below equation:

$\mathrm{\Φ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^{T}y\left(v_2\right)$

$\mathrm{\Ψ}(v_1,v_2)=\frac{1}{m-1}y{\left(v_1\right)}^T\mathrm{Ny}\left(v_2\right)$

In formula: y (v) is input spectrum matrix, �� (v1, v2) is the relevant spectrum picture matrix of the synchronization generated, and �� (v1, v2) is the asynchronous correlation spectrum image array generated.

4. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterized in that, in described step S4, reconstruct refers to and each layer of wavelet coefficient after the decomposition of the two-dimensional correlation spectra of same sample is reconstructed respectively.

5. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that the root-mean-square error in described step S5 is RMSECV, and formula is as follows:

$\mathrm{RMSECV}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{{\mathrm{sumC}}_{\mathrm{REF}}^2}}$

In formula: C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

6. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that the RMSEP in described step S6 is predicted root mean square error, below equation obtain:

$\mathrm{RMSEP}=\sqrt{\frac{\mathrm{sum}{(C_{\mathrm{NIR}}-C_{\mathrm{REF}})}^2}{n}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

7. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that the correlation coefficient in described step S6 is R, below equation obtain:

$R=\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}{C}_{\mathrm{REF}}-\frac{\mathrm{\Σ}{C}_{\mathrm{NIR}}\mathrm{\Σ}{C}_{\mathrm{REF}}}{n}}{\sqrt{(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left(\mathrm{\Σ}{C}_{\mathrm{NIR}}\right)}^2}{n})(\mathrm{\Σ}{C_{\mathrm{NIR}}}^2-\frac{{\left(\mathrm{\Σ}{C}_{\mathrm{REF}}\right)}^2}{n})}}$

In formula: n is sample number, C_NIRIt is a certain actual attribute of sample, is generally concentration; C_REFFor the sample properties doped.

8. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterized in that, Model Fusion in described step S6 refers to that each layer of two-dimensional wavelet transformation coefficient is reconstructed image carries out NPLS modeling, is predicted the outcome and predicted root mean square error.

9. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that the weights in described step S6 are obtained by below equation:

$W_I=\frac{{\left(\frac{1}{{\mathrm{RMSECV}}_i}\right)}^2}{{\mathrm{\Σ}}_i^m{\left(\frac{1}{\mathrm{RMSE}{\mathrm{CV}}_i}\right)}^2}$

Wherein, RMSECV_iIt it is the predicted root mean square error after i-th submodel cross validation.

10. a kind of two-dimensional correlation spectra multi-scale Modeling method being suitable to olive oil doping identification according to claim 1, it is characterised in that submodel is merged by described step S6 by below equation:

$C=\underset{i}{\overset{m}{\mathrm{\Σ}}}{W}_i\·C_{\mathrm{iREF}}$

In formula: C_iREFBeing predicting the outcome of submodel, m is the yardstick decomposed, and C is predicting the outcome after Model Fusion, namely final model prediction final result.