CN107543838B - A kind of adulterated magnetic resonance detection method for planting butter cream in dilute cream - Google Patents

Abstract

The present invention provides the nuclear magnetic resonance discrimination methods for planting butter cream adulterated in a kind of dilute cream.Step includes: that (1) collects cream samples, and carries out authenticity examination to sample；(2) adulterated cream samples are prepared using by the cream samples of authenticity examination, acquires sample¹H-NMR；(3) using after data processing¹H-NMR data establish PLS-DA qualitative model and PCA-SVM returns Quantitative Analysis Model；(4) cream samples whether acquiring unknown adulterated¹H-NMR data are detected using qualitative, the quantitative model of foundation, to obtain the qualification result of unknown cream samples., can be in fast qualitative and quantitative detection dilute cream the case where adulterated plant butter cream this invention ensures that the accuracy of modeling sample, technically reliable, it is easy to operate, used model calculating speed is fast, and identification result is accurate, and the quality monitoring for baking goods such as cream cakes provides technical support.

Description

A kind of adulterated magnetic resonance detection method for planting butter cream in dilute cream

Technical field

The invention belongs to field of food detection.Specifically, the present invention relates to the adulterated mirror method for distinguishing of food quality and purposes. More specifically, the present invention relates to the plant butter cream content for planting butter cream and doping whether is adulterated in detection/identification dilute cream Method and purposes.Particularly, it the present invention relates to the use of nuclear magnetic resonance (NMR) in conjunction with offset minimum binary-techniques of discriminant analysis (PLS- DA it) and Principal component analysis-support vector machine (PCA-SVM) Return Law, and then qualitatively and quantitatively detects and plants rouge in dilute cream The method and purposes of cream content.

Background technique

Dilute cream (cream) is the production of processed manufactured fat content 10.0%~80.0% using animal cream as raw material Product [1].Wherein, the dilute cream of the butterfat containing 30%-40% occupies very big specific gravity in consumption, from flowing shape after dismissing Become non-current shape and there is plasticity, therefore also referred to as whipped cream (whip cream) [2], is production decorative cakes etc. The important source material of Western-style bakery.Planting butter cream (non-dairy whip topping) is with edible hydrogenated oil, sweetener It is all very much like in appearance, characteristic and purposes with dilute cream Deng the margarine [3] for primary raw material, but price is remote low In dilute cream.In the application of practical baking goods, on the one hand businessman is to cater to consumer and pursue healthy diet to declare to make at heart It is 100% animal dilute cream, on the other hand, is mostly substituted with dilute cream with butter cream mixture is planted for pursuit interests 100% whipping cream, the additive amount for planting butter cream are even more to lean on artificial estimation.In face of bake market this confusion, in disappear Cream type used in bakery processor (such as cake room) obligated explicit mark is appealed by association.But it is domestic at present related Cream type and adulterated quantitative examination criteria are not yet put into effect in baking goods, the missing of related national standard, also to effectively supervision band Carry out difficulty, and as network bakes the gradually growth of the Western-style bakery consumption such as rise and the cream cake in shop, these Problem will be protruded more and more.

Currently, being the methods of sense organ identification, gas-chromatography to the conventional method of butterfat detection of adulterations, wherein sense organ identification is It is checked by quality of the features such as color, taste, character to butterfat, and gas-chromatography rule is by cream samples The fatty acid that the fat extracted performs the derivatization out is analyzed, to distinguish the quality of butterfat.However, before these methods Manage that cumbersome, detection time is long, while measurement result is affected by human factors big [4-5].

In recent years, also there is scholar using the method for the spectral techniques combination Chemical Measurements such as infrared, Raman, fluorescence for oil Rouge content carries out adulterated quantitative identification research [6-8] in 80% or more butter.Infrared, Raman, fluorescence these spectral techniques are all It is the method by acquiring the Chemical Measurements such as map combination PCA, PLS, KNN of whipped fat, adulterated identification is carried out to butter. But since these spectral techniques are limited by its testing principle, map is to each component accounts energy in the COMPLEX MIXEDs objects system such as butterfat Power is limited, it is difficult to realize and judge the authenticity of modeling sample that there is the false sample of introducing to cause the true of calibration model Property reduce and/or exclude special authentic specimen in modeling sample to cause the universality impaired risks of calibration model.

Meanwhile the traditional chemometrics algorithm such as PCA, PLS, KNN is when handling regression modeling problem, is all with classics Statistics mathematical theory be foundation, be conceived to the basic point of maximum likelihood, it is desirable that " residual sum of squares (RSS) " is minimum, thus usually needs Want training sample number that could reveal [21-24] by exact close to its validity when infinity.But because in real work In to can be obtained sample size often very limited, model overfitting can be made, be that the generalization ability of model is deteriorated.

Nuclear magnetic resonance method has high throughput, weight compared with the traditional analysis such as liquid chromatography, gas chromatography method [9-12] The advantages that existing property is good, easy to operate, and structural information is enriched has inherent advantage [13- to the global analysis of the complex systems such as food 16], have been used in the analysis of various plants oil and other lipid components [18-20].Support vector machines (support vector Machine, SVM) pattern recognition problem to be solved transforms into a quadratic programming optimization problem by method, theoretically protects Globally optimal solution will be obtained by demonstrate,proving it, and technical ability handles nonlinear problem, and effectively can prevent and limit overfitting, particularly suitable In the data processing [25-28] of small sample set.

But food is many kinds of, matrix is complicated, and being collected into from the horse's mouth food samples is a time and effort consuming Thing.Detection method in the prior art may introduce false sample and the authenticity of calibration model is caused to reduce and/or incite somebody to action Special authentic specimen excludes to cause the universality impaired risks of calibration model in modeling sample.

At the same time, these existing spectral method of detection are to be mixed for fat content in 80% or more butter It is assumed that amount identifies research, and fat-extraction step will be passed through.Therefore, in the prior art not with dilute cream and possible work Based on the plant butter cream of adulterated additive, using configuring adulterated butter product close to by the way of actual conditions and open Hair is suitable for the detection method of dilute cream detection.

Summary of the invention

Problems to be solved by the invention

Dilute cream is mixed effectively to solve the problems, such as to bake the existing butter cream that may will plant in the market, the present invention provides One kind being based on nuclear magnetic resonance technique, in conjunction with the adulterated butter for planting butter cream of dilute cream of the PLS-DA and PCA-SVM Return Law Qualitative and quantitative detecting method.This method is easy to operate, and a large amount of samples can be handled in the short time, avoids human error, conclusion It is scientific reliable, the fast slowdown monitoring suitable for a large amount of actual samples.

The solution to the problem

In order to achieve the above object, the present invention collects the dilute cream for baking common brand from monitoring area and plants butter cream sample Product, and prepared based on the sample by authenticity examination close to the adulterated butter used is actually baked, measure it¹H-NMR establishes adulteration qualitative and quantitative model using PLS-DA the and PCA-SVM Return Law, to realize to plant adulterated in dilute cream The fast and accurately qualitative discrimination of butter cream and quantitative detection.

The present invention provides following method and solves the above problems:

The present invention provides a kind of for the adulterated nuclear magnetic resonance discrimination method for planting butter cream of dilute cream comprising following step It is rapid:

(1) standard cream samples and adulterated cream samples are pre-processed；

(2) pretreatment sample obtained in acquisition step (1)¹H-NMR spectrum counts the spectrogram of collected hydrogen spectrum According to processing；

(3) using data obtained in step (2) establish the adulterated cream samples qualitative discrimination model and quantitative mirror Other model；

(4) it verifies the qualitative discrimination model of the adulterated cream samples or quantitatively identifies the reliability of model；

(5) using establish it is qualitative, quantitatively identify model practical cream samples are identified；

Preferably, the standard cream samples are dilute cream and plant butter cream, and the adulterated cream samples are according to certain The butter sample of dilute cream and plant butter cream that ratio is prepared.

Nuclear magnetic resonance discrimination method according to the present invention, wherein the preparation of step (1) the adulterated cream samples walks It suddenly include: after grinding bead is added in the adulterated cream samples, to be put into oscillator and shake, make the adulterated cream in bottle Sample is sufficiently mixed；

Preferably, the preprocess method of step (1) the standard cream samples and/or adulterated cream samples are as follows: weigh 1mL is added in 2mL EP pipe in the standard cream samples of 500mg and/or the adulterated cream samples prepared CDCl₃, it being placed in homogenizer, frequency is put into a centrifuge after being 30Hz homogeneous 40s, under the conditions of 4 DEG C, revolving speed 8000r/min It is centrifuged 10min；The clear liquid obtained after 600 μ L centrifugation is pipetted in 5mm nuclear magnetic tube, it is to be measured；Wherein, the adulterated cream samples The preparation method comprises the following steps: weighing two kinds of standard cream in proportion, the quality summation that each mix ratio selects weighed two kinds of cream is 10g is put into plastic bottle, and the stainless-steel grinding pearl that diameter is 5mm is added, is put into multitube turbula shaker after being sealed with lid 1min is shaken, two in bottle kinds of cream are sufficiently mixed；

Preferably, the specific detection parameters that nuclear magnetic resonance spectroscopy detects in step (2) are as follows: pulse train noesyig1d, inspection Testing temperature is 297K,¹90 ° of pulse width P1 of H are 10.04 μ s, and spectrum width SWH is 6002.40Hz, and centre frequency O1P is 2400.52Hz, pulse delay time D1 are 10s, and incorporation time D8 is 0.01s,¹³90 ° of pulse PCPD2 of decoupling sequence of C are 260 μ s, it is 4 that sky, which sweeps number DS, and scanning times NS is 32；

Preferably, the data processing method in step (2) specifically: measure¹H-NMR map uses Bruker The processing of Topspin3.2 software, transformation points are 64K, and LB 1.00Hz is handled with exponential window function, and baseline and phasing are equal It is carried out using manual mode, TMS is internal standard signal；Treated map MestReNova software, with 0.005 integration segment pair of δ Chemical shift section δ 0.40~8.00 carries out subsection integral, carries out area after the signal in 7.21~7.30 region δ in removal spectrum and returns One change processing obtains sample nuclear magnetic spectrum and converts the exemplary two dimensional matrix to be formed, and wherein each row represents a sample, and each column represents Intensity of the sample in same chemical shift integrates relative value.

Nuclear magnetic resonance discrimination method according to the present invention, the method that wherein step (3) establishes the qualitative discrimination model are PLS-DA。

Nuclear magnetic resonance discrimination method according to the present invention, the method that wherein step (3) establishes the quantitatively identification model are The PCA-SVM Return Law.

Nuclear magnetic resonance discrimination method according to the present invention further includes analysis for the standard cream before step (1) The step of sample authenticity；

Preferably, if sample is there are abnormal component, give up the sample there are abnormal component.

Nuclear magnetic resonance discrimination method according to the present invention, analysis are wrapped for the step of standard cream samples authenticity It includes: acquiring the standard cream samples¹H-NMR obtains dimensional matrix data through data processing, by this dimensional matrix data into Row PCA analysis is considered as doubtful authenticity abnormal sample to the sample beyond 95% confidence interval；It compares the abnormal sample and sets Believe similar cream in section¹H-NMR finds out difference signal peak, in conjunction with 2D-NMR technology, carries out structure solution to difference signal Analysis, to judge doubtful sample with the presence or absence of true sexual abnormality.

Nuclear magnetic resonance discrimination method according to the present invention, wherein acquisition step (2) is described¹After H-NMR, the hydrogen is composed It is converted into two-dimensional matrix, establishes PCA-SVM regression model according to the following steps:

(a) using the subsection integral value of map as independent variable, with the relative amount of dilute cream fat contained in adulterated sample Dependent variable is exported as fitting, is imported in software；The dependent variable is calculated according to formula (1):

Z=aK/ (aK+bJ) ... (1)

Wherein, Z indicates the relative amount of contained dilute cream fat in adulterated sample；A indicates dilute cream sample Commercial goods labels The fat content of mark；B indicates to plant the fat content of butter cream sample Commercial goods labels mark；K indicates dilute cream in adulterated sample Mass percent；J indicates the mass percent that butter cream is planted in adulterated sample；

(b) PCA analysis is carried out to the argument data in step (a), dimension-reduction treatment is carried out to original argument；

(c) use radial basis function for kernel function, using the new characteristic variable that dimensionality reduction obtains as input, with step (a) Obtained in Z value be fitting output dependent variable, establish SVM regression model；

(d) the penalty parameter c value and kernel functional parameter g for the SVM regression model established using the method optimization of cross validation Value, to optimize the obtained c value, g value as model parameter, data set in training step (c) data establishes PCA-SVM Model；

Preferably, the c value is 256, and the g value is 0.0625.

Nuclear magnetic resonance discrimination method according to the present invention, wherein specific to practical cream samples Qualitive test in step (5) Include the following steps:

(1) cream samples to be measured are pre-processed according to step (1)；

(2) it is operated according to step (2), obtained preprocessing solution¹H-NMR；

(3) using the obtained qualitative discrimination model of any one of claim 1-7, identify whether the sample is that sterling is dilute Cream.

Nuclear magnetic resonance discrimination method according to the present invention, which is characterized in that the qualitative mirror of practical cream samples in step (5) Do not further include following steps:

(4) if identifying that the sample is adulterated cream, pass through PCA-SVM regression model as claimed in claim 7, Identify the adulterated amount of plant butter cream of the sample.

Nuclear magnetic resonance discrimination method according to the present invention, wherein the adulterated amount of plant butter cream for identifying the sample includes such as Lower step:

(a) it is described to acquire step (2) in claim 1¹After H-NMR, two-dimensional matrix is converted by hydrogen spectrum, is imported into In the PCA-SVM model established by claim 7, due to:

Z=aK/ (aK+bJ) ... (1)

K+J=1 ... (2)

Wherein, Z indicates the relative amount of contained dilute cream fat in adulterated sample；A indicates dilute cream sample Commercial goods labels The fat content of mark；B indicates to plant the fat content of butter cream sample Commercial goods labels mark；K indicates dilute cream in adulterated sample Mass percent；J indicates the mass percent that butter cream is planted in adulterated sample；

Therefore, (3) K=Zb/ (a+bZ-aZ) ...

A=35%, the b=20% in formula (3) are set, is obtained:

K=4Z/ (7-3Z) ... (4)

The adulterated quantitative analysis results for planting butter cream in dilute cream are calculated by formula (4).

Specifically, of the present invention, illustratively steps are as follows:

1. the authenticity examination of modeling sample

1.1 sample collections: collecting common dilute cream on the market and plants butter cream sample.

1.2 sample pretreatments: weighing the cream samples of about 500mg in 2mL EP pipe, and 1mL CDCl is added₃, it is placed in It is put into a centrifuge after homogeneous 40s (30Hz) in matter device, is centrifuged 10min (8000r/min) under the conditions of 4 DEG C.Pipette 600 μ L centrifugation The clear liquid obtained afterwards is to be measured in 5mm nuclear magnetic tube.

1.3 testing conditions: by machine testing on the sample handled well in 1.2, the hydrogen nuclear magnetic resonance modal data of sample is obtained.Core Magnetic Instrument measuring condition are as follows: pulse train noesyig1d, detection temperature are 297K,¹90 ° of pulse width P1 of H are 10.04 μ s, Spectrum width SWH is 6002.40Hz, and centre frequency O1P is 2400.52Hz, and pulse delay time D1 is 10s, and incorporation time D8 is 0.01s,¹³90 ° of pulse PCPD2 of decoupling sequence of C are 260 μ s, and it is 4 that sky, which sweeps number DS, and scanning times NS is 32.

1.4 data processings and the acquisition of two-dimensional matrix: it measures¹H-NMR spectrum uses at Bruker Topspin3.2 software Reason, transformation points are 64K, and LB 1.00Hz handles with exponential window function, baseline and phasing be all made of manual mode into Row, TMS are internal standard signal (δ 0.00).Treated map MestReNova (version 6.0.1, Spain) software, with δ 0.005 integration segment carries out subsection integral to chemical shift section δ 0.40~8.00, the letter in 7.21~7.30 region δ in removal spectrum Area normalization processing is carried out after number, is obtained sample nuclear magnetic spectrum and is converted the exemplary two dimensional matrix to be formed, wherein each row represents one A sample, intensity of each column representative sample in same chemical shift integrate relative value.

1.5 sample authenticity examinations: data obtained in 1.4 are imported into 11.0 software of SIMCA-p and carry out PCA analysis, choosing Data Standard graduation conversion is carried out with centralization method (Center, Ctr).Sample beyond 95% confidence level is considered as doubtful abnormal sample Product compare the hydrogen spectrogram of sample in the sample and credibility interval, difference signal peak are found out, in conjunction with a variety of 2D-NMR technologies, to difference Xor signal carries out component resolving.According to parsing result, to those because the sample that rotten or type is not inconsistent will be used as abnormal sample It rejects, to those because being wanted containing the sample for allowing the material composition added in the national standards such as food additives and causing analysis abnormal Retain.It will be used to following qualitative, quantitative model foundation by the cream samples of abnormality inspection.

2. the foundation and application of adulteration qualitative model

The preparation of 2.1 adulterated samples:

The preparation of the adulterated sample of the dilute cream of gradient containing different quality, respectively by plant butter cream mass fraction 15%, 20%, 50%, 70%, 100% gradient prepares the adulterated dilute cream sample for planting butter cream, obtains dilute cream content different experiments sample； 6 kinds of laboratory sample are formed with sterling dilute cream again, 10 samples of every kind of gradient, totally 60 samples are spare.

Each quality summation for selecting weighed two kinds of cream is 10g, is put into the plastic bottle of 15ml, and diameter is added and is about The stainless-steel grinding pearl of 5mm, is put into multitube turbula shaker after being sealed with lid and shakes 1min, fills two in bottle kinds of cream Divide mixing.

2.2 data obtain:

60 samples are subjected to pre-treatments according in 1.2, upper machine testing, carries out at data according to 1.4 under the conditions of 1.3 Reason, obtains modeling two-dimensional matrix.

The 2.3 qualitative foundation for distinguishing model:

The two-dimensional matrix obtained in 2.2 is imported into 11.0 software of SIMCA-p, the Y of addition representative sample type becomes Amount, numerical value " 1 " represent the adulterated dilute cream sample for planting butter cream, and numerical value " 2 " represents pure dilute cream sample, select unit variance method (Unitvariance, UV) method converts to data scale.It will be by the pure dilute cream sample of Standard graduation conversion using PLS-DA method The intensity integral relative value matrix of product, slight integral opposite value matrix and the class variable for planting the adulterated dilute cream sample of butter cream Y is fitted, and obtains qualitative model.

The certificate authenticity of 2.4 qualitative models:

By arrangement experiment random repeatedly (n=200) change classified variable Y put in order to obtain it is corresponding different random Contribution rate of accumulative total value (R²) and predictive ability value (Q²), it tests to model validation.Variable Y sequence is changed into model and suitable Sequence has not been changed the obtained R of model²And Q²Regression fit is done between value, if all Q²In R²Under, and Q²Regression straight line Intersection point with y-axis illustrates that qualitative model is reliably effective, can be used in negative semiaxis.

The Qualitive test of 2.5 cream samples:

Dilute cream sample to Qualitive test is operated according to 1.2-1.4 step, obtains sample to be tested nuclear magnetic spectrum The matrix data formed is converted, is conducted into the qualitative discrimination model established to 2.3.The class variable Y of model prediction is 1 When between ± 0.5, cream to be measured is determined to plant the adulterated dilute cream of butter cream；Y determines that cream to be measured is pure dilute at 2 ± 0.5 Cream.

3. the foundation and application of adulterated quantitative model

The adulterated nuclear magnetic resonance method of discrimination for planting butter cream in above-mentioned dilute cream further includes quantitative after being qualitatively judged Determine, the quantitative judgement includes:

The preparation of 3.1 adulterated samples:

Respectively in mass fraction from the adulterated ratio of 5%-95% (being divided into 10% between each point) and 0%, 100% It weighs and plants butter cream and dilute cream in the plastic bottle of 15ml, each quality summation for selecting weighed two kinds of cream is 10g.It puts Enter the stainless-steel grinding pearl that diameter is about 5mm, is put into after being sealed with lid in multitube turbula shaker and shakes 1min, made in bottle Two kinds of cream are sufficiently mixed.

The acquisition of 3.2 training sets and test set sample:

14 dilute creams are divided into two groups at random with randperm function in Matlab, every group of 7 samples；Equally, will 11 plant butter creams are divided into two groups, one group of 6 sample, one group of 5 sample.It takes first group of 7 sample in dilute cream and plants rouge milk First group of 6 sample in oil prepare sample according to the method in 3.1 adulterated sample preparations, and totally 213 samples are as training set (training set) sample；It takes second group of 7 sample in dilute cream and plants second group of 5 sample in butter cream, mixed according to 3.1 Method in false sample preparation prepares sample, and totally 112 samples are as test set (testing set) sample.

The foundation of 3.3 quantitative models:

(1) 213 training set samples in 3.2 are obtained into a sample map conversion according to 1.2~1.4 step operations Subsection integral relative intensity two-dimensional matrix.

(2) using the subsection integral value of map as independent variable, with the relative amount of dilute cream fat contained in adulterated sample Dependent variable is exported as fitting, is imported in Matlab software.Dependent variable is calculated according to formula (1):

Z=aK/ (aK+bJ) (1)

Wherein, Z: the relative amount of contained dilute cream fat in adulterated sample；A: dilute cream sample Commercial goods labels mark Fat content；B: the fat content of butter cream sample Commercial goods labels mark is planted；K indicates the quality percentage of dilute cream in adulterated sample Number；J indicates the mass percent that butter cream is planted in adulterated sample.

(3) PCA analysis is carried out to the argument data in 3.3 (2), obtains the master for explaining original argument 99% degree Ingredient carries out dimension-reduction treatment to original argument.

(4) use radial basis function for kernel function, using the new characteristic variable that dimensionality reduction obtains as input, in 3.3 (2) Obtained Z value is fitting output dependent variable, establishes SVM regression model.

(5) penalty parameter c and kernel functional parameter g for the SVM regression model established using the method optimization of cross validation, with Optimize obtained c, g value trains the data set in 3.3 (4) data, establish quantitative model as model parameter.

The evaluation of 3.4 quantitative models:

Using established quantitative model, predicts 112 samples of test set, obtain nuclear magnetic resonance spectroscopy combination PCA- SVM method predicted value and actual value it is almost the same as a result, 112 sample predicted values and actual value R²For 97.43%, RMSEP It is 5.87%, the prediction effect for verifying model is good.

The quantitative identification of 3.5 cream samples:

Dilute cream sample to quantitative identification is operated according to 1.2-1.4 step, obtains sample to be tested nuclear magnetic spectrum The matrix data formed is converted, is conducted into the quantitative model established to 3.3, contained dilute milk in available sample to be tested The relative amount Z value of oil and fat.Due to

Z=aK/ (aK+bJ) (1)

K+J=1 (2)

Wherein, Z: the relative amount of contained dilute cream fat in sample to be tested；A: the fat content of dilute cream sample；B: it plants The fat content of butter cream sample；K indicates the mass percent of dilute cream in sample to be tested；J indicates to plant rouge milk in sample to be tested The mass percent of oil.

Solving equations obtain:

K=Zb/ (a+bZ-aZ) (3)

Because the fat content for baking common dilute cream on the market is with fat content between 35%~38%, and mostly 35% sample is in the majority；The fat content of butter cream is planted between 18%~23%, and is mostly that 20% sample is in the majority with fat content； So a=35%, b=20% in setting formula 3, obtain:

K=4Z/ (7-3Z) (4)

K value is mass fraction shared by dilute cream in sample to be tested, and (1-K) value is to plant shared by butter cream in sample to be tested The adulterated quantitative analysis for planting butter cream in dilute cream can be obtained by formula 4 in mass fraction.

The effect of invention

The present invention in terms of existing technologies, at least has the advantages that

(1) sample pretreatment is simple, only needs deuterated solvent dissolution, centrifugation, easy to operation, one-time detection can Realize the qualitative and quantitative analysis to sample, it can be achieved that high-throughput sample detection.

(2) the PCA-SVM Return Law is used to plant the adulterated quantitative analysis of butter cream in dilute cream by the present invention, can accelerate model Calculating speed improves the prediction accuracy of small sample quantities model.

(3) in the preferred scheme, present invention finds the adulterated plant butter cream mixtures of dilute cream in qualitative, quantitative model The pretreated necessity prepared, and optimize the pretreating scheme of aforementioned preparation.It is a discovery of the invention that mixed pre-treatment step After carrying out certain optimization, the accuracy of prediction result is significantly improved.Grinding bead is added without in pre-treatment step to be mixed Sample carry out under the premise of sufficiently vibrating, the RMSECV of quantitative model is greater than 7.0%, RMSEP and is greater than 15%, model accuracy Model after substantially less than optimizing.

(4) in the preferred scheme, the present invention is in calculating adulterated sample when the relative amount of contained dilute cream fat, Calculation formula (1) and (4) are introduced in modeling, and then are omitted fatty in the extraction cream that there must be in other methods Step.

(5) in the preferred scheme, the present invention has carried out the authenticity of modeling sample when establishing quantitative detection model It checks, in conjunction with a variety of 2D-NMR technologies, doubtful exceptional sample is confirmed.By carrying out specificity to sample before modeling It checks, can reduce the false sample of introducing and the authenticity of calibration model is caused to reduce and/or exclude special authentic specimen The universality impaired risks of calibration model is caused in modeling sample, and then improves model accuracy in practical applications.

Detailed description of the invention

Fig. 1 qualitative, quantitative model Establishing process figure for adulterated cream.

Fig. 2 is adulteration qualitative, the quantitative analysis flow chart of cream samples to be measured.

Fig. 3 is dilute cream, plants butter cream CDCl₃Extract¹H-NMR map.

Fig. 4 is dilute cream, plants butter cream CDCl₃Extract¹H-NMR map PCA analyzes PC1/PC2 shot chart.

Fig. 5 is the arrangement experiment of PLS-DA adulteration qualitative model.

Fig. 6 is the adulterated quantitative model predicted value of PCA-SVM and actual value correlation.Wherein, (a) is training set data；(b) For test set data.

Fig. 7 is the adulterated predicted value of 21 cream samples to be measured and true value relationship figure.

Specific embodiment

Below with reference to attached drawing various exemplary embodiments, feature and the aspect that the present invention will be described in detail.It is dedicated herein Word " exemplary " mean " be used as example, embodiment or illustrative ".Here as any embodiment illustrated by " exemplary " It should not necessarily be construed as preferred or advantageous over other embodiments.

In addition, in order to better illustrate the present invention, numerous details is given in specific embodiment below. It will be appreciated by those skilled in the art that without certain details, the present invention equally be can be implemented.In other example, Method well known to those skilled in the art, means, equipment and step are not described in detail, in order to highlight master of the invention Purport.

Unless otherwise defined, the technical and scientific term used in the present invention has general in the technical field of the invention The normally understood identical meanings of logical technical staff institute.

Term " dilute cream " (cream) refers to, using animal cream as raw material, it is processed made of fat content 10.0%~ 80.0% product.It becomes non-current shape from flowing shape after dismissing and has plasticity, therefore also referred to as beats Cream (whip cream).

Term " planting butter cream " (non-dairy whip topping) refers to, based on edible hydrogenated oil, sweetener etc. Want the margarine of raw material.

Term " principal component analysis " (PCA) refers to, try for primal variable to be reassembled into one group it is new mutual unrelated Several generalized variables, while can therefrom take out several less generalized variables according to actual needs and reflect as much as possible originally The statistical method of the information of variable.

Term " offset minimum binary " (PLS) refers to that the quadratic sum by minimizing error finds the best letter of one group of data Number matching.Some absolutely not known true value are acquired with most simple method, and enable the sum of square-error for minimum.

Term " support vector machines " (SVM) is built upon the VC dimension theory and Structural risk minization principle of Statistical Learning Theory On the basis of, according to limited sample information in the complexity (i.e. to the study precision of specific training sample) of model and study energy Seek best compromise between power (identifying the ability of arbitrary sample without error), in the hope of obtaining best Generalization Ability.

Embodiment

Embodiment is exemplified below to illustrate the present invention, it will be appreciated by those skilled in the art that the example is merely exemplary Illustrate, and the explanation of non-exclusive.

Test main material

The cream samples (table 1) of 25 parts of different brands, the place of production are acquired in the market, wherein dilute cream (fat content 35.0%~38.0%) 14 parts and 11 parts of plant butter cream (fat content 18%~23%).4 DEG C of dilute cream storages；Plant butter cream- 20 DEG C of storages, are melted using preceding room temperature, avoid multigelation.All samples are used before the shelf-life.

Deuterated chloroform (CDCl₃, deuterated degree: 99.8%, CIL Corp., the U.S.)；Norell 5mm nuclear magnetic tube (U.S. Norell Company).

Table 1 is used for¹25 cream samples information of H-NMR analysis

Number	The place of production	Type	Number	The place of production	Type
						C-1	France	Dilute cream	N-1	TaiWan, China	Plant butter cream
C-2	Italy	Dilute cream	N-2	Chinese Suzhou	Plant butter cream
						C-3	Germany	Dilute cream	N-3	Chinese Tianjin	Plant butter cream
C-4	Chinese Qingdao	Dilute cream	N-4	Chinese Foshan	Plant butter cream
						C-5	New Zealand	Dilute cream	N-5	Chinese Yancheng	Plant butter cream
C-6	France	Dilute cream	N-6	Henan China	Plant butter cream
						C-7	Britain	Dilute cream	N-7	Chinese Suzhou	Plant butter cream
C-8	France	Dilute cream	N-8	Chinese Yancheng	Plant butter cream
						C-9	Denmark	Dilute cream	N-9	South Korea	Plant butter cream
C-10	Germany	Dilute cream	N-10	Chinese Foshan	Plant butter cream
						C-11	Australia	Dilute cream	N-11	Chinese Jiangmen	Plant butter cream
C-12	France	Dilute cream
						C-13	Ireland	Dilute cream
C-14	Italy	Dilute cream

Test key instrument equipment

Bruker AVANCE 600MHZ superconduction fourier transform NMR instrument is (equipped with BBO probe and topspin3.2 Processing software, Bruker company, Switzerland)；XS204 electronic balance (Mettler Toledo company, Switzerland)；Centrifuge 5424R centrifuge (German Eppendorf company)；TARGIN multitube turbula shaker (Beijing Ta Jin Science and Technology Ltd.)； TissueLyser II homogenizer (German Qiagen company).

The preparation of the adulterated sample of embodiment 1

Qualitative model: it is mixed respectively by the gradient preparation for planting butter cream mass fraction 15%, 20%, 50%, 70%, 100% It is planted the dilute cream sample of butter cream, obtains dilute cream content different experiments sample；Laboratory sample 6 is formed with sterling dilute cream again Kind, 10 samples of every kind of gradient, totally 60 samples are spare.

Quantitative model: respectively by mass fraction from 5%-95% (being divided into 10% between each point) and 0%, 100% Adulterated ratio prepare butter sample.

Qualitative, quantitative with each quality summation for selecting weighed two kinds of cream of butter is 10g, is put into the modeling of 15ml Expect in bottle, the stainless-steel grinding pearl that diameter is about 5mm is added, is put into multitube turbula shaker and shakes after being sealed with lid 1min is sufficiently mixed two in bottle kinds of cream.

The preparation of 2 sample solution of embodiment

The cream samples prepared in embodiment 1 about 500mg is weighed in 2mL EP pipe, 1mL CDCl is added₃, it is placed in It is put into a centrifuge after homogeneous 40s (30Hz) in matter device, is centrifuged 10min (8000r/min) under the conditions of 4 DEG C.Pipette 600 μ L centrifugation The clear liquid obtained afterwards is to be measured in 5mm nuclear magnetic tube.

The acquisition of embodiment 3 quantitative model training set and test set sample

14 dilute creams are divided into two groups at random with randperm function in Matlab, every group of 7 samples；Equally, will 11 plant butter creams are divided into two groups, one group of 6 sample, one group of 5 sample.It takes first group of 7 sample in dilute cream and plants rouge milk First group of 6 sample in oil prepare sample according to the method in the adulterated sample preparation of embodiment 1, and totally 213 samples are as training Collect (training set) sample；It takes second group of 7 sample in dilute cream and plants second group of 5 sample in butter cream, according to reality The method applied in the adulterated sample preparation of example 1 prepares sample, and totally 112 samples are as test set (testing set) sample.

Embodiment 4¹The establishment of H-NMR spectrum determination condition

Nuclear Magnetic Resonance 1H carrier frequency is 600.13MHz, uses Bruker standard pulse sequence noesyig1d, detection Temperature is 297K, and 90 ° of pulse width P1 of 1H are 10.04 μ s, and spectrum width SWH is 6002.40Hz, and centre frequency O1P is 2400.52Hz, pulse delay time D1 are 10s, and incorporation time D8 is 0.01s, and 90 ° of pulse PCPD2 of decoupling sequence of 13C are 260 μ s, scanning times NS are 32, and it is 4 that sky, which sweeps number DS,.

5 data processing of embodiment and the acquisition of two-dimensional matrix

It measures¹H-NMR spectrum using Bruker Topspin3.2 software handle, transformation points be 64K, LB 1.00Hz, It is handled with exponential window function, baseline and phasing are all made of manual mode progress, and TMS is internal standard signal (0.00).After processing Map MestReNova (version 6.0.1, Spain) software, with 0.005 integration segment of δ to chemical shift section δ 0.40 ~8.00 carry out subsection integral, carry out area normalization processing after the signal in 7.21~7.30 region δ in removal spectrum, obtain sample Nuclear magnetic spectrum converts the exemplary two dimensional matrix to be formed, and wherein each row represents a sample, and each column representative sample is in same chemical potential Intensity in shifting integrates relative value.

The authenticity examination of 6 modeling sample of embodiment

Be collected into 14 sterling dilute creams and 11 plant butter creams are operated according to embodiment 2,4,5, the data waited until It imports 11.0 software of SIMCA-p and carries out PCA analysis, select centralization method (Center, Ctr) to carry out data Standard graduation conversion, exceed The sample of 95% confidence level is considered as doubtful abnormal sample.By PCA shot chart (Fig. 4) as it can be seen that planting butter cream group occurs one It is significantly away from the abnormal sample of other samples, the chemical component for representing the sample, which plants butter cream sample with other, has conspicuousness poor It is different.It is original by consulting¹H-NMR map simultaneously combines a variety of 2D-NMR maps to find, compared with other plant butter cream sample, this sample The trend of abnormal deviation can thus be shown in product on PCA shot chart containing vanillic aldehyde.Vanillic aldehyde is a kind of with milk fragrance The edible spices of breath are widely used in the food such as cream, candy, therefore, although this sample is in PCA shot chart far from it He plants butter cream sample, but he represents the feature of a kind of plant butter cream sample containing vanillic aldehyde, therefore subsequent quantitative This sample should be retained in adulteration assay, exceptional sample rejecting should not be done.

The foundation of 7 adulteration qualitative model of embodiment, certificate authenticity will obtain 60 adulteration qualitative aggregate samples in embodiment 1 Product are pressedIt is operated according to embodiment 2,4,5, obtained data are imported into 11.0 software of SIMCA-p, add the Y of representative sample type Variable, numerical value " 1 " represent the adulterated dilute cream sample for planting butter cream, and numerical value " 2 " represents pure dilute cream sample, select unit variance Method (Unitvariance, UV) method converts to data scale.It will be by the pure dilute cream sample of Standard graduation conversion using PLS-DA The intensity integral relative value matrix of product, slight integral opposite value matrix and the class variable for planting the adulterated dilute cream sample of butter cream Y is fitted, and obtains qualitative model.When the class variable Y of model prediction is between 1 ± 0.5, cream to be measured is determined to plant rouge milk The adulterated dilute cream of oil；Y determines that cream to be measured is pure dilute cream at 2 ± 0.5.

By arrangement experiment random repeatedly (n=200) change classified variable Y put in order to obtain it is corresponding different random Contribution rate of accumulative total value (R²) and predictive ability value (Q²), it tests to model validation.Variable Y sequence is changed into model and suitable Sequence has not been changed the obtained R of model²And Q²Regression fit is done between value, if all Q²In R²Under, and Q²Regression straight line Intersection point with y-axis illustrates that qualitative model is reliably effective, can be used in negative semiaxis.Fig. 5 shows, established model it is all Q²In R²Under, and Q²Regression straight line and y-axis intersection point in negative semiaxis, illustrate that established qualitative model is reliably effective.

The foundation of the adulterated quantitative model of embodiment 8

By in embodiment 3 using method prepared by adulterated quantitative model involved in embodiment 1 obtain 213 it is adulterated quantitative mixed It closes sample to operate according to embodiment 2,4,5, obtains the subsection integral relative intensity two-dimensional matrix of sample map conversion, according to Following steps establish quantitative model:

(1) using the subsection integral value of map as independent variable, with the relative amount of dilute cream fat contained in adulterated sample Dependent variable is exported as fitting, is imported in Matlab software.Dependent variable is calculated according to formula (1):

Z=aK/ (aK+bJ) (1)

Wherein, Z: the relative amount of contained dilute cream fat in adulterated sample；A: dilute cream sample Commercial goods labels mark Fat content；B: the fat content of butter cream sample Commercial goods labels mark is planted；K indicates the quality percentage of dilute cream in adulterated sample Number；J indicates the mass percent that butter cream is planted in adulterated sample.

(2) PCA analysis is carried out to the argument data in example 8 (1), will obtained to 99% explanation degree of original argument Principal component, i.e., to original argument carry out dimension-reduction treatment, be reduced to 13 from 1520 original input independents variable.

(3) use radial basis function for kernel function, using the new characteristic variable that dimensionality reduction obtains as input, with example 8 (1) Obtained in Z value be fitting output dependent variable, establish SVM regression model.

(4) penalty parameter c and kernel functional parameter g for the SVM regression model established using the method optimization of cross validation, it is excellent Best c value after change is 256, and best g value is 0.0625.To optimize obtained c, g value is as model parameter, 3.3 (4) of training number Data set in establishes PCA-SVM and returns quantitative model.

The evaluation of 9 quantitative model of embodiment

Model-evaluation index: the coefficient of determination of training set validation-cross root-mean-square error (RMSECV) and test set are used R², evaluation index of the predicted root mean square error (RMSEP) as regression model.RMSECV is used to evaluate the feasibility of modeling method And the predictive ability of gained model, for RMSEP for evaluating model built to the predictive ability of external samples, the two values are smaller, Show that the accuracy of model is higher, predictive ability is better；R², it is square of correlation coefficient r, R²Closer to 1, illustrate desired value It is better that correlation is surveyed between predicted value.

By in embodiment 8 (1) to data the method that PLS is returned, SVM is returned be respectively adopted model.It adopts respectively The test set of 112 samples is predicted with the adulterated quantitative model of PLS, SVM and PCA-SVM of foundation, the results are shown in Table 1.Knot Fruit shows, the RMSECV value of SVM model is less than PLS model, and Training R²Value is greater than PLS model, illustrates SVM model Fitting precision is better than PLS model；Meanwhile the RMSEP value of SVM model is less than PLS model, and Testing R²Value is greater than PLS Model illustrates that the predictive ability of SVM model to external sample is higher than PLS model.It is both SVM model, by PCA dimensionality reduction SVM model RMSECV, RMSEP, Testing R²Value is superior to the SVM model of not dimensionality reduction, illustrates that PCA dimensionality reduction can guarantee Under the premise of former variable information is constant, by Data Dimensionality Reduction, so that original multidimensional problem greatly simplifies, when effective less operation Between, improve precision of prediction.

2 PLS, SVM and PCA-SVM Parameters in Mathematical Model of table

The adulteration qualitative of 10 sample to be tested of embodiment, quantitative identification

By it is known whether be it is adulterated plant butter cream and adulterated ratio to be identified 30 part cream samples according to embodiment 2, 4,5 steps are operated, and are obtained sample to be tested nuclear magnetic spectrum and are converted the matrix data to be formed, and are conducted into embodiment 7 and are built Determine in vertical qualitative model.Table 3 is the calculated value of 30 parts of cream samples class variables.Sample of the calculated value between 1 ± 0.5 For adulterated plant butter cream sample, predicted value is dilute cream sample between 2 ± 0.5.According to this decision criteria, 21 adulterated samples Product, 9 dilute cream samples have all obtained correct classification, and identifying accuracy is 100%.

The cream samples Y value table to be measured of table 3

It will be determined as that the corresponding matrix data of adulterated sample is imported into the quantitative model that embodiment 8 is built, due to:

Z=aK/ (aK+bJ) (1)

K+J=1 (2)

Wherein, Z: the relative amount of contained dilute cream fat in sample to be tested；A: the fat content of dilute cream sample；B: it plants The fat content of butter cream sample；K indicates the mass percent of dilute cream in sample to be tested；J indicates to plant rouge milk in sample to be tested The mass percent of oil.

So solving equations obtain:

K=Zb/ (a+bZ-aZ) (3)

Because the fat content for baking common dilute cream on the market is with fat content between 35%~38%, and mostly 35% sample is in the majority；The fat content of butter cream is planted between 18%~23%, and is mostly that 20% sample is in the majority with fat content； So a=35%, b=20% in setting formula 3, obtain:

K=4Z/ (7-3Z) (4)

K value is mass fraction shared by dilute cream in sample to be tested, and (1-K) value is to plant shared by butter cream in sample to be tested The adulterated quantitative analysis results for planting butter cream in dilute cream can be obtained by formula (4) in mass fraction.It is calculated according to formula (4) The adulterated mass fraction of plant butter cream of 21 adulterated cream samples, is as a result shown in Fig. 7.As seen from Figure 7, model predication value and true Real adulterated value is almost the same, the predicted value of 21 samples and the R of actual value²For 97.50%, RMSEP 5.48, illustrate model Prediction effect is good, is able to satisfy the detection accuracy requirement of routine monitoring.

Comparative example 1

In the preparation process of the adulterated sample of embodiment 1, it is added without stainless-steel grinding pearl and sample to be mixed is filled Under the premise of dividing oscillation, the experimentation of embodiment 9 is repeated, establishes model using PCA-SVM homing method.

By evaluating aforementioned quantitative model, it is found that its RMSECV is greater than 7.0%, RMSEP and is greater than 15%, model is quasi- True property is substantially less than the model after vibrating.

Industrial availability

The country is not yet put into effect in relation to cream type in baking goods and adulterated quantitative examination criteria at present, related national standard Missing also brings difficulty to effectively supervision, and as network bakes the Western-style bakeries such as rise and the cream cake in shop The gradually growth of consumption, these problems will be protruded more and more.The present invention has carried out abnormality inspection to modeling sample first, knot The chemical component difference that a variety of 2D-NMR technologies analyze the sample that peels off is closed, is established in dilute cream using the PCA-SVM Return Law Plant the adulterated Quantitative Analysis Model of butter cream, and the performance for the model established with traditional PLS algorithm and simple SVM regression algorithm into Comparison is gone, the predictive ability of the stability of the quantitative model as the result is shown based on PCA-SVM algorithm, accuracy and model is equal Better than PLS, SVM algorithm, the quality monitoring for baking goods such as cream cakes on standard market provides technical support.

The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.

Bibliography

[1] 19646-2010 national food safety standard dilute cream of Ministry of Health of the People's Republic of China .GB, cream and nothing The Beijing water cream: China Standards Press, 2010.

[2] Wu Tao, Wu Hui, Wu Jianfeng wait dilute cream and its apply China Dairy Industry, 2005,33 (2): 44-47.

[3] Ministry of Commerce, People's Republic of China (PRC) .SB/T 10419-2007 plants Beijing butter cream: China Standards Press, 2007.

[4]Krause A,Miracle R,Sanders T,et al.The effect of refrigerated and frozen storage on butter flavor and texture.Journal of Dairy Science,2008,91 (2):455-465.

[5]Gutierrez R,Vega S,Díaz,G,et al.Detection of non-milk fat in milk fat by gas chromatography and linear discriminant analysis.Journal of Dairy Science,2009,92(5):1846-1855.

[6]Ntakatsane M,Liu X,Zhou P.Short communication:rapid detection of milk fat adulteration with vegetable oil by fluorescence spectroscopy.Journal of Dairy Science,2013,96(4):2130-2136.

[7]Heussen P,Janssen H,Samwel I.The use of multivariate modelling of near infra-red spectra to predict the butter fat content of spreads.Analytica Chimica Acta,2007,595(1-2):176-181.

[8]Reyhan S,Ismail H,Hüseyin E,et al.Determination of butter adulteration with margarine using Raman spectroscopy.Food Chemistry,2013,141 (4):4397-4403.

[9] Zhao Haizhen, Lu Zhaoxin, Bie little Mei wait the fatty acid position in cerols from Lard by High Performance Liquid Chromatography Set distribution chromatography, 2005,23 (2): 142-145.

[10] 17377-2008 animal and plant fat of People's Republic of China's national quality supervision and inspection general bureau .GB/T fat Beijing gas chromatographic analysis of sour methyl esters: China Standards Press, 2008.

[11]ZOU X Q,HUANG J H,JIN Q Z,et al.Lipid composition analysis of milk fats from different mammalian species:potential for use as human milk fat substitutes.Journal of Agricultural Food Chemistry,2013,61:7070-7080.

[12] Pan Yushi, Re Naguli wood are husky, Chen Yue, wait a .GC-MS analysis Xinjiang Altay fine breed of chicken with thick brownish feathers and Nanjing chicken cock skin Fatty acid forms Food Science, 2015,35 (16): 122-126. in lipid

[13] Li Aiping, Li Zhenyu, Xing Jie wait nuclear magnetic resonance metabonomic technology to detect vinegar chemical component food section It learns, 2013,34 (12): 247-253.

[14] Yan Zhengli, Yang Mingsheng, Li Tianbao wait glucose in honey, fructose and sucrose NMR Quantitative Study food Product science, 2009,30 (14): 253-255.

[15]MARIA R M,COLNAGO L A,FORATO L A,et al.Fast and simple nuclear magnetic resonance method to measure conjugated linoleic acid in beef.Journal of Agricultural Food Chemistry,2010,58(11):6562-6564.

[16]SOPELANA P,ARIZABALETA I,IBARGOITIA M,et al.Characterisation of the lipidic components of margarines by 1H nuclear magnetic resonance.Food Chemistry,2013(141):3357–3364.

[17]ANGELOS P,APOSTOLOS S,et al.Classification of edible oils by employing 31P and 1H NMR spectroscopy in combination with multivariate statistical analysis.A proposal for the detection of seed oil adulteration in virgin olive oils.Journal of Agricultural Food Chemistry,2003,51:5715-5722.

[18]MANNINA L,PATUMI M,PROIETTI N,et al.Geographical characterization of Italian extra virgin olive oils using high-field 1H-NMR spectroscopy.Journal of Agricultural Food Chemistry,2001,49:2687-2696.

[19]ELISABETTA S,GABRIELLA P,GIULIO C,et al.Identification of production chain of Asiagod’Allevo cheese by nuclear magnetic resonance spectroscopy and principal component analysis.Journal of Agricultural Food Chemistry,2008,56:7208-7214.

[20] Li Wei, Jiang Jie, Lu Yong wait the content food section of total conjugated linoleic acid in .NMR hydrogen spectrum quantitative determination cheese It learns, 2015,36 (10): 134-138.

[21]Blanco M,Coello J,Iturriaga H,et al.Calibration in non-linear near infrared reflectance spectroscopy:a comparison of several methods[J] .AnalyticaChimicaActa.1999,384(2):207-213.

[22]Despagne F,Massart D L,Chabot P.Anal.Chem.,Development of a Robust Calibration Model for Nonlinear In-Line Process Data[J].Analytical Chemistry.2000,72:1657-1665.

[23]Blanco M,Pages J.Classification and quantitation of finishing oils by near infrared spectroscopy.Analytica Chimica Acta.2002,463(2):295- 303.

[24] Wang Chunyan, Shi Xiaofeng, Li Wendong wait to be based on principal component and support vector machines concentration parameter synchronous fluorescence light Spectrum oil kind identifies analysis test journal, 2014,33 (3): 289-294.

[25] Gu Yaxiang, Ding Shifei support vector machines progress computer science, 2011,38 (2): 14-17.

[26] Mei Hu, Liang Guizhao, Zhou Yuan support vector machines are used for the research Science Bulletin of quantitative structure activity relationship modeling, 2005,24(6):1703-1711.

[27] Liu Xu, Chen Huacai, Liu Taiang wait .PCA-SVR combination algorithm in near-infrared spectrum analysis tobacco component Using spectroscopy and spectrum analysis, 2007,27 (12): 2460-2463.

[28] yellow rice wine of Fourier Transform Near Infrared of Jiang Shiquan, Zhou Xingcai, the Jiang Shiping based on PCA and LS-SVM The spectrographic laboratory identification model wine storage time, 2012,29 (2): 806-811.

Claims (7)

1. a kind of for the adulterated nuclear magnetic resonance discrimination method for planting butter cream of dilute cream, which is characterized in that it includes the following steps:

(1) standard cream samples and adulterated cream samples are pre-processed；

(2) pretreatment sample obtained in acquisition step (1)¹H-NMR spectrum, to collected¹The spectrogram of H-NMR spectrum is counted According to processing；

(3) it establishes the qualitative discrimination model of the adulterated cream samples using data obtained in step (2) and quantitatively identifies mould Type, wherein the method for establishing the qualitative discrimination model is PLS-DA, obtains PLS-DA model；It establishes and described quantitatively identifies mould The method of type is the PCA-SVM Return Law, obtains PCA-SVM regression model；

(4) it verifies the qualitative discrimination model of the adulterated cream samples or quantitatively identifies the reliability of model；

(5) using establish it is qualitative, quantitatively identify model practical cream samples are identified；

Before step (1), further include the steps that analysis for the standard cream samples authenticity；If sample exist it is abnormal at Point, then give up the sample there are abnormal component；

Wherein, acquisition step (2) is described¹It, will be described after H-NMR spectrum¹H-NMR spectrum is converted into two-dimensional matrix, builds according to the following steps Found the PCA-SVM regression model:

(a) using the subsection integral value of map as independent variable, using in adulterated sample the relative amount of contained dilute cream fat as Fitting output dependent variable, imports in software；The dependent variable is calculated according to formula (1):

Z=aK/ (aK+bJ) ... (1)

Wherein, Z indicates the relative amount of contained dilute cream fat in adulterated sample；A indicates dilute cream sample Commercial goods labels mark Fat content；B indicates to plant the fat content of butter cream sample Commercial goods labels mark；K indicates the matter of dilute cream in adulterated sample Measure percentage；J indicates the mass percent that butter cream is planted in adulterated sample；

(b) PCA analysis is carried out to the argument data in step (a), dimension-reduction treatment is carried out to original argument；

(c) use radial basis function for kernel function, using the new characteristic variable that dimensionality reduction obtains as input, to obtain in step (a) The Z value arrived is fitting output dependent variable, establishes SVM regression model；

(d) the penalty parameter c value and kernel functional parameter g value for the SVM regression model established using the method optimization of cross validation, with Optimize the obtained c value, g value as model parameter, the data set in training step (c) data is established the PCA-SVM times Return model.

2. nuclear magnetic resonance discrimination method as described in claim 1, which is characterized in that step (1) the adulterated cream samples Preparation step includes: after grinding bead is added in the adulterated cream samples, to be put into oscillator and shake, and make in bottle described mixes False cream samples are sufficiently mixed.

3. nuclear magnetic resonance discrimination method as described in claim 1, analysis is directed to the step of the standard cream samples authenticity It suddenly include: the acquisition standard cream samples¹H-NMR spectrum, obtains dimensional matrix data through data processing, by this two-dimensional matrix Data carry out PCA analysis, are considered as doubtful authenticity abnormal sample to the sample beyond 95% confidence interval；Compare the abnormal sample Similar cream in product and confidence interval¹H-NMR spectrum, finds out difference signal peak, in conjunction with 2D-NMR technology, carries out to difference signal Structure elucidation, to judge doubtful sample with the presence or absence of true sexual abnormality.

4. nuclear magnetic resonance discrimination method as described in claim 1, which is characterized in that fixed to practical cream samples in step (5) Property identify specifically comprise the following steps:

(1) cream samples to be measured are pre-processed according to step (1)；

(2) it is operated according to step (2), obtained preprocessing solution¹H-NMR spectrum；

(3) the PLS-DA model is utilized, identifies whether the sample is sterling dilute cream.

5. nuclear magnetic resonance discrimination method as claimed in claim 4, which is characterized in that fixed to practical cream samples in step (5) It further includes following steps that amount, which identifies:

(4) if identifying that the sample is adulterated cream, pass through the PCA-SVM regression model, identify the plant of the sample The adulterated amount of butter cream.

6. nuclear magnetic resonance discrimination method as claimed in claim 5, wherein the adulterated amount of plant butter cream for identifying the sample includes Following steps:

(a) acquisition step (2) is described¹It, will be described after H-NMR spectrum¹H-NMR spectrum is converted into two-dimensional matrix, imported into the PCA- In SVM regression model, due to:

Z=aK/ (aK+bJ) ... (1)

K+J=1 ... (2)

Wherein, Z indicates the relative amount of contained dilute cream fat in adulterated sample；A indicates dilute cream sample Commercial goods labels mark Fat content；B indicates to plant the fat content of butter cream sample Commercial goods labels mark；K indicates the matter of dilute cream in adulterated sample Measure percentage；J indicates the mass percent that butter cream is planted in adulterated sample；

Therefore, (3) K=Zb/ (a+bZ-aZ) ...

A=35%, the b=20% in formula (3) are set, is obtained:

K=4Z/ (7-3Z) ... (4)

The adulterated quantitative analysis results for planting butter cream in dilute cream are calculated by formula (4).

7. nuclear magnetic resonance discrimination method as described in claim 1, wherein the standard cream samples are dilute cream and plant rouge milk Oil, the adulterated cream samples are the dilute cream prepared according to a certain percentage and the butter sample for planting butter cream.