CN114239692B - Method and device for identifying fresh milk fat adulteration - Google Patents

Abstract

The invention discloses a method and device for identifying raw milk fat adulteration. The method is: construct a classification model based on support vector machine, train, test and evaluate the classification model to obtain the optimal raw milk fat adulteration discrimination model; collect the chromatogram data of the raw milk sample to be tested, and It is introduced into the optimal raw milk fat adulteration discrimination model to identify the raw milk sample to be tested and determine whether the raw milk to be tested is adulterated with fat. The device includes an optimization module and a judgment module. The invention can efficiently, quickly and simply identify raw milk fat adulteration, thereby achieving the purpose of early online monitoring of milk quality, with high detection efficiency and accurate detection results.

Description

A method and device for identifying adulterated raw milk fat

Technical field

The invention relates to a method and device for identifying raw milk fat adulteration, and belongs to the technical field of food testing.

Background technique

The fat content in raw milk is about 3.1% to 4.0%. Its main components are triglycerides (about 98%), phospholipids, sterols, etc. It is rich in essential fatty acids needed by the human body, linoleic acid, in the form of smaller particles. The fat globules are dispersed in the emulsion and are easily absorbed by the human body. They are a high-quality fat and have a good smell, giving the milk a pleasant aroma. In recent years, as the demand for milk fat has continued to grow, the price of milk fat has been on an upward trend. Some unscrupulous manufacturers see this business opportunity and adulterate it with cheap vegetable oils to reduce production costs and increase profit margins.

Currently, methods used to detect foreign fat in milk fat include determining its physical and chemical properties, detecting the composition of unsaponifiable matter, and identifying water-soluble or non-water-soluble volatile fatty acids. In addition, methods such as thin layer chromatography, gas chromatography, high performance liquid chromatography, and infrared spectroscopy based on their chemical properties are also included. Although these methods have been proven to be effective in detecting milk fat, most detection methods can only guarantee their effectiveness when the adulterant content is high enough, and these usually require more complex sample pre-processing methods and will affect the samples to be tested. Causes irreversible damage. Therefore, there is an urgent need to develop a non-destructive, high-throughput method for detecting raw milk fat adulteration, so as to achieve the purpose of early online monitoring of milk quality.

Contents of the invention

The technical problem to be solved by the present invention is: a method for identifying adulterated raw milk fat.

In order to solve the above technical problems, the present invention provides a method for identifying raw milk fat adulteration, which includes the following steps:

Step 1): Construct a classification model based on support vector machines, train, test and evaluate the classification model to obtain the optimal raw milk fat adulteration discrimination model;

Step 2): Collect the chromatogram data of the raw milk sample to be tested, import it into the optimal raw milk fat adulteration discrimination model, identify the raw milk sample to be tested, and determine the raw milk sample to be tested. Is milk adulterated with fat?

Preferably, the construction method of the classification model in step 1) is as follows: using a fast gas-phase electronic nose to detect different raw milk samples to obtain chromatograms of different raw milk samples. The peak areas at different retention times are The independent variables and the types of different raw milk samples are the dependent variables to construct a data set of different raw milk samples. The data set is preprocessed and the preprocessed data set is divided into a training set and a test set according to a predetermined ratio. .

More preferably, the classification model is a support vector machine (SVM) classification model built using Python language. The input vector is mapped to a high-dimensional feature space through a nonlinear mapping function. The nonlinear mapping function is: k( _xi ,x )=exp(-γ{|xx _i |}) ² , where γ is the kernel function parameter, _xi is the i-th sample vector, and x is the support vector.

Further, the optimal regression hyperplane is found in the high-dimensional feature space to minimize the target loss function. The expression of the optimal regression hyperplane is:

It is determined by solving the following quadratic convex programming problem and b:

The constraints are:

In the above formula: is the weight coefficient matrix; b is the threshold;/> is the mapping function; x is the support vector; ξ _i ,/> is the slack variable; C is the penalty factor, C>0; ε is the allowable error; x _i is the i-th sample vector; y _i is the output value of x _i ;/> is the loss value of the i-th sample vector.

Preferably, the evaluation method in step 1) is to evaluate the obtained optimal raw milk fat adulteration discrimination model and use the discrimination accuracy rate Accuracy and F1 score as indicators, where:

In the formula: TP is a positive sample predicted by the model as a positive class, TN is a negative sample predicted by the model as a negative class, FP is a negative sample predicted by the model as a positive class, and FN is a positive sample predicted by the model as a negative class. .

The invention also provides a device for identifying raw milk fat adulteration, which includes:

Optimization module: Construct a classification model based on support vector machines, train, test and evaluate the classification model to obtain the optimal raw milk fat adulteration discrimination model;

Identification module: Collect the chromatogram data of the raw milk sample to be tested, import it into the optimal raw milk fat adulteration discrimination model, identify the raw milk sample to be tested, and determine the raw milk to be tested Whether fat is adulterated.

Each of the above modules is run by a microcontroller.

Preferably, the optimization module uses a fast gas-phase electronic nose to detect different raw milk samples to obtain chromatograms of different raw milk samples. The peak areas of different retention times are used as independent variables. A data set of different raw milk samples is constructed with the category as the dependent variable, the data set is preprocessed, and the preprocessed data set is divided into a training set and a test set according to a predetermined ratio.

More preferably, the optimization module uses the support vector machine SVM classification model built in Python language to map the input vector to the high-dimensional feature space through a nonlinear mapping function. The nonlinear mapping function is: k( _xi ,x) =exp(-γ{|xx _i |}) ² , where γ is the kernel function parameter, _xi is the i-th sample vector, and x is the support vector.

Further, the optimization module searches for the optimal regression hyperplane in the high-dimensional feature space to minimize the target loss function. The expression of the optimal regression hyperplane is:

It is determined by solving the following quadratic convex programming problem and b:

The constraints are:

In the above formula: is the weight coefficient matrix; b is the threshold;/> is the mapping function; x is the support vector; ξ _i ,/> All are slack variables; C is the penalty factor, C>0; ε is the allowable error; x _i is the i-th sample vector; y _i is the output value of x _i ;/> is the loss value of the i-th sample vector.

Preferably, it also includes an evaluation module, which is used to evaluate the optimal raw milk fat adulteration discrimination model using the accuracy of discrimination and F1 score as indicators, and the accuracy of discrimination and F1 score as indicators, in:

In the formula: TP is a positive sample predicted by the model as a positive class, TN is a negative sample predicted by the model as a negative class, FP is a negative sample predicted by the model as a positive class, and FN is a positive sample predicted by the model as a negative class. .

The present invention obtains the optimal raw milk fat adulteration discrimination model by creating and training a classification model based on a support vector machine; then collects the chromatogram data of the raw milk sample to be tested and imports it into the optimal raw milk In the fat adulteration discrimination model, the raw milk sample to be tested is identified to determine whether the raw milk to be tested is fat adulterated, which can efficiently, quickly and simply identify the fat adulteration of raw milk, thereby achieving early online monitoring of milk quality. The purpose of monitoring is to ensure high detection efficiency and accurate detection results.

Description of the drawings

Figure 1 is a flow chart of a method for identifying raw milk fat adulteration provided by the present invention;

Figure 2 is a module connection diagram of a device for identifying raw milk fat adulteration provided by the present invention.

Detailed ways

In order to make the present invention more obvious and understandable, preferred embodiments are described in detail below along with the accompanying drawings.

Example 1

As shown in Figure 1, a method for identifying adulteration of raw milk fat provided by the present invention:

S1: Construct a classification model based on support vector machine, train, test and evaluate the classification model to obtain the optimal raw milk fat adulteration discrimination model.

In this step, fresh raw milk samples are first obtained and raw milk samples mixed with different vegetable oils are prepared; then a data set of different raw milk samples is constructed, the data set is preprocessed, and the preprocessed milk samples are preprocessed according to a predetermined ratio. The processed data set is divided into training set and test set. In one embodiment, the predetermined ratio is 7:3.

The preprocessing method uses min-max standardization (normalization), the specific formula:

In the formula: x _i is the eigenvalue of the i-th sample, x _max , x _min , are the maximum value, minimum value and average value of the sample features respectively, and x' is the preprocessed sample feature value.

Preferably, the Python language is used to construct a support vector machine (SVM) classification model, and the input vector is mapped to the high-dimensional feature space through a nonlinear mapping function. The nonlinear mapping function is:

k(x _i ,x)=exp(-γ{|xx _i |}) ²

Among them, γ is the kernel function parameter, _xi is the i-th sample vector, and x is the support vector.

SVM converts the input sample data into a high-dimensional feature space through the nonlinear changes defined by the kernel function, and searches for the linear relationship between the input variables and the output variables in this high-dimensional space. In this embodiment, the radial feature commonly used to deal with nonlinear problems is selected. Basis functions (RBF) serve as mapping functions.

Preferably, the optimal regression hyperplane is found in the high-dimensional feature space to minimize the target loss function. The expression of the optimal regression hyperplane is:

It is determined by solving the following quadratic convex programming problem and b:

The constraints are:

In the above formula: is the weight coefficient matrix; b is the threshold;/> is the mapping function; x is the support vector; ξ _i ,/> All are slack variables; C is the penalty factor, C>0; ε is the allowable error; x _i is the i-th sample vector; y _i is the output value of x _i ;/> is the loss value of the i-th sample vector.

In the process of finding the optimal hyperplane, the smoothness of the functional relationship is ensured by minimizing the sum of squares of the weight coefficients, while allowing an error less than ε.

Preferably, among the determined key parameters, the penalty factor C is 1.3 and the allowable error ε is 0.001.

Preferably, the optimal raw milk fat adulteration discrimination model is evaluated using the discrimination accuracy rate Accuracy and F1 score as indicators, where:

In the formula: TP is a positive sample predicted by the model as a positive class, TN is a negative sample predicted by the model as a negative class, FP is a negative sample predicted by the model as a positive class, and FN is a positive sample predicted by the model as a negative class. .

During the model training process, the 10-fold cross-validation method was used to evaluate the discriminative effect of a specific combination of parameters, and the optimal parameter combination was selected accordingly.

S2: Collect the chromatogram data of the raw milk sample to be tested, import it into the optimal raw milk fat adulteration discrimination model, identify the raw milk sample to be tested, and determine whether the raw milk to be tested is Fat adulteration.

Specifically, a fast gas-phase electronic nose was used to detect different raw milk samples, and chromatograms of different raw milk samples were obtained. The peak areas of different retention times were used as independent variables, and the types of different raw milk samples were used as dependent variables. Construct a data set of different raw milk samples, preprocess the data set, and divide the preprocessed data set into a training set and a test set according to a predetermined ratio.

The model of the fast gas-phase electronic nose is Herales II of French Alpha MOS Company, and the chromatographic column models used are MXT-5 and MXT-1701 respectively; accordingly, the following method is used when constructing the data set: from the model to MXT- The spectra obtained from the chromatographic columns of 5 and MXT-1701 are combined and arranged according to the retention time. Substances with different response value peak areas but similar retention times are regarded as the same substance. The peak areas of different retention times are used as independent variables, and different products are generated. The type of fresh milk sample creates a data set for the dependent variable.

The conditions for using rapid gas phase electronic nose detection are:

Sample volume: 5g; sample incubation temperature: 50℃; sample incubation time: 20min; injection volume: 5000μL; injection speed: 125μL/s; injection method: headspace injection; Tenax trap collection temperature: 40℃; Tenax trap collection time: 50s; carrier gas: hydrogen; split flow: 10mL/min; sampler temperature: 200℃; temperature rising program: 80℃ constant temperature for 0s, 3℃/s temperature rise to 250℃, 250℃ constant temperature for 21s; detection Device temperature: 260℃; FID gain: FID1/FID2.

A specific experimental process of this embodiment is given below.

Simulation of adulterated raw milk samples:

Take an appropriate amount of skim milk sample in a beaker, add different vegetable oil samples (3.1%, w/w) singly, stir thoroughly and homogenize to obtain a uniform milk sample mixed with different vegetable oils.

Rapid gas phase electronic nose detection:

Accurately weigh 5g of the sample to be tested into a 20mL sample bottle, and then place it in an orderly manner on the sample rack that comes with the instrument for its robotic arm to detect the sample in an orderly and accurate manner. Set the sampling sequence through the software and use a fast gas phase electronic nose. The volatile compounds of the sample to be tested are detected under the following conditions: the sample bottle is closed with a leak-proof cap and covered with a silicone/PTFE septum. The sample was incubated at 50°C for 20 min, then the autosampler injected 5000 μL of sample from the headspace into the GC at a rate of 125 μL/s, and the analytes were collected in a Tenax trap at 40°C for 50 s. After rapid heating, the analytes are separated and transferred to two parallel short GC columns: non-polar column (MXT-5: 5% biphenyl, 95% methylpolysiloxane, 10m × 0.180mm × 0.4 μm) and weakly polar chromatography column (MXT-1701: 14% cyanopropyl-phenyl, 86% methylpolysiloxane, 10m×0.180mm×0.4μm). Hydrogen was used as carrier gas. The system was operated at a constant pressure of 80 kPa, and the head-column split flow rate was 10 mL/min. The temperature conditions are: the sampler temperature is 200°C; the temperature rising program includes constant temperature at 80°C for 0s, 3°C/s heating to 250°C, constant temperature at 250°C for 21s; 260°C flame ionization detection (FID1/FID2). Each sample was tested in six replicates to obtain better parallelism and model performance.

Data preprocessing and data set creation:

The spectra obtained from the two fast chromatography columns were combined and arranged according to the retention time. Substances with different response values but similar retention times were regarded as the same substance. Using the peak areas of different retention times as independent variables, the values of different raw milk samples were Category creates a data set for the dependent variable, preprocesses the data set, and then randomly divides the data set into a training set and a test set in a ratio of 7:3.

Establishment of discriminant model:

Use the Python language to build a support vector machine (SVM) classification model on the Pycharm (version: 2021.2.1) platform. SVM is based on the principle of structural risk minimization and maps the input vector to a high-dimensional feature space through nonlinear mapping to find in this space The optimal regression hyperplane minimizes the target loss function.

SVM modeling results:

The key parameters determined are: penalty factor C is 1.3, allowable error ε is 0.001 and kernel function parameter γ is RBF. The optimized SVM discriminant model has a discrimination accuracy of 95.65% on the test set and an F1 score of 0.9778. The results show that the model has excellent discriminative performance.

Application of the raw milk fat adulteration identification model:

Randomly prepare 30 raw milk samples containing different vegetable oils, use an electronic nose to detect them, obtain a blind sample data set, and import it into the SVM discrimination built in the early stage to identify the types of raw milk samples. The results show that the SVM discrimination model The discrimination accuracy on the blind sample data set is 93.47%, and the F1 score is 0.9523.

This example uses a fast gas-phase electronic nose (FGC E-nose) combined with chemometrics to quickly identify raw milk fat adulteration without complicated sample pre-processing steps. The determination process is simple and fast, and has good practical application value. ; This embodiment can perform non-destructive, high-throughput identification of whether raw milk is adulterated with vegetable oil and its type, and can be used for rapid detection of fat adulteration in raw milk, providing a reference for raw milk quality control in the dairy industry.

Example 2

As shown in Figure 2, a device for identifying raw milk fat adulteration provided by the present invention includes:

An optimization module, used to construct a classification model based on support vector machines, train, test and evaluate the classification model to obtain an optimal raw milk fat adulteration discrimination model;

The identification module is used to collect the chromatogram data of the raw milk sample to be tested, import it into the optimal raw milk fat adulteration discrimination model, identify the raw milk sample to be tested, and determine the raw milk sample to be tested. Is fresh milk adulterated with fat?

Each of the above modules is run by a microcontroller.

Preferably, the optimization module is used for:

A rapid gas-phase electronic nose was used to detect different raw milk samples, and the chromatograms of different raw milk samples were obtained. The peak areas of different retention times were used as independent variables, and the types of different raw milk samples were used as dependent variables to construct different biomass. A data set of fresh milk samples is prepared, and the data set is preprocessed, and the preprocessed data set is divided into a training set and a test set according to a predetermined ratio.

Preferably, the optimization module is used for:

Use Python language to build a support vector machine SVM classification model, and map the input vector to a high-dimensional feature space through a nonlinear mapping function. The nonlinear mapping function is:

k(x _i ,x)=exp(-γ{|xx _i |}) ²

Among them, γ is the kernel function parameter, _xi is the i-th sample vector, and x is the support vector.

Preferably, the optimization module is used for:

Find the optimal regression hyperplane in the high-dimensional feature space to minimize the target loss function. The expression of the optimal regression hyperplane is:

It is determined by solving the following quadratic convex programming problem and b:

The constraints are:

In the above formula: is the weight coefficient matrix; b is the threshold;/> is the mapping function; x is the support vector; ξ _i ,/> All are slack variables; C is the penalty factor, C>0; ε is the allowable error; x _i is the i-th sample vector; y _i is the output value of x _i ;/> is the loss value of the i-th sample vector.

Preferably, the device further includes an evaluation module, the evaluation module is used for:

The optimal raw milk fat adulteration discrimination model was evaluated using the discrimination accuracy rate Accuracy and F1 score as indicators, where:

In the formula: TP is a positive sample predicted by the model as a positive class, TN is a negative sample predicted by the model as a negative class, FP is a negative sample predicted by the model as a positive class, and FN is a positive sample predicted by the model as a negative class. .

The specific implementation process of the functions implemented by each module in Embodiment 2 is the same as the implementation process of each step in Embodiment 1, and will not be described again here.

Claims (2)

1. A method of identifying a fresh milk fat adulteration comprising the steps of:

step 1): constructing a classification model based on a support vector machine, and training, testing and evaluating the classification model to obtain an optimal fresh milk fat adulteration discrimination model;

the construction method of the classification model comprises the following steps: detecting different raw milk samples by adopting a rapid gas-phase electronic nose to obtain chromatograms of the different raw milk samples, constructing data sets of the different raw milk samples by taking peak areas of different retention times as independent variables and types of the different raw milk samples as dependent variables, preprocessing the data sets, and dividing the preprocessed data sets into a training set and a testing set according to a preset proportion; the classification model is a support direction constructed by using Python languageThe vector machine SVM classification model maps an input vector to a high-dimensional feature space through a nonlinear mapping function, wherein the nonlinear mapping function is as follows: k (x) _i ,x)＝exp(-γ{|x-x _i |}) ² Wherein, gamma is a kernel function parameter, x _i The i sample vector is the ith sample vector, and x is a support vector; searching an optimal regression hyperplane in the high-dimensional feature space so as to minimize a target loss function, wherein the expression of the optimal regression hyperplane is as follows:

determination by solving the following quadratic convex programming problemAnd b:

the constraint conditions are as follows:

in the above formula:is a weight coefficient matrix; b is a threshold; />Is a mapping function; x is a support vector; zeta type toy _i 、/>Are relaxation variables; c is penalty factor, C>0; epsilon is the tolerance; x is x _i Is the i-th sample vector; y is _i Is x _i Output value of (2)；/>A loss value for the i-th sample vector;

the evaluation method is to evaluate the obtained optimal fresh milk fat adulteration judgment model, and adopts judgment Accuracy Accurcy and F1 fraction as indexes, wherein:

wherein: TP is a positive sample predicted by the model as a positive class, TN is a negative sample predicted by the model as a negative class, FP is a negative sample predicted by the model as a positive class, FN is a positive sample predicted by the model as a negative class;

step 2): and collecting chromatogram data of the fresh milk sample to be detected, introducing the chromatogram data into the optimal fresh milk fat adulteration judging model, identifying the fresh milk sample to be detected, and determining whether the fresh milk to be detected is fat adulterated.

2. An apparatus for identifying the adulteration of fresh milk fat comprising:

and an optimization module: constructing a classification model based on a support vector machine, and training, testing and evaluating the classification model to obtain an optimal fresh milk fat adulteration discrimination model; the optimizing module adopts a rapid gas-phase electronic nose to detect different raw milk samples to obtain chromatograms of the different raw milk samples, takes peak areas of different retention times as independent variables and types of the different raw milk samples as dependent variables to construct data sets of the different raw milk samples, preprocesses the data sets, and divides the preprocessed data sets into training sets and test sets according to a preset proportion; support Vector Machine (SVM) constructed by using Python language by using optimization moduleAnd the classification model maps the input vector to a high-dimensional feature space through a nonlinear mapping function, wherein the nonlinear mapping function is as follows: k (x) _i ,x)＝exp(-γ{|x-x _i |}) ² Wherein, gamma is a kernel function parameter, x _i The i sample vector is the ith sample vector, and x is a support vector; the optimization module searches an optimal regression hyperplane in the high-dimensional feature space so as to minimize a target loss function, and the expression of the optimal regression hyperplane is as follows:

determination by solving the following quadratic convex programming problemAnd b:

the constraint conditions are as follows:

in the above formula:is a weight coefficient matrix; b is a threshold; />Is a mapping function; x is a support vector; />Are relaxation variables; c is penalty factor, C>0; epsilon is the tolerance; x is x _i Is the i-th sample vector; y is _i Is x _i Output value of (2); />A loss value for the i-th sample vector;

and a judging module: collecting chromatogram data of a fresh milk sample to be detected, introducing the chromatogram data into the optimal fresh milk fat adulteration discrimination model, discriminating the fresh milk sample to be detected, and determining whether the fresh milk to be detected is fat adulterated;

and an evaluation module: the method is used for evaluating the obtained optimal fresh milk fat adulteration judgment model by taking the judgment Accuracy Accury and the F1 fraction as indexes, wherein the judgment Accuracy Accury and the F1 fraction are used as indexes, and the method comprises the following steps:

wherein: TP is a positive sample that is model predicted as a positive class, TN is a negative sample that is model predicted as a negative class, FP is a negative sample that is model predicted as a positive class, and FN is a positive sample that is model predicted as a negative class.