Rapid method for the identification of pasteurized milk, ultra-pasteurized milk and formula milk powder

Abstract

The present application relates to a kind of pasteurized milk, high-temperature sterilized milk and formula milk powder fast identification method, which comprises that pasteurized milk, high-temperature sterilized milk and formula milk powder are made into samples to be detected respectively;Under different current frequency, the dielectric properties of the sample to be detected are collected using LCR tester, and dielectric property spectrum data are obtained;Select the dielectric property data of several frequency points, and carry out dimension reduction processing by principal component analysis;After dimension reduction, the data is divided into training set and test set according to the principle of random sampling;On the training set, Gaussian naive bayes algorithm is used to establish the mathematical model for distinguishing pasteurized milk, high-temperature sterilized milk and formula milk powder.The method uses simple LCR measurement, quickly obtains the dielectric property spectrum of milk, the dielectric property spectrum has numerous independent information, can establish milk type identification model under very small sample size, greatly reduces detection time and detection cost.

Description

Technical Field

[0001] This invention relates to the field of dairy product identification technology, specifically to a rapid identification method for pasteurized milk, UHT milk, and formula milk powder. Background Technology

[0002] Milk is an excellent source of nutrition, providing energy and containing abundant essential nutrients, including protein, fat, carbohydrates, sodium, calcium, and vitamins. With its rich nutritional value and low price, milk has become an indispensable food in modern life. Due to the large production volume of milk, various processing methods have been invented to extend its shelf life and meet transportation requirements. Milk sold on the market is mainly divided into three types: UHT milk, pasteurized milk, and formula milk powder.

[0003] With the development of food processing technology, more and more food processors are using milk as the main raw material to produce dairy products. However, some commercial fraud issues exist in the market. Some unscrupulous merchants advertise that they use high-quality, fresh pasteurized milk as raw material, but in reality, they add other types of milk, or even spoiled milk. More seriously, diabetic patients may unknowingly consume formula milk powder that may contain high levels of sugar. Therefore, there is an urgent need for a rapid, simple, and accurate milk testing method for food supervision.

[0004] Previous studies have proposed various methods for detecting and analyzing milk. Nuclear magnetic resonance spectroscopy has demonstrated the impact of high-pressure storage on the chemical composition of pasteurized milk. Chromatography-mass spectrometry (GC-MS) can analyze and distinguish between organic and conventional UHT milk. Labeled peptides and stable isotope techniques can differentiate between UHT and reconstituted milk. Fourier transform infrared spectroscopy and fluorescence spectroscopy can identify the fat content and animal origin of milk. These studies are based on various principles, including physics, biology, chemistry, and spectroscopy, and can analyze the differences between different types of milk. However, their practicality is limited by drawbacks such as high instrument costs, cumbersome and time-consuming sample pretreatment, and the need for large sample sizes to establish identification models. Summary of the Invention

[0005] This invention aims to address at least one of the technical problems existing in current milk testing technologies. Therefore, one objective of this invention is to provide a rapid identification method for pasteurized milk, UHT milk, and formula milk powder. This method overcomes the shortcomings of existing technologies, such as cumbersome sample pretreatment, the need for large sample sizes, and expensive instruments. It utilizes an ultra-small sample size to establish a mathematical model based on dielectric properties and machine learning, enabling a simple, fast, and affordable identification of pasteurized milk, UHT milk, and formula milk powder.

[0006] Specifically, the present invention provides the following technical solution:

[0007] According to embodiments of the present invention, a rapid identification method for pasteurized milk, UHT milk, and formula milk powder is provided, comprising the following steps:

[0008] Pasteurized milk, UHT milk, and formula milk powder were prepared into samples for testing.

[0009] At different current frequencies, the dielectric properties of the sample under test are collected using an LCR meter to obtain dielectric property spectrum data;

[0010] Dielectric property data at several frequency points were selected and dimensionality reduction was performed using principal component analysis.

[0011] The dimensionality-reduced data is divided into training and testing sets according to the principle of random sampling.

[0012] A mathematical model was built on the training set to distinguish between pasteurized milk, UHT milk, and formula milk powder.

[0013]

[0014]

[0015] In equation (1), x i σ represents the feature of the i-th principal component after dimensionality reduction by principal component analysis; ci and μ ci These represent categories y. c (c=0,1,2) characteristic x i The corresponding standard deviation and expected value;

[0016] After calculating the conditional probability of each feature, the posterior probability is maximized using equation (2), and the condition with the highest probability is taken as the identification result; "0" represents pasteurized milk, "1" represents ultra-high temperature sterilized milk, and "2" represents formula milk powder.

[0017] This invention provides a rapid identification method for pasteurized milk, UHT milk, and formula milk powder. The method uses simple LCR measurement to quickly obtain the dielectric property spectrum of milk. The dielectric property spectrum contains a wealth of independent information, enabling the establishment of a milk type identification model with a very small sample size, which greatly reduces detection time and detection cost.

[0018] In addition, the rapid identification method for pasteurized milk, UHT milk, and formula milk powder proposed in the above embodiments of the present invention may also have the following additional technical features:

[0019] Optionally,

[0020]

[0021]

[0022] Optionally, the formula for classifying pasteurized milk, UHT milk, and formula milk powder is (2):

[0023] logP(y0) = log(13 / 37), logP(y1) = log(12 / 37), logP(y2) = log(12 / 37), and let...

[0024]

[0025] Find the feature x respectively i The conditional probability (k) corresponding to y = 0, y = 1, y = 2 c After that, substitute it into formula (2) and compare. The result is selected based on the magnitude of the result, and the largest value is chosen as the species identification result, thereby maximizing the probability calculation result.

[0026] Optionally, the dielectric characteristics include capacitance Cs, capacitive reactance X, loss factor D, impedance Z, loss angle θ, conductance G, and susceptance B.

[0027] Optionally, the training set comprises 70% of the dataset, and the test set comprises 30%.

[0028] Optionally, data from six frequency points—60Hz, 80Hz, 5200Hz, 6700Hz, 10000Hz, and 14000Hz—can be selected and dimensionality reduced using Anaconda software.

[0029] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0030] Figure 1 The LCR tester is used in the experiment in this embodiment of the invention. The test clip of the LCR tester is clamped on two copper plates. The copper plates are fixed on the lid of the container. The copper plates are inserted into the plastic container containing milk. The plastic container is immersed in a constant temperature water bath at 35°C.

[0031] Figure 2 The dielectric characteristic spectrum of the series capacitor Cs of this invention is shown, with the x-axis representing frequency and the y-axis representing the intensity of Cs.

[0032] Figure 3 The confusion matrix of the test set of the optimal model in the embodiments of the present invention;

[0033] Figure 4 This represents the classification boundary of the Gaussian Naive Bayes algorithm in this embodiment of the invention. Detailed Implementation

[0034] The technical solution of the present invention is illustrated below through specific examples. It should be understood that the one or more method steps mentioned in the present invention do not preclude the existence of other method steps before or after the combined steps, or the insertion of other method steps between these explicitly mentioned steps; it should also be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the present invention. Furthermore, unless otherwise stated, the numbering of each method step is merely a convenient tool for identifying each method step, and not for limiting the order of the method steps or defining the scope of the present invention. Changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the present invention.

[0035] To better understand the above technical solutions, exemplary embodiments of the present invention are described in more detail below. While exemplary embodiments of the present invention are shown, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present invention and to fully convey the scope of the invention to those skilled in the art.

[0036] The test materials used in this invention are all common commercial products and can be purchased on the market.

[0037] Instruments, equipment and software:

[0038] The TH2816A LCR instrument from Changzhou Tonghui Electronics Co., Ltd. in China was selected, consisting of two copper electrodes and a plastic container.

[0039] ZSBB-712 water bath manufactured by China Zhicheng Company (operate according to the product instruction manual);

[0040] Anaconda software from Continuum Analytics, a US company;

[0041] SPSS software from International Business Machines, Inc., USA.

[0042] The present invention will now be described with reference to specific embodiments. It should be noted that these embodiments are merely descriptive and do not limit the present invention in any way.

[0043] Example 1: Model Establishment

[0044] (1) Select milk samples

[0045] Eighteen UHT milk samples and eighteen pasteurized milk samples were each 40 mL in size and stored in plastic containers measuring 8 cm x 6 cm x 6 cm, numbered 1–18 and 19–36 respectively. Eighteen formula milk powder samples were mixed at a ratio of 25 g to 150 mL of distilled water, and 40 mL of the resulting solution was poured into plastic containers and placed in a 35°C water bath, numbered 37–54. The plastic containers containing the samples to be tested were sealed with lids secured with copper plates, ensuring good contact between the copper plates and the milk, and placed in a 35°C water bath. Figure 1 (A galvanic cell apparatus can be used for the container). Place the container in a water bath for about 20 minutes. Once the milk sample temperature has stabilized, begin the measurement.

[0046] (2) Dielectric property data acquisition

[0047] The dielectric properties of the sample under test in step (1) were measured and the data were recorded using an LCR meter. Because high-frequency currents easily pass through the sample, the dielectric properties do not change much. Therefore, the number of high-frequency sampling points can be appropriately reduced, and the number of low-frequency sampling points can be appropriately increased. In this experiment, 25 frequencies were selected for data measurement: 50 Hz, 60 Hz, 80 Hz, 130 Hz, 170 Hz, 220 Hz, 280 Hz, 350 Hz, 450 Hz, 580 Hz, 740 Hz, 1000 Hz, 1200 Hz, 1500 Hz, 2000 Hz, 2500 Hz, 3200 Hz, 4200 Hz, 5200 Hz, 6700 Hz, 8500 Hz, 10000 Hz, 14000 Hz, and 20000 Hz.

[0048] (3) Select dielectric property data

[0049] Importing the data recorded in step (1) into SPSS software and performing a one-way ANOVA revealed significant differences in the series capacitance Cs (p = 0.000), capacitive reactance X (p = 0.000), loss coefficient D (p = 0.000), impedance Z (p = 0.000), loss angle θ (p = 0.000), conductance G (p = 0.000), and susceptance B (p = 0.000) of the three types of milk (p < 0.05 indicates a significant difference). Taking the series capacitance as an example, to visually represent and analyze the differences in dielectric properties of the milk, the x-axis represents the frequency, and the y-axis represents the series capacitance value, as shown below. Figure 2 As shown, Cs exhibits a sharp decreasing trend in the range of 50Hz to 2000Hz because, within a certain range, the higher the frequency, the less the capacitor impedes the alternating current. This trend gradually disappears and becomes gradual after reaching the resonant point (approximately 2000Hz). Figure 2The Cs (conductivity) of the formula milk powder is significantly higher than that of the other two types of milk. The formula C = εS / 4πkd indicates that Cs is mainly affected by the dielectric constant, which is a parameter related to the degree of polarization. Under the condition that the contact area and electrode spacing remain constant, the value of the dielectric constant is directly proportional to the value of Cs. There is a positive correlation between the dielectric constant and conductivity of a liquid; the strength of conductivity depends on the content of metal ions, such as K+, Na+, and Ca+. Formula milk powder usually contains K+, Na+, and Ca+, which can increase the polarity of the liquid, so its conductivity is significantly higher than that of the other two types of milk, making its Cs unique. The main reason for the difference between pasteurized milk and UHT milk is the difference in production process and storage time. During the production process, UHT milk is produced at a higher temperature and stored for a longer time than pasteurized milk, causing the fat globules to deform and the casein micelles to rupture, thus leading to… Figure 2 Medium- and high-temperature sterilized milk has a higher Cs content.

[0050] Among the significantly different dielectric properties Cs (p = 0.000), capacitive reactance X (p = 0.000), loss coefficient D (p = 0.000), impedance Z (p = 0.000), loss angle θ (p = 0.000), conductance G (p = 0.000), and susceptance B (p = 0.000), data from two low-frequency, two mid-frequency, and two high-frequency frequencies were selected as modeling data. Specifically, data from six frequency points—60 Hz, 80 Hz, 5200 Hz, 6700 Hz, 10000 Hz, and 14000 Hz—were chosen.

[0051] (4) Dimensionality reduction of data

[0052] The dielectric property data selected in step (3) is used to reduce the principal component analysis to 4 dimensions using Anaconda software to prevent overfitting of the modeling results.

[0053] (5) The data after dimensionality reduction in step (4) is used as the modeling dataset and divided into training set and test set according to the principle of random sampling. The two sets account for 70% and 30% of the dataset, respectively.

[0054] (6) Establishment of the identification model

[0055] Partial Least Squares (PLS-DA), Gaussian Naive Bayes (GNB), and Support Vector Machine (SVM) algorithms were combined with principal component analysis to build classification models using training set data and to predict samples in the test set. The modeling results of PLS-DA, GNB, and SVM algorithms under different preprocessing conditions are shown in Table 1.

[0056] Table 1. Modeling results of GNB, SVM, and PLS-DA with and without data dimensionality reduction.

[0057]

[0058]

[0059] (7) Selection and determination of the optimal model

[0060] Using the dielectric properties of the training set as input and the categories of pasteurized milk, UHT milk, and formula milk powder as output, the segmented dataset is fed into the Gaussian Naive Bayes algorithm for training. In this discriminative model, accuracy is the probability of a correct judgment out of all judgments, and a value closer to 1 is better.

[0061] As shown in Table 1, all models achieved excellent results in the classification tests of pasteurized milk, UHT milk, and formula milk powder. Among them, Gaussian Naive Bayes is particularly effective for multi-class classification tasks, especially for data with a Gaussian distribution and a large amount of independent information. It performs best with small sample sizes, achieving a test set accuracy of 1, and the training and test set R0 values ​​are also high. 2 The values ​​reached 0.808 and 0.748, therefore the Gaussian Naive Bayes was selected as the optimal model for this embodiment, and its mathematical model is as follows:

[0062]

[0063]

[0064] (1) In the formula, x i This represents the i-th principal component feature after dimensionality reduction through principal component analysis.

[0065] σ ci and μ ci These represent categories y. c (c=0,1,2) characteristic x i The corresponding standard deviation and expectation are calculated. After the conditional probability of each feature is obtained, the posterior probability is maximized by equation (2). The condition with the highest probability is taken as the identification result. "0" represents pasteurized milk, "1" represents ultra-high temperature sterilized milk, and "2" represents formula milk powder.

[0066] Using the selected optimal classification model, predict all samples in the test set (n=16). In the Gaussian Naive Bayes algorithm model, the accuracy on data without PCA dimensionality reduction is 87.5%, and the R-values ​​for both the training and test sets are [missing data]. 2 The values ​​were 1 and -48.938 respectively, indicating that the model was also overfitting. Therefore, PCA was used to reduce the dimensionality of the data, and the results showed that the accuracy could reach 100%, with R... 2 =0.808 (training set) and 0.748 (test set) are used as the confusion matrix to measure the model's performance on the test set, and the results are as follows. Figure 3As shown, the horizontal axis represents the predicted label, and the vertical axis represents the true label. The squares in the matrix where the predicted label and the true label overlap are the correct classifications. "0" represents pasteurized milk, "1" represents UHT milk, and "2" represents formula milk powder. The test results are [[6 0 0][0 6 0][0 0 4]], where the columns represent the test results and the rows represent the actual classifications. As can be seen from the confusion matrix, no sample was misidentified, indicating that the model has a good classification effect on the test set.

[0067] Figure 4 To visualize the classification of the test set, the GNB classification boundaries were displayed intuitively, with regions of the same symbol shape representing sample regions of the same category. "0" represents pasteurized milk, "1" represents UHT milk, and "2" represents formula milk powder. Except for a few samples that were misclassified (because only the two most important features from principal component analysis were used as the basis for the visualization results), the vast majority of different types of milk samples were clearly divided into three distinct regions.

[0068] Example 2: Application of the model of the present invention

[0069] Using the dielectric property data acquisition and data dimensionality reduction techniques described in Example 1, 28 samples were measured and processed. The optimal model selected was used for identification, and the results are shown in Table 2.

[0070] Table 2 shows the application results of the model in this embodiment.

[0071]

[0072]

[0073] In summary, this invention addresses the issues of sample consumption and detection thresholds in existing milk type identification models by establishing a mathematical model based on dielectric properties and machine learning using an extremely small sample size. This provides a method for rapidly identifying pasteurized milk, UHT milk, and formula milk powder. It overcomes the shortcomings of existing technologies, such as cumbersome and time-consuming sample pretreatment, the need for large sample sizes, and expensive instruments, thus benefiting food quality detection and supervision.

[0074] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. The illustrative expressions of the above terms in this specification should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. In addition, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0075] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for rapid discrimination of pasteurized milk, UHT milk and powdered milk formula, characterized by, The method comprises the following steps: Prepare the samples to be tested from pasteurized milk, UHT milk and formula milk powder respectively; Under different current frequencies, the dielectric properties of the samples to be tested are collected by using an LCR tester to obtain dielectric property spectrum data; Select dielectric property data at several frequency points and perform dimension reduction processing by principal component analysis; Divide the dimension-reduced data into a training set and a test set according to the principle of random sampling; Establish a mathematical model for distinguishing pasteurized milk, UHT milk and formula milk powder by using a Gaussian naive Bayes algorithm on the training set: In formula (1), x i represents the i-th principal component feature after dimensionality reduction by principal component analysis, σ ci and μ ci respectively represent the corresponding standard deviation and expectation of the feature x c under the category y i (c=0, 1, 2). After the conditional probability of each feature is calculated, the posterior probability is calculated by formula (2) to obtain the identification result; "0" represents pasteurized milk, "1" represents UHT milk, and "2" represents formula milk powder.

2. The method for rapidly identifying pasteurized milk, UHT milk and formula milk powder according to claim 1, characterized in that, 3. The method for rapid discrimination of pasteurized milk, UHT milk and powdered milk according to claim 1, wherein The formula (2) of the mathematical model for classifying pasteurized milk, UHT milk and formula milk powder: log P(y0) = log(13 / 37), log P(y1) = log(12 / 37), log P(y2) = log(12 / 37), and let The feature x is respectively solved i The corresponding conditional probability (k c ) is brought into formula (2) and compared The result of the largest is selected as the category identification result, so as to obtain the maximized probability calculation result.

4. The method for rapidly discriminating pasteurized milk, UHT milk and powdered milk according to claim 1, wherein The dielectric properties include capacitance Cs, capacitive reactance X, loss coefficient D, impedance Z, loss angle θ, conductance G and susceptance B.

5. The method for rapid discrimination of pasteurized milk, UHT milk and powdered milk according to claim 1, wherein The training set accounts for 70% of the data set, and the test set accounts for 30% of the data set.

6. The method for rapid discrimination of pasteurized milk, UHT milk and powdered milk according to claim 1, wherein Select data at 60 Hz, 80 Hz, 5,200 Hz, 6,700 Hz, 10,000 Hz and 14,000 Hz, and perform dimension reduction processing by principal component analysis using Anaconda software.

Images