Non-directional identification method for authenticity of egg white powder
Through the LOF non-directional identification model based on near-infrared spectroscopy, the problem of difficult to quickly and accurately identify unknown dopants in egg white powder in the prior art is solved, and efficient and accurate dopant detection is achieved, which is suitable for authenticity identification of egg white powder with a wide variety of doped types on the market.
Patent Information
- Application Number
- CN202411893941.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-20
- Publication Date
- 2025-05-02
AI Technical Summary
The prior art is difficult to quickly and accurately identify unknown dopants in egg white powder, and the detection range of directional identification models is limited and cannot effectively cover all possible dopants.
A non-directional identification model based on near-infrared spectroscopy is adopted to establish a detection system that can identify various dopants in general through spectral data acquisition, preprocessing, feature wavelength screening and model establishment.
The rapid and accurate identification of egg white powder dopants is achieved, with the detection rate reaching 93.57%, the detection rate of egg white powder with doping concentration exceeding 15% can reach 100%, and the minimum detection limit for multi-component doping species can reach 0.1%.
Smart Images
Figure CN119920349A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the field of non-destructive detection of agricultural products, and in particular relates to a non-directional identification method for the authenticity of egg white powder. Background Art
[0002] Egg powder products include whole egg powder, egg yolk powder, egg white powder, etc. Because its nutritional ingredients are suitable for human needs, it has high-quality protein, balanced minerals and vitamins and is easy to carry, it has become an ideal substitute for eggs and has been widely used in food production, cosmetics industry, medical manufacturing and other fields. Among them, the protein content of egg white powder can be as high as 80%, and it has the characteristics of low fat and low carbon water, which is particularly suitable for people in the growth period, the elderly or sports. At present, in the process of egg powder production, some companies add cheap plant protein powder or flour, etc., which not only disrupts the order of the egg powder market, but also brings great hidden dangers to food safety. Compared with whole egg powder and egg yolk powder, egg white powder has a higher protein content, is more expensive on the market, and its adulteration is more serious. Therefore, studying the authenticity identification technology of egg white powder is of great significance to ensure the quality and safety of egg powder.
[0003] At present, the authenticity identification of egg white powder mainly includes sensory recognition method, biomolecular technology based on DNA molecule and protein molecule detection, and spectral detection technology. Among them, sensory recognition method and biomolecular technology based on DNA molecule and protein molecule detection have the problems of strong subjectivity, high cost, complex operation and long time consumption, and cannot meet the requirements of rapid detection. At present, the modeling method of food adulteration identification based on near-infrared spectroscopy is still a directional identification model, which can only effectively detect specific adulterants. However, there are many adulterants in egg white powder on the market, and the directional identification model can detect limited types of adulteration, which cannot effectively cover all possible adulterants. There is room for improvement in its detection range. Faced with the numerous unknown adulterants in the egg powder market, it is urgent to develop a fast, accurate and universal detection method. Summary of the invention
[0004] In order to solve the above problems, the present invention aims to provide a non-directional identification method for the authenticity of egg white powder.
[0005] The technical solution of the present invention is as follows: A non-directional identification method for the authenticity of egg white powder, the method comprising the following steps: S1: Collect egg white powder samples; S2: Spectral data acquisition; S3: Sample division and expansion; S4: spectral preprocessing; S5: spectral characteristic wavelength screening; S6: Modeling; S7: Model evaluation resulted in a non-directional identification model for the authenticity of egg white powder.
[0006] Preferably, the egg white powder sample in step S1 includes a pure egg white powder sample and an impure egg white powder sample; There are 2800-3200 pure egg white powder samples; the protein content of pure egg white powder samples is ≥78.0%, and the moisture content is ≤9.0%; The doped egg white powder samples were prepared by selecting 7 kinds of extender dopants and nitrogen-rich compounds, including starch, soy protein isolate, melamine, urea, glycine, maltodextrin and dicyanic acid; the doping of egg white powder was divided into single-component doping and multi-component doping, wherein the multi-component was two-component doping and three-component doping, and the doping concentration was 0.05-0.15%, 0.4-0.6%, 0.8-1.2%, 4-6%, 8-12%, 13-17%, 18-22%, 28-32%; preferably, the doping concentration was 0.1%, 0.5%, 1%, 5%, 10%, 15%, 20%, 30%; For starch, soy protein isolate, melamine, urea, and glycine egg white powder adulterants, 160-180 parts of each adulterant were used in a single component, 350-370 parts in a two-component, and 210-225 parts in a three-component. 2-4 pure adulterant samples were prepared for each of the five common egg white powder adulterants; For the latter two maltodextrins and dicyandiamide and the other typical adulterants in powdered foods, 46-50 parts of each adulterant were used in a single component, 46-50 parts in a two-component, and 4-8 parts in samples of the two pure adulterants.
[0007] Further preferably, the egg white powder sample in step S1 includes a pure egg white powder sample and an impure egg white powder sample; Pure egg white powder samples were obtained from egg powder brands in five provinces, namely Anhui, Jiangsu, Hubei, Liaoning and Jilin. Pure egg white powder samples from different production batches were selected, and 600 samples were prepared in each province, and finally a total of 3,000 pure egg white powder samples were prepared; the protein content of pure egg white powder samples was ≥78.0%, and the moisture content was ≤9.0%; The doped egg white powder samples were prepared by selecting 7 kinds of extenders and nitrogen-rich compounds, including starch, soy protein isolate, melamine, urea, glycine, maltodextrin and dicyanic acid; the doping of egg white powder was divided into single-component doping and multi-component doping, among which the multi-component was two-component doping and three-component doping, and the doping concentration was 0.1%, 0.5%, 1%, 5%, 10%, 15%, 20% and 30%; For the five common egg white powder adulterants, starch, soy protein isolate, melamine, urea, and glycine, 168 portions of each adulterant were prepared in the single component, for a total of 840 portions, 360 portions in the two components, and 216 portions in the three components. Three samples of pure adulterants were prepared for each of the five common egg white powder adulterants, for a total of 15 portions. Finally, 1431 samples of the first five adulterated egg white powders were prepared; For the remaining typical adulterants of powdered foods, the last two maltodextrins and dicyanic acid, there were 48 samples of each adulterant in the single component, for a total of 96 samples, 48 samples in the two components, 6 samples of the two pure adulterants, for a total of 12 samples, and 156 samples of the last two adulterated egg white powders were finally prepared; All pure egg white powder and adulterants were sieved through a standard sieve with a pore size of 1 mm to ensure that the sample particles were uniform in size. The adulterated egg white powder was placed in a blender and stirred for 2 min to ensure that the adulterant particles were evenly mixed in the egg white powder sample. All prepared samples were placed in sample bags of 120 mm × 80 mm and stored in a refrigerator at 4 °C.
[0008] The reason for setting up pure egg white powder from five provinces: 1. The spectral curves of pure egg white powder in each province will be different due to production materials, production processes, storage time, etc., mainly to enhance the robustness and generalization of the model. 2. The reason for setting seven adulterants and dividing them into two categories, A and B: The first five are common egg white powder adulterants on the market, mainly to verify the accuracy of model detection, and the last two are typical adulterants of other powdered foods, mainly to verify the accuracy of the model for unknown adulterants. 3. The reason for designing two components and three components: In order to circumvent the detection of regulatory authorities, the doping on the market may be a mixture of two or three substances, so this experiment will set up two-component and three-component doping, and verifying the performance of the model for multi-component doping can prove the generalization of the model. 4. The reason for designing a gradient: The doping on the market is mainly to dope low-cost substances to make profits, and the doping concentration is generally within a certain range (the doping concentration is too low, the cost cannot make profits, and the doping concentration is too high, which is easy to be detected by regulatory authorities). This experiment was set to eight gradients in order to intuitively demonstrate the detection ability of the model. The low concentration was set to demonstrate the sensitivity of the model, and the high concentration was set to demonstrate the generalization ability of the model.
[0009] Preferably, in the spectral data collection step S2, the spectral data is collected by diffuse reflection using an Antaris II Fourier near-infrared spectrometer, and the spectral data collection range is 10000~4000 cm -1 , the spectral resolution is 3.85cm -1 The scanning times were 32 times, the ambient temperature was 25 °C, each sample was scanned 3 times, and the average spectrum was taken as the original spectrum data of the sample.
[0010] Preferably, in the sample division and expansion of step S3, 3000 pure egg white powder samples are randomly divided according to the Kennard-Stone algorithm and divided into a training set and a test set at a ratio of 2:1; the test set has two groups, A and B, wherein the test set A consists of 1000 pure egg white powder samples and 1431 egg white powder samples doped with the first five common adulterants of egg white powder to verify the accuracy of the model; the test set B consists of 156 egg white powder samples doped with the last two other typical adulterants; The synthetic minority class oversampling technology is used to expand the training set data, resulting in a total of 4,000 new training set data, 2,431 test set A data, and 156 test set B data.
[0011] Preferably, in the step S5 of screening the characteristic wavelengths of the spectrum, a competitive adaptive reweighted sampling algorithm is used to extract the characteristic wavelengths of the spectrum.
[0012] Preferably, in the modeling process of step S6, each piece of spectral data is taken as a data point, wherein the spectrum of pure egg white powder is a normal point, and the spectrum of doped egg white powder is an abnormal point. On the basis of the LOF algorithm, spectral preprocessing and wavelength screening processing are added. During model training, the local reachable density threshold of the training set is determined to establish a non-directional doping identification model; during model testing, after the spectral data is preprocessed and wavelength screened, the density of the data points in the test set is calculated to screen out abnormal points, and finally the discrimination result of the test set is output.
[0013] Preferably, in the model evaluation index of step S7, the model is evaluated from three aspects: detection accuracy, detection precision and detection speed.
[0014] Further preferably, the detection accuracy index includes accuracy, precision, recall and F1 score to evaluate the comprehensive performance of the model; there are three indicators for the test results, namely, false positive rate ( F FPR , false positive rate), corresponding to the false detection rate, refers to the proportion of pure egg white powder that is misjudged as an adulterated sample. The calculation formula is shown in (1); the false negative rate ( F FNR , falsenegative rate), corresponding to the missed detection rate, refers to the proportion of adulterated egg white powder misjudged as pure egg white powder, and the calculation formula is shown in (2); the total accuracy rate ( A AR , accuracy rate), evaluates the model’s ability to distinguish the authenticity of samples, and the calculation formula is shown in (3): (1); (2); (3); In formulas (1) to (3), the pure egg white powder sample is taken as the negative sample, and the adulterated egg white powder sample is taken as the positive sample; T P is the number of true positives, i.e., the number of adulterated egg white powders correctly judged as adulterated samples; F N is the number of false negatives, i.e., the number of adulterated egg white powder samples that were mistakenly identified as pure egg white powder; T N is the number of true negatives, that is, the number of pure egg white powder samples correctly judged as pure egg white powder; F P is the number of false positives, i.e., the number of pure egg white powders mistakenly identified as adulterated samples.
[0015] Further preferably, the detection accuracy is the minimum detection limit ( L LRC , lowest recognition concentration), the calculation formula is shown in (4), where x1~x n It is the doping concentration of the missed samples in this doping type. This indicator can evaluate the detection accuracy of the model for doped samples.
[0016] (4); Further preferably, the detection speed indicator is the average detection time ( A ATS , average time spent), that is, the average time spent by the model for each sample in batch detection, in seconds (s). This indicator can evaluate the detection efficiency of the model for samples. Compared with the prior art, the present invention has the following beneficial effects: A LOF non-directional identification model based on near-infrared spectroscopy was proposed, in which the MSC+CARS processing method has an advantage in model establishment. The final LOF non-directional identification model can achieve a detection rate of 93.57% for adulterated egg white powder, and a detection rate of 100% for egg white powder with an adulteration concentration exceeding 15%. For some multi-component adulterated species, its minimum detection limit ( L LRC ) can reach 0.1%, and the total accuracy of test sets A and B ( A AR ) were 93.62% and 93.59% respectively, and the average detection time ( A ATS) can reach 0.0011 s. This model has better performance than other non-directional algorithms, and compared with the directional identification model, this model can identify a wider range of adulteration types and is more suitable for the authenticity identification of egg white powder with a variety of adulteration types on the market. BRIEF DESCRIPTION OF THE DRAWINGS Figure 1 Spectral acquisition; Figure 2 LOF model flow chart; DETAILED DESCRIPTION The following will be combined with the specific embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. The embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work belong to the protection scope of the present invention.
[0017] Example 1 1 Experimental part 1.1 Sample preparation This embodiment uses pure egg white powder from egg powder brands in five provinces, including Anhui, Jiangsu, Hubei, Liaoning and Jilin. In order to ensure the differences between pure egg white powder samples, pure egg white powder samples from different production batches were selected. 600 samples were prepared in each province, and 3,000 pure egg white powder samples were finally prepared. In order to verify the credibility of the sample, the protein content in the egg powder was determined by the Kjeldahl nitrogen determination method in GB 5009.5-2016 "Determination of Protein in Food", and the moisture content was calculated by the direct drying method in GB 5009.3-2016 "Determination of Moisture in Food in National Food Safety Standards". 2g of pure egg white powder samples were taken from each brand for determination, and the protein content of the five pure egg white powder samples was finally measured to be ≥78.0%, and the moisture content was ≤9.0%. Referring to the physical and chemical indicators in GB / T42237-2022 "General Rules for Egg Powder Quality", the samples were all within the quality standards.
[0018] Generally, the adulterants of egg white powder on the market are divided into two categories: one is the common extender adulterants (such as starch), which are difficult to distinguish from pure egg white powder in morphology and can be used to pass off inferior products as good ones; the second is nitrogen-rich compounds (such as melamine). Since the Kjeldahl method is usually used to measure the protein content of the sample in practice, such adulterants can be used to forge high-quality egg white powder. The doped egg white powder sample of this embodiment is prepared by selecting 7 common extender adulterants and nitrogen-rich compounds, including starch, soy protein isolate, melamine, urea, glycine, maltodextrin and dicyanic acid. Among them, the first five are common adulterants of egg white powder, and the latter two are typical adulterants of other powdered foods. The doping of egg white powder can be divided into single-component doping and multi-component doping, among which multi-component is two-component doping and three-component doping. The concentration of doping is 0.1%, 0.5%, 1%, 5%, 10%, 15%, 20%, and 30%. For the first five common adulterants in egg white powder, there were 168 portions of each adulterant in the single component, a total of 840 portions, 360 portions in the two components, 216 portions in the three components, 3 pure adulterant samples were prepared for each of the five common adulterants, a total of 15 portions, and finally 1431 samples of the first five adulterated egg white powders were prepared. For the last two other typical adulterants in powdered food, there were 48 portions of each adulterant in the single component, a total of 96 portions, 48 portions in the two components, 6 samples of the two pure adulterants, a total of 12 portions, and finally 156 samples of the last two adulterated egg white powders were prepared.
[0019] All pure egg white powder and adulterants were sieved through a standard sieve with a pore size of 1 mm to ensure that the sample particles were uniform in size. The adulterated egg white powder was placed in a blender and stirred for 2 min to ensure that the adulterant particles were evenly mixed in the egg white powder sample. All prepared samples were placed in sample bags of 120 mm × 80 mm and stored in a refrigerator at 4 °C.
[0020] 1.2 Spectral data acquisition The spectral data were collected using an Antaris II Fourier near-infrared spectrometer (Thermo Scientific, USA) in diffuse reflectance mode. The spectral data collection range was 10000~4000 cm -1 (1000~2500 nm), spectral resolution is 3.85 cm -1 The scanning number was 32. The ambient temperature was 25 °C, and each sample was scanned 3 times. The average spectrum was taken as the original spectrum data of the sample, and the Python 3.7.1 software was used to perform corresponding data analysis on the saved spectrum data.
[0021] The near infrared spectrum acquisition process of egg white powder sample. Each time the sample spectrum is collected, the air should be used as a blank control to measure the blank background spectrum. The ambient temperature during the measurement is 25°C. After the blank background spectrum is measured, the sample in the sample bag is poured into a test vessel with a transparent bottom, placed in the detection area and fully covered with the transparent area at the bottom of the sample box to ensure that the light source cannot penetrate the sample and avoid the influence of external light sources. The Thermo Result Integration software is used on the computer to read and save the spectrum measured by the spectrometer. The spectral data is stored in *.csv format as a 1557*2 table. Each sample is scanned 3 times, and the average spectrum is taken as the original spectral data of the sample. The collected samples are put back into the sample bag and stored in a refrigerator. The final saved spectral data is processed using Python 3.7.1 for corresponding data analysis.
[0022] 1.3 Sample Division and Expansion Since the training set consists only of pure egg white powder samples, 3000 pure egg white powder samples were randomly divided according to the Kennard-Stone algorithm and divided into training set and test set in a ratio of 2:1. The test set consists of two groups, A and B. Test set A consists of 1000 pure egg white powder samples and 1431 adulterated egg white powder samples of the first five common adulterants of egg white powder to verify the accuracy of the model; test set B consists of 156 adulterated egg white powder samples of the last two typical adulterants to verify the detection performance of the model for unknown adulterants. In order to improve the problem of low model accuracy caused by insufficient sample diversity, the synthetic minority over-sampling technique (SMOTE) was used to expand the training set data, and finally a total of 4000 new training set data, 2431 test set A, and 156 test set B were obtained. The detailed information of sample division is shown in Table 1.
[0023] Table 1 Sample grouping details
[0024] 1.4 Spectral preprocessing During the spectrum collection process, the collected spectral data will contain a large amount of random noise. In order to reduce such influence and highlight the useful spectral information, this embodiment will preprocess the spectral data and select the multiplicative scatter correction (MSC) method for spectral preprocessing.
[0025] 1.5 Spectral characteristic wavelength screening There are a large number of variables in the original spectral data, and there is a lot of noise and invalid information between these variables. If the full spectrum data is used to establish a non-directional identification model, it will not only increase the amount of model calculation, but also affect the accuracy of the model. Therefore, this embodiment will use the competitive adaptive reweighted sampling algorithm (CARS) to extract the characteristic wavelengths of the spectrum.
[0026] 1.6 Modeling Methods LOF is a density-based anomaly detection algorithm that takes into account both the local and global properties of the data set. It can effectively detect outliers in local areas. The data set does not need to be pre-labeled and can be directly applied to unlabeled data sets, making it suitable for unsupervised learning tasks.
[0027] For the identification of non-directional doping of egg white powder, this embodiment uses each spectral data as a data point, wherein the spectrum of pure egg white powder is a normal point, and the spectrum of doped egg white powder is an abnormal point. On the basis of the LOF algorithm, the pre-processing of the spectrum and the wavelength screening process are added. The process of model training and testing is as follows: Figure 2 As shown. During model training, the local reachable density threshold of the training set is determined to establish a non-directional doping identification model. During model testing, after the spectral data is preprocessed and wavelength screened, the density of the data points in the test set is calculated to screen out abnormal points, and finally the discrimination results of the test set are output.
[0028] 1.7 Model Evaluation Metrics The model established in this embodiment is a qualitative discrimination model, which is used to determine whether the sample is real egg white powder. Therefore, this paper will evaluate the model from three aspects: detection accuracy, detection precision and detection speed.
[0029] 1) Detection accuracy index: For model performance, the overall performance of the model is evaluated by accuracy, precision, recall and F1 score. There are three indicators for the test results: false positive rate ( F FPR , false positive rate), corresponding to the false detection rate, refers to the proportion of pure egg white powder that is misjudged as an adulterated sample. The calculation formula is shown in (1); the false negative rate ( F FNR , false negative rate), corresponding to the missed detection rate, refers to the proportion of adulterated egg white powder misjudged as pure egg white powder, and the calculation formula is shown in (2); the total accuracy rate ( A AR, accuracyrate), evaluates the model’s ability to identify the authenticity of samples, and the calculation formula is shown in (3).
[0030] (1) (2) (3) In formulas (1) to (3), the pure egg white powder sample is taken as the negative sample, and the adulterated egg white powder sample is taken as the positive sample. T P is the number of true positives, i.e., the number of adulterated egg white powders correctly judged as adulterated samples; F N is the number of false negatives, i.e., the number of adulterated egg white powder samples that were mistakenly identified as pure egg white powder; T N is the number of true negatives, that is, the number of pure egg white powder samples correctly judged as pure egg white powder; F P is the number of false positives, i.e., the number of pure egg white powders mistakenly identified as adulterated samples.
[0031] 2) Detection accuracy index: Minimum detection limit ( L LRC , lowest recognition concentration), that is, the minimum doping concentration at which the doping species can be completely detected, expressed in percentage (%), and its calculation formula is shown in (4), where x1~x n It is the doping concentration of the missed samples in this doping type. This indicator can evaluate the detection accuracy of the model for doped samples.
[0032] (4) 3) Detection speed indicator: average detection time ( A ATS , average time spent), that is, the average time spent by the model for each sample in batch detection, in seconds (s). This indicator can evaluate the detection efficiency of the model for samples.
[0033] 2 Results 2.1 LOF non-directional discrimination model test results After the spectral data were preprocessed by MSC spectrum and screened by CARS spectrum characteristic wavelength, the training set was input into the LOF non-directional identification model for model training, where the model parameter K was set to 100. After the model training was completed, the test set A and test set B were input into the model for testing.
[0034] Table 2 shows the authenticity identification results of different doping concentrations of test set A. FFPR The total accuracy of the whole test set A is 6.3%. A AR It can reach 93.62%.
[0035] Table 2 Authenticity identification results of different doping concentrations
[0036] The specific results of authenticity identification of doping types are shown in Table 3. For each doping type, the number of false positives and the minimum detection limit at different concentrations are counted. L LRC The highest is 15%. L LRC The lowest values can reach 5%, 0.1%, and 0.1% respectively. For multi-component doping, since the characteristic areas of the multi-component doping spectrum are more than those of a single component, they are more likely to be judged as abnormal points. Therefore, this model has a higher detection accuracy for multi-component doping. L LRC The minimum can reach 0.1%.
[0037] Table 3 Authenticity identification results and minimum detection limits of different types of adulteration
[0038] The authenticity identification results of test set B are shown in Table 4. Compared with the extender, the spectral differentiation of nitrogen-rich compounds is greater. Therefore, in the identification of single-component doping, dicyanic acid has a better identification effect, and its minimum detection limit can reach 0.5%. A AR can reach 98.15%; for the entire test set B, A AR It can reach 93.59%, which is not significantly different from the accuracy of test set A, which shows that the LOF non-directional identification model can also maintain good discrimination performance for other unknown adulterants.
[0039] Table 4 Authenticity identification results and minimum detection limits of test set B
[0040] In summary, the model has good accuracy in detecting the authenticity of egg white powder, and has good discrimination performance for both single-component and multi-component mixed adulterations. In addition, the model is highly versatile and can maintain good detection performance for other unknown adulterants.
[0041] 2.2 Performance comparison of different models In order to compare the performance of the LOF non-directional identification model with other non-directional algorithms, three unsupervised methods, namely isolation forest (IF), support vector data description (SVDD), and deep isolation forest (DIF), were used to establish the model. For IF and DIF, the number of trees was set to 50 and the abnormality ratio was set to 0.1; for SVDD, the penalty coefficient C was set to 5 and the gamma was set to 10. The model was trained and tested using the same training set and test set A. The final experimental results are shown in Table 5. Compared with the other three algorithms, LOF has the best accuracy, recall rate, and F1 score. Among them, SVDD has the best precision, and there is no false detection for pure egg white powder samples, but SVDD has too low accuracy in detecting adulterated samples, and its comprehensive model performance is not as good as LOF. In addition, LOF has the fastest detection speed. A ATS It can reach 0.0011s, meeting the requirements of rapid detection.
[0042] Table 5 Identification results of different non-directional identification models
[0043] In order to compare the performance of the LOF non-directional discrimination model trained with different training sets, the original training set, X training set (pure egg white powder + a small amount of test set A doped egg white powder), and Y training set (pure egg white powder + a small amount of test set B doped egg white powder) were used for model training. The model names were LOF, LOFX, and LOFY, respectively, and the model was tested with the same test set A. The final experimental results are shown in Table 6. Compared with LOFX and LOFY, LOF has the best recall rate and F1 score, so the training of the LOF model is optimal only with pure egg white powder samples.
[0044] Table 6 LOF identification results with different training set compositions
[0045] In order to verify the superiority of the LOF non-directional identification model in identifying unknown adulterants compared with the directional identification model, the LOF non-directional identification model was compared with the traditional support vector machine (SVM) and ensemble learning (EL) directional identification models. The SVM set gamma to 10; the EL used the Boosting method. The final results are shown in Table 7. For test set A, since the SVM and EL models have learned the doping types of the doped samples in test set A, their A ARCompared with the LOF model, it is higher. However, for test set B, since the directional model has not learned the doping types in test set B, its F FNR It has increased significantly, A AR The LOF non-directional identification model has a lower accuracy than the LOF non-directional identification model in test set A and test set B. A AR They can all be stabilized to more than 93%, avoiding the problem of a significant drop in model performance due to unknown adulterants.
[0046] Table 7 Results of non-directional identification model and directional identification model
[0047] In summary, compared with other non-directional models, the LOF model has the best comprehensive performance. Compared with the directional identification model, the model has better and more stable identification effect on unknown adulterants, can effectively identify a wider range of adulterants, and is more suitable for authenticity identification of egg white powder with a variety of adulterants on the market.
[0048] The final LOF non-directional identification model can achieve a detection rate of 93.57% for adulterated egg white powder, and a detection rate of 100% for egg white powder with an adulteration concentration exceeding 15%. L LRC It can reach 0.1%, for test sets A and B A AR They are 93.62% and 93.59% respectively. A ATS It can reach 0.0011 s. Compared with other non-directional algorithms, this model has better performance, and compared with the directional identification model, this model can identify a wider range of adulteration types, and is more suitable for the authenticity identification of egg white powder with a variety of adulteration types on the market. The technical solution of the present invention is explained through the above embodiments, but the present invention is not limited to the above embodiments, that is, it does not mean that the present invention must rely on the above specific embodiments to be implemented. Any improvements made by technicians in the relevant field based on the present invention, or equivalent replacement of the materials selected by the present invention, etc., fall within the scope of protection of the patent.
Claims
1. A non-directional identification method for the authenticity of egg white powder, characterized in that: The method comprises the following steps: S1: Collect egg white powder samples; S2: Spectral data acquisition; S3: Sample division and expansion; S4: spectral preprocessing; S5: spectral characteristic wavelength screening; S6: Modeling; S7: Model evaluation resulted in a non-directional identification model for the authenticity of egg white powder.
2. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: The egg white powder samples in step S1 include pure egg white powder samples and doped egg white powder samples; There are 2800-3200 pure egg white powder samples; the protein content of pure egg white powder samples is ≥78.0%, and the moisture content is ≤9.0%; The doped egg white powder samples were prepared by selecting 7 kinds of extender dopants and nitrogen-rich compounds, including starch, soy protein isolate, melamine, urea, glycine, maltodextrin and dicyanic acid; the doping of egg white powder was divided into single-component doping and multi-component doping, wherein the multi-component was two-component doping and three-component doping, and the doping concentration was 0.05-0.15%, 0.4-0.6%, 0.8-1.2%, 4-6%, 8-12%, 13-17%, 18-22%, 28-32%; preferably, the doping concentration was 0.1%, 0.5%, 1%, 5%, 10%, 15%, 20%, 30%; For starch, soy protein isolate, melamine, urea, and glycine egg white powder adulterants, 160-180 parts of each adulterant were used in a single component, 350-370 parts in a two-component, and 210-225 parts in a three-component. 2-4 pure adulterant samples were prepared for each of the five common egg white powder adulterants; For the latter two maltodextrins and dicyandiamide and the other typical adulterants in powdered foods, 46-50 parts of each adulterant were used in a single component, 46-50 parts in a two-component, and 4-8 parts in samples of the two pure adulterants.
3. The non-directional authenticity identification method of egg white powder according to claim 2, characterized in that: The egg white powder samples in step S1 include pure egg white powder samples and doped egg white powder samples; Pure egg white powder samples were obtained from egg powder brands in several provinces. Pure egg white powder samples from different production batches were selected, and 600 samples were prepared in each province, and finally a total of 3,000 pure egg white powder samples were prepared; the protein content of pure egg white powder samples was ≥78.0%, and the moisture content was ≤9.0%; The doped egg white powder samples were prepared by selecting 7 kinds of extender dopants and nitrogen-rich compounds, including starch, soy protein isolate, melamine, urea, glycine, maltodextrin and dicyanic acid; the doping of egg white powder was divided into single-component doping and multi-component doping, wherein the multi-component was two-component doping and three-component doping, and the doping concentration was 0.05-0.15%, 0.4-0.6%, 0.8-1.2%, 4-6%, 8-12%, 13-17%, 18-22%, 28-32%; preferably, the doping concentration was 0.1%, 0.5%, 1%, 5%, 10%, 15%, 20%, 30%; For starch, soy protein isolate, melamine, urea, and glycine egg white powder adulterants, 168 portions of each adulterant were prepared in the single component, for a total of 840 portions, 360 portions in the two components, and 216 portions in the three components. Three pure adulterant samples were prepared for each of the five common egg white powder adulterants, for a total of 15 portions. Finally, 1431 samples of the first five adulterated egg white powders were prepared; For the remaining typical adulterants of powdered foods, the last two maltodextrins and dicyanic acid, there were 48 samples of each adulterant in the single component, for a total of 96 samples, 48 samples in the two components, 6 samples of the two pure adulterants, for a total of 12 samples, and 156 samples of the last two adulterated egg white powders were finally prepared; All pure egg white powder and adulterants were sieved through a standard sieve with a pore size of 1-2 mm to ensure that the sample particles were uniform in size; the adulterated egg white powder was placed in a blender and stirred for 1.5-2.5 min to ensure that the adulterant particles were evenly mixed in the egg white powder sample; all prepared samples were placed in sample bags of 120 mm×80 mm and stored in a refrigerator at 4°C.
4. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: In the step S2, the spectral data is collected by using an Antaris II Fourier near-infrared spectrometer in a diffuse reflectance mode, and the spectral data collection range is 10000~4000 cm -1 , the spectral resolution is 3.85 cm -1 The scanning times were 32 times, the ambient temperature was 25°C, each sample was scanned 3 times, and the average spectrum was taken as the original spectrum data of the sample.
5. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: In the sample division and expansion of step S3, 3000 pure egg white powder samples are randomly divided according to the Kennard-Stone algorithm and divided into a training set and a test set at a ratio of 2:1; the test set has two groups, A and B, wherein the test set A consists of 1000 pure egg white powder samples and 1431 egg white powder samples doped with the first five common adulterants of egg white powder to verify the accuracy of the model; the test set B consists of 156 egg white powder samples doped with the last two and other typical adulterants; The synthetic minority class oversampling technology is used to expand the training set data, resulting in a total of 4,000 new training set data, 2,431 test set A data, and 156 test set B data.
6. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: In the step S5 of screening the characteristic wavelengths of the spectrum, a competitive adaptive reweighted sampling algorithm is used to extract the characteristic wavelengths of the spectrum.
7. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: In the modeling process of step S6, each piece of spectral data is taken as a data point, wherein the spectrum of pure egg white powder is a normal point, and the spectrum of doped egg white powder is an abnormal point. On the basis of the LOF algorithm, the spectrum preprocessing and wavelength screening processing are added. During the model training, the local reachable density threshold of the training set is determined to establish a non-directional doping identification model; During model testing, after the spectral data is preprocessed and wavelength screened, the density of the data points in the test set is calculated to screen out abnormal points, and finally the discrimination results of the test set are output.
8. The non-directional authenticity identification method of egg white powder according to claim 1, characterized in that: In the model evaluation index of step S7, the model is evaluated from three aspects: detection accuracy, detection precision and detection speed.
9. The non-directional authenticity identification method of egg white powder according to claim 8, characterized in that: The detection accuracy indicators include accuracy, precision, recall and F1 score to evaluate the overall performance of the model; there are three indicators for the test results, namely, false positive rate F FPR , false positive rate, corresponding to the false detection rate, refers to the proportion of pure egg white powder being misjudged as adulterated samples, and the calculation formula is shown in (1); false negative rate F FNR , false negative rate, corresponding to missed detection rate, refers to the proportion of adulterated egg white powder misjudged as pure egg white powder, and the calculation formula is shown in (2); the total accuracy rate A AR , accuracy rate, evaluates the model's ability to distinguish the authenticity of samples. The calculation formula is shown in (3): (1)； (2)； (3)； In formulas (1) to (3), the pure egg white powder sample is taken as the negative sample, and the adulterated egg white powder sample is taken as the positive sample; T P is the number of true positives, i.e., the number of adulterated egg white powders correctly judged as adulterated samples; F N is the number of false negatives, i.e., the number of adulterated egg white powder samples that were mistakenly identified as pure egg white powder; T N is the number of true negatives, that is, the number of pure egg white powder samples correctly judged as pure egg white powder; F P is the number of false positives, i.e., the number of pure egg white powders mistakenly identified as adulterated samples.
10. The method for non-directionally identifying the authenticity of egg white powder according to claim 9, characterized in that: Detection accuracy is the minimum detection limit L LRC , the lowest recognition concentration, calculated as shown in (4), where x1~x n is the doping concentration of the missed samples in this doping type. This indicator evaluates the detection accuracy of the model for doped samples: (4)； The detection speed indicator is the average detection time A ATS , average time spent, that is, the average time spent by the model for each sample in batch testing, in seconds.