Low-cost tea oil true and false identification method

Abstract

The invention discloses a low-cost tea oil true and false identification method. The method comprises the following steps: 1) preparing an adulterated tea oil sample; (2) acquiring a surface enhanced Raman spectrum; 3) edible oil classification and camellia oil authentic identification model establishment; 4) identifying the camellia oil sample; according to the method, the NPAg sheet coated with an epoxy resin solution with the concentration of 0.02% and provided with internal standard molecules is used as an SERS substrate, SERS spectrums of pure camellia oil and adulterated camellia oil are collected respectively for machine learning analysis training, and the obtained camellia oil identification model can detect adulterated camellia oil even under the condition that the adulteration rate is 0.5%; wherein the k-nearest neighbor (KNN) model shows excellent performance, the accuracy rate of a test set reaches 97.24%, the recall rate reaches 99%, and the f1 score reaches 98%. The method provided by the invention has the advantages of high accuracy, low cost and no pollution to the environment, and is suitable for further popularization and application.

Description

Technical Field

[0001] The invention belongs to the technical field of analytical chemistry, and in particular relates to a low-cost method for identifying the authenticity of tea oil. Background Art

[0002] Adulteration of edible vegetable oils is an urgent problem that poses a major threat to the health and safety of consumers. Camellia oil is known for its high nutritional value and multiple medicinal benefits, with an unsaturated fatty acid content of up to 90%. It has earned nicknames such as "health oil", "longevity oil" and "confinement oil" among the public. Camellia oil is rich in nutrients and is loved by consumers. However, due to its relatively high cost, it has become a target for unscrupulous vendors to make huge profits by adulterating low-priced oils. The problem of camellia oil adulteration is on the rise, and illegal activities often involve complex multi-variety adulteration, which makes it increasingly challenging to detect camellia oil by traditional methods. Therefore, an efficient and reliable method is urgently needed to identify the authenticity of camellia oil. Such a method is extremely important for both the quality assessment of camellia oil and the supervision work of market supervision agencies.

[0003] Raman spectroscopy has a wide range of uses in the field of edible oil adulteration detection. Raman spectroscopy has the unique ability to detect changes in the molecular composition of liquids without the need for sample preparation. In addition, Raman spectroscopy has been used to check for adulterated edible oils and effectively distinguish inferior oils from high-quality olive oils. However, Raman spectroscopy generates a large amount of data, and it is challenging to extract qualitative or quantitative information due to the inherent similarities in the Raman spectra of many edible oils. Surface enhanced Raman spectroscopy (SERS) has emerged as an effective analytical technique that can provide not only structural vibration information but also valuable analytical molecular concentration information. SERS technology can significantly amplify subtle differences between different types of edible oils. When combined with statistical tools such as machine learning, efficient characterization and differentiation can be achieved.

[0004] Surface-Enhanced Raman Scattering (SERS) technology can provide highly sensitive and selective molecular spectral information, but analyzing this data may be challenging because the SERS signal is interfered by many influencing factors. Machine learning methods are combined with surface-enhanced Raman Scattering (SERS), and machine learning methods can help extract and interpret these complex data, accurately identify and classify substances, and improve analysis efficiency and accuracy. Therefore, the present invention establishes a low-cost method for distinguishing the authenticity of camellia oil to achieve the identification of the adulteration of camellia oil with other oils. Summary of the invention

[0005] In order to overcome the problems existing in the prior art, the purpose of the present invention is to provide a low-cost method for identifying the authenticity of tea oil, which amplifies the differences in different edible oils through SERS technology and combines statistical tools such as machine learning to achieve the identification of different types of edible oils.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] A low-cost method for distinguishing the authenticity of tea oil, comprising:

[0008] 1) Preparation of adulterated tea oil samples

[0009] The D-optimal mixing design in the mixture design was used to prepare adulterated oils with different adulteration ratios to simulate the adulterated samples of camellia oil.

[0010] 2) Surface enhanced Raman spectroscopy acquisition

[0011] The surface enhanced Raman spectra of authentic camellia oil and adulterated camellia oil were collected respectively;

[0012] 3) Classification of edible oils and establishment of camellia oil authentication model

[0013] The surface enhanced Raman spectra of all samples were selected as machine learning analysis data, and the algorithm with classification function in the sklearn library of Anaconda3-2023.07 analysis software was used to process the surface Raman spectra data of the collected samples to establish a camellia oil authentication model;

[0014] 4) Identification of camellia oil samples

[0015] The surface enhanced Raman spectrum of the sample to be tested is collected, and then the collected surface enhanced Raman spectrum data is analyzed at 2228 cm -1 The internal standard is used for calibration, and then the calibrated surface-enhanced Raman spectroscopy data is input into the camellia oil authentication model to identify the sample.

[0016] As a possible implementation manner, further, the surface enhanced Raman is a nano-gold or silver substrate containing an internal standard with an enhancement effect; when collecting Raman spectra, the sample oil is dropped on the nano-gold or silver substrate containing an internal standard with an enhancement effect for detection; wherein the internal standard molecule is used to calibrate the differences between samples in a complex detection environment in order to facilitate a more accurate comparison.

[0017] As a possible implementation manner, further, the internal standard molecule is 4MBN.

[0018] As a possible implementation, further, the surface enhanced Raman spectroscopy collection conditions are as follows:

[0019] The Raman spectrum is collected in the range of 0-3000cm -1 , using a 785nm excitation light source, an integrated power of 200Mw, an integration time of 5000ms, and scanning each sample 5-10 times.

[0020] As a possible implementation mode, further, algorithms with classification functions in the sklearn library include DT, RF, KNN, SVM, XGBoost, and MLP.

[0021] As a possible implementation mode, further, the adulterated camellia oil in step 1) is authentic camellia oil mixed with peanut oil, soybean oil, corn oil, palm oil, rice oil, sunflower oil and linseed oil respectively, and the mixed mass percentage is 0.5% to 35%.

[0022] As a possible implementation mode, further, the method for preparing the nanosilver substrate containing the internal standard comprises the following steps:

[0023] S1. Preparation of nanosilver sheet NPAg sheet: Grinding the silver foil with 240 sandpaper until the protective layer on the surface of the silver foil is removed and a grid pattern appears on the surface of the silver foil to form an NPAg sheet;

[0024] S2. Preparation of NPAg sheet@4MBN: Immerse the NPAg sheet in 10 -3 mol / L 4MBN ethanol solution for 30 minutes, then taken out and air-dried at room temperature to form 4MBN-coupled silver foil NPAg sheet@4MBN;

[0025] S3. Preparation of 0.02% ER: weigh 0.3 g epoxy resin (ER) and 0.1 g polyetheramine, keeping a ratio of 3:1, and stir the mixture with a spray gun for 5-10 minutes until the mixture becomes completely transparent without any filamentous substances;

[0026] Subsequently, the mixture was placed in a vacuum oven at 35-40°C and evacuated for 5-10 minutes to remove bubbles. The mixture was heated to 35-40°C using a magnetic stirrer, and then 100 μL of epoxy resin and 900 μL of acetone were extracted using a pipette to prepare a 1% volume fraction solution. The 1% epoxy resin solution was further diluted with acetone to obtain a solution with a lower volume fraction. This process was repeated as needed to prepare a 0.02% volume fraction epoxy resin solution.

[0027] S4. Preparation of NPAg sheet@4MBN@0.02%ER: Immerse the dried NPAg sheet@4MBN in a 0.02% epoxy resin solution for 2-3 seconds, take it out and place it with the polished surface facing up, and let it air-dry until solidified, thereby forming a transparent protective layer to obtain NPAg sheet@4MBN@0.02%ER.

[0028] As a possible implementation, further, step 3) specifically includes the following steps:

[0029] Data processing: The collected surface enhanced Raman spectroscopy data were processed at 2228 cm -1 For internal standard calibration, each data collected was selected from 400-3000cm -1 Spectral range, and then normalize the data respectively;

[0030] Model training and evaluation: The raw spectral data points are randomly divided into training and test sets for machine learning analysis;

[0031] The training set and the test set data are used to train the model of the algorithm with the classification function, the algorithm performs a grid search, and a five-fold cross-validation method is used to select hyperparameters;

[0032] The main performance criteria for evaluating the models include accuracy, confusion matrix, precision, recall, f1 score, ROC curve, and AUC.

[0033] As a preferred implementation, preferably, the data volume ratio of the training set to the test set is 4:1.

[0034] The main reason for using the internal standard molecule 4MBN in the present invention is that the thickness of commercial silver foil is usually 10 μm, and its surface is easily contaminated by ambient air, resulting in reduced optical reflectivity and chemical reactivity. In order to manufacture surface enhanced Raman scattering substrates and (bio) chemical and optical sensors, the present invention uses sandpaper to destroy the reaction layer on the surface of the silver foil, and finally obtains the NPAg sheet. Since sandpaper polishing is affected by the intensity and polishing time, the silver foil has uneven characteristics after polishing. For this reason, the present invention introduces 4-mercaptobenzonitrile (4MBN) as an internal standard molecule. The characteristic peak of 4MBN can be perfectly staggered with the main characteristic peaks of camellia oil or other edible oils, generating signals in the Raman silent zone, and will not change the Raman peak position of the oil itself, effectively reducing the complexity caused by the SERS signal fluctuation.

[0035] The present invention utilizes the SERS substrate material NPAg sheet@0.02%ER to significantly amplify the subtle differences between different types of edible oils. The material is a thin and transparent epoxy resin protective layer coated on a polished silver foil modified with an internal standard molecule. This method ensures that the test substance oil is isolated from the silver nanoparticles on the surface of the NPAg sheet to avoid direct contact. Therefore, the high transmittance of epoxy resin can be effectively utilized to focus the laser beam on the silver nanoparticles. In addition, the transparency of the substrate allows reverse excitation and collection through the substrate, resulting in clear and well-defined SERS spectral peaks.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1) The method of the present invention is simple, with high classification and identification accuracy, and can achieve adulteration identification of camellia oil at a doping level of 0.5%. The surface enhanced Raman substrate prepared by the present invention has the characteristic of being reusable, and the internal standard molecule 4MBN can be used for calibration, making the test results more accurate, and expanding the identification method of adulteration of camellia oil and other edible oils;

[0038] 2) The present invention uses machine learning analysis by using SERS spectra of pure camellia oil and adulterated camellia oil on NPAg sheets (polished commercial silver foil) coated with 4-thiobenzonitrile (4MBN) at a concentration of 0.02%. This material is not only easy to synthesize, but also corrosion-resistant and reusable. The use of benzonitrile (4MBN) to generate signals in the Raman quiet zone further improves the stability of spectral acquisition and ensures more accurate results. This method can detect adulterated camellia oil even with an adulteration rate as low as 0.5%. The k-nearest neighbor (KNN) model showed excellent performance, with a test set accuracy of 97.24%, a recall rate of 99%, and an f1 score of 98%. Therefore, the NPAg sheet@4MBN@0.02%ER strategy designed to amplify the differences in edible oil composition has become an effective tool for quickly verifying the authenticity of such oils. This method has high accuracy, is environmentally friendly, and the classification result information is visualized, which expands the identification method of camellia oil and other edible oil adulteration. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the embodiments described below are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on these embodiments without creative work belong to the protection scope of the present invention.

[0040] Figure 1Figure 1 shows the enhancement effect of NPAg sheet@ER encapsulated with different epoxy resin concentrations on CAO52. Figure A shows the effect of NPAg sheet@ER encapsulated with different epoxy resin concentrations on the Raman spectrum of CAO52; Figure B shows the effect of NPAg sheet@ER encapsulated with different epoxy resin concentrations on the Raman spectrum of CAO52 1442cm -1 The influence of Raman characteristic peak signal intensity.

[0041] Figure 2 Comparison of the reinforcement effects of different substrate materials on CAO52.

[0042] Figure 3 The scanning electron micrographs and water contact angles of Ag sheet, NPAg sheet and NPAg sheet@4MBN@0.02%ER. Figures A, D, G and J are the scanning electron micrographs and water contact angles of Ag sheet, Figures B, E, H and K are the scanning electron micrographs and water contact angles of NPAg sheet, and Figures C, F, I and L are the scanning electron micrographs and water contact angles of NPAg sheet@4MBN@0.02%ER.

[0043] Figure 4 These are the normal Raman spectra and enhanced Raman spectra (after calibration) of CAO52, CO, LO, PAO, PO, RBO, SO, and SSO. Figure A is the normal Raman spectra of CAO52, CO, LO, PAO, PO, RBO, SO, and SSO; Figure B is the enhanced Raman spectra of CAO52, CO, LO, PAO, PO, RBO, SO, and SSO (after calibration); Figure C is the normal Raman spectra of CAO52, CO, LO, PAO, PO, RBO, SO, and SSO at 1442 cm -1 The signal intensity line diagram of the Raman characteristic peak at 1442 cm; Figure D is the common Raman spectra of CAO52, CO, LO, PAO, PO, RBO, SO and SSO at 1442 cm -1 Signal intensity line diagram of the Raman characteristic peak.

[0044] Figure 5 The test results of the machine learning models of ordinary Raman, SERS (uncalibrated) and SERS (calibrated). Figures A, B and C are the accuracy of the machine learning models of ordinary Raman, SERS (uncalibrated) and SERS (calibrated) on the test set; Figures D, E and F are the ROC curves of the machine learning models of ordinary Raman, SERS (uncalibrated) and SERS (calibrated); Figures G and H are the confusion matrices of the machine learning model of SERS (calibrated) on the training set and test set.

[0045] Figure 6Performance evaluation of NPAg sheet@4MBN@0.02%ER SERS under NaOH (A), acid: HCl (B), salt: KCl (C), low temperature: -80℃ (D), oil (E), and high temperature: 80℃ (F).

[0046] Figure 7 The uniformity and repeatability performance evaluation of NPAg sheet@4MBN@0.02%ER SERS. Figures A and B are uniformity evaluations; Figures C and D are repeatability evaluations. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0048] This embodiment provides a low-cost method for identifying the authenticity of tea oil, which specifically includes the following steps:

[0049] S1. Preparation of adulterated samples: A set of 38 representative adulterated oils were prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the preparation of adulterated samples, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0050] S2. Preparation of nanosilver sheet NPAg sheet: Grind the silver foil (Ag sheet) with 240 sandpaper for about 1 minute to remove the protective layer on the surface of the silver foil. Continue grinding until a grid pattern appears on the surface of the silver foil to form an NPAg sheet.

[0051] S3, Preparation of NPAg sheet@4MBN: Immerse the NPAg sheet in 10 -3 mol / L 4MBN ethanol solution for about 30 minutes, and then taken out and air-dried at room temperature. This process formed a 4MBN-coupled silver foil (called NPAg sheet@4MBN), which was then placed in a centrifuge tube for use.

[0052] S4. Preparation of 0.02% ER: Weigh 0.3 g of epoxy resin (ER) and 0.1 g of polyetheramine, maintaining a ratio of about 3:1. Stir the mixture with a spray gun tip for 5-10 minutes until the mixture becomes completely transparent and there is no filamentous substance. Subsequently, the mixture is placed in a vacuum oven at 35-40°C and evacuated for another 5-10 minutes to remove bubbles. Use a magnetic stirrer to heat the mixture to 35-40°C, and then use a pipette to extract 100 μL of epoxy resin and 900 μL of acetone to make a solution with a volume fraction of 1%. Further dilute the 1% epoxy resin solution with acetone to obtain a solution with a lower volume fraction. Repeat this process as needed to prepare the epoxy resin solution of the desired volume fraction.

[0053] See attached Figure 1 As shown in the figure, it can be seen that when the volume fraction is 0.02%, the substrate has the best reinforcing effect on camellia oil, so the preferred volume fraction of the epoxy resin solution is 0.02%.

[0054] S5. Preparation of NPAg sheet@4MBN@0.02%ER: Immerse the dried NPAg sheet@4MBN in a 0.02% epoxy resin solution for 2-3 seconds. After taking it out, place it with the polished surface facing up and let it air-dry until solidified, thereby forming a thin and transparent protective layer to obtain a reinforced substrate: NPAg sheet@4MBN@0.02%ER.

[0055] S6. Surface enhanced Raman spectroscopy acquisition and data processing: The acquired spectral data are divided into three different data sets: Data set 1 represents the common Raman data set (i.e., a drop of oil sample is dropped on the aluminum sheet and the Raman spectroscopy test is performed), Data set 2 corresponds to the surface enhanced Raman data set (i.e., a drop of oil sample is dropped on the enhanced substrate: NPAg sheet@4MBN@0.02%ER and the Raman spectroscopy test is performed), and Data set 3 is based on Data set 2 with a Raman spectrum of 2228 cm -1 The surface enhanced Raman data set was calibrated for internal standards. In terms of data preprocessing, each data set was selected from 400-3000cm -1 spectral range, and normalize the data respectively to prepare for subsequent machine learning training.

[0056] The Raman spectrum acquisition range is 0-3000cm -1 , using 785nm excitation light source, integrated power of 200Mw, integration time set to 5000ms, and scanning each sample 5-10 times;

[0057] S7. Machine learning model evaluation: 1085 (217×5) raw spectral data points collected from datasets 1, 2, and 3 were randomly divided into two subsets: a training set for machine learning analysis (including 80% of the data used to develop the model) and a test set (including 20% ​​of the data used to evaluate the predictive performance of the built model). Six different machine learning methods were used for model training, including decision tree (DT), random forest (RF), k-nearest neighbor (KNN), support vector machine (SVM), extreme gradient boosting (XGBoost), and multilayer perceptron classifier (MLP). In addition, each algorithm was grid searched and hyperparameters were selected using a five-fold cross-validation method to obtain accurate results for both the training and test datasets. The main performance criteria for evaluating the models included accuracy, confusion matrix, precision, recall, f1 score, ROC curve, and AUC.

[0058] S8. Identification of camellia oil samples: The surface enhanced Raman spectrum of the sample to be tested was collected, and then the collected surface enhanced Raman spectrum data was analyzed at 2228 cm -1 The internal standard is used for calibration, and then the calibrated surface-enhanced Raman spectroscopy data is input into the camellia oil authentication model to identify the sample.

[0059] Example 1

[0060] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0061] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0062] 3) Establishment of the authentication model for different edible oil classifications and camellia oil adulterated with other edible oils: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the SVM algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0063] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0064] Ordinary Raman: SVM (93.09%)

[0065] Enhanced Raman (uncalibrated): SVM (88.94%)

[0066] Enhanced Raman (calibration): SVM (89.40%)

[0067] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the SVM training; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the SVM training; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the SVM training; -1 The surface enhanced Raman data set calibrated for the internal standard (ie, the above-mentioned data set 3) was brought into the model obtained by SVM training.

[0068] Example 2

[0069] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0070] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0071] 3) Establishment of the model for identifying counterfeit products of different edible oil classifications and camellia oil adulterated with other edible oils: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the MLP algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0072] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0073] Ordinary Raman: MLP (90.78%)

[0074] Enhanced Raman (uncalibrated): MLP (92.17%)

[0075] Enhanced Raman (calibration): MLP (91.71%)

[0076] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the MLP training; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the MLP training; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the MLP training; -1 The surface enhanced Raman data set calibrated for the internal standard (ie, data set 3 mentioned above) was brought into the model obtained by MLP training.

[0077] Example 3

[0078] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0079] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0080] 3) Establishment of the authentication model for different edible oil classifications and camellia oil adulterated with other edible oils: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the KNN algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0081] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0082] Ordinary Raman: KNN (89.86%)

[0083] Enhanced Raman (uncalibrated): KNN (95.85%)

[0084] Enhanced Raman (calibration): KNN (97.24%)

[0085] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the KNN training; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the KNN training; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the KNN training; -1 The surface enhanced Raman data set calibrated for the internal standard (ie, the above-mentioned data set 3) was brought into the model obtained by KNN training.

[0086] Example 4

[0087] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0088] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0089] 3) Establishment of the authentication model for different edible oil classifications and camellia oil adulterated with other edible oils: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the RF algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0090] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0091] Ordinary Raman: RF (88.94%)

[0092] Enhanced Raman (uncalibrated): RF (91.71%)

[0093] Enhanced Raman (calibration): RF (91.71%)

[0094] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the RF training; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the RF training; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the RF training; -1 The surface enhanced Raman data set calibrated for the internal standard (ie, data set 3 mentioned above) was brought into the model obtained by RF training.

[0095] Example 5

[0096] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0097] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0098] 3) Establishment of the identification model for different edible oil classifications and other edible oils added to camellia oil: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the XGBoost algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0099] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0100] Normal Raman: XGBoost (88.02%)

[0101] Enhanced Raman (uncalibrated): XGBoost (94.01%)

[0102] Enhanced Raman (calibration): XGBoost (94.01%)

[0103] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the XGBoost training model; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the XGBoost training model; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the XGBoost training model. -1 The surface enhanced Raman data set calibrated for the internal standard (i.e., data set 3 mentioned above) was brought into the model trained by XGBoost.

[0104] Example 6

[0105] 1) Selection of different types of edible oil samples: To ensure a wide range of samples, edible oils of different brands and batches were used. Eight edible oils, including camellia oil, palm oil, corn oil, soybean oil, peanut oil, rice oil, linseed oil, and sunflower oil, were from different well-known manufacturers. Some of these oils were extracted with n-hexane to verify their authenticity. A total of 62 samples were included in this study, including 55 camellia oil samples, one palm oil (PAO), corn oil (CO), soybean oil (SO), peanut oil (RO), rice oil (RBO), linseed oil (LO), and sunflower oil (SSO) samples. The detailed information of these edible oil samples is shown in Table 1.

[0106] 2) Preparation of adulterated camellia oil: A set of 38 representative adulterated oils were carefully prepared using the D-optimal mixing design according to the proportions listed in Table 2. This process aims to derive the best mixing ratio for the adulteration detection model. To ensure uniformity, the adulterated edible oils were vigorously shaken for 2 hours. Next, three different amounts of camellia oil were randomly selected. Subsequently, three different adulterated vegetable oils were randomly selected, and the adulteration level of each oil ranged from 0.5% to 35%, and the specific proportions are detailed in Table 3. A total of 117 samples were obtained in this process. All samples were collected and stored during the period, resulting in a total of 217 (62+38+117=217) samples, which were stored in a refrigerator at 4°C. All samples were analyzed within one week of collection.

[0107] 3) Establishment of the identification model for different edible oil classifications and camellia oil adulterated with other edible oils: The spectrum of each sample was collected 5 times, and a total of 1085 (217×5=1085) Raman spectra were obtained, which were divided into training sets (80%, 868 spectra) and test sets (20%, 217 spectra). 868 spectra (including all categories of edible oils) were randomly selected from the Raman spectral data of all edible oil samples to generate a training set to establish a machine learning model. The machine learning model in the Anaconda3-2023.07 analysis software was used to process the data, and JupyterLab 64 was used as the compilation tool. At the same time, sklearn was called to import the DT algorithm model. In order to facilitate the establishment of the model, the type of edible oil was considered to be the dependent variable "Y value", the spectral feature data of the edible oil was considered to be the independent variable "X value", the authentic camellia oil was numbered 1, and the adulterated camellia oil and other vegetable oils were numbered 0. In the confusion matrix, the data in each square on the diagonal represents the number of predicted values ​​that match the true value, and the percentage represents the accuracy of the predicted value.

[0108] 4) Classification and identification of unknown edible oil samples: 217 spectra of the test set were brought into the model for training. The accuracy of the test set was as follows:

[0109] Ordinary Raman: DT (82.03%);

[0110] Enhanced Raman (uncalibrated): DT (83.41%);

[0111] Enhanced Raman (calibration): DT (88.48%)

[0112] Among them, ordinary Raman refers to the model obtained by bringing the ordinary Raman data set (i.e., the above-mentioned data set 1) into the DT training; enhanced Raman (uncalibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the DT training; enhanced Raman (calibrated) refers to the model obtained by bringing the surface enhanced Raman data set (i.e., the above-mentioned data set 2) into the DT training; -1 The surface enhanced Raman data set calibrated for the internal standard (ie, data set 3 mentioned above) was brought into the model obtained by DT training.

[0113] In summary, the present invention uses surface enhanced Raman technology to measure the spectral data of different edible oils and adulterated camellia oils. The spectral data of all edible oil samples in the training set are used as analysis data and imported into the Anaconda3-2023.07 analysis software to establish an edible oil classification and camellia oil authentication model. The model has an accuracy of 100% in identifying and classifying all edible oil samples in the training set; the spectral data of the test set samples are brought into the above identification model for analysis to classify edible oil types and authenticate camellia oil. The present invention can accurately classify different edible oils, and can identify adulterated camellia oil. At the same time, the minimum adulteration level that the model can identify is 0.5%, and the accuracy of the test set can reach up to 97.24%.

[0114] See attached Figure 2 As shown, the present invention compared aluminum sheets, NPAg sheets and NPAg sheet@0.02%ER with camellia oil and rhodamine 6G (R6G) to evaluate their enhancement effects. It is noteworthy that the NPAg sheet impregnated with 0.02% epoxy resin showed a superior SERS enhancement effect compared with the NPAg sheet without epoxy resin. It is speculated that the direct contact between the NPAg sheet and the tested CAO52 played a key role. The peroxide in the camellia oil may also have reacted to oxidize the nanosilver particles in the NPAg sheet, thereby weakening its enhancement effect. The NPAg sheet@0.02%ER is coated with a thin and transparent epoxy resin protective layer. This method ensures that the test substance CAO52 is isolated from the silver nanoparticles on the surface of the NPAg sheet and avoids direct contact. Therefore, the high transmittance of epoxy resin can be effectively utilized to focus the laser beam on the silver nanoparticles. In addition, the transparency of the substrate allows reverse excitation and collection through the substrate, resulting in clear and well-defined SERS spectral peaks.

[0115] See attached Figure 3As shown in the figure, the morphology and structure of silver flakes (A, D, G), NPAg flakes (B, E, H) and NPAg flakes@4MBN@0.02%ER (C, F, I) are shown by scanning electron microscopy images. Figure 3 In (A, D, G), the surface of the Ag sheet appears relatively smooth, with only some minor scratches. Figure 3 In (B, E, H), the morphological changes of the silver foil after polishing can be observed; the impact of the sandpaper is clearly visible, which destroys the protective layer on the surface of the silver foil and forms a grid-like pattern. This process ultimately produces NPAg sheets. Figure 3 As shown in (C, F, I), the surface appearance of NPAg sheet@4MBN@0.02%ER changes significantly after being immersed in epoxy resin; however, it still maintains transparency and the texture is still clearly visible after polishing, indicating that this substrate has excellent optical transparency. Figure 3 In (J, K, L), the water contact angles of Ag sheet, NPAg sheet, and NPAg sheet@4MBN@0.02%ER are slightly different, but they generally show hydrophilic properties. Figure 3 As shown in (L), after epoxy resin encapsulation, the contact angle is significantly reduced. Since the substrate is hydrophilic, it will interact with the water-soluble components in the oil. This interaction plays a key role in the SERS effect and amplifies the differences in the oil.

[0116] See attached Figure 4 As shown, 845 and 1078cm -1 The peaks at 1268 and 1654 cm correspond to the CC stretching vibrations in the methylene group; -1 The peak at 1301 cm is related to the vibration of C=C and CH bonds in unsaturated fatty acids; -1 The nearby peaks are related to CH bending vibrations in the methylene groups; 1442 cm -1 The peak at 1745 cm is related to the CH shear vibration in the methylene group; finally, -1 The peaks at correspond to the C=C=O stretching vibrations of the ester groups in triglycerides. These characteristic peaks are ubiquitous in almost all edible vegetable oils. Given the high similarity between most edible oils and their corresponding Raman spectra, visual inspection alone is insufficient to distinguish molecules with closely related structures and spectral peak positions. While the differences between these oils can be amplified by SERS, a strategy focused on enhancing spectral data addresses the challenges posed by the limited number of training samples, thereby achieving effective identification of pure camellia oil.

[0117] See attached Figure 5 As shown, from Figure 5(A, B, C) It can be seen that there are obvious differences between the results obtained by the ordinary Raman, SERS (uncalibrated) and SERS (calibrated) methods. The six machine learning algorithms all showed high accuracy on the test set. The accuracy of ordinary Raman on the test set is as follows: SVM (93.09%) > MLP (90.78%) > KNN (89.86%) > RF (88.94%) > XGBoost (88.02%) > DT (82.03%). For SERS (uncalibrated), the accuracy is KNN (95.85%) > XGBoost (94.01%) > MLP (92.17%) > RF (91.71%) > SVM (88.94%) > DT (83.41%). Finally, using the SERS (calibration) method, the accuracy results are as follows: KNN (97.24%) > XGBoost (94.01%) > RF (91.71%) = MLP (91.71%) > SVM (89.40%) = DT (88.48%). On the test set, the accuracy of ordinary Raman exceeded 82.03%, the accuracy of SERS (uncalibrated) exceeded 83.41%, and the accuracy of SERS (calibrated) exceeded 88.48%. It is worth noting that the calibrated machine learning has significantly improved the accuracy on both the training set and the test set, which highlights the beneficial effect of using 4MBN as a calibration internal standard molecule because it can improve the stability of the measurement. Among them, the K-nearest neighbor (KNN) algorithm showed excellent classification performance. The calibrated KNN model achieved an accuracy of 97.24% and an AUC of 0.9992 in the test set. This performance exceeds that of the conventional Raman and uncalibrated enhanced Raman methods, whose accuracies in the test set are 89.86% < 95.85% < 97.24% and AUC values ​​are 0.9642 < 0.9892 < 0.9992, respectively.

[0118] See attached Figure 6 As shown in the figure, it can be seen that the NPAg sheet@4MBN@0.02%ER substrate material is 2228cm after being treated in harsh environments and recycled and washed. -1 The signal fluctuation at the point is not large, indicating that the material has good alkali resistance, acid resistance, salt resistance, low temperature resistance, high temperature resistance, oil resistance and reusability.

[0119] See attached Figure 7 As shown, in Figure 7 (A, B) characteristic peak 2228cm -1 The vibration at 100 nm showed stability, with a relative standard deviation (RSD) value of 14.64%, which is lower than the critical value of 20%. This observation indicates that the material has good uniformity. Figure 7As can be seen from (C, D), when different batches of substrates are tested, the Raman characteristic peaks and Raman intensities do not change substantially, which indicates that the substrate has excellent reproducibility. This vibration stability and the characteristic peak at 2228cm -1 The low RSD (relative standard deviation) value of 3.99% strongly demonstrated the repeatability of the NPAg sheet@4MBN@0.02% ER substrate prepared by the present invention.

[0120] Table 1 is the detailed information of the sources of camellia oil and other vegetable oils in the present invention;

[0121]

[0122]

[0123]

[0124] Table 2 Composition of adulterated vegetable oil in the present invention

[0125]

[0126]

[0127] Table 3 is the composition of the adulterated camellia oil in the present invention

[0128]

[0129] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A low-cost method for identifying the authenticity of tea oil. It is characterized in that include: 1) Preparation of adulterated tea oil samples The D-optimal mixing design in the mixture design was used to prepare adulterated oils with different adulteration ratios to simulate the adulterated samples of camellia oil. 2) Surface enhanced Raman spectroscopy acquisition The surface enhanced Raman spectra of authentic camellia oil and adulterated camellia oil were collected respectively; 3) Classification of edible oils and establishment of camellia oil authentication model The surface enhanced Raman spectra of all samples were selected as machine learning analysis data, and the algorithm with classification function in the sklearn library of Anaconda3-2023.07 analysis software was used to process the surface Raman spectra data of the collected samples to establish a camellia oil authentication model; 4) Identification of camellia oil samples The surface enhanced Raman spectrum of the sample to be tested is collected, and then the collected surface enhanced Raman spectrum data is analyzed at 2228 cm -1 The internal standard is used for calibration, and then the calibrated surface-enhanced Raman spectroscopy data is input into the camellia oil authentication model to identify the sample.

2. A low-cost method for distinguishing the authenticity of tea oil according to claim 1, It is characterized in that The surface enhanced Raman is a nano-gold or silver substrate containing an internal standard with an enhancement effect; when collecting Raman spectra, the sample oil is dropped on the nano-gold or silver substrate containing an internal standard with an enhancement effect for detection.

3. A low-cost method for distinguishing the authenticity of tea oil according to claim 2, It is characterized in that The internal standard molecule is 4MBN.

4. A low-cost method for distinguishing the authenticity of tea oil according to claim 2, It is characterized in that The surface enhanced Raman spectroscopy acquisition conditions are as follows: The Raman spectrum is collected in the range of 0-3000cm -1 , using a 785nm excitation light source, an integrated power of 200Mw, an integration time of 5000ms, and scanning each sample 5-10 times.

5. A low-cost method for distinguishing the authenticity of tea oil according to claim 1, It is characterized in that The algorithms with classification functions in the sklearn library include DT, RF, KNN, SVM, XGBoost, and MLP.

6. A low-cost method for distinguishing the authenticity of tea oil according to claim 1, It is characterized in that The adulterated camellia oil in step 1) is authentic camellia oil mixed with peanut oil, soybean oil, corn oil, palm oil, rice oil, sunflower seed oil and linseed oil respectively, and the mixed mass percentage is 0.5% to 35%.

7. A low-cost method for distinguishing the authenticity of tea oil according to claim 3, It is characterized in that The method for preparing the nanosilver substrate containing an internal standard comprises the following steps: S1. Preparation of nanosilver sheet NPAg sheet: Grinding the silver foil with 240 sandpaper until the protective layer on the surface of the silver foil is removed and a grid pattern appears on the surface of the silver foil to form an NPAg sheet; S2. Preparation of NPAg sheet@4MBN: Immerse the NPAg sheet in 10 -3 mol / L 4MBN ethanol solution for 30 minutes, then taken out and air-dried at room temperature to form 4MBN-coupled silver foil NPAg sheet@4MBN; S3. Preparation of 0.02% ER: weigh 0.3 g epoxy resin and 0.1 g polyetheramine, keeping a ratio of 3:1, and stir the mixture with a spray gun for 5-10 minutes until the mixture becomes completely transparent without any filamentous substances; Subsequently, the mixture was placed in a vacuum oven at 35-40°C and evacuated for 5-10 minutes to remove bubbles. The mixture was heated to 35-40°C using a magnetic stirrer, and then 100 μL of epoxy resin and 900 μL of acetone were extracted using a pipette to prepare a 1% volume fraction solution. The 1% epoxy resin solution was further diluted with acetone to obtain a solution with a lower volume fraction. This process was repeated as needed to prepare a 0.02% volume fraction epoxy resin solution. S4. Preparation of NPAg sheet@4MBN@0.02%ER: Immerse the dried NPAg sheet@4MBN in a 0.02% epoxy resin solution for 2-3 seconds, take it out and place it with the polished surface facing up, and let it air-dry until solidified, thereby forming a transparent protective layer to obtain NPAg sheet@4MBN@0.02%ER.

8. A low-cost method for distinguishing the authenticity of tea oil according to claim 5, It is characterized in that Step 3) specifically includes the following steps: Data processing: The collected surface enhanced Raman spectroscopy data were processed at 2228 cm -1 For internal standard calibration, each data collected was selected from 400-3000cm -1 Spectral range, and then normalize the data respectively; Model training and evaluation: The raw spectral data points are randomly divided into training and test sets for machine learning analysis; Model training is performed on an algorithm with classification function, the algorithm performs grid search and uses a five-fold cross validation method for hyperparameter selection; The main performance criteria for evaluating the models include accuracy, confusion matrix, precision, recall, f1 score, ROC curve, and AUC.

9. A low-cost method for distinguishing the authenticity of tea oil according to claim 8, It is characterized in that The data volume ratio of the training set and the test set is 4:1.

Images