A method for rapid quantitative detection of matcha adulterants

Abstract

The present invention belongs to the field of laser spectroscopy technology, and discloses a method for rapid quantitative analysis of adulteration of matcha tea, including 1: preparing matcha tea samples with different adulteration concentrations, collecting spectral images and ablation image information of the samples; 2: preprocessing the collected LIBS spectrum and ablation image data to reduce the interference caused by noise and irrelevant factors; 3: extracting LIBS spectrum image features using XGBoost, and extracting image texture features using grayscale co-occurrence matrix; 4: fusing the extracted spectral features and image features, and using the fused information as a new feature of the sample; 5: using the fused feature information as input, and attaching data labels to matcha teas with different adulteration concentrations, and establishing a quantitative analysis model of the degree of adulteration of matcha tea based on LIBS spectrum and ablation image. The present invention realizes rapid quantitative detection of adulteration of matcha tea, and has the characteristics of low cost, simple operation, high efficiency, etc.

Description

Technical Field

[0001] The invention belongs to the technical field of laser spectroscopy, and in particular relates to a method for rapid quantitative detection of adulterants in matcha. Background Art

[0002] Matcha refers to a micro-powdered tea product made from fresh leaves of tea trees that are covered and cultivated, steamed (or hot aired), and dried, and then ground and processed. Matcha has high nutritional value and is a green and healthy food that can enhance emotional cognition, reduce stress and anxiety, improve the body's resistance and slow down aging. The demand for matcha is increasing year by year, and the market demand is large, with great room for development. How to detect the quality of matcha has become a demand.

[0003] Laser-induced breakdown spectroscopy (LIBS) is an atomic emission spectroscopy technique that can perform qualitative and quantitative analysis of the elemental composition in solid, liquid, and gas samples. It uses the plasma generated by a pulsed laser to ablate and excite the substances in the sample, and obtains the spectrum through a spectrometer to identify the elemental components in the sample. At present, due to the advantages of LIBS such as simple sample pretreatment, rapid detection, and real-time analysis, laser-induced breakdown technology has been widely used in different fields such as geological exploration, pharmaceuticals, environmental testing, and food quality testing.

[0004] Due to the morphology of matcha samples, the powder is easy to splash, which will affect the collection of spectral information. Most studies on powder adulteration detection use the direct tableting method to press the sample into flake particles to reduce the impact of powder splashing, but the sample preparation process is cumbersome and inefficient. In addition, due to the influence of factors such as laser energy and experimental environment, LIBS spectra are unstable. These fluctuations will affect the performance of quantitative analysis, which may limit the further promotion of LIBS in food adulteration. Summary of the invention

[0005] The object of the present invention is to provide a method for rapid and quantitative detection of matcha adulterants to solve the above-mentioned technical problems.

[0006] In order to solve the above technical problems, the specific technical scheme of a method for rapid quantitative detection of adulterants in matcha of the present invention is as follows:

[0007] A method for rapid quantitative analysis of matcha adulteration comprises the following steps:

[0008] Step 1: Prepare matcha samples with different adulteration concentrations, collect spectral images and ablation image information of the samples, and divide the samples into training sets and prediction sets;

[0009] Step 2: Preprocess the collected LIBS spectrum and ablation image data to reduce the interference caused by noise and irrelevant factors;

[0010] Step 3: Use XGBoost to extract LIBS spectral image features, and use gray-level co-occurrence matrix GLCM to extract image texture features. The spectral features are recorded as W1 and the image features are recorded as W2.

[0011] Step 4: Fuse the extracted spectral features and image features, and use the fused information as the new features of the sample, denoted as W;

[0012] Step 5: Use the fused feature information as input and label the data of matcha with different adulteration concentrations. Build a quantitative analysis model of matcha adulteration based on LIBS spectrum and ablation image to predict the adulteration concentration of matcha.

[0013] Furthermore, the sample in step 1 is specifically prepared by mixing matcha and adulterants in proportion and homogenizing them using a mixer, and the total mass of each sample is X g; the total number of samples is M+N, the number of training groups is M, and the number of prediction groups is N.

[0014] Furthermore, in the step 1, a proper amount of matcha mixture is weighed from each sample and dissolved in distilled water at room temperature, with the ratio of the two being 0.005-0.01; a magnetic stirrer is used to fully stir the solution so that the matcha particles are evenly suspended in the solution; a pipette is used to measure Y μl of the solution and transfer it to a filtration device; the filtration device includes a vacuum pump, a filter cup and a filter bottle; and the final sample for testing is obtained after filtration.

[0015] Furthermore, in step 1, the instrument parameters for collecting spectral signals are: laser pulse energy is 80 mJ; pulse frequency is 1 Hz; delay time is 2 μs; integration time is 10 μs; and spectral information of the sample is obtained within the range of 320-428 nm.

[0016] Furthermore, in step 1, the sample obtained from the filtration device is first allowed to stand for 3-5 minutes, and then n×n points are collected in a square area for each sample, with each point spaced 1 mm apart. Each point is struck by a laser once and a spectrum is collected; an image of the sample is collected using a camera of a self-designed LIBS rapid detection system.

[0017] Furthermore, the step 2 includes the following specific steps:

[0018] Step 2.1: First, average the spectra of n×n points collected for each sample to represent the spectral information of the sample;

[0019] Step 2.2: Perform standard normalization on the spectral information of the M+N group of samples to obtain the spectral matrix information, recorded as α(α1, α2, α3...α M+N );

[0020] Step 2.3: Extract the region of interest from the image, that is, select the range that includes the sample and all ablation pits, and the size of the region of interest is m×m pixels.

[0021] Furthermore, the step 3 includes the following specific steps:

[0022] Step 3.1: Use XGBoost to select feature variables in the entire spectrum according to the importance value. Adjust the parameters through ten interactive validations to determine the number of feature variables that can achieve the optimal model performance and obtain the spectral feature W1. The two key hyperparameters are n_estimators and learning rate, and their values ​​are 100 and 0.1 respectively.

[0023] Step 3.2: Extract six texture features of contrast, uniformity, energy, correlation, and angular second-order moment through GLCM analysis, and integrate the values ​​of these texture features into a matrix information representing the texture features of the sample, and then obtain the image features W2 of all samples.

[0024] Furthermore, in step 4, the spectral feature W1 and the image texture feature W2 are spliced ​​together to obtain the characteristic information W of the fused sample.

[0025] Furthermore, in step 5, a model result based on the fusion information as input is obtained, and the performance of regression models based on different inputs is compared to obtain the optimal quantitative analysis model of the corresponding adulteration concentration index.

[0026] Furthermore, in step 5, the correlation coefficient and root mean square error are used as indicators for evaluating model performance, the R value measures the relationship between the predicted adulteration content and the actual value, and the RMSE value measures the prediction error.

[0027] The method for rapid quantitative detection of matcha adulterants of the present invention has the following advantages:

[0028] (1) Rapid quantitative detection of matcha adulteration was achieved.

[0029] (2) A powder sample preparation method is proposed, which has the characteristics of low cost, simple operation, high efficiency, etc.; it avoids the cumbersome operation of the tableting method. In particular, the sample preparation proposed by the present invention can be made using only 0.1g of matcha raw material, while tableting requires about 0.5g of matcha raw material.

[0030] (3) The spectral information and image information of the tested samples are collected simultaneously, and the characteristic information of the samples is obtained at multiple levels to better ensure the accuracy and reliability of the model.

[0031] (4) The spectral fusion modeling method based on LIBS can provide assistance in the field of food quality inspection and also has potential practical application value in other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the rapid quantitative analysis of matcha adulteration based on LIBS spectrum fusion of the present invention;

[0033] Figure 2 The average spectra of matcha and adulterants are shown;

[0034] FIG3( a ) is a diagram showing the quantitative analysis results of the content of adulterants in Fuding Matcha based on the full spectrum;

[0035] FIG3( b ) is a diagram showing the quantitative analysis results of the adulterant content of Jiukeng Matcha based on the full spectrum;

[0036] FIG3(c) is a diagram showing the quantitative analysis results of the content of adulterants in Fuding Matcha based on spectral characteristics;

[0037] FIG3( d ) is a diagram showing the quantitative analysis results of the adulterant content of Jiukeng Matcha based on spectral characteristics;

[0038] FIG3(e) is a diagram showing the quantitative analysis results of the content of adulterants in Fuding Matcha based on the fusion of spectral features and image texture features;

[0039] Figure 3(f) is a quantitative analysis result of the adulterant content of Jiukeng Matcha based on the fusion of spectral features and image texture features. DETAILED DESCRIPTION

[0040] In order to better understand the purpose, structure and function of the present invention, the method for rapid quantitative detection of matcha adulteration of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0041] like Figure 1 As shown, a method for rapid quantitative analysis of matcha adulteration based on LIBS spectrum and image fusion includes the following steps:

[0042] Step 1: Prepare matcha samples with different adulteration concentrations, collect spectral images and ablation image information of the samples, and divide the samples into training sets and prediction sets;

[0043] The instrument parameters for collecting spectral signals are: laser pulse energy of 80 mJ; pulse frequency of 1 Hz; delay time of 2 μs; integration time of 10 μs; and spectral information of the sample is obtained in the range of 320-428 nm.

[0044] The sample is specifically prepared by dissolving, centrifugally stirring, and filtering the mixed sample of matcha and adulterants: first, weigh 0.1g of each mixed sample with an electronic balance and dissolve it in 20g of distilled water; then stir it thoroughly with a magnetic stirrer; use a pipette to transfer 400μl of the solution to a filtration device; after filtration, obtain matcha precipitation on a filter membrane with a diameter of 25mm; place the obtained precipitation on a 25mm×25mm flat plate for 3-5min, and then place it on an XYZ moving platform. In particular, the sample preparation method proposed in the present invention can prepare a set of samples in an average of 2 minutes. The total number of samples is M+N, the number of training sets is M, and the number of prediction sets is N. For each sample, n×n points are collected in a square area, each point is 1mm apart, each point is hit by a laser once and the spectrum is collected; the image of the sample is collected using a camera of a self-designed LIBS rapid detection system.

[0045] Step 2: Preprocess the collected LIBS spectrum and ablation image data to reduce the interference caused by noise and irrelevant factors; the preprocessing method includes:

[0046] Step 2.1: The spectra of all points collected for each sample are averaged to represent the spectral information of the sample;

[0047] Step 2.2: Perform standard normalization on the spectral information of all samples to obtain full spectrum information;

[0048] Step 2.3: Extract the region of interest (ROI) of the ablation image, that is, select the range that includes the sample and all ablation pits.

[0049] Step 3: Use XGBoost to extract LIBS spectral image features, and use gray-level co-occurrence matrix (GLCM) to extract image texture features. The spectral features are recorded as W1 and the image features are recorded as W2.

[0050] XGBoost and GLCM were used to extract features from LIBS spectra and ablation images, respectively.

[0051] Step 3.1: Use XGBoost to select feature variables in the entire spectrum according to the importance value, adjust the parameters through ten interactive verifications to determine the number of feature variables that can achieve the optimal model performance and obtain the spectral feature W1;

[0052] Step 3.2: Extract six texture features of contrast, uniformity, energy, correlation, and angular second-order moment through GLCM analysis, and integrate the values ​​of these texture features into a matrix information representing the texture features of the sample, and then obtain the image features W2 of all samples.

[0053] Step 4: Fuse the extracted spectral features and image features, and use the fused information as the new features of the sample, denoted as W;

[0054] The spectral feature W1 and the image feature W2 are concatenated together to obtain a new matrix W representing the sample information.

[0055] Step 5: Use the fused feature information as input and label the data of matcha with different adulteration concentrations. Build a quantitative analysis model of matcha adulteration based on LIBS spectrum and ablation image to predict the adulteration concentration of matcha.

[0056] The model results based on the fusion information as input are obtained, and the performance of the regression models based on different inputs is compared to obtain the optimal quantitative analysis model of the corresponding adulteration concentration index. The present invention uses the correlation coefficient (R) and the root mean square error (RMSE) as indicators for evaluating the performance of the model, the R value measures the relationship between the predicted adulteration content and the actual value, and the RMSE value measures the prediction error.

[0057] Example:

[0058] 1. First-grade matcha (Fuding), second-grade matcha (Jiukeng) and barley grass powder were prepared, and sensory evaluation of the two types of matcha was performed. Therefore, we regarded the first-grade matcha and the second-grade matcha as pure matcha, and the barley grass powder as an adulterant. The preparation method of matcha adulterants is to mix barley grass powder with first-grade matcha and second-grade matcha. In order to establish a model to quantify the content of adulterants, first-grade matcha and second-grade matcha were adulterated with barley grass powder at 21 different percentages (0%, 5%, 10%, 15%, 20%, 25%, 30%, 35%, 40%, 45%, 50%, 55%, 60%, 65%, 70%, 80%, 85%, 90%, 95% and 100%). In addition, adulterated samples for external prediction were prepared at 13 different adulteration rates, namely 0%, 8%, 16%, 24%, 32%, 40%, 48%, 56%, 64%, 72%, 80%, 88% and 96%. Each adulteration concentration was tested in triplicate, so there were 63 mixed samples for training calibration and 39 mixed samples for prediction. After mixing, all samples were placed in a centrifuge tube and homogenized using an adjustable mixer. Due to the physical properties of the powder, the powder will splash during laser ablation, which may affect the stability of the LIBS signal. Therefore, we proposed a novel sample preparation method for LIBS analysis. First, 0.1 g of each mixed sample was weighed using an electronic balance and dissolved in 20 g of distilled water; then the mixture was stirred thoroughly using a magnetic stirrer; 400 μl of the solution was transferred to a filtration device using a pipette; after filtration, a matcha precipitate was obtained on a filter membrane with a diameter of 25 mm; the obtained precipitate was placed on a 25 mm × 25 mm plate, which was placed on an XYZ moving platform.

[0059] 2. Collect a spectrum of 8×8 points for each sample, with a total of 64 points, and collect images of the sample after ablation.

[0060] 3. Preprocess the collected spectral information and image information: average the spectrum of 64 points of each sample to obtain a spectral information representing the sample, such as Figure 2 The following table shows the average spectra of first-grade matcha, second-grade matcha and barley grass powder. The spectra of all samples after averaging are normalized to obtain the spectral matrix information α (α1, α2, α3...α 102 ); extract the region of interest (ROI) of the image, that is, select the range including the sample and all the ablation pits, and the size of the region of interest is 730×730 pixels.

[0061] 4. Use XGBoost and GLCM to extract features from LIBS spectra and ablation images respectively. 1. Use XGBoost to select feature variables in the entire spectrum according to the importance value, and adjust the parameters through ten interactive verifications to determine that the number of feature variables that can achieve the best model performance is 20. The two key hyperparameters are n_estimators (nt) and learning rate (lr), and their values ​​are 100 and 0.1 respectively. The spectral feature extracted by XGBoost is W1, and the matrix size is (102×20). Table 1 shows the top 10 spectral features with the most important values ​​selected by XGBoost; 2. GLCM analysis can provide some second-order statistical information to describe the grayscale relationship in the neighborhood around the pixels in the area of ​​interest. Six texture features, namely contrast, uniformity, energy, correlation, and angular second-order moment, are extracted from the GLCM calculation. In particular, texture features are extracted from 5 distances from 1 to 5 pixels and 4 directions (0°, 45°, 90°, 135°) for each image, and then the values ​​of these texture features are integrated into a matrix information representing the texture features of the samples. The texture features of all samples are W2, with a size of (102×24).

[0062] Table 1: Feature variables selected for matcha adulteration analysis using XGBoost.

[0063]

[0064] Note: a Important variables selected from Fuding Matcha. b Important variables selected from Jiukeng Matcha.

[0065] 5. The spectral features and image features of the sample are fused. Specifically, the spectral features are in front and the image features are in the back. The matrix W obtained after fusion has a size of 102×44.

[0066] 6. Input the fused data W into the support vector regression (SVR) model, specifically: use the training set data to train and build the model, adjust the parameters, and obtain a model with the best performance. Then use the data of the prediction set as the evaluation criteria, and use the correlation coefficient (R) and the root mean square error (RMSE) as indicators for evaluating model performance. The R value measures the relationship between the predicted adulteration content and the actual value, and the RMSE value measures the prediction error. The performance of the model with fused data as input is compared with the model with only spectral data as input. It is found that after feature fusion, the performance of the SVR model is significantly improved. Figure 3(a)-3(f) The performance comparison of the models based on different inputs for two types of matcha adulteration is shown.

[0067] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A method for rapid quantitative analysis of matcha adulteration, characterized in that: The steps include: Step 1: Prepare matcha samples with different adulteration concentrations, collect spectral images and ablation image information of the samples, and divide the samples into training sets and prediction sets; Step 2: Preprocess the collected LIBS spectrum and ablation image data to reduce the interference caused by noise and irrelevant factors; Step 3: Use XGBoost to extract LIBS spectral image features, and use gray-level co-occurrence matrix GLCM to extract image texture features. The spectral features are recorded as W1 and the image features are recorded as W2. Step 4: Fuse the extracted spectral features and image features, and use the fused information as the new features of the sample, denoted as W; Step 5: Use the fused feature information as input and label the data of matcha with different adulteration concentrations. Build a quantitative analysis model of matcha adulteration based on LIBS spectrum and ablation image to predict the adulteration concentration of matcha.

2. The method for rapid quantitative analysis of matcha adulteration according to claim 1, characterized in that: The sample in step 1 is specifically prepared by mixing matcha and adulterants in proportion and homogenizing them using a mixer. The total mass of each sample is Xg; the total number of samples is M+N, the number of training groups is M, and the number of prediction set groups is N.

3. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: In the step 1, a proper amount of the matcha mixture is weighed from each sample and dissolved in distilled water at room temperature, with the ratio of the two being 0.005-0.01; the solution is fully stirred using a magnetic stirrer, and the matcha particles are evenly suspended in the solution; Use a pipette to measure Y μl of solution and transfer it to a filtration device; the filtration device includes a vacuum pump, a filter cup and a filter bottle; after filtration, the final sample for detection is obtained.

4. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: In step 1, the instrument parameters for collecting spectral signals are: laser pulse energy of 80 mJ; pulse frequency of 1 Hz; delay time of 2 μs; integration time of 10 μs; and spectral information of the sample is obtained within the range of 320-428 nm.

5. The method for rapid quantitative analysis of matcha adulteration according to claim 1, characterized in that: In the step 1, the sample obtained from the filtration device is first allowed to stand for 3-5 minutes, and then n×n points are collected in a square area for each sample, with each point spaced 1 mm apart, and each point is struck once by the laser and the spectrum is collected; The image of the sample was collected using the camera of the self-designed LIBS rapid detection system.

6. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: The step 2 comprises the following specific steps: Step 2.1: First, average the spectra of n×n points collected for each sample to represent the spectral information of the sample; Step 2.2: Perform standard normalization on the spectral information of the M+N group of samples to obtain the spectral matrix information, recorded as α(α1, α2, α3...α M+N ); Step 2.3: Extract the region of interest from the image, that is, select the range that includes the sample and all ablation pits. The size of the region of interest is m×m pixels.

7. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: The step 3 comprises the following specific steps: Step 3.1: Use XGBoost to select feature variables in the entire spectrum according to the importance value. Adjust the parameters through ten interactive validations to determine the number of feature variables that can achieve the optimal model performance and obtain the spectral feature W1. The two key hyperparameters are n_estimators and learning rate, and their values ​​are 100 and 0.1 respectively. Step 3.2: Extract six texture features of contrast, uniformity, energy, correlation, and angular second-order moment through GLCM analysis, and integrate the values ​​of these texture features into a matrix information representing the texture features of the sample, and then obtain the image features W2 of all samples.

8. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: In step 4, the spectral feature W1 and the image texture feature W2 are spliced ​​together to obtain the characteristic information W of the fused sample.

9. The method for rapid quantitative analysis of matcha adulteration according to claim 1, characterized in that: In step 5, a model result based on the fusion information as input is obtained, and the performance of regression models based on different inputs is compared to obtain the best quantitative analysis model for the corresponding adulteration concentration index.

10. The method for rapid quantitative analysis of adulteration of matcha according to claim 1, characterized in that: In step 5, the correlation coefficient and the root mean square error are used as indicators for evaluating the performance of the model, the R value measures the relationship between the predicted adulteration content and the actual value, and the RMSE value measures the prediction error.

Images