A multi-element rapid detection system and method in food based on multi-voltage portable XRF

Abstract

The application discloses a kind of based on multi-voltage portable XRF in food Multi-element rapid detection system and method, method includes: step one, set sample pretreatment and measurement specification;Step two, multi-voltage excitation mechanism;Step three, background deduction and signal correction method;Step four, multi-model prediction mechanism;Step five, intelligent matrix matching and FP method trigger mechanism.The application also provides a kind of based on multi-voltage portable XRF in food Multi-element rapid detection system.The application detection element range is wide, detection limit is low, adaptability is strong, can be fast in situ detection, anti-interference ability is strong, intelligent degree is high.

Description

TECHNICAL FIELD

[0001] The application relates to a food detection and analysis system and method, in particular to a multi-element rapid detection system and method for food based on a multi-voltage portable XRF, and belongs to the technical field of food detection and analysis. BACKGROUND

[0002] At present, the detection of element content in food mainly depends on large laboratory instruments (such as ICP-MS, AAS, etc.), which has problems such as complex sample pretreatment, long detection period and inability to conduct on-site rapid screening. Although portable X-ray fluorescence spectrometers (XRF spectrometers) have been used for element detection in the fields of environment and geology, they still have the following deficiencies in food detection: the excitation efficiency of light elements (such as P, S, Cl, K and Ca) in food is low, and it is difficult to accurately detect them; the food matrix is complex and diverse, and a single excitation condition cannot meet the detection requirements of light elements and heavy metal elements. The existing technology for XRF detection of food is mostly aimed at the detection of specific elements in food. For example, a Chinese invention patent with the publication number CN115389542A provides a method for detecting the content of aluminum elements in food. The original aluminum-containing sample is prepared into a powder film sample, and a very thin powder sample can be obtained. When the single-color focused X-ray fluorescence analysis equipment is used to irradiate the very thin powder sample, the excited aluminum characteristic X-rays and scattered rays are emitted from the powder sample and enter the equipment, without additional scattered ray background sources, which can effectively improve the signal-to-background ratio of aluminum element characteristic X-rays.

[0003] At the same time, the existing technology also lacks special background subtraction, signal correction and matrix matching methods for food samples, resulting in low detection accuracy and limited application range. SUMMARY

[0004] The purpose of the application is to provide a multi-element rapid detection system and method for food based on a multi-voltage portable XRF, which realizes rapid, accurate and on-site detection of light elements (P, S, Cl, K and Ca) and main elements (V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Cd, Hg and Pb) in food through multi-voltage cooperative excitation, background subtraction and signal correction, multi-model prediction and matrix matching technologies.

[0005] The application is implemented as follows:

[0006] A multi-element rapid detection method for food based on a multi-voltage portable XRF, comprising:

[0007] Step 1: Set sample pretreatment and measurement specifications

[0008] Dry, crush and sieve the sample to be tested through a 200-mesh sieve;

[0009] The sample to be tested is packaged using a special sample cup and Mylar film;

[0010] The total measurement time is set to 30 seconds to ensure signal stability.

[0011] Step two, multi-voltage excitation mechanism:

[0012] 1. Low voltage (8kV) without filter to excite light elements such as P, S, Cl, K, Ca;

[0013] 2. High voltage (40kV) with filter to excite main elements such as V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Cd, Hg, Pb;

[0014] 3. Voltage switching to maximize element excitation efficiency.

[0015] Step three, background subtraction and signal correction method:

[0016] 1. Use pure glucose as background standard substance;

[0017] 2. Low voltage based on Ag-La peak ratio to subtract background;

[0018] 3. High voltage based on Ag-Ka Compton scattering peak ratio to subtract background;

[0019] 4. Use Compton scattering normalization to correct measurement gap and matrix effect.

[0020] Step four, multi-model prediction mechanism:

[0021] 1. Light elements use standard curve method (linear / quadratic fitting) + basic parameter method (FP method);

[0022] 2. Main elements use standard curve method (linear + quadratic fitting), combined with interference factor matrix to strip overlapping peak interference;

[0023] 3. Built-in 6 common food matrix models (rice, wheat, red beans, cabbage leaves, rape leaves, Sanqi) and global model, select the optimal light element prediction model through spectral similarity matching.

[0024] Step four specifically includes: multi-model construction and intelligent prediction

[0025] Multi-model construction:

[0026] Light element model library: Because of the variety of food and the large fluctuation range of light element content (generally 0.01%-10%), and the signal of light element is seriously affected by the matrix effect, multiple typical food matrices (rice, wheat, red beans, cabbage, rape, and Sanqi) are selected to establish the calibration model. Each matrix model contains two kinds of relationships: quadratic fitting and linear fitting.

[0027] The specific light element modeling process (taking rice matrix as an example) is as follows: First, purchase GBW series of rice standard materials, dilute the standard materials with glucose (purity > 99.5%) to different gradients to prepare calibration samples, and the specific steps of sample packaging, measurement, and data processing are as follows:

[0028] ①After grinding or crushing, the sample is passed through a 200-mesh sieve and packaged with a sample cup and a Mylar film as a calibration sample;

[0029] ②The calibration sample is measured at high and low voltages for 30 s (low voltage spectrum is used to calculate light element peak area, and high voltage spectrum is used for Compton scattering normalization processing of peak area);

[0030] ③The blank material glucose is also measured at high and low voltages for 30 s;

[0031] ④Determine the channel address interval corresponding to the La peak energy of Ag in the low voltage spectrum and the channel address interval of the Compton scattering peak (hereinafter referred to as Compton scattering peak) corresponding to the Ka peak energy;

[0032] ⑤Low voltage spectrum background subtraction: subtract the low voltage spectrum of glucose multiplied by the proportionality coefficient from the low voltage spectrum, and the proportionality coefficient is obtained by the La peak area of Ag in the calibration sample and the glucose sample.

[0033] ⑥Light element peak area calculation: calculate the light element peak area according to the channel address interval corresponding to the Ka peak energy of the light element.

[0034] ⑦Compton scattering peak area calculation: calculate the peak area of Compton scattering in the high voltage spectrum according to the channel address interval corresponding to the Compton scattering peak energy;

[0035] ⑧Compton scattering normalization: correlate the calculated light element peak area to the Compton scattering peak area for normalization processing, and finally obtain the normalized light element peak area;

[0036] ⑨ Model Fitting: Then, the normalized peak area and the corresponding actual concentration of the element (the actual concentration is obtained from the GBW standard material certification and dilution gradient) are fitted and regressed to establish a standard curve (first-order and second-order fitting) as described in step four. The above is the modeling method for the rice matrix model; the modeling process for other food matrices is the same as for rice. In particular, first-order and second-order fitting models for six food matrices were finally established, and the parameters of these models (first-order coefficients, second-order coefficients, zero-order coefficients, slope, intercept, etc.) will be built into the prediction model.

[0037] ⑩ Global Model Construction: In order to deal with other situations, the normalized peak area of ​​the elements in all the above calibration samples and their actual element content are linearly fitted to obtain the linear fitting model used for the global model. The parameters of the linear fitting model (slope, intercept, etc.) are also built into the prediction model.

[0038] Food matrix fingerprint database: Following steps ①-⑧, obtain the normalized peak areas of light elements for six original GBW standard substances. This yields a column vector (denoted as s2) containing the normalized peak areas of light elements. There are six column vectors in total (one for each food matrix), which will be incorporated into the prediction model. The format of this column vector is as follows:

[0039]

[0040] Wherein, Normal_P represents the normalized peak area of ​​element P, Normal_S represents the normalized peak area of ​​element S, Normal_Cl represents the normalized peak area of ​​element Cl, Normal_K represents the normalized peak area of ​​element K, and Normal_Ca represents the normalized peak area of ​​element Ca.

[0041] Unified principal element model: Since the content of principal elements in food is generally low (usually < 0.1%), the matrix effect generally has little impact on the signal of principal elements at this content, therefore, matrix differentiation is no longer required. The specific modeling process is as follows:

[0042] ① Calibration Sample Preparation: First, purchase GBW series standard substances and multi-element standard liquids (containing all major elements to be analyzed) certified by a metrology institution. Measure a certain volume of the multi-element standard liquid and add it to a certain mass of GBW standard substances. After thorough stirring, freeze-dry the mixture. Then, thoroughly pulverize the dried solid. Next, mix the sample powder with glucose in different mass ratios, grind thoroughly, mix well, and pass through a 200-mesh sieve to prepare major element calibration samples with different concentration gradients.

[0043] ② Calibration sample measurement: After all the main element calibration samples have been prepared, measure the spectra of these calibration samples under high voltage for 30 seconds.

[0044] ③ High voltage spectral background subtraction: Subtract the high voltage spectrum of glucose multiplied by a scaling factor from the high voltage spectrum of the calibration sample. The scaling factor is obtained by calculating the area of ​​the Compton scattering peak in the calibration sample and the glucose sample.

[0045] ④ Calculation of the peak area of ​​the main element: The peak area of ​​the main element is calculated based on the channel address interval corresponding to the Ka peak energy of the main element;

[0046] ⑤ Compton scattering peak area calculation: Calculate the peak area of ​​Compton scattering in the high voltage spectrum based on the channel address range corresponding to the Compton scattering energy of the Ka peak of Ag.

[0047] ⑥Compton scattering normalization: The calculated principal element peak area is correlated with the Compton scattering peak area and normalized to obtain the normalized principal element peak area.

[0048] ⑦ Overlapping Peak Interference Removal of Main Elements: After calculating the normalized peak area, it needs to be multiplied by an interference factor to obtain the net intensity after eliminating overlapping peak interference. The interference factor actually indicates the mutual interference coefficient between main elements, and its calculation method is as follows: First, purchase the pure substance corresponding to the main element to be analyzed (e.g., for Fe, purchase iron powder with a purity > 99.99%). Measure the spectrum of these pure substances under high voltage and then calculate the peak area corresponding to the Ka peak of all main elements in each pure substance. Finally, an n-order square matrix (n being the number of main elements) containing the peak areas of each element is obtained. The interference factor matrix is ​​obtained by performing a series of matrix operations on this matrix. Then, matrix operations are performed on the normalized peak area and the inverse matrix of the interference factor to obtain the net peak area after eliminating overlapping interference.

[0049] ⑧ Model Fitting: Then, the net peak area and the corresponding actual concentration of the element (the actual concentration is determined by the amount of multi-element standard liquid and GBW standard substance mixed) are fitted and regressed to establish the standard curve (first-order fitting and second-order fitting) as described in step four. The relevant parameters of these standard curves (first-order coefficient, second-order coefficient, zero-order coefficient, slope, intercept, etc.) will also be built into the prediction model.

[0050] Ultralight element content prediction model: This model is built to predict the ultralight element content in unknown samples. This will determine whether to use the basic parameter method for light element concentration prediction. The specific process is as follows:

[0051] ① Obtain the actual content of ultralight elements based on the purchased GBW series standard materials;

[0052] ② Obtain the spectra of these standard substances under high voltage;

[0053] ③ Determine the channel address range corresponding to the energy of the Compton scattering peak in the spectrum;

[0054] ④ Calculate the peak area of ​​the Compton scattering peak;

[0055] ⑤ Perform a second fitting between the actual content and peak area of ​​the ultralight elements to obtain the second fitting parameters. Finally, the second fitting parameters for predicting the content of ultralight elements will also be built into the prediction model.

[0056] Fundamental parameter (FP) method model construction: The FP method model requires accurate knowledge of the physical parameters of the target element and relevant instrument parameters. Then, the actual content of the target element is calculated iteratively by calculating the spectral line intensities of the element in the unknown sample. The specific process is as follows:

[0057] ① A pseudo-element C (atomic number 6) was set to replace the ultralight element. Since the content of main elements in food is low and the voltage of 8 kV when measuring light elements is not enough to excite most main elements, the elements finally used for FP method iterative calculation include C, P, S, Cl, K, and Ca.

[0058] ② Physical parameter acquisition: Physical parameters such as fluorescence yield, mass absorption coefficient, absorption limit transition factor, and spectral line fraction of elements are obtained from public authoritative databases, and instrument parameters such as filters, targets, X-ray tubes, and detectors of instruments are obtained from instrument manufacturers. These parameters are all built into the prediction model.

[0059] ③ FP method establishment: Based on the obtained normalized peak area of ​​light elements, primary fluorescence and secondary fluorescence formulas, calculation iteration is performed. After reaching the convergence target, the prediction result is output. The entire FP method model is built into the prediction model.

[0060] ④ Calibration Factor Acquisition: Due to the non-negligible error of FP, the final output results will be additionally calibrated. First, the FP method is used to verify the GBW standard material with known element content. Based on the FP method results and the actual element content of the standard material, a linear fitting model is established. The fitting parameters of this model (i.e., the calibration factor) are used to calibrate the FP method prediction results for unknown samples. This calibration factor is also built into the prediction model.

[0061] Intelligent prediction process:

[0062] Light element prediction:

[0063] A. Spectral measurement, spectral processing, and spectral calculation of the sample to be tested: To reduce time consumption, the total measurement time for the sample to be tested was fixed at 30 s while ensuring measurement accuracy, of which 20 s was for spectral measurement under low voltage and 10 s was for spectral measurement under high voltage. The spectral processing and data calculation of the sample to be tested were strictly carried out according to step four - multi-model construction - light element model library - ④⑤⑥⑦⑧. Finally, the normalized peak area (column vector, denoted as s1) and Compton scattering peak area of ​​light elements in the unknown sample were obtained. Substituting the Compton scattering peak area into the ultralight element content prediction model, the ultralight element content can be obtained.

[0064] B. Matrix Matching: Calculate the similarity (correlation coefficient R², Euclidean distance) between the peak area of ​​light elements in the sample and the fingerprint vector s² of each food matrix, and select the best matching model. The specific process is as follows:

[0065] ① Calculate the correlation coefficient and Euclidean distance of the six fingerprint vectors s2 and s1 respectively. The closer the correlation coefficient is to 1, the higher the similarity between the two vectors, which means that the matrix of the unknown sample is close to the matrix of the corresponding standard material. The smaller the Euclidean distance, the higher the similarity between the two vectors, which means that the matrix of the unknown sample is close to the matrix of the corresponding standard material.

[0066] ④ Set matching conditions: If the optimal matching results of the correlation coefficient and Euclidean distance point to the same standard material matrix, then the subsequent predictions are based on that standard material matrix; otherwise, it is first determined whether the correlation coefficient is greater than 0.99. If it is, the subsequent predictions are based on the optimal matching result of the correlation coefficient. If none of the above conditions are met, it means that the unknown sample cannot be preferentially matched to the 6 standard material matrices. Therefore, the subsequent predictions for this sample will use a global linear fitting model as a fallback.

[0067] C. Preferred Quadratic Fit: Concentration prediction is performed using the quadratic fitting formula of the matching model. The specific process is as follows:

[0068] ①The column vector s1 containing the peak areas of each light element has been obtained according to step A;

[0069] ② Based on the above multi-model construction steps, a prediction model containing the quadratic fitting parameters (quadratic fitting: first-order coefficient, quadratic coefficient, zero-order coefficient) of each model has been established;

[0070] ③ Perform algebraic operations on the peak area of ​​each element in s1 and the corresponding quadratic fitting parameters to finally output the predicted concentration.

[0071] D. Linear Fitting Backup: If the quadratic fitting result exceeds the calibration range by a certain degree (more than 20%), then the linear fitting result of this model will be used instead. The specific process is as follows:

[0072] ①The column vector s1 containing the peak areas of each element has been obtained according to step A;

[0073] ②Based on the above multi-model construction steps, a prediction model containing the linear fitting parameters (slope, intercept) of each model has been established;

[0074] ③ Perform algebraic operations on the peak area of ​​each element in s1 and the corresponding linear fitting parameters of that element to finally output the predicted concentration.

[0075] E. Global Model as a Safety Net: If no matching matrix is ​​available, a global linear model is used for prediction. The specific process is as follows:

[0076] ①The column vector s1 containing the peak areas of each element has been obtained according to step A;

[0077] ②Based on the above multi-model construction steps, a prediction model containing global model linear fitting parameters (slope, intercept) has been established;

[0078] ③ Perform algebraic operations on the peak area of ​​each element in s1 and the corresponding global linear fitting parameters of that element to finally output the predicted concentration.

[0079] F. FP Method as a fallback: If the predicted concentration of the above model significantly exceeds the calibration range (exceeding 50%) or the predicted ultralight element content is ≤ 96%, the FP method is triggered for prediction. The original results are discarded, and the FP method results are corrected using a calibration factor. The specific process is as follows:

[0080] ① Determination of ultralight element content: If the ultralight element content predicted in step A is ≤ 96%, skip the model matching process and directly use the FP method for prediction.

[0081] ② Model matching prediction result judgment: Determine whether the concentration of the prediction results in steps C, D, and E seriously exceeds the calibration range (exceeding 50%). If it does, perform FP method prediction and discard the original prediction results.

[0082] ② FP method result calibration: Perform algebraic operations on the prediction results and calibration factors of the FP method to calibrate the prediction results.

[0083] G. Light element output result processing: The final prediction result will be compared with the pre-set LOD and LOQ, and results that are lower than LOD or LOQ will be marked.

[0084] Primary element prediction:

[0085] A. Based on the intelligent prediction process – light element prediction – step A, the low-voltage spectrum and high-voltage spectrum of the sample to be tested have been obtained; based on the obtained high-voltage spectrum, the net peak area of ​​the main element is obtained by strictly following steps ③-⑦ of step four – multi-model construction – unified model of main element.

[0086] A. Substitute the net peak area into the quadratic fitting model and output the prediction results.

[0087] B. If the predicted concentration is abnormally high (> 2000 ppm), switch to the linear fitting model for prediction.

[0088] C. Main element output result processing: The final prediction result will be compared with the pre-set LOD and LOQ, and results that are lower than LOD or LOQ will be marked.

[0089] The present invention also provides a rapid detection system for multiple elements in food based on multi-voltage portable XRF, which adopts the rapid detection method for multiple elements in food based on multi-voltage portable XRF provided by the present invention.

[0090] A further step is:

[0091] The multi-element rapid detection system for food based on multi-voltage portable XRF has the following built-in files:

[0092] The spectra of glucose at high and low voltages; the energy channels of each element; various correction parameters (slope, intercept, basic X-ray fluorescence parameters, etc.).

[0093] This invention has at least the following outstanding technical effects:

[0094] 1. Wide range of elements to be detected: Covers both nutrients and harmful heavy metals, meeting the needs of food safety testing;

[0095] 2. Low detection limits: LOD (limit of quantitation) for light elements is approximately 50 ppm, and LOD for major elements is approximately 10 ppm;

[0096] 3. High adaptability: Through multi-matrix model matching and FP embedding, it can adapt to various food types;

[0097] 4. Enables rapid on-site testing: Especially for powder samples, no complex pretreatment is required, and dual voltage measurement can be completed within 30 seconds;

[0098] 5. Strong anti-interference capability: Background subtraction and Compton normalization effectively suppress matrix effects and measurement gap influence;

[0099] High level of intelligence: Automatically matches models and determines whether to enable FP method, and is easy to operate. Attached Figure Description

[0100] Figure 1 This is a flowchart of a rapid detection method for multiple elements in food based on multi-voltage portable XRF according to an embodiment of the present invention;

[0101] Figure 2 This is a raw spectral image of tea leaves in one embodiment of the present invention;

[0102] Figure 3 This is a spectral image of tea leaves after background subtraction in one embodiment of the present invention. Detailed Implementation

[0103] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0104] Example 1

[0105] As attached Figure 1 As shown, this embodiment provides a rapid detection method for multiple elements in food based on multi-voltage portable XRF, including:

[0106] Step 1: Set sample pretreatment and measurement specifications

[0107] The sample to be tested is dried, crushed, and passed through a 200-mesh sieve;

[0108] The sample to be tested was encapsulated using a special sample cup and Mylar membrane;

[0109] Set the total measurement time to 30 seconds to ensure signal stability;

[0110] Step 2: Multi-voltage excitation

[0111] (1) Light elements are excited using low voltage (8KV / 25μA) without filters;

[0112] (2) High voltage (40KV / 30μA) with filter is used to excite the main elements;

[0113] (3) Maximize element excitation efficiency by switching voltage;

[0114] Step 3: Background Subtraction and Signal Correction

[0115] (1) Pure glucose was used as the background standard.

[0116] (2) Background subtraction based on Ag-La peak ratio at low voltage;

[0117] (3) Background subtraction based on the proportion of Ag-Ka Compton scattering peaks under high voltage;

[0118] (4) Use Compton scattering normalization to correct for measurement gap and matrix effects;

[0119] Step 4: Multi-model prediction mechanism

[0120] (1) Light elements are detected using the standard curve method + basic parameter method;

[0121] (2) The principal element is obtained by using the standard curve method and combining it with the interference factor matrix to remove overlapping peaks;

[0122] (3) It has six common food matrix models and a global model built in, and selects the optimal model by matching spectral similarity.

[0123] Step 5: Smart matrix matching and FP method triggering mechanism:

[0124] (1) Based on the similarity matching between the peak area of ​​light elements and the matrix model;

[0125] (2) When the predicted concentration exceeds the calibration range or the content of ultralight elements is ≤96%, the FP method will be automatically activated;

[0126] Finally, multi-element prediction results are obtained.

[0127] The detailed process of the present invention will be further described below with a more specific embodiment.

[0128] Example 2

[0129] Elemental analysis of green tea leaves

[0130] ① Construction of the calibration model: A complete prediction model was established strictly following steps one through four in the invention description. This model includes the interference factor matrix of the principal element, the concentration curves s² of the six food matrices, the linear and quadratic fitting parameters for each prediction model, the lower limit of quantitation (LOQ) and lower limit of detection (LOD) for each prediction model, the upper limit of calibration concentration for each prediction model, the calibration factor for FP method result calibration, the channel address interval required for calculating the peak area of ​​each element, the quadratic fitting parameters for predicting the content of ultralight elements, and the physical and instrumental parameters of the elements predicted by the FP method. Table 1 shows the LOD and LOQ of the global linear model for light elements and the unified model for principal elements in the prediction model constructed in this invention.

[0131] Table 1. LOD and LOQ of the light element global linear model and the principal element unified model in the prediction models.

[0132]

[0133] ② Sample Preparation and Measurement: Take 50 g of green tea leaves, grind them thoroughly, sieve them through a 200-mesh sieve, mix them thoroughly, place them in a special measurement cup, and cover them with a special Mylar membrane. Prepare three parallel samples in total. Measure the spectrum of the tea sample at low voltage (8 kV) (denoted as Tea_Spectra_Low, measurement time 20 s), and measure the spectrum of the tea sample at 40 kV (denoted as Tea_Spectra_High, measurement time 10 s). Measure the spectra of pure glucose at 8 kV and 40 kV, denoted as BG1 and BG2. The original spectra of the tea obtained are attached. Figure 2 As shown.

[0134] ③ Background subtraction and signal correction of the spectral spectrum of the sample:

[0135] The sample spectral background subtraction and signal correction formulas shown in Table 2 were applied.

[0136] Calculate the La peak area (Ag_Tea_La) of Ag in the sample according to formula (1);

[0137] Calculate the Ka Compton scattering peak area (Ag_Tea_Ka_Compton) of Ag in the sample according to formula (2).

[0138] Calculate the background subtraction coefficient (Ratio1) of the voltage spectrum of the sample under test according to formula (3).

[0139] The background subtraction coefficient (Ratio2) of the high voltage spectrum of the sample to be tested is calculated according to formula (4).

[0140] The low-voltage spectral background is subtracted using the proportional method according to formula (5);

[0141] The high-voltage spectral background is subtracted using the proportional method according to formula (6);

[0142] Calculate the normalized peak area (Peak_Area_Light_i) of light elements according to formula (7);

[0143] Calculate the peak area (Peak_Area_Main_i) of the principal element normalization according to formula (8);

[0144] The interference of overlapping peaks of the principal elements is stripped according to formula (9);

[0145] Table 2 Formulas for Spectral Background Subtraction and Signal Correction of Samples

[0146]

[0147] Wherein, Sum represents summation; BG1 is the blank for glucose under low voltage in the prediction model; BG2 is the blank for glucose under high voltage in the prediction model; Ka_Start represents the starting channel address of the Ka Compton scattering peak of Ag; Ka_End represents the ending channel address of the Ka Compton scattering peak of Ag; La_Start represents the starting channel address of the La peak of Ag; La_End represents the ending channel address of the La peak of Ag; Net_Tea_Spectra_Low represents the net low-voltage spectrum after background subtraction; Net_Tea_Spectra_High represents the net high-voltage spectrum after background subtraction; i_Start represents the starting channel address of the Ka peak of the element; i_End represents the starting channel address of the Ka peak of the element; Peak_Area_Light_i represents the normalized peak area of ​​the light element; Peak_Area_Main_i represents the normalized peak area of ​​the main element; Peak_Area_Main_Net_i represents the peak area of ​​the main element after removing overlapping interference. The net peak area after perturbation; Factor_Food_Main is the perturbation factor matrix built into the model; Inv represents the matrix inversion operation; specifically, in this invention, 1215:1271 represents the address interval corresponding to Ka of Pb; 1129:1187 represents the address interval corresponding to Ka of Hg; 2228:2301 represents the address interval corresponding to Ka of Cd; 1003:1059 represents the address interval corresponding to Ka of As; 829:871 represents... The following table shows the address ranges corresponding to the Ka values ​​of Zn; 770:812 represents the address range corresponding to the Ka value of Cu; 717:752 represents the address range corresponding to the Ka value of Ni; 665:701 represents the address range corresponding to the Ka value of Co; 609:647 represents the address range corresponding to the Ka value of Fe; 563:597 represents the address range corresponding to the Ka value of Mn; 516:550 represents the address range corresponding to the Ka value of Cr; 470:504 represents the address range corresponding to the Ka value of V; 345:382 represents the address range corresponding to the Ka value of Ca; 311:342 represents the address range corresponding to the Ka value of K; 242:277 represents the address range corresponding to the Ka value of Cl; 213:242 represents the address range corresponding to the Ka value of S; and 186:210 represents the address range corresponding to the Ka value of P.

[0148] The tea spectrum after background removal is shown in the attached image. Figure 3 As shown.

[0149] The results of calculating Peak_Area_Light_i and Peak_Area_Main_Net_i are shown in Table 3:

[0150] Table 3. Calculation results of net peak area of ​​main element and normalized peak area of ​​light element.

[0151]

[0152] ④ Light element prediction

[0153] <1> Since the predicted content of ultralight elements in this sample is 97.51%, the direct triggering condition of the FP method is not met.

[0154] <2> The results of the light elements are combined into a column vector s1 as [3.0490 8.6918 0.2120 1.1069 0.7616]. T Compared with the food fingerprint vector s2 built into the prediction model, the calculated correlation coefficients and Euclidean distances are [0.8943, 0.9497, 0.9841, 0.8054, 0.2826, 0.9919], respectively. T And [7.3893 6.0607 1.3168 3.4585 6.22230.8726] T It is obvious that the highest correlation coefficient is in the sixth column (0.9919 > 0.99), and the lowest Euclidean distance is also in the sixth column (0.8726). Therefore, the sixth matrix model will be used for prediction.

[0155] <3> The peak areas of light elements were first applied to a quadratic fitting model, and the predicted concentrations of Ca, K, Cl, S, and P were obtained as [2506 14008 882 2442 4452]. T (Unit: ppm) The calibration ranges set in the prediction model are [2450 13170 420 1000 2030]. T Since the predicted results for Cl, S, and P significantly exceeded the calibration range (exceeding 50%), the FP method was still triggered for prediction, while the original prediction results were discarded. The final FP method prediction results for Ca, K, Cl, S, and P are [2906 15608 1065 2732 5153]. T (Unit: ppm) The above results are greater than the model's built-in LOD and LOQ.

[0156] ⑤ Master element prediction

[0157] <1> The peak areas of the principal elements were input into the quadratic fitting model of the principal elements. The predicted results for V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Cd, Hg, and Pb were [4 5 1342 27 0.3 12 25 49 3 1 0 0]. T, it can be seen that the results of Hg and Pb are 0 ppm, the result of Cd is 3 ppm, and the results of V, Cr, and Co are 4 ppm, 5 ppm, and 0.3 ppm;

[0158] <2>The contents of elements such as Hg, Pb, Cd, V, Cr, and Co are lower than the LOD built into the prediction model. Therefore, the final output result is "<LOD", so the final output concentration of the main elements is [<LOD <LOD 1342 27 <LOD 12 25 49 <LOD <LOD<LOD <LOD] T , and the sorted results are shown in Table 4:

[0159] Table 4 Final Output Concentration of Main Elements

[0160]

[0161] Although the present invention has been described herein with reference to illustrative embodiments of the present invention, the above embodiments are only preferred embodiments of the present invention. The embodiments of the present invention are not limited by the above embodiments. It should be understood that those skilled in the art can design many other modifications and embodiments, and these modifications and embodiments will fall within the scope of the principles and spirit disclosed in this application.

Claims

1. A rapid detection method for multiple elements in food based on multi-voltage portable XRF, characterized in that... include: Step 1: Set sample pretreatment and measurement specifications The sample to be tested is dried, crushed, and passed through a 200-mesh sieve; The sample to be tested was encapsulated using a special sample cup and Mylar membrane; Set the total measurement time to 30 seconds to ensure signal stability; Step 2: Multi-voltage excitation (1) Low voltage and filterless excitation of light elements; (2) High-voltage band filter is used to excite the main elements; (3) Maximize element excitation efficiency by switching voltage; The low voltage is 8kV and the high voltage is 40kV. Step 3: Background Subtraction and Signal Correction (1) Pure glucose was used as the background standard. (2) Background subtraction based on Ag-La peak ratio at low voltage; (3) Background subtraction based on the proportion of Ag-Ka Compton scattering peaks under high voltage; (4) Use Compton scattering normalization to correct for measurement gap and matrix effects; Step 4: Multi-model prediction mechanism (1) Light elements are detected using the standard curve method + basic parameter method; (2) The principal element is obtained by using the standard curve method and combining it with the interference factor matrix to remove overlapping peaks; (3) It has six common food matrix models and a global model built in, and selects the optimal model by matching spectral similarity. The intelligent prediction process includes: Light element prediction: a. Matrix matching: Calculate the similarity between the peak area of ​​light elements in the unknown sample and the reference spectrum of each matrix model, and select the best matching model; b. Quadratic fitting is preferred: use the quadratic fitting formula of the matching model for concentration prediction; c. Linear fitting backup: If the quadratic fitting result exceeds the calibration range and results in inaccuracy, then the linear fitting result of this model shall be used instead; d. FP method fallback: If the predicted concentration is significantly outside the calibration range or the predicted ultralight element content is ≤96%, the system will automatically switch to the FP method for prediction and use the calibration factor to correct the FP method results. e. Global model as a safety net: If there is no matching matrix, a global linear model is used for prediction; Primary element prediction: a. Quadratic fitting model is preferred; b. If the predicted concentration is abnormally high, exceeding 2000 ppm, switch to the linear fitting model for verification and output; Detection limit judgment: For all prediction results, compare them with the pre-set LOD and LOQ, and mark the results that are lower than LOD or LOQ; Step 4: Smart matrix matching and FP method triggering mechanism: (1) Based on the similarity matching between the peak area of ​​light elements and the matrix model; (2) When the predicted concentration exceeds the calibration range or the content of ultralight elements is ≤96%, the FP method will be automatically enabled.

2. The rapid detection method for multiple elements in food based on multi-voltage portable XRF according to claim 1, characterized in that: In step two, the light elements include P, S, Cl, K, and Ca; The main elements include V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Cd, Hg, and Pb.

3. The rapid detection method for multiple elements in food based on multi-voltage portable XRF according to claim 1, characterized in that: In step four, the standard curve method is obtained by first-order fitting and second-order fitting.

4. The rapid detection method for multiple elements in food based on multi-voltage portable XRF according to claim 3, characterized in that: In step four, the built-in common food matrix models include rice, wheat, beans, cabbage leaves, rapeseed leaves, and Panax notoginseng models.

5. The rapid detection method for multiple elements in food based on multi-voltage portable XRF according to claim 3 or 4, characterized in that: Step four specifically includes: Model construction: Light element model library: Calibration models for at least 6 typical food matrices are established for P, S, Cl, K, and Ca; each matrix model includes both quadratic and linear fitting relationships; Principal Element Unified Model: For the principal element, a unified quadratic fitting and linear fitting calibration curve is established, and an interference factor matrix is ​​introduced for spectral line overlap correction; Basic parameter method model: Establish a basic parameter method model suitable for iterative calculation of light elements.

6. The rapid detection method for multiple elements in food based on multi-voltage portable XRF according to claim 2, characterized in that: Step four, the intelligent matrix matching and FP method triggering mechanism, specifically includes: Establish a matrix model fingerprint library: A calibration model is established in advance using a variety of typical food standard substances, and the "feature fingerprint" of each matrix model is extracted; the "feature fingerprint" is the standard peak area vector of each light element after background subtraction and Compton normalization; that is, s2 = [Norm_P, Norm_S, Norm_Cl, Norm_K, Norm_Ca]^T; Unknown sample feature extraction: For unknown samples, the same background subtraction and Compton normalization process is used to calculate the normalized peak area vector s1 of its light elements; Similarity matching: Calculate the similarity between the feature vector s1 of the unknown sample and the feature vector s2 of each matrix in the model fingerprint database; Multiple matching algorithms are used for parallel computation, including: Euclidean distance: the smaller the value, the more similar they are; Correlation coefficient: the higher the value, the more similar they are. The optimal matching matrix type is determined by combining the results of the two algorithms; If both algorithms target the same matrix, then the matrix model should be selected directly. If the models pointed to by the high-confidence algorithm are different, but the confidence of a certain algorithm is extremely high, reaching R² ≥ 0.99, then the model pointed to by the high-confidence algorithm shall be selected. If none of the conditions are met, the model is classified as "unknown matrix" and the global calibration model is activated. Model Invocation: Based on the matching results, invoke the corresponding quadratic / linear calibration model to predict the elemental concentration of the unknown sample.

7. A rapid detection system for multiple elements in food based on multi-voltage portable XRF, characterized in that... The rapid detection method for multiple elements in food based on multi-voltage portable XRF, as described in any one of claims 1 to 6, was adopted.

8. The rapid detection system for multiple elements in food based on multi-voltage portable XRF according to claim 7, characterized in that: The multi-element rapid detection system for food based on multi-voltage portable XRF has the following built-in files: The spectrum of glucose at high and low voltages; the energy channels of each element.

Images