Tracing model construction method and tracing method for intra-city-level, intra-province-level and inter-province-level rice

Through multi-source data fusion technology, traceability models for rice at the municipal, provincial and inter-provincial levels were constructed, which solved the problem of insufficient traceability accuracy in existing technologies and achieved higher model accuracy.

CN120612101APending Publication Date: 2025-09-09SHANGHAI ACAD OF AGRI SCI
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510751543.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-06-06
Publication Date
2025-09-09

AI Technical Summary

Technical Problem

Existing technologies lack systematic research in rice traceability, especially the insufficient accuracy of municipal-level traceability models and limited application of multi-source data fusion strategies.

Method used

By integrating multi-source data such as stable isotope ratios, near-infrared spectroscopy, mineral elements and metabolomics information, and using competitive adaptive reweighted sampling, orthogonal partial least squares discriminant analysis and other techniques, traceability models for rice at the municipal, provincial and inter-provincial levels are constructed.

Benefits of technology

The accuracy of rice traceability models at the municipal, provincial and inter-provincial levels has been improved, especially the accuracy of models within small areas has been significantly improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120612101A_ABST
    Figure CN120612101A_ABST
Patent Text Reader

Abstract

The invention belongs to the technical field of rice traceablility, and provides a traceablility model construction method and traceablility method for city-level, provincial-level and inter-provincial-level rice. According to the method for constructing the traceability model of the city-level rice, the stable isotope ratio information, the near infrared spectrum information, the mineral element information and the metabonomics information are fused in the middle level for the model of a small area, and the accuracy of the model is improved. When the provincial rice traceability model is constructed, the stable isotope ratio information and the near infrared spectrum information are fused in the middle stage, and the standardization processing method and the modeling mode are combined, so that the accuracy of the provincial rice traceability model is improved. When the traceability model of the inter-provincial rice is constructed, the stable isotope ratio information and the near infrared spectrum information are fused in the middle stage, and the accuracy of the traceability model of the inter-provincial rice is improved by combining a standardized processing method and a modeling mode.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of rice traceability, and in particular to a method for constructing a traceability model and a traceability method for rice at the municipal, provincial and inter-provincial levels. Background Art

[0002] In recent years, rice origin traceability technology has made significant progress. 13 C, δ 15 N) is widely used to distinguish agricultural products from different geographical origins because it is significantly affected by environmental factors. 13 Combined analysis of C and mineral elements enabled the identification of the origin of Jilin rice. Near-infrared spectroscopy, with its rapid and non-destructive nature, has been used for rapid traceability of rice in Northeast China. Combined with the partial least squares discrimination analysis (PLS-DA) algorithm, the classification accuracy of Heilongjiang and Jilin rice was increased to 92%. The composition and content of mineral elements in rice are closely related to its growing environment. Studies have successfully traced the origin of Songjiang rice using mineral element fingerprinting. Furthermore, metabolomics can distinguish metabolic signatures from different origins by analyzing differences in rice metabolites. Studies have also found significant differences in the metabolite profiles of rice from Jiangsu and Heilongjiang, enabling origin identification based on metabolite signatures. However, existing studies have mostly focused on a single technology or a single geographic scale, lacking systematic research on intra-city traceability models, and the application of multi-source data fusion strategies remains limited. Summary of the Invention

[0003] In view of this, the object of the present invention is to provide a method for constructing a traceability model for rice at the municipal, provincial, and inter-provincial levels and a traceability method. The traceability model for municipal-level rice obtained by the construction method provided by the present invention has high accuracy.

[0004] In order to achieve the above-mentioned object of the invention, the present invention provides the following technical solutions:

[0005] The present invention provides a method for constructing a traceability model for city-level rice, comprising the following steps:

[0006] Obtain rice samples from different areas of a city;

[0007] performing stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination, and metabolomics determination on the rice sample to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information, and metabolomics information;

[0008] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0009] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0010] Using a competitive adaptive reweighted sampling method (CARS) to filter characteristic information of the preprocessed near-infrared spectral information, to obtain characteristic near-infrared spectral information;

[0011] Using orthogonal partial least squares discriminant analysis (OPLS-DA) to screen the characteristic information of the mineral element information to obtain characteristic mineral element information;

[0012] Standardizing the characteristic mineral element information to obtain standardized characteristic mineral element information;

[0013] Using orthogonal partial least squares discriminant analysis (OPLS-DA) to screen the characteristic information of the metabolomics information to obtain characteristic metabolomics information;

[0014] Standardizing the characteristic metabolomics information to obtain standardized characteristic metabolomics information;

[0015] The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standardized characteristic metabolomics information are integrated to perform modeling to obtain a traceability model for the municipal-grade rice;

[0016] The method of the normalization process is Z-Score normalization (Zero-mean normalization);

[0017] The preprocessing method is raw spectrum (RAW), standard normal variate (SNV), multiplicative scatter correction (MSC) or Savitzky-Golay smoothing (SG);

[0018] The modeling method is linear discriminant analysis (LDA) or support vector machine (SVM).

[0019] Preferably, the stable isotope ratio includes δ 13 C, δ 15 N, δ 2H and δ 18 O.

[0020] Preferably, the parameters of the near infrared spectroscopy include: a spectral range of 10000 to 4000 cm -1 , the spectral resolution is 8cm -1 , the number of scans is 32 times.

[0021] Preferably, the mineral elements include Na, Ca, K, Mg, Al, Ba, Mn, Fe, V, Cr, Sr, Co, Ni, Se, As, Pb and Cd.

[0022] Preferably, the metabolomics includes volatile metabolites and non-volatile metabolites.

[0023] The present invention also provides a method for tracing the source of city-level rice, comprising the following steps:

[0024] The rice to be tested is subjected to stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination and metabolomics determination to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information and metabolomics information;

[0025] Based on the standardized stable isotope ratio information result in the construction method described in the above technical solution, a standardization process is performed to obtain standardized stable isotope ratio information;

[0026] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0027] Based on the standardized characteristic mineral element information results in the construction method described in the above technical solution, measurement and standardization processing are performed to obtain standardized characteristic mineral element information;

[0028] Based on the standardized characteristic metabolomics information results in the construction method described in the above technical solution, measurement and standardization processing are performed to obtain standardized characteristic metabolomics information;

[0029] The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standardized characteristic metabolomics information are substituted into the corresponding municipal-grade rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0030] The present invention also provides a method for constructing a traceability model for provincial-level rice, comprising the following steps:

[0031] Obtain rice samples from different regions within the same province;

[0032] The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information;

[0033] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0034] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0035] Using the CARS method to screen the characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information;

[0036] The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain a traceability model for the provincial-level rice;

[0037] The method of the standardization process is Z-Score standardization;

[0038] The preprocessing methods are raw spectrum (RAW), standard normal transformation (SNV), multivariate scatter correction (MSC) or Savitzky-Golay smoothing (SG);

[0039] The modeling method is random forest (RF).

[0040] The present invention also provides a method for tracing the source of provincial-level rice, comprising the following steps:

[0041] Performing stable isotope ratio measurement and near-infrared spectrum measurement on the rice to be tested to obtain stable isotope ratio information and near-infrared spectrum information;

[0042] Based on the standardized stable isotope ratio information result in the construction method described in the above technical solution, a standardization process is performed to obtain standardized stable isotope ratio information;

[0043] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0044] The standardized stable isotope ratio information and characteristic near-infrared spectrum information are substituted into the corresponding provincial-level rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0045] The present invention also provides a method for constructing a provincial rice traceability model, comprising the following steps:

[0046] Obtain rice samples from different provinces;

[0047] The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information;

[0048] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0049] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0050] Using the CARS method to screen the characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information;

[0051] The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain the provincial-level rice traceability model;

[0052] The method of the standardization process is Z-Score standardization;

[0053] The preprocessing methods are raw spectrum (RAW), standard normal transformation (SNV), multivariate scatter correction (MSC) or Savitzky-Golay smoothing (SG);

[0054] The modeling method is random forest (RF) or K-nearest neighbor (KNN).

[0055] The present invention also provides a provincial rice traceability method, comprising the following steps:

[0056] Performing stable isotope ratio measurement and near-infrared spectrum measurement on the rice to be tested to obtain stable isotope ratio information and near-infrared spectrum information;

[0057] Based on the standardized stable isotope ratio information result in the construction method described in the above technical solution, a standardization process is performed to obtain standardized stable isotope ratio information;

[0058] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0059] The standardized stable isotope ratio information and characteristic near-infrared spectrum information are substituted into the provincial rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0060] The present invention provides a method for constructing a traceability model for municipal-level rice. When constructing a model for a small region, the method integrates stable isotope ratio information, near-infrared spectroscopy information, mineral element information, and metabolomics information, thereby improving the accuracy of the model.

[0061] The present invention provides a method for constructing a traceability model for provincial-level rice. This method integrates stable isotope ratio information and near-infrared spectroscopy information when constructing the traceability model for provincial-level rice, and combines this with standardized processing methods and modeling approaches to improve the accuracy of the traceability model for provincial-level rice.

[0062] The present invention provides a method for constructing a provincial rice traceability model. This method integrates stable isotope ratio information and near-infrared spectroscopy information, combined with standardized processing methods and modeling approaches, to improve the accuracy of the provincial rice traceability model. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 The δ of rice in 5 different provinces of China 13 C, δ 15 N, δ 2 H and δ 18 O value distribution;

[0064] Figure 2 The δ 13 C, δ 15 N, δ 2 H and δ 18 O value distribution;

[0065] Figure 3 The δ 13 C, δ 15 N, δ 2 H and δ 18 O value distribution;

[0066] Figure 4 The near-infrared spectra of rice from different origins;

[0067] Figure 5 Select variable results for CARS;

[0068] Figure 6 It is the characteristic variable spectrum extracted by CARS algorithm. DETAILED DESCRIPTION

[0069] The present invention provides a method for constructing a traceability model for city-level rice, comprising the following steps:

[0070] Obtain rice samples from different areas of a city;

[0071] performing stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination, and metabolomics determination on the rice sample to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information, and metabolomics information;

[0072] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0073] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0074] Using the CARS method to screen the characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information;

[0075] Utilizing OPLS-DA to screen characteristic information of the mineral element information to obtain characteristic mineral element information;

[0076] Standardizing the characteristic mineral element information to obtain standardized characteristic mineral element information;

[0077] OPLS-DA is used to screen characteristic information of the metabolomics information to obtain characteristic metabolomics information;

[0078] Standardizing the characteristic metabolomics information to obtain standardized characteristic metabolomics information;

[0079] The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standard characteristic metabolomics information are integrated to perform modeling to obtain a traceability model for the municipal-grade rice;

[0080] The method of the standardization process is Z-Score standardization;

[0081] The pretreatment method is RAW, SNV, MSC or SG;

[0082] The modeling method is LDA or SVM.

[0083] The present invention obtains rice samples from different regions within a city. The present invention does not specifically limit the method for obtaining the rice samples, but preferably uses rice varieties that are most representative of the region. In one embodiment of the present invention, when establishing a rice traceability model for Panjin City, rice from Dawa District and Panshan County is preferably selected.

[0084] After obtaining the rice sample, the present invention performs stable isotope ratio determination, near-infrared spectrum determination, mineral element determination and metabolomics determination on the rice sample to obtain stable isotope ratio information, near-infrared spectrum information, mineral element information and metabolomics information.

[0085] In the present invention, the stable isotope ratio preferably includes δ 13 C, δ 15 N, δ 2 H and δ 18O.

[0086] In the present invention, the calculation formula of the stable isotope ratio is shown in Formula 1:

[0087]

[0088] In formula 1, Rsample is the abundance ratio of heavy isotopes to light isotopes in the measured sample, that is, 13 C / 12 C. 15 N / 14 The N;R standard is the abundance ratio of heavy isotopes to light isotopes in international standard samples.

[0089] In the present invention, the stable isotope ratio is preferably processed and determined using multi-point calibration; the stable isotope ratio δ 13 C and δ 15 The N value is preferably calibrated using USGS40, USGS90 and USGS91 standard material values; the stable isotope ratio δ 2 H and δ 18 The O value is preferably calibrated using USGS55, USGS90 and USGS91 standard material values; the instrument accuracy is preferably δ 13 C≤0.1‰, δ 15 N≤0.2‰, δ 2 H≤2.0‰, δ 18 O≤0.5‰.

[0090] In the present invention, when the element is carbon or nitrogen, the determination method preferably includes: placing rice powder in a tin cup, and then placing it in an elemental analyzer, and detecting carbon and nitrogen in C / N mode. In the present invention, during the detection process, one standard quality control is inserted into every 10 samples. In the present invention, the combustion furnace temperature of the elemental analyzer is set to 980°C, the chromatographic column temperature is 60°C, the carrier gas flow rate is 180mL / min, and the reference gas flow rate is 60mL / min. In the present invention, the principle of detection by the elemental analyzer is: the carbon and nitrogen elements in the rice powder are converted into CO2 and N2 respectively through a combustion reaction, and are detected after appropriate dilution.

[0091] In the present invention, when the element is hydrogen or oxygen, the determination method preferably includes: placing rice powder in a silver cup, and then placing it in an elemental analyzer, and realizing the detection of hydrogen and oxygen in H / O mode. In the present invention, during the detection process, one standard quality control is inserted into every 10 samples. In the present invention, the combustion furnace temperature of the elemental analyzer is set to 1450°C, the carrier gas flow rate is 100mL / min, and the reference gas flow rate is 100mL / min. In the present invention, the principle of detection by the elemental analyzer is: the hydrogen element, oxygen and nitrogen elements in the rice powder are converted into H2 and CO respectively through a high-temperature cracking process, and are detected after appropriate dilution.

[0092] In the present invention, the parameters of the near infrared spectroscopy include: the spectral range is preferably 10000~4000cm -1 , the spectral resolution is preferably 8cm -1 , the number of scans is preferably 32 times. In the present invention, when performing the near-infrared spectroscopy measurement, an internal blank is preferably used as a reference. In the present invention, the near-infrared spectroscopy measurement preferably includes the following steps: mixing the sample, and then placing it in a rotating sample cup and scanning it three times. In the present invention, OMNIC 9 is preferably used to record near-infrared spectral data and store it in absorbance format. The average value of three repetitions of each sample spectral data is taken for subsequent analysis; during the operation of the instrument, the laboratory temperature is controlled to be constant at 25°C.

[0093] In the present invention, the mineral elements preferably include Na, Ca, K, Mg, Al, Ba, Mn, Fe, V, Cr, Sr, Co, Ni, Se, As, Pb and Cd. In the present invention, the determination of the mineral elements is preferably carried out with reference to GB 5009.268-2016 "National Food Safety Standard - Determination of Multiple Elements in Food".

[0094] In the present invention, the metabolomics preferably includes volatile metabolites and non-volatile metabolites.

[0095] In the present invention, the determination of volatile metabolites preferably comprises the following steps:

[0096] The method comprises placing rice powder to be tested, a saturated sodium chloride solution and an o-dichlorobenzene solution in a headspace bottle and performing equilibrium; then inserting an extraction head into the headspace bottle and performing extraction; after extraction, inserting the extraction head into a gas chromatography-mass spectrometer (GC-MS) inlet, performing desorption, and then performing gas chromatography-mass spectrometry detection.

[0097] In the present invention, the solvent of the o-dichlorobenzene solution is preferably methanol. In the present invention, the concentration of the o-dichlorobenzene solution is preferably 20 mg / L.

[0098] In the present invention, the ratio of the rice powder to be tested, the saturated sodium chloride solution and the o-dichlorobenzene solution is preferably 1.0 g: 5 mL: 10 μL.

[0099] In the present invention, the equilibrium temperature is preferably 70° C., and the equilibrium time is preferably 20 minutes.

[0100] In the present invention, the extraction head is preferably 50 / 30 μm DVB / CAR / PDMS purchased from Merck, Germany.

[0101] In the present invention, the extraction temperature is preferably 70° C., and the extraction time is preferably 30 minutes.

[0102] In the present invention, the desorption temperature is preferably 250° C., and the desorption time is preferably 5 minutes.

[0103] In the present invention, the gas chromatograph-mass spectrometer is preferably 7890A-5975C, Agilent Technologies, Santa Clara, CA, USA.

[0104] In the present invention, the gas chromatography-mass spectrometry detection includes gas chromatography detection and mass spectrometry detection.

[0105] In the present invention, the gas chromatography detection parameters include: the chromatographic column is preferably an HP-INNOWAX chromatographic column (60m×0.25mm, 0.25μm, Agilent); the gas chromatography temperature program is preferably: 40°C for 2 minutes, then increased to 130°C at 3°C / min, then increased to 220°C at 10°C / min and held for 5 minutes; the carrier gas is helium (purity 99.999%), with a flow rate of 1 mL / min; splitless injection; solvent delay is 3 minutes; ion source temperature is 230°C; quadrupole temperature is 150°C; transfer line temperature is 240°C; electron ionization source mode is used, electron energy is 70eV; and the mass scan range is m / z 30-350.

[0106] Data analysis preferably includes: data processing using a chromatography workstation (Agilent) and ChromaTOF version 4.71 (LECO) software, and preliminary identification after comparison of the mass spectrum with the NIST 17 database. The linear retention index (RI) is calculated using the retention time of n-alkanes C7-C40, and the compound is qualitatively compared with the published database NIST Chemistry WebBook (https: / / webbook.nist.gov / chemical / ) and RI reference values ​​reported in the literature. Odor descriptions are based on the Good Scents Company online database (http: / / www.thegoodscentscompany.com / ). The relative content of volatile compounds is calculated using the area normalization method using o-dichlorobenzene as the internal standard.

[0107] In the present invention, the determination of the non-volatile metabolites preferably comprises the following steps:

[0108] Weigh 50 mg of rice sample and dissolve it in 1.2 mL of 70% methanol solution containing an internal standard (L-2-chlorophenylalanine, 0.25 mg / mL). Vortex the sample for 30 seconds every 30 minutes for six times to facilitate extraction and then store at 4°C. The mixture is then centrifuged at 12,000 rpm for 3 minutes. The supernatant is aspirated and filtered through a 0.22 μm microporous membrane and stored in a vial to obtain the sample for the instrument.

[0109] The sample was subjected to ultra-high performance liquid chromatography time-of-flight mass spectrometry detection.

[0110] In the present invention, all rice samples were mixed as quality control (QC) samples, and one QC sample was inserted into every 10 test samples to monitor the reproducibility of the assay.

[0111] In the present invention, the ultra-high performance liquid chromatography time-of-flight mass spectrometry is preferably performed on an ultra-high performance liquid chromatography time-of-flight mass spectrometer, and the ultra-high performance liquid chromatography time-of-flight mass spectrometer is preferably UPLC-Triple TOF 6600, Sciex, Foster City, CA, USA.

[0112] In the present invention, the ultra-high performance liquid chromatography time-of-flight mass spectrometry detection includes chromatographic detection and mass spectrometric detection; the parameters of the chromatographic detection include: a T3 chromatographic column (Waters ACQUITY Premier HSS T3 chromatographic column, 1.8 μm, 2.1 mm×100 mm), a column temperature of 40° C.; a mobile phase of 0.1% formic acid aqueous solution (A) and 0.1% formic acid acetonitrile solution (B), and a gradient elution program as follows: 5% to 20% B (0 to 2 min), 20% to 60% B (2 to 5 min), 60% to 99% B (5 to 6 min), 99% B (6 to 7.5 min), 99% to 5% B (7.5 to 7.6 min), and 5% B (7.6 to 10 min); an injection volume of 4 μL and a flow rate of 0.4 mL / min.

[0113] In the present invention, data acquisition is preferably performed using the full scan mode (IDA) (scanning mass range of 50 to 1000 Da) of AnalystTF 1.7.1 software (Sciex, Concord, ON, Canada).

[0114] In the present invention, the mass spectrometry detection parameters include: ESI source parameters: ion source gas 1 (GAS1) is 50 psi; ion source gas 2 (GAS2) is 60 psi; curtain gas (CUR) is 35 psi; ion source temperature is 550 ° C (ESI + ) / 450℃(ESI - ); ionization voltage (IS) is 5000 V (ESI + ) / -4000V(ESI - ); declustering potential (DP) was 60 / -60 V; MS1 collision energy (CE) was 10 / -10 V; MS2 collision energy (CE) was 30 / -30 V; and collision energy step size (CES) was 15 V.

[0115] Qualitative and quantitative analysis of compounds preferably includes peak extraction, alignment, and retention time correction using the XCMS program. Peaks with a missing rate >50% in each sample group are filtered, blank values ​​are filled using KNN, and peak areas are corrected using the SVR method. Metabolite identification is performed by searching the MWDB 2.0 database developed by Wuhan Maiwei Biotechnology Co., Ltd., integrated public libraries, predicted libraries, and metDNA methods.

[0116] After obtaining the stable isotope ratio information, the present invention performs a normalization process on the stable isotope ratio information to obtain normalized stable isotope ratio information. In the present invention, the normalization process is performed by Z-Score normalization.

[0117] After obtaining the near-infrared spectrum information, the present invention preprocesses the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information. In the present invention, the preprocessing method is preferably RAW, SNV, MSC or SG.

[0118] After obtaining the pre-processed near-infrared spectral information, the present invention uses the CARS method to screen the characteristic information of the pre-processed near-infrared spectral information to obtain characteristic near-infrared spectral information. In the present invention, the CARS method is preferably performed using the CARS algorithm code in Matlab R2024a software.

[0119] After obtaining the mineral element information, the present invention uses OPLS-DA to screen the characteristic information of the mineral element information to obtain characteristic mineral element information. In the present invention, the VIP value of the OPLS-DA is preferably greater than 1.

[0120] After obtaining the characteristic mineral element information, the present invention performs a standardization process on the characteristic mineral element information to obtain standardized characteristic mineral element information. In the present invention, the standardization process is performed by Z-Score standardization.

[0121] After obtaining the metabolomics information, the present invention uses OPLS-DA to screen the characteristic information of the metabolomics information to obtain characteristic metabolomics information. In the present invention, the VIP value of the OPLS-DA is preferably greater than 1.

[0122] After obtaining the characteristic metabolomics information, the present invention performs a normalization process on the characteristic metabolomics information to obtain normalized characteristic metabolomics information. In the present invention, the normalization process is performed by Z-Score normalization.

[0123] After obtaining standardized stable isotope ratio information, characteristic near-infrared spectral information, standardized characteristic mineral element information, and standardized characteristic metabolomics information, the present invention fuses the standardized stable isotope ratio information, characteristic near-infrared spectral information, standardized characteristic mineral element information, and standardized characteristic metabolomics information to perform modeling to obtain a traceability model for the municipal-level rice. In the present invention, the modeling method is preferably LDA (Linear Discriminant Analysis, Lineardiscriminant analysis, LDA) or SVM (Support Vector Machine, Support vector classification, SVM).

[0124] The present invention also provides a method for tracing the source of city-level rice, comprising the following steps:

[0125] The rice to be tested is subjected to stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination and metabolomics determination to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information and metabolomics information;

[0126] Based on the standardized stable isotope ratio information result in the construction method described in the above technical solution, a standardization process is performed to obtain standardized stable isotope ratio information;

[0127] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0128] Based on the standardized characteristic mineral element information results in the construction method described in the above technical solution, measurement and standardization processing are performed to obtain standardized characteristic mineral element information;

[0129] Based on the standardized characteristic metabolomics information results in the construction method described in the above technical solution, measurement and standardization processing are performed to obtain standardized characteristic metabolomics information;

[0130] The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standardized characteristic metabolomics information are substituted into the corresponding municipal-grade rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0131] In the present invention, in the traceability method, the methods of stable isotope ratio determination and near-infrared spectroscopy determination are preferably consistent with the above technical solution and will not be described in detail here.

[0132] After obtaining the near-infrared spectral information, the present invention performs screening based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution to obtain characteristic near-infrared spectral information. In the present invention, the screening based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution means that the modeling method has already screened out the characteristic infrared spectral information, i.e., the characteristic band; the information of the corresponding characteristic band can be directly extracted.

[0133] After obtaining the stable isotope ratio information, the present invention performs a normalization process based on the standardized stable isotope ratio information results in the construction method described in the above technical solution to obtain the standardized stable isotope ratio information. In the present invention, the method of the normalization process is consistent with the parameters in the traceability model construction method and will not be repeated here.

[0134] The present invention is based on the standardized characteristic mineral element information results in the construction method described in the above technical solution, and performs measurement and standardization processing to obtain standardized characteristic mineral element information. In the present invention, the measurement and standardization processing of the standardized characteristic mineral element information results in the construction method described in the above technical solution refers to directly measuring the characteristic mineral element information based on the characteristic mineral element information results screened in the traceability model construction method, and processing according to the measurement and standardization processing in the construction method to obtain standardized characteristic mineral element information.

[0135] The present invention is based on the standardized characteristic metabolomics information results in the construction method described in the above technical solution, and performs measurement and standardization processing to obtain standardized characteristic metabolomics information. In the present invention, the standardized characteristic metabolomics information results in the construction method described in the above technical solution, and performs measurement and standardization processing refers to: based on the characteristic metabolomics information screened in the traceability model construction method, directly measuring the characteristic metabolomics information, and processing it according to the measurement and standardization processing in the construction method to obtain standardized characteristic metabolomics information.

[0136] The present invention does not specifically limit the substitution method, and operations well known to those skilled in the art can be used.

[0137] The present invention also provides a method for constructing a traceability model for provincial-level rice, comprising the following steps:

[0138] Obtain rice samples from different regions within the same province;

[0139] The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information;

[0140] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0141] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0142] Using the CARS method to screen the characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information;

[0143] The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain a traceability model for the provincial-level rice;

[0144] The method of the standardization process is Z-Score standardization;

[0145] The pretreatment method is RAW, SNV, MSC or SG;

[0146] The modeling method is RF (Random forest, RF).

[0147] In the present invention, the parameters of the stable isotope ratio measurement and near-infrared spectroscopy measurement are preferably consistent with the above technical solution and will not be repeated here.

[0148] In the present invention, the parameters for screening the characteristic information of the near-infrared spectral information using the CARS method are preferably consistent with the above technical solution.

[0149] The present invention also provides a method for tracing the source of provincial-level rice, comprising the following steps:

[0150] Performing stable isotope ratio measurement and near-infrared spectrum measurement on the rice to be tested to obtain stable isotope ratio information and near-infrared spectrum information;

[0151] Based on the standardized stable isotope ratio information result in the construction method described in the above technical solution, a standardization process is performed to obtain standardized stable isotope ratio information;

[0152] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0153] The standardized stable isotope ratio information and characteristic near-infrared spectrum information are substituted into the corresponding provincial-level rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0154] In the present invention, in the traceability method, the stable isotope ratio determination, the near-infrared spectrum determination, and the standardized stable isotope ratio information results in the construction method described in the above technical solution are standardized to obtain standardized stable isotope ratio information, and the characteristic near-infrared spectrum information results in the construction method described in the above technical solution are screened to obtain characteristic near-infrared spectrum information, which is preferably consistent with the above technical solution and will not be repeated here.

[0155] The present invention also provides a method for constructing a provincial rice traceability model, comprising the following steps:

[0156] Obtain rice samples from different provinces;

[0157] The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information;

[0158] Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information;

[0159] Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information;

[0160] Using the CARS method to screen the characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information;

[0161] The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain the provincial-level rice traceability model;

[0162] The method of the standardization process is Z-Score standardization;

[0163] The pretreatment method is RAW, SNV, MSC or SG;

[0164] The modeling method is RF or KNN.

[0165] In the present invention, the parameters of the stable isotope ratio measurement and near-infrared spectroscopy measurement are preferably consistent with the above technical solution and will not be repeated here.

[0166] In the present invention, the parameters for screening the characteristic information of the near-infrared spectral information using the CARS method are preferably consistent with the above technical solution.

[0167] The present invention also provides a provincial rice traceability method, comprising the following steps:

[0168] The rice to be tested is subjected to stable isotope ratio measurement and near-infrared spectrum measurement to obtain provincial-level stable isotope ratio and near-infrared spectrum information;

[0169] Based on the standardized stable isotope ratio provincial level result in the construction method described in the above technical solution, a standardized stable isotope ratio provincial level is obtained;

[0170] Based on the characteristic near-infrared spectral information results in the construction method described in the above technical solution, screening is performed to obtain characteristic near-infrared spectral information;

[0171] The standardized stable isotope ratio provincial level and characteristic near-infrared spectrum information are substituted into the provincial rice traceability model obtained by the construction method described in the above technical solution to achieve traceability of the sample to be tested.

[0172] In the present invention, in the traceability method, the stable isotope ratio is determined, and the standardized stable isotope ratio result in the construction method described in the above technical solution is standardized to obtain the standardized stable isotope ratio; the characteristic near-infrared spectral information result in the construction method described in the above technical solution is screened, and the method for obtaining the characteristic near-infrared spectral information is preferably consistent with the above technical solution and will not be repeated here.

[0173] The following detailed description of the traceability model construction method and traceability method for municipal, provincial and inter-provincial rice provided by the present invention is given in conjunction with the embodiments, but they should not be understood as limiting the scope of protection of the present invention.

[0174] Example

[0175] 1. Materials and Methods

[0176] 1.1 Test samples

[0177] A total of 94 geographical indication rice samples were collected from different provinces for the stable isotope ratios, including 15 from Heilongjiang Province, 20 from Jilin Province, 34 from Liaoning Province (24 from Panjin, 5 from Donggang, and 5 from Yingkou), 15 from Shanghai, and 10 from Jiangsu.

[0178] 1.2 Materials and Reagents

[0179] Stable isotope ratio reference materials USGS40, USGS90, USGS91, and USGS55 (Reston Stable Isotope Ratio Laboratory, USA); 4 mm × 4 mm × 11 mm tin cups and 3.3 mm × 5 mm × 0.03 mL silver cups (Elimenta, Germany).

[0180] 1.3 Instruments and Equipment

[0181] Stable isotope ratio instrument: DELTA V Advantage stable isotope ratio mass spectrometer and ConFlo IV continuous flow interface device (Thermo Fisher Scientific); Nicolet is50 Fourier transform near-infrared spectrometer (Thermo Fisher Scientific, USA).

[0182] 1.4 Experimental methods

[0183] (1) Determination of stable isotope ratio: About 3.5 mg of rice powder was weighed into a tin cup for C and N element determination, and about 0.20 mg was weighed into a silver cup for H and O element determination. One standard quality control was inserted into every 10 samples. In the C / N mode, the C and N elements in the rice were converted into CO2 and N2 respectively through combustion reaction, and then tested after appropriate dilution. The combustion furnace temperature of the element analyzer was set to 980 °C, the column temperature was 60 °C, the carrier gas flow rate was 180 mL / min, and the reference gas flow rate was 60 mL / min. In the H / O mode, the H and O elements in the rice were converted into H2 and CO respectively through a high-temperature pyrolysis process, and then tested after appropriate dilution. The high-temperature pyrolysis furnace temperature of the element analyzer was set to 1450 °C, the carrier gas flow rate was 100 mL / min, and the reference gas flow rate was 100 mL / min. δ represents the stable isotope ratio. The standard material with a determined stable isotope ratio was used as a reference. The stable isotope ratio calculation formula is shown in Equation 1:

[0184]

[0185] In formula 1, Rsample is the abundance ratio of heavy isotopes to light isotopes in the measured sample, that is, 13 C / 12 C. 15 N / 14 The N;R standard is the abundance ratio of heavy isotopes to light isotopes in international standard samples. All test data are processed and fixed using multi-point calibration. 13 C and δ 15 N value correction: USGS40, USGS90 and USGS91 standard material values, δ 2 H and δ 18 O value correction: USGS55, USGS90 and USGS91 standard material values. Instrument accuracy δ 13 C≤0.1‰, δ 15 N≤0.2‰, δ 2 H≤2.0‰, δ 18 O≤0.5‰.

[0186] (2) Near infrared spectroscopy: 5-10 g of rice powder passed through a 100-mesh sieve was placed in a sample cup and scanned by near infrared spectroscopy. The spectral range was 10,000-4,000 cm -1 , the spectral resolution is 8cm -1 , scanned 32 times, using an internal blank as a reference. The sample was mixed before each scan and then placed in a rotating sample cup for three scans. Near-infrared spectral data were recorded using an OMNIC 9 and stored in absorbance format. The average of three replicates for each sample was used for subsequent analysis. The laboratory temperature was maintained at a constant 25°C during instrument operation.

[0187] (3) Mineral elements (ICP-MS)

[0188] ① Mineral Elements in Samples. Seventeen elements, including Na, Ca, K, Mg, Al, Ba, Mn, Fe, V, Cr, Sr, Co, Ni, Se, As, Pb, and Cd, were determined in rice according to GB 5009.268-2016, "National Food Safety Standard - Determination of Multiple Elements in Food." Weigh 0.5 g of sample into a microwave digestion tube, add 6 mL of nitric acid, cover, and incubate overnight. Digest and remove the acid using a microwave digester. The microwave digestion procedure was followed according to the standard. After cooling, dilute to 25 mL with deionized water. Before loading, filter the sample solution through a 0.45 μm filter. The calibration curve and ICP-MS analysis conditions were established according to the standard.

[0189] (4) Metabolomics technology (GC-MS / UPLC-QTOF-MS)

[0190] ① Determination of volatile metabolites (GC-MS). 1.0 g of rice flour was accurately weighed and placed in a 20 mL headspace vial. 5 mL of saturated sodium chloride and 10 μL of o-dichlorobenzene (20 mg / L) were added. The vial was sealed and equilibrated at 70°C for 20 min. A pretreated SPME tip (50 / 30 μm DVB / CAR / PDMS, Merck, Germany) was inserted into the headspace vial and extracted at 70°C for 30 min. The tip was then inserted into the GC-MS inlet and desorbed at 250°C for 5 min. Separation and identification were performed using a gas chromatography-mass spectrometer (7890A-5975C, Agilent Technologies, Santa Clara, CA, USA) using an HP-INNOWAX column (60 m × 0.25 mm, 0.25 μm, Agilent). The temperature program was as follows: 40°C for 2 min, then increased to 130°C at 3°C / min, and then increased to 220°C at 10°C / min and held for 5 min. The carrier gas was helium (99.999% purity) at a flow rate of 1 mL / min, with splitless injection and a solvent delay of 3 min. The ion source temperature was 230°C, the quadrupole temperature was 150°C, and the transfer line temperature was 240°C. The electron ionization source was operated in electron energy mode at 70 eV, and the mass scan range was m / z 30-350. Data were processed using an Agilent chromatographic workstation and ChromaTOF version 4.71 (LECO) software. Mass spectra were compared with the NIST 17 database for preliminary identification. The linear retention index (RI) was calculated using the retention times of n-alkanes C7–C40. Compound identification was performed by comparing the RI values ​​with those in the published NIST Chemistry WebBook (https: / / webbook.nist.gov / chemical / ) and literature. Odor descriptions were based on the Good Scents Company online database (http: / / www.thegoodscentscompany.com / ). The relative content of volatile compounds was calculated using area normalization using o-dichlorobenzene as the internal standard.

[0191] ② Non-volatile metabolite determination (UPLC-QTOF-MS): 50 mg of rice sample was accurately weighed and dissolved in 1.2 mL of 70% methanol containing an internal standard (L-2-chlorophenylalanine, 0.25 mg / mL). The sample was vortexed for 30 seconds every 30 minutes for six times to facilitate extraction and then stored at 4°C. The mixture was centrifuged at 12,000 rpm for 3 minutes. The supernatant was filtered through a 0.22 μm microporous filter and stored in a vial for UPLC-MS / MS analysis. All rice samples were combined as a quality control (QC) sample, and one QC sample was inserted into every 10 test samples to monitor assay reproducibility. Qualitative analysis of non-volatile organic compounds in rice was performed using a UPLC-QTOF-MS system (UPLC-Triple TOF 6600, Sciex, Foster City, CA, USA). Chromatographic separation was performed on a T3 column (Waters ACQUITY Premier HSS T3 column, 1.8 μm, 2.1 mm × 100 mm) at 40°C. The mobile phase consisted of 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B), with a gradient of 5% to 20% B (0 to 2 min), 20% to 60% B (2 to 5 min), 60% to 99% B (5 to 6 min), 99% B (6 to 7.5 min), 99% to 5% B (7.5 to 7.6 min), and 5% B (7.6 to 10 min). The injection volume was 4 μL, and the flow rate was 0.4 mL / min. Data were acquired using Analyst TF 1.7.1 software (Sciex, Concord, ON, Canada) in full scan mode (IDA) (scanning mass range 50 to 1000 Da). The ESI source parameters were as follows: ion source gas 1 (GAS1), 50 psi; ion source gas 2 (GAS2), 60 psi; curtain gas (CUR), 35 psi; ion source temperature, 550°C (ESI + ) / 450℃(ESI - ); ionization voltage (IS), 5000V (ESI + ) / -4000V(ESI -); declustering potential (DP), 60 / -60 V; MS1 collision energy (CE), 10 / -10 V; MS2 collision energy (CE), 30 / -30 V; collision energy step size (CES), 15 V. Peak extraction, alignment, and retention time correction were performed using the XCMS program. Peaks with a missing rate >50% in each sample group were filtered, and blank values ​​were filled using KNN. Peak areas were corrected using the SVR method. Metabolite identification was performed using the MWDB 2.0 database developed by Wuhan Maiwei Biotechnology Co., Ltd., an integrated public library, a predicted library, and the metDNA method.

[0192] 1.5 Data Processing

[0193] Descriptive and differential analyses were performed using SPSS 26.0, and boxplots and near-infrared spectra were drawn using Origin 2021. The machine learning algorithm was developed using Python (version 3.8), Jupyter Notebook, and scikit-learn (version 1.0.2). The LDA, RF, SVM, and KNN algorithms were used to train and construct the model. The CARS algorithm in Matlab R2024a was used to extract spectral features, and VIP values ​​for mineral elements and metabolomics data were calculated using OPLS-DA analysis.

[0194] 2. Results and Analysis

[0195] 2.1 Panjin rice origin traceability based on stable isotope ratio technology

[0196] 2.1.1 Characteristic analysis of stable isotope ratios in rice from different origins

[0197] Figure 1 The δ of rice in 5 different provinces of China 13 C, δ 15 N, δ 2 H and δ 18 O value distribution, Figure 2 The δ 13 C, δ 15 N, δ 2 H and δ 18 O value distribution, Figure 3 The δ 13 C, δ 15 N, δ 2 H and δ 18 O value distribution; Note: HLJ: Heilongjiang Province; JN: Jilin Province; PJ: Panjin City; JS: Jiangsu Province; SH: Shanghai City; Others: other cities in Liaoning Province; DW: Dawa District; PS: Panshan County.

[0198] Depend on Figure 1 It can be seen that the δ 13 C, δ 15 N, δ 2 H and δ 18 There are significant differences in O values. 13 The C value is between -28.82‰ and -25.97‰. The δ 13 The C value is significantly higher than that of rice from Jiangsu and Shanghai. 13 The C value is related to the photosynthesis pathway. The δ 13 The high C value may be related to its climate. The climate in Panjin is cool, and the stomatal conductance of japonica rice decreases during growth, resulting in reduced carbon isotope fractionation during photosynthesis. 15 N value found that the δ 15 The average N values ​​were higher than 4‰, significantly higher than those of rice from Jiangsu and Shanghai. 15 N was the highest (6.21±1.02‰), which may be related to the fertilization method. 2 The H value ranged from -74.40‰ to -30.25‰, with the most significant variation. Among them, the δ 2 The H value is significantly higher in the Northeast, which may be due to the difference in climatic conditions such as rainfall, temperature, and humidity between the north and south. 18 The O value is between 20.41‰ and 27.27‰. The δ 18 The average O value was 25.24±0.67‰, which was significantly higher than that of rice from other provinces.

[0199] Figure 2 and Figure 3 The results show that in Liaoning Province, the stable isotope ratios of Panjin rice and other rice were compared, and the δ 15 N, δ 18 The O value of Panjin rice and rice from other areas of Liaoning Province is significantly higher than that of rice from other areas of Liaoning Province. 13 C, δ 15 The average H values ​​were close, with no significant difference. Further narrowing the sample range, the stable isotope ratios of Panjin rice were relatively stable, with no significant difference between the rice from Panjin Dawa District and Panshan County.

[0200] 2.1.2 Analysis of machine learning models based on stable isotope ratio technology

[0201] LDA is a discriminant analysis method that maximizes inter-class separation and minimizes intra-class variance in a low-dimensional space. It transforms multidimensional data into low-dimensional data through projection, mapping the original high-dimensional data into a low-dimensional space. This method aims to ensure that data points of different classes are separated in this low-dimensional space, while data points of the same class are closely clustered. RF is an ensemble learning method that generates multiple decision trees through bootstrap resampling. Each tree randomly selects a subset of features for splitting, and ultimately outputs a prediction by voting or averaging the results. SVM is a supervised learning classifier. Its core concept is to find a hyperplane that maximizes the class margin to effectively distinguish data samples from different classes. Kernel functions (such as linear and polynomial functions) are used to map data into a high-dimensional space, solving nonlinear separability problems. Its objective function is optimized using convex quadratic programming to ensure a global optimal solution. KNN makes classification decisions based on the distance between unknown and known samples. Distance metrics such as Euclidean distance are commonly used. The k closest samples are selected and the class of the unknown sample is determined based on the classification of these k samples.

[0202] Before building the classification model, the data were randomly divided into training set (70%) and test set (30%), and then the δ 13 C, δ 15 N, δ 2 H and δ 18 The data were Z-score normalized to ensure that different feature variables were on the same scale. Four machine learning models were used to classify rice from five different provinces, rice from different regions in Liaoning Province, and rice from different regions in Panjin City. The performance of the machine learning models was evaluated using 5-fold cross-validation. The results of the four machine learning models are shown in Table 1.

[0203] For stable isotope ratio data of rice from five different provinces, the test set accuracy of the four models, LDA, RF, SVM, and KNN, ranged from 65% to 81%. The RF model achieved the highest test set accuracy, at 80.77%. In the stable isotope ratio traceability model for Panjin rice and rice from other production areas in Liaoning Province, the LDA and KNN models achieved 100% test set accuracy, while the KNN model achieved higher accuracy and more stable model stability using 5-fold cross-validation. The origin traceability model for rice from different production areas in Panjin City performed poorly, with test set accuracy below 50% and 5-fold cross-validation accuracy below 70%. This may be due to the similar stable isotope ratios of rice from the two production areas, making effective differentiation difficult. Furthermore, the small sample size of this dataset may have contributed to the low model accuracy. Model accuracy can be further improved through the use of other traceability technologies or through integration with other technologies.

[0204] Table 1 Classification accuracy of stable isotope ratio model

[0205]

[0206] 2.2 Panjin rice origin traceability based on near-infrared spectroscopy

[0207] 2.2.1 Near-infrared spectroscopy analysis of rice from different origins

[0208] Figure 4 Near-infrared spectra of rice from different origins, where (a) is the spectrum of all samples, (b) is the average spectrum of five different provinces, (c) is the average spectrum of Liaoning Province, and (d) is the average spectrum of Panjin City.

[0209] from Figure 4 It can be seen that the near-infrared spectra of rice from different regions show similar trends, while the absorbance varies across different bands. The absorption peaks in different bands of the near-infrared spectra are related to chemical components in rice, such as starch, protein, and fat, and the height of the absorption peaks represents the content of these substances.

[0210] 4283~4386cm in the near infrared spectrum -1 The band corresponds to the combined frequency of CH bond stretching and CH2 deformation, which mainly reflects the content of carbohydrates such as starch. -1 and 4762cm -1 The bands are related to the combined vibration of aldehyde group (C=O) and hydroxyl group (OH), further characterizing the composition of carbohydrates. -1 、5095~5025cm -1 、6667~6526cm -1 The spectral bands correspond to the combined vibration of NH plane bending and CN stretching, the combined vibration of NH stretching and plane bending, and the frequency doubling of NH symmetric / antisymmetric vibration, all of which are closely related to protein content. -1 、4329cm -1 4673cm -1 The bands at are related to the second harmonic of CH bending vibration and the combined vibration of CH stretching and CO stretching, reflecting the characteristics of lipids. Figure 4 (b) The average near-infrared spectra of the five provinces show that in the range of 4000-5000 cm -1 and 6000~7000cm -1 The absorption peaks of the two bands are slightly different, which may be related to the protein and lipid components in rice. Figure 4 (c)) found that at 4000~7000cm -1 The range shows more significant separation. Figure 4In (d), the near-infrared spectra of rice from two producing areas in Panjin City are relatively close.

[0211] 2.2.2 Analysis of machine learning models based on near-infrared spectroscopy technology

[0212] Near-infrared spectral data were preprocessed using the SG, MSC, and SNV methods. Based on this, origin traceability models were constructed using four machine learning algorithms: LDA, RF, SVM, and KNN. To comprehensively evaluate the performance of these models, their accuracy on the training and test sets was calculated, and a five-fold cross-validation was performed using stratified sampling. The results are shown in Tables 2–4. As can be seen from Tables 2–4, the accuracy of the preprocessed near-infrared spectral data for rice from five different provinces improved significantly compared to the original spectral data. In particular, the test set accuracy of the RF and KNN models significantly improved after SNV and MSC processing, while the five-fold cross-validation accuracy of the KNN model increased from 50% to 70.91%. This indicates that different near-infrared spectral preprocessing methods have a certain impact on the classification accuracy of the models. Comparing the accuracy of the four models, the LDA model performed relatively stably, with test set accuracy exceeding 75%. The SNV-LDA and MSC-LDA models achieved the highest test set accuracy, reaching 88.46%. The RF and KNN models for near-infrared spectral data of rice from five different provinces showed unstable performance and low accuracy. The origin traceability models for rice from different producing areas within Liaoning Province achieved high accuracy, with all four models achieving test set accuracy exceeding 80%, with the highest accuracy reaching 100%. The accuracy of the models constructed based on near-infrared spectral data of rice from different producing areas in Panjin City declined compared to the previous two groups, likely due to the lack of significant differences in the near-infrared spectra between samples from Dawa District and Panshan County. The RF model had the lowest accuracy, indicating that it is not suitable for near-infrared spectral origin traceability of rice from small production areas. For small-area origin classification, the LDA model still performed best. In summary, among the four machine learning classification models constructed based on near-infrared spectral data of rice from different production areas, the LDA model achieved high accuracy and was relatively stable. Most LDA models achieved accuracy exceeding 80%, with the highest reaching 100%.

[0213] Table 2 Classification accuracy of near infrared spectroscopy model

[0214]

[0215] Table 3 Classification accuracy of near infrared spectroscopy model

[0216]

[0217]

[0218] Table 4 Classification accuracy of near infrared spectroscopy model

[0219]

[0220] 2.2.3 Machine Learning Model Analysis Based on Near-Infrared Spectra after CARS Feature Extraction

[0221] Near infrared spectroscopy is often accompanied by a large amount of redundant data. Characteristic wavelength screening can effectively reduce collinear interference. The CARS algorithm uses Monte Carlo sampling and partial least squares regression (PLS) regression coefficients to screen key wavelengths, significantly reducing the data dimension while retaining a strong correlation with the target variable. The present invention uses the CARS algorithm to extract characteristic wavelengths. The Monte Carlo method samples and iterates 1000 times. The extraction process is as follows: Figure 5 As shown, Figure 5 The results of variable selection for CARS, where (a) is the variable optimization process, (b) is the RMSE change trend, and (c) is the regression coefficient change. Taking the original spectra of 5 different provinces as an example, Figure 6 is the characteristic variable spectrum extracted by CARS algorithm. Figure 5 From the variable optimization process in (a), it can be seen that when the number of variables reaches 26, the root mean square error RMSECV reaches the lowest value of 0.2936, Q 2 The maximum value reached was 0.9501. Ultimately, 52 wavelength variables were selected for model development. The total number of original near-infrared spectra wavelengths was 1557. After CARS processing, the number of characteristic wavelengths extracted is shown in Table 5. The near-infrared spectra of the five provinces differed significantly, with a relatively high number of extracted wavelengths, ranging from 35 to 60. The number of extracted wavelengths for Panjin rice and rice from other production areas in Liaoning Province was relatively low, with no more than 20 wavelengths.

[0222] Table 5 Number of characteristic wavelengths extracted by CARS

[0223] Extract characteristic wavelengths 5 provinces Liaoning Province Panjin City RAW-CARS 52 20 18 SNV-CARS 35 20 35 MSC-CARS 46 20 10 SG-CARS 60 13 30

[0224] Four machine learning models were constructed using near-infrared spectral data after CARS feature extraction. The results are shown in Tables 6 to 8. After CARS processing, the SNV-CARS LDA model improved from 88.18% to 98.33% in 5-fold cross-validation for traceability models from five different provinces, demonstrating significant improvement in model stability. However, performance remained poor in the RF model, with no improvement in model accuracy after feature variable extraction. Accuracy decreased in both the SVM and KNN models. In the rice traceability model from different origins in Liaoning Province, the LDA model achieved 100% accuracy on the test set using all three preprocessing methods. However, the accuracy of the LDA model on the raw spectra decreased after feature wavelength extraction. While the accuracy of some models improved in the RF model, the accuracy of the SVM model test set significantly decreased after CARS feature extraction. This suggests that this feature extraction method is not suitable for SVM models. Other feature extraction methods, such as SPA and UV, may be considered. For rice origin traceability models from two producing areas in Panjin City, the RAW-CARS and SG-CARS models in LDA achieved 100% test set accuracy, and their five-fold cross-validation accuracy increased from 81.67% to 93.33%, demonstrating improved model stability. The RF model had low accuracy when run on spectra without feature extraction. However, after feature extraction using the CARS method, model accuracy improved across several preprocessing methods, with the SNV and MSC models' test set accuracy increasing from 62.50% to 87.50%. Similar to the other two groups, the accuracy of the SVM rice origin traceability model for different producing areas in Panjin City decreased. However, the accuracy of the KNN model improved. In summary, the CARS feature extraction method is not suitable for the SVM model, as the test set accuracy of all three data sets decreased after feature extraction. Overall, the CARS method demonstrates positive effects and potential in improving model accuracy.

[0225] Table 6 Classification accuracy of near-infrared spectroscopy model after CARS feature extraction

[0226]

[0227]

[0228] Table 7 Classification accuracy of near-infrared spectroscopy model after CARS feature extraction

[0229]

[0230] Table 8 Classification accuracy of near-infrared spectroscopy model after CARS feature extraction

[0231]

[0232]

[0233] 2.3 Panjin rice origin traceability based on multi-source data fusion

[0234] 2.3.1 Fusion model of stable isotope ratio and near-infrared spectral data

[0235] As an information processing strategy, data fusion focuses on organically integrating and deeply analyzing information from different data channels, aiming to obtain more comprehensive and accurate data sets and improve accuracy and stability. Generally speaking, data fusion strategies can be divided into three levels: low, medium, and high. Low-level data fusion refers to the direct merging of individual data sets; medium-level data fusion involves extracting characteristic variables from different data sets and then integrating them into a unified data set; and high-level data fusion is performed at the decision-making level. Due to the extensive data processing, low-level fusion techniques are very time-consuming. Medium-level fusion can reduce data redundancy and improve model stability. High-level fusion is more complex and requires consideration of complex correlations between different sources. Therefore, a medium-level data fusion strategy is adopted to fuse multi-source data sets.

[0236] Stable isotope ratios and feature-extracted near-infrared spectral data were fused and trained using four models. The model analysis results are shown in Tables 9 to 11. The best performing models after intermediate-level fusion of stable isotope ratio and near-infrared spectral data from rice from five different provinces were the RF-MSC-CARS and KNN-SG-CARS models, achieving a test set accuracy of 80.77%. Compared to the LDA traceability model based solely on stable isotope ratios, the intermediate-level fusion models achieved higher 5-fold cross-validation accuracy and improved model stability. Among the rice traceability models from different producing areas in Liaoning Province, the best model was the RF model, with test set accuracy ranging from 81.82% to 100.00%. The accuracy of the LDA, SVM, and KNN models after intermediate-level fusion was significantly reduced, possibly due to the introduction of noise during the fusion process or the removal of significant valid variables while removing a large amount of redundant information. For the Panjin rice traceability model from the Dawa and Panshan rice producing areas, the LDA model after intermediate fusion showed lower test set accuracy compared to the model constructed using only near-infrared spectral data, and the model performance was unstable. The SVM model with intermediate fusion showed lower test set accuracy compared to the model constructed using unextracted spectral data, while the accuracy of the model using feature-extracted spectral data and single stable isotope ratio data improved. However, the test set accuracy and 5-fold cross-validation accuracy of the model after intermediate fusion were still not ideal.

[0237] Table 9 Classification accuracy of near-infrared spectral data fusion model after stable isotope ratio and CARS feature extraction

[0238]

[0239]

[0240] Table 10 Classification accuracy of near-infrared spectral data fusion model after stable isotope ratio and CARS feature extraction

[0241]

[0242] Table 11 Classification accuracy of near-infrared spectral data fusion model after stable isotope ratio and CARS feature extraction

[0243]

[0244]

[0245] 2.3.2 Analysis of the Intermediate Fusion Model of Mineral Elements and Metabolomics Data of Rice from Different Production Areas in Panjin City

[0246] The accuracy of the origin traceability model of rice from different producing areas in Panjin City did not achieve very good results. This may be because the two producing areas are relatively close, and the differences in their stable isotope ratio values ​​and near-infrared spectra are small, resulting in poor model effects. In order to improve the accuracy of the model, the present invention continues to introduce mineral elements and metabolomics data to construct an origin traceability model. Before data fusion, the mineral elements and metabolomics data of rice from two different producing areas in Panjin City are firstly extracted, the data are Z-Score standardized, and the VIP value of OPLS-DA analysis is used to screen characteristic variables. Through VIP>1, Na, Ca, Cr and Pb were screened out from 17 mineral elements, 33 were screened out from 86 volatile metabolites, and 287 were screened out from 840 non-volatile metabolites. A total of 324 variables were used as characteristic variables for intermediate data fusion. Four machine learning models were used for training to obtain training results, as shown in Table 12. Table 12 shows that, after intermediate data fusion, the three mineral element and metabolomics models, with the exception of the LDA model, achieved high accuracy on the test set, with the RF and SVM models achieving 100% accuracy. This suggests that rice from different producing areas in Panjin City exhibits differences in mineral element and metabolomics profiles, which can be classified using machine learning models. However, the LDA model is not suitable for this dataset, or further adjustment of model parameters is required to achieve higher accuracy.

[0247] Table 12 Classification accuracy of the model based on the fusion of mineral elements and metabolomics data of rice from different producing areas in Panjin City

[0248] Model Train accuracy (%) Test accuracy (%) 5-fold cross validation accuracy (%) LDA 68.75 37.50 63.33 RF 100.00 100.00 100.00 Support Vector Machine 100.00 100.00 100.00 KNN 100.00 87.50 100.00

[0249] 2.3.3 Analysis of the Four Data Fusion Models of Stable Isotope Ratios, Near-Infrared Spectroscopy, Mineral Elements, and Metabolomics for Rice from Different Producing Areas in Panjin City

[0250] The present invention further fuses four kinds of data, namely stable isotope ratio, near infrared spectrum, mineral elements and metabolomics, to observe whether the improvement of LDA model accuracy can be achieved. The stable isotope ratio of rice from different producing areas in Panjin City, the near infrared spectrum after feature extraction, mineral elements and metabolomics data are subjected to intermediate fusion. The results are shown in Table 13. The intermediate data fusion of the four kinds of data can effectively improve the model accuracy of LDA, with an accuracy of up to 100%. As for the RF model, the 5-fold cross-validation accuracy of the four kinds of data fusion model is reduced compared to the fusion of mineral elements and metabolomics data, and the model stability is slightly reduced. The test set and 5-fold cross-validation accuracy of the SVM model remain at 100%, and the model stability is good. Compared with the intermediate data fusion model of stable isotope ratio and near infrared spectrum, the accuracy is significantly improved. Compared with the intermediate data fusion model of mineral elements and metabolomics, the accuracy of the KNN model of the intermediate data fusion of the four kinds of data is not improved, but is significantly higher than the stable isotope ratio and near infrared spectrum data fusion model.

[0251] Table 13 Classification accuracy of the model based on the fusion of stable isotope ratio, near infrared spectroscopy, mineral elements and metabolomics data of rice from different producing areas in Panjin City

[0252]

[0253] (1) Stable isotope analysis showed that the δ 13 C, δ 15 N, δ 2 H and δ 18 There are significant differences in O values. 13 C, δ 15 N, δ 18 The O value is significantly higher than that of other provinces, which may be related to its climate and fertilization methods. 2 H and δ 18 The north-south difference in O is affected by rainfall, temperature and humidity. 15 N and δ 18 O is significantly higher than other areas in the province, but δ 13 C and δ 2 There was no significant difference in H. There was no significant difference in the stable isotope values ​​of rice between Dawa District and Panshan County, indicating that the environment and planting conditions within the small area were similar. Among the four machine learning models based on stable isotope technology, the RF model had the highest test set accuracy (80.77%) among the four models for distinguishing Panjin rice from rice producing areas in other provinces. For the classification of production areas within Liaoning Province, the LDA and KNN models had a test set accuracy of up to 100%. However, the test set accuracy of the machine learning model for classifying rice from different production areas in Panjin City was low, and other technologies need to be combined to improve the model accuracy.

[0254] (2) Near infrared spectroscopy analysis showed that there were differences in the near infrared spectral curves of rice from different origins, especially in the range of 4000-7000 cm -1 The differences in starch, protein, and lipid content may be related to these differences. Preprocessing near-infrared spectra can improve model performance. SNV and MSC preprocessing significantly improved the model accuracy for near-infrared spectra across five provinces. After CARS feature extraction, the accuracy of some models was further improved. The SNV-CARS-LDA model performed particularly well in Panjin rice origin traceability across three dimensions, with a test set accuracy of 75% to 100% and a 5-fold cross-validation accuracy of 93.33% to 100%. However, the spectral differences between the two producing areas in Panjin City are relatively small, resulting in poor adaptability of the SVM model and generally low accuracy. Among the multi-scale Panjin rice origin traceability models, all four models performed better when comparing different producing areas within Liaoning Province. Among the four models, the LDA test set accuracy exceeded 75% for the majority, with the highest reaching 100%, making it more stable than the other three models.

[0255] (3) The intermediate data fusion strategy of multi-source data can significantly improve the classification performance of the traceability model of rice from different origins. Among them, the performance of RF and SVM models was significantly improved after the stable isotope and near-infrared spectral data were fused by the intermediate data strategy. The test set accuracy of the Panjin rice RF model constructed at the inter-provincial scale was increased to 80.77%, and the test set accuracy of the KNN model was significantly improved to between 70% and 81%. Among the traceability models at the provincial scale, the RF model performed best, with the test set accuracy reaching 100%. The intermediate data fusion in Panjin City can effectively improve the accuracy of the SVM model. This shows that multi-source data fusion can effectively make up for the limitations of a single technology. Some of the classification models of production areas in Panjin City still have poor results, and mineral element and metabolomics data are further introduced for intermediate data fusion. It was found that the test set accuracy of the RF and SVM models of the classification models of different production areas in Panjin City can reach 100%, while the accuracy of the LDA model is low. After further considering the integration of four types of data: stable isotopes, near-infrared spectroscopy, mineral elements, and metabolomics, it was found that the accuracy of the LDA model test set was increased to 100%.

[0256] In summary, stable isotope and near-infrared spectroscopy technology combined with machine learning models can achieve effective traceability of production areas at the provincial level and within the province, while small-area classification can achieve high accuracy after relying on highly specific data such as mineral elements and metabolomics.

[0257] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A method for constructing a traceability model for city-level rice, characterized in that: The following steps are involved: Obtain rice samples from different areas of a city; performing stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination, and metabolomics determination on the rice sample to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information, and metabolomics information; Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information; Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information; Using a competitive adaptive reweighted sampling method to filter characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information; Using orthogonal partial least squares discriminant analysis to screen characteristic information of the mineral element information to obtain characteristic mineral element information; Standardizing the characteristic mineral element information to obtain standardized characteristic mineral element information; Using orthogonal partial least squares discriminant analysis to screen characteristic information of the metabolomics information to obtain characteristic metabolomics information; Standardizing the characteristic metabolomics information to obtain standardized characteristic metabolomics information; The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standardized characteristic metabolomics information are integrated to perform modeling to obtain a traceability model for the municipal-grade rice; The method of the standardization process is Z-Score standardization; The preprocessing method is raw spectrum, standard normal transformation, multivariate scatter correction or Savitzky-Golay smoothing; The modeling method is linear discriminant analysis or support vector machine.

2. The method for constructing a traceability model according to claim 1, wherein: The stable isotope ratio includes δ 13 C, δ 15 N, δ 2 H and δ 18 O.

3. The method for constructing a traceability model according to claim 1, wherein: The parameters of the near infrared spectroscopy include: the spectral range is 10000~4000cm -1 , the spectral resolution is 8cm -1 , the number of scans is 32 times.

4. The method for constructing a traceability model according to claim 1, wherein: The mineral elements include Na, Ca, K, Mg, Al, Ba, Mn, Fe, V, Cr, Sr, Co, Ni, Se, As, Pb and Cd.

5. The method for constructing a traceability model according to claim 1, wherein: The metabolomics includes volatile metabolites and non-volatile metabolites.

6. A method for tracing the origin of city-level rice, characterized in that: The following steps are involved: The rice to be tested is subjected to stable isotope ratio determination, near-infrared spectroscopy determination, mineral element determination and metabolomics determination to obtain stable isotope ratio information, near-infrared spectroscopy information, mineral element information and metabolomics information; Based on the standardized stable isotope ratio information result in the construction method according to any one of claims 1 to 5, performing a standardization process to obtain standardized stable isotope ratio information; Based on the characteristic near-infrared spectral information results in the construction method according to any one of claims 1 to 5, screening is performed to obtain characteristic near-infrared spectral information; Based on the standardized characteristic mineral element information result in the construction method according to any one of claims 1 to 5, measurement and standardization processing are performed to obtain standardized characteristic mineral element information; Based on the standardized characteristic metabolomics information result in the construction method according to any one of claims 1 to 5, performing measurement and standardization processing to obtain standardized characteristic metabolomics information; The standardized stable isotope ratio information, characteristic near-infrared spectrum information, standardized characteristic mineral element information and standardized characteristic metabolomics information are substituted into the corresponding municipal-grade rice traceability model obtained by the construction method described in any one of claims 1 to 5 to achieve traceability of the sample to be tested.

7. A method for constructing a traceability model for provincial-level rice, characterized in that: The following steps are involved: Obtain rice samples from different regions within the same province; The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information; Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information; Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information; Using a competitive adaptive reweighted sampling method to filter characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information; The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain a traceability model for the provincial-level rice; The method of the standardization process is Z-Score standardization; The preprocessing method is raw spectrum, standard normal transformation, multivariate scatter correction or Savitzky-Golay smoothing; The modeling method is random forest.

8. A method for tracing the source of provincial-level rice, characterized in that: The following steps are involved: Performing stable isotope ratio measurement and near-infrared spectrum measurement on the rice to be tested to obtain stable isotope ratio and near-infrared spectrum information; Based on the standardized stable isotope ratio information result in the construction method according to claim 7, performing a standardization process to obtain standardized stable isotope ratio information; Based on the characteristic near-infrared spectral information results in the construction method according to claim 7, screening is performed to obtain characteristic near-infrared spectral information; Substituting the standardized stable isotope ratio information and characteristic near-infrared spectrum information into the corresponding provincial-level rice traceability model obtained by the construction method according to claim 7, the traceability of the sample to be tested is achieved.

9. A method for constructing a provincial rice traceability model, characterized in that: The following steps are involved: Obtain rice samples from different provinces; The rice sample is subjected to stable isotope ratio measurement and near infrared spectrum measurement to obtain stable isotope ratio information and near infrared spectrum information; Standardizing the stable isotope ratio information to obtain standardized stable isotope ratio information; Preprocessing the near-infrared spectrum information to obtain preprocessed near-infrared spectrum information; Using a competitive adaptive reweighted sampling method to filter characteristic information of the preprocessed near-infrared spectral information to obtain characteristic near-infrared spectral information; The standardized stable isotope ratio information and the characteristic near-infrared spectrum information are integrated to perform modeling to obtain the provincial-level rice traceability model; The method of the standardization process is Z-Score standardization; The preprocessing method is raw spectrum, standard normal transformation, multivariate scatter correction or Savitzky-Golay smoothing; The modeling method is random forest or K-nearest neighbor.

10. A provincial rice traceability method, characterized in that: The following steps are involved: Performing stable isotope ratio measurement and near-infrared spectrum measurement on the rice to be tested to obtain stable isotope ratio and near-infrared spectrum information; Based on the standardized stable isotope ratio information result in the construction method according to claim 9, performing a standardization process to obtain standardized stable isotope ratio information; Based on the characteristic near-infrared spectral information result in the construction method according to claim 9, screening is performed to obtain characteristic near-infrared spectral information; Substituting the standardized stable isotope ratio information and characteristic near-infrared spectrum information into the provincial rice traceability model obtained by the construction method according to claim 9, the traceability of the sample to be tested is achieved.