Method for predicting phytic acid content and glucosinolate content in horseradish tree leaves

Abstract

The invention discloses a method for predicting the content of phytic acid and glucosinolate in horseradish tree leaves. According to the method, a near infrared spectrum analyzer is used for collecting spectral data of the moringa oleifera leaf powder, the spectral data are combined with data measured by a chemical analysis method for processing, a prediction model for the content of phytic acid and glucosinolate in the moringa oleifera leaves is constructed, the prediction effect on the content of phytic acid and glucosinolate is good, the credibility is high, and the method is suitable for popularization and application. The method is efficient, rapid, simple and low in cost.

Description

Technical Field

[0001] The present invention belongs to the technical field of analysis and detection of Moringa oleifera leaves, and particularly relates to a method for predicting the contents of phytic acid and glucosinolate in Moringa oleifera leaves. Background Art

[0002] Near infrared spectral analysis technique (NIRS) is an indirect analysis technique that can achieve qualitative or quantitative analysis of unknown samples by establishing a calibration model. Near infrared light refers to the electromagnetic wave between visible light and mid-infrared light in the wavelength range of 700 - 2500 nm. In this region, the wavelengths can record the overtone and combination frequency absorption information of the chemical bond vibrations of different hydrogen-containing groups (such as O-H, C-H, N-H, S-H, P-H, etc.). The type and quantity of groups will directly affect data such as spectral peak position, peak intensity, and peak shape. Through these data, the internal information of the sample can be judged and obtained, which lays the theoretical foundation for the quantitative and qualitative analysis of NIRS.

[0003] The origin of near infrared spectral technology can be traced back to 1800 when the British physicist F.W. Herschel discovered infrared light. In 1881, W. Abney and E.R. Festing used a Hilger spectrometer to take 48 near infrared absorption spectra (700 - 1200 nm) and found that the near infrared spectral band was related to hydrogen-containing groups. However, it was not until 1974 that the Swedish chemist S. Wold and Professor B.R. Kowalski of the University of Washington in the United States established the discipline of chemometrics that near infrared spectral technology began to be applied and was taken seriously in the mid-1980s. In the past few decades, near infrared spectroscopy has been one of the most rapidly developing analysis technologies. As the three cores of modern near infrared spectral analysis technology, the basic theory of vibration spectroscopy, spectral instrument hardware, and chemometrics have all made very great progress, which has also laid a good foundation for the development of near infrared spectral technology.

[0004] Due to the characteristics of high measurement efficiency, low cost, good reproducibility, etc. of near infrared spectral technology, this technology has been widely applied in many fields, such as agriculture, food, medicine, geology, etc. For example, predicting the crude protein and crude fat of forage grass, detecting the quality of chicken breast meat, predicting the post-harvest quality of blueberries, detecting the total phenol content of Chinese prickly ash, detecting the quality traits of passion fruit pulp, detecting kidney conditions, detecting the quantitative analysis of traditional Chinese medicine and information such as the origin of medicinal materials, predicting soil properties, detecting the state of building material limestone, etc.

[0005] In agriculture, near-infrared analysis technology is mainly used to predict and determine the quality of agricultural products and the nutrient content of forage, such as crude protein and moisture in corn, crude protein and acid detergent fiber in dried forage, etc. However, there are few studies on the construction of near-infrared models for anti-nutritional components in forage. There is no report on the research of prediction methods for the content of phytic acid and glucosinolate in Moringa oleifera leaves. Summary of the Invention

[0006] In view of the above deficiencies of the prior art, the present invention constructs for the first time a near-infrared prediction model for anti-nutritional factors such as phytic acid and glucosinolate in Moringa oleifera leaves. The near-infrared spectrum analyzer is used to collect spectral data of Moringa oleifera leaf powder, which is combined with the data measured by chemical analysis methods, providing a method for predicting the content of phytic acid and glucosinolate in Moringa oleifera leaves, establishing an evaluation system for the later rapid detection of the content of anti-nutritional factors in Moringa oleifera leaves, and thus providing technical support for breeding Moringa oleifera families or individuals with low anti-nutritional factors.

[0007] A method for predicting the content of phytic acid and glucosinolate in Moringa oleifera leaves of the present invention includes the following steps:

[0008] S1. Collect fresh Moringa oleifera leaves, cut them into sections, dry, crush, and sieve them to obtain Moringa oleifera leaf powder samples;

[0009] S2. Use a near-infrared analyzer to collect spectral data of Moringa oleifera leaf powder. The spectral scanning wavelength range is 950 - 1650 nm to obtain the original spectral data of the samples;

[0010] S3. Use the ferric chloride method to determine the phytic acid content of Moringa oleifera leaf powder and the palladium chloride colorimetric method to determine the glucosinolate content of Moringa oleifera leaf powder as the sample reference values;

[0011] S4. Correlate the sample reference values and the original spectral data, and divide the samples into a calibration set and a validation set;

[0012] S5. Preprocess the original spectral data of the calibration set by the differential second derivative method, select wavelength characteristic variables by the competitive adaptive reweighted sampling algorithm, and construct a prediction model for the content of phytic acid and glucosinolate by the partial least squares method;

[0013] S6. Draw a relationship curve between the reference values of the validation set and the predicted values obtained by using the prediction model, and evaluate the difference level between the reference values and the predicted values through paired t-tests. If P > 0.05, it indicates that the prediction model has a good prediction effect and high credibility;

[0014] S7. Use the prediction model to predict and determine the content of phytic acid and glucosinolate in the Moringa oleifera leaves to be analyzed.

[0015] Preferably, the step S1 is: collecting fresh moringa leaves, cutting them into small sections of 2-3 cm, drying at 65 °C, pulverizing and sieving through a 60-mesh sieve to obtain a moringa leaf powder sample.

[0016] Preferably, the step S2 is: taking 3 g of moringa leaf powder from each sample and putting it into the sample cup of a near-infrared analyzer, collecting the spectrum using a DA7200 type near-infrared analyzer, the sample measurement method being diffuse reflection, the spot diameter being 3.5 cm, the spectral scanning wavelength range being 950-1650 nm, the ambient temperature being 23±2 °C and the relative humidity being 50%, each sample being scanned continuously three times, the sample loading being repeated three times, and taking the average value as the original sample spectrum data.

[0017] Preferably, the step S4 is: importing the reference value data of step S3 into the software Simplicity TM , ensuring that the reference value of each sample corresponds one-to-one with the original spectrum data, exporting it in the JCAMP-DX format, observing, analyzing and processing the original near-infrared spectrum using the chemometric software TheUnscrambler 9.8; based on the sample division method of X-Y distance, eliminating abnormal samples, dividing the most diverse subset in the overall samples into the calibration set, and the remaining samples being the validation set, the quantity ratio of the calibration set to the validation set being 10-11:1.

[0018] Preferably, the wavelength characteristic variables for the phytic acid content prediction model are the wavelength points of 985, 990, 995, 1010, 1025, 1035, 1040, 1045, 1100, 1130, 1135, 1150, 1160, 1165, 1180, 1195, 1205, 1210, 1235, 1255, 1310, 1320, 1340, 1345, 1365, 1370, 1400, 1405, 1410, 1445, 1450, 1460, 1470, 1490, 1500, 1510, 1520, 1535, 1545, 1550, 1570, 1600, 1605, 1640 nm.

[0019] Preferably, the wavelength characteristic variables for the glucosinolate content prediction model are the wavelength points at 950, 960, 965, 990, 1005, 1015, 1050, 1065, 1120, 1130, 1140, 1185, 1190, 1205, 1225, 1230, 1235, 1240, 1250, 1270, 1275, 1280, 1290, 1305, 1315, 1340, 1355, 1370, 1415, 1450, 1455, 1465, 1470, 1475, 1485, 1495, 1530, 1540, 1545, 1555, 1580, 1620, 1630, 1640 nm.

[0020] Preferably, the prediction models for phytic acid and glucosinolate contents constructed by partial least squares method include the following steps: eliminating abnormal samples according to the studentized residuals and Influence Plot leverage values of the samples, controlling the number of eliminated samples to be less than 10% of the total sample amount, and constructing a prediction model by partial least squares method; evaluating the prediction effect of the model by combining the determination coefficient of the calibration set, the root mean square error of the calibration set, the determination coefficient of the prediction set, the root mean square error of the prediction set, and the prediction relative analysis error to obtain the optimal modeling parameters.

[0021] Preferably, before the collection in step S2, the sample temperature is balanced to the temperature of the operation room where the near-infrared analyzer is located, and the instrument is preheated for 30 min and waits for the light source to be stable.

[0022] The beneficial effects of the present invention are as follows:

[0023] The method proposed by the present invention is to collect spectral data of moringa leaf powder using a near-infrared spectrometer, combine it with the data measured by chemical analysis methods for processing, construct prediction models for the contents of anti-nutritional factors such as phytic acid and glucosinolate in moringa, and evaluate the models.

[0024] The existing methods for determining the content of anti-nutritional factors in moringa are mainly chemical analysis methods, which are cumbersome, labor-intensive, time-consuming, and costly, and are not suitable for large-scale use. The near-infrared prediction model established by the present invention and the prediction method based on this model are an efficient, fast, simple, and low-cost method. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is the result graph of the CARS characteristic variables of phytic acid.

[0026] Figure 2 It is the position of the wavelength points screened by CARS for the phytic acid model. The blue curve is the independent variable formed by PLS, and the red dots on the line are the wavelength points screened by CARS.

[0027] Figure 3 It is the result graph of the characteristic variables of glucosinolate CARS.

[0028] Figure 4 It is the position of the wavelength points screened by glucosinolate model CARS. The blue curve is the independent variable formed by PLS, and the red dots on the line are the wavelength points screened by CARS.

[0029] Figure 5 It is the comparison graph of the CARS-PLS prediction results of the phytic acid calibration set.

[0030] Figure 6 It is the comparison graph of the prediction results of the phytic acid validation set.

[0031] Figure 7 It is the comparison graph of the CARS-PLS prediction results of the glucosinolate calibration set.

[0032] Figure 8 It is the comparison graph of the prediction results of the glucosinolate validation set. Detailed implementation manners

[0033] The following examples are further illustrations of the present invention rather than limitations thereof.

[0034] Example 1

[0035] 1 Collection of representative Moringa oleifera leaf samples and sample pretreatment

[0036] The Moringa oleifera leaves were collected from the Qilin North Base of South China Agricultural University in Tianhe District, Guangzhou. In July 2022, 120 Moringa oleifera leaf samples were collected, and 32 Moringa oleifera leaves were collected in other months, with a total of 152 leaf samples.

[0037] After the fresh leaves were collected, they were cut into small sections of 2 - 3 cm with a guillotine, then dried at a constant temperature of 65 °C, pulverized with a plant pulverizer after drying, and passed through a 60-mesh sieve to obtain Moringa oleifera leaf powder, which was respectively packed into self-sealing bags in units of plants and marked for testing.

[0038] When selecting representative samples, it is necessary to consider the high and low ranges of component contents and the spatial distribution range. The samples used for modeling will affect the quantitative analysis and determination range of the final model. The sample pretreatment method also needs to be unified, and there are differences in the spectral data collected for samples with different particle sizes.

[0039] 2 Near-infrared spectral data collection

[0040] When measuring near-infrared spectra, different samples need to select appropriate spectral measurement methods according to their characteristics to obtain higher-quality sample spectra. At the same time, according to the requirements of the instrument, it is necessary to ensure that the environmental temperature and humidity are within an appropriate range and not near heat sources or cold air vents, because temperature changes and environmental humidity will affect the measurement of spectral data. During operation, it is necessary to strictly control conditions such as sample preparation, sample loading, and instrument parameters to reduce errors.

[0041] Before the experiment, the samples need to be placed in the operation room where the near-infrared spectrometer is located for more than 24 hours to make the sample temperature consistent with the environmental temperature to reduce the influence of temperature changes on the data. Before collection, the machine needs to be preheated for 30 minutes and wait for the light source to stabilize.

[0042] When collecting spectra, take 3 g of each sample and put it into the sample cup of the near-infrared spectrometer, spread it evenly, and ensure that each sample loading operation is the same. Use a near-infrared analyzer (model: DA7200, Perten Company, Sweden) to collect spectra. The experimental conditions for collecting spectra are as follows: the sample measurement method is diffuse reflection, the spot diameter is 3.5 cm, the spectral scanning wavelength range is 950 - 1650 nm, the environmental temperature is 23(±2)°C, and the environmental relative humidity is about 50%. Each sample is scanned continuously three times, loaded three times repeatedly, and the average value is taken as the original data of the sample spectrum.

[0043] 3 Determination of chemical reference values

[0044] To establish a quantitative model, reference values are needed, that is, the measured values of the nutrient contents (phytic acid content, glucosinolate content) determined by chemical analysis methods. To ensure the reliability of the chemical reference values, the determination should be carried out as soon as possible after collecting the spectra. The phytic acid content is determined by the ferric chloride method, and the glucosinolate content is determined by the palladium chloride colorimetric method.

[0045] 4 Correlate chemical reference values and spectral raw data

[0046] Import the measured value data of the nutrient content of the measured samples into the software Simplicity TM , ensure that the measured value of each sample corresponds one-to-one with the near-infrared spectral raw data, export it in JCAMP-DX format, and use the chemometric software TheUnscrambler 9.8 (Camo Company, Norway) to observe, analyze, and process the near-infrared spectral raw map.

[0047] 5 Divide the calibration set and the validation set

[0048] Before building a prediction model, it is necessary to partition the sample set. The calibration set is used to train the model and optimize the parameters, while the validation set is used to verify the performance of the model. Therefore, the calibration set needs to cover the spatial range of the overall sample as much as possible. In quantitative analysis, the Sample set partitioning method based on the joint x-y distances (SPXY) is usually used for sample set partitioning. The principle of this method is to gradually select samples according to the differences of samples in the x (instrument response) and y (prediction parameter) spaces, and partition the most diverse subset from the overall samples as the calibration set, and the remaining samples are the validation set.

[0049] The calculation method of the SPXY method is as follows: First, calculate the Euclidean distance d x (p,q) between samples p and q and the distance d y (p,q) of the corresponding sample reference value y:

[0050]

[0051] where x p (j), x q (j), y p and y q are the j-th variables and the corresponding reference values of the p-th and q-th samples respectively. Then, by integrating the distances between the spectral values and the reference values, an evaluation index of the sample distance is calculated:

[0052]

[0053] For the phytic acid content prediction model, after removing the anomalies, there are 144 samples, among which: 130 samples in the calibration set and 14 samples in the validation set.

[0054] For the glucosinolate content prediction model, after removing the anomalies, there are 143 samples, among which: 130 samples in the calibration set and 13 samples in the validation set.

[0055] 6 Near-infrared spectroscopy preprocessing

[0056] The original spectrum during instrument acquisition not only contains information on the sample composition, but also contains noise generated by various factors. Preprocessing the spectrum can effectively address the impact of spectral noise, data screening, spectral range optimization, and other factors on the spectrum. Selecting an appropriate preprocessing method is a crucial step in establishing a prediction model with good predictive ability and high stability. Preprocessing methods include Normalization, Smoothing, Savitzky Golay (SG) convolution smoothing, 1st Derivative, 2nd Derivative, Standard Normal Variate Transformation (SNV), Multiplicative Scatter Correction (MSC), Differential 2nd Derivative, etc.

[0057] In this embodiment, the PLS partial least squares method is used to construct models for the phytic acid content and glucosinolate content respectively, and their respective numerical indicators are viewed and compared. According to the lower the RMSEC root mean square value of the calibration set and the closer the Correlation correlation coefficient of the calibration set is to 1, the more accurate the model prediction. The optimal preprocessing method is selected. The results are shown in Table 1 and Table 2; among them, the best preprocessing effect is the Differential 2nd Derivative. After processing, and are closer to 1, and RMSEC and RMSEP are closer to 0, and at the same time, there is a higher RPD value.

[0058] Table 1 Comparison of phytic acid PLS model results for different spectral preprocessings of Moringa oleifera leaves

[0059]

[0060] Table 2 Comparison of glucosinolate PLS model results for different spectral preprocessings of Moringa oleifera leaves

[0061]

[0062] 7 Methods for selecting characteristic variables of spectral data

[0063] Traditionally, the wavelength selection methods mainly include the correlation coefficient method, the analysis of variance method, the stepwise regression method, the uninformative variables elimination method, the interval partial least squares method, the competitive adaptive reweighted sampling (CARS) method, the genetic algorithms method, etc. Among them, the genetic algorithms method is a widely used wavelength selection method.

[0064] This embodiment mainly uses the competitive adaptive reweighted sampling (CARS) method.

[0065] For the phytic acid content prediction model, after multiple CARS variable selections, it is determined that when the iterative sampling number is 28, the model has a better effect ( Figure 1 ), and 44 wavelength points are screened out ( Figure 2 , Table 3), and the wavelength point numbers are 8, 9, 10, 13, 16, 18, 19, 20, 31, 37, 38, 41, 43, 44, 47, 50, 52, 53, 58, 62, 73, 75, 79, 80, 84, 85, 91, 92, 93, 100, 101, 103, 105, 109, 111, 113, 115, 118, 120, 121, 125, 131, 132, 139, a total of 44 points, accounting for 31.21% of the total number of wavelength points (there are 141 original spectral wavelength points). After screening by CARS, the data dimension is effectively reduced, and the model construction operation speed is improved.

[0066] For the glucosinolate content prediction model, after multiple CARS variable selections, it is determined that when the iterative sampling number is 25, the model has a better effect ( Figure 3 ), and 44 wavelength points are screened out ( Figure 4 , Table 3), and the wavelength point numbers are 1, 3, 4, 9, 12, 14, 21, 24, 35, 37, 39, 48, 49, 52, 56, 57, 58, 59, 61, 65, 66, 67, 69, 72, 74, 79, 82, 85, 94, 101, 102, 104, 105, 106, 108, 110, 117, 119, 120, 122, 127, 135, 137, 139, a total of 44 points, accounting for 31.21% of the total number of wavelength points (there are 141 original spectral wavelength points). After screening by CARS, the data dimension is effectively reduced, and the model construction operation speed is improved.

[0067] Table 3 Wavelength points screened

[0068]

[0069]

[0070] Near-infrared model construction method

[0071] In terms of the modeling method, the partial least squares method has strong anti-interference ability. It can participate in the establishment of the multivariate calibration model with all wavelengths, which is more effective than performing multiple regression on each dependent variable one by one. The analysis results are more reliable, the prediction accuracy is higher and more stable. This embodiment uses this modeling method.

[0072] According to the studentized residual of the sample and the leverage value of the Influence Plot, abnormal samples are removed. At the same time, it is controlled that the number of removed samples should be less than 10% of the total sample volume until the root mean square value of the calibration set drops to a lower level and the correlation coefficient of the calibration set approaches 1. Thus, a phytic acid content prediction model and a glucosinolate content prediction model are constructed. The key modeling parameters are shown in Table 4.

[0073] Table 4 Key modeling parameters of the constructed phytic acid and glucosinolate content prediction models

[0074]

[0075] Draw the relationship curve between the measured values and the predicted values of the prediction model for phytic acid and glucosinolate in the calibration set of Moringa oleifera leaves ( Figure 5 、 Figure 7 ).

[0076] Evaluation of near-infrared model

[0077] The model usually uses the coefficient of determination for calibration ( ), the root mean square error of the calibration set (RMSEC), the coefficient of determination for prediction ( ), and the root mean square error of the prediction set (RMSEP) as the main indicators to measure the prediction effect of the near-infrared model. When the model has a high and When the RMSEC and RMSEP are lower, the performance of the model is better. In addition, the model can comprehensively predict the Residual predictive deviation (RPD) index to further evaluate the prediction effect of the model. In this embodiment, five indicators, namely RMSEC, RMSEP, and RPD, are used for evaluation.

[0078] To describe the accuracy of the phytic acid content prediction model, a relationship curve between the measured values and the predicted values of the phytic acid validation set is plotted ( Figure 6 ). According to the modeling results, for the phytic acid model validation set of Moringa oleifera leaves, is 0.8903, RMSEP is 0.2311, and RPD is 2.7194. The specific values of the measured and predicted values of 14 validation set samples are shown in Table 5.

[0079] Table 5 Prediction results of phytic acid content in Moringa oleifera leaf validation set

[0080]

[0081]

[0082] To describe the accuracy of the glucosinolate content prediction model, a relationship curve between the measured values and the predicted values of the glucosinolate validation set is plotted ( Figure 8 ). According to the modeling results, in the glucosinolate model validation set of Moringa oleifera leaves, is 0.8213, RMSEP is 0.2228, and RPD is 2.5122. The specific values of the measured and predicted values of 13 validation set samples are shown in Table 6.

[0083] Table 6 Prediction results of glucosinolate content in Moringa oleifera leaf validation set

[0084] Test Number Measured Value (%) Predicted Value (%) Absolute Error (%) Relative Error (%) 1 2.8394 2.9851 0.1457 0.0513 2 3.4290 3.3231 0.1059 0.0309 3 2.7990 2.5860 0.2130 0.0761 4 2.0589 1.9390 0.1199 0.0582 5 2.6220 3.0140 0.3920 0.1495 6 1.8552 2.1024 0.2472 0.1332 7 3.0341 3.1030 0.0689 0.0227 8 3.0993 2.6000 0.4993 0.1611 9 1.8689 1.9708 0.1019 0.0545 10 2.1970 2.1510 0.0460 0.0209 11 1.9163 1.7846 0.1317 0.0687 12 1.7891 2.0213 0.2322 0.1298 13 1.7872 1.7672 0.0200 0.0112

[0085] 10 Validation of the near-infrared model prediction method (t-test)

[0086] To test the reliability of the phytic acid content prediction model, in this embodiment, the predicted values and measured values of 14 samples in the validation set are compared and analyzed. The results are shown in Table 5. The minimum absolute error between the predicted and measured values of the phytic acid content of 14 samples is 0.0062%, and the maximum is 0.3860%. Through paired t-test, the correlation coefficient between the predicted and measured values of 14 validation set samples reaches 0.958. Compared with the given significance level of 0.05, the test results show that the difference between the two is not significant (P = 0.168 > 0.05), indicating that the phytic acid content prediction model has good prediction effect and high credibility.

[0087] To test the reliability of the glucosinolate content prediction model, in this embodiment, the predicted values and measured values of 13 samples in the validation set were compared and analyzed. As can be seen from Table 6, the minimum absolute error between the predicted value and the measured value of the glucosinolate content of the 13 validation set samples is 0.0200%, and the maximum is 0.4993%. Through paired t-test, the correlation coefficient between the predicted values and the measured values of the 13 validation set samples reaches 0.918. Compared with the given significance level of 0.05, the test results show that the difference between the two is not significant (P = 0.951 > 0.05), indicating that the glucosinolate content prediction model has a good prediction effect and high credibility.

Claims

1. A method for predicting the contents of phytic acid and glucosinolate in Moringa oleifera leaves, characterized in that, Including the following steps: S1. Collect fresh Moringa oleifera leaves, cut them into sections, dry, pulverize, and sieve to obtain Moringa oleifera leaf powder samples; S2. Use a near-infrared analyzer to collect spectral data of the Moringa oleifera leaf powder. The spectral scanning wavelength range is 950 - 1650 nm to obtain the original sample spectral data; S3. Use the ferric chloride method to determine the phytic acid content of the Moringa oleifera leaf powder and the palladium chloride colorimetric method to determine the glucosinolate content of the Moringa oleifera leaf powder as the sample reference values; S4. Correlate the sample reference values and the original spectral data, and divide the samples into a calibration set and a validation set; S5. Preprocess the original spectral data of the calibration set using the differential second derivative method, select wavelength characteristic variables using the competitive adaptive reweighted sampling algorithm, and construct prediction models for phytic acid and glucosinolate contents using the partial least squares method; S6. Plot the relationship curve between the reference values of the validation set and the predicted values obtained using the prediction model, and evaluate the difference level between the reference values and the predicted values through paired t-tests. If P > 0.05, it indicates that the prediction model has a good prediction effect and high credibility; S7. Use the prediction model to predict and determine the phytic acid and glucosinolate contents of the Moringa oleifera leaves to be analyzed.

2. The method according to claim 1, wherein The step S1 is as follows: Collect fresh Moringa oleifera leaves, cut them into small sections of 2 - 3 cm, dry them at 65 °C, pulverize them, and sieve through a 60-mesh sieve to obtain Moringa oleifera leaf powder samples.

3. The method according to claim 1, characterized in that, The step S2 is as follows: Put 3 g of Moringa oleifera leaf powder of each sample into the sample cup of a near-infrared analyzer. Use a DA7200 type near-infrared analyzer to collect spectra. The sample measurement method is diffuse reflection, the spot diameter is 3.5 cm, the spectral scanning wavelength range is 950 - 1650 nm, the ambient temperature is 23 ± 2 °C, and the relative humidity is 50%. Each sample is continuously scanned three times, and the sample is loaded three times repeatedly. Take the average value as the original sample spectral data.

4. The method according to claim 1, characterized in that, The step S4 is as follows: Import the reference value data of step S3 into the software Simplicity TM , ensure that the reference value of each sample corresponds one-to-one with the original spectral data, export it in the JCAMP-DX format, and use the chemometric software The Unscrambler 9.8 to observe, analyze and process the original near-infrared spectrum; Based on the sample division method of X-Y distance, eliminate abnormal samples, divide the most diverse subset in the overall samples into the calibration set, and the remaining samples are the validation set. The quantity ratio of the calibration set to the validation set is 10-11:

1.

5. The method according to claim 1, wherein The wavelength characteristic variables for the phytic acid content prediction model are the wavelength points of 985, 990, 995, 1010, 1025, 1035, 1040, 1045, 1100, 1130, 1135, 1150, 1160, 1165, 1180, 1195, 1205, 1210, 1235, 1255, 1310, 1320, 1340, 1345, 1365, 1370, 1400, 1405, 1410, 1445, 1450, 1460, 1470, 1490, 1500, 1510, 1520, 1535, 1545, 1550, 1570, 1600, 1605, 1640 nm.

6. The method according to claim 1, characterized in that, The wavelength characteristic variables for the glucosinolate content prediction model are the wavelength points of 950, 960, 965, 990, 1005, 1015, 1050, 1065, 1120, 1130, 1140, 1185, 1190, 1205, 1225, 1230, 1235, 1240, 1250, 1270, 1275, 1280, 1290, 1305, 1315, 1340, 1355, 1370, 1415, 1450, 1455, 1465, 1470, 1475, 1485, 1495, 1530, 1540, 1545, 1555, 1580, 1620, 1630, 1640 nm.

7. The method according to claim 1, wherein The prediction models for phytic acid and glucosinolate contents constructed by partial least squares method include the following steps: eliminating abnormal samples according to the studentized residuals and Influence Plot leverage values of the samples, controlling the number of eliminated samples to be less than 10% of the total sample quantity, and constructing a prediction model by partial least squares method; evaluating the prediction effect of the model by combining the determination coefficient of the calibration set, the root mean square error of the calibration set, the determination coefficient of the prediction set, the root mean square error of the prediction set, and the prediction relative analysis error to obtain the optimal modeling parameters.

8. The method according to claim 1, characterized in that Before the collection in step S2, the temperature of the sample is balanced to the temperature of the operation room where the near-infrared analyzer is located, and the instrument is preheated for 30 min and waits for the light source to be stable.

Images