Method for predicting moisture content in flavors and fragrances based on characteristic wave band weighted near infrared spectrum

Abstract

The invention discloses a method for predicting moisture content in flavors and fragrances based on a characteristic wave band weighted near infrared spectrum, which comprises the following steps of: obtaining a plurality of characteristic peak wave bands of moisture on the basis of full wave number prediction; multiplying the infrared data of the characteristic peak wave band by an amplification coefficient to carry out weighted amplification, and adding a full-wave band spectrum of the essence and the perfume to obtain a full-wave-number spectrum with weighted characteristic wave bands; carrying out segmented modeling on different moisture contents; by comparing the distances from the sample points to the blending average center point of each group of samples, selecting the group with the minimum distance as a preliminary prediction interval of the moisture content of the samples to obtain a preliminary prediction result, and then performing accurate prediction to obtain the moisture content in the flavors and fragrances. According to the method, the prediction accuracy and precision can be greatly improved, and the demand quantity of modeling samples is reduced.

Description

Technical Field

[0001] The invention relates to a method for quickly predicting moisture in flavors and spices, and in particular to a method for predicting moisture content in flavors and spices based on near-infrared spectroscopy with weighted characteristic bands. Background Art

[0002] Flavors and fragrances are important components of fine chemical products and are widely used in food, cosmetics, daily necessities and other industries. As a key indicator for quality control of flavors and fragrances, moisture content directly affects the stability, safety and shelf life of the products. The accuracy and efficiency of the moisture content determination method directly affect the overall quality of the product. Near-infrared spectroscopy technology has become an important means of moisture content determination due to its advantages such as rapidity, non-destructiveness and accuracy.

[0003] Infrared spectroscopy is a molecular vibration-rotation spectrum that can obtain the overall chemical information of the sample. Due to its rapid, non-destructive and simple characteristics, it is widely used in the quality control and analytical evaluation of traditional Chinese medicine, agricultural products, and wine. The near-infrared spectral region is consistent with the absorption region of the combined frequency and frequency harmonics of the vibration of hydrogen-containing groups (OH, NH, CH) in organic molecules. By scanning the near-infrared spectrum of the sample, the characteristic information of the hydrogen-containing groups in the organic molecules in the sample can be obtained. At present, there are two main ways to predict chemical components by near-infrared quantitative analysis: one is based on characteristic band modeling prediction; the other is based on full wave number modeling prediction. There are the following deficiencies for the above two quantitative predictions: for single characteristic band modeling prediction, in the analysis of complex systems, since the near-infrared only corresponds to the hydrogen-containing group, it is difficult to accurately select the characteristic band of the predicted component, and the selection process may cause information loss, thereby causing prediction deviation. For the full-band prediction model, it is usually required that the source of the modeling sample and the predicted sample, such as type, variety, origin and processing method, is basically consistent, and the better the consistency, the better the prediction accuracy. However, there are many types of flavors and fragrances, and the water content ranges widely, so it is difficult for a simple model to make accurate predictions. Therefore, using the full-band prediction results will introduce systematic deviations.

[0004] For model prediction analysis, a large amount of sample data is usually required in the modeling process to ensure accurate model prediction. Since large-scale sample collection and testing is troublesome and sometimes difficult to achieve, in order to reduce the demand for samples, it is usually hoped that accurate prediction of components can be achieved through small sample modeling. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a two-step method based on characteristic band weighted near infrared spectroscopy technology to quickly predict the moisture content in flavors and fragrances.

[0006] The purpose of the present invention is achieved through the following technical solutions:

[0007] A two-step method based on characteristic band weighted near infrared spectroscopy technology to quickly predict the moisture content in flavors and fragrances includes the following steps:

[0008] Step (1): Near infrared scanning

[0009] Sample solution, scanned by near-infrared diffuse reflectance spectroscopy; the sample solution was stirred evenly and poured into the sample cup, the sample solution depth was ≥10mm; the number of scans was 64 times, and the resolution was 1cm -1 The scanning range was 4 000-10 000 cm-1, and the average value of three tests was used to eliminate the error. The original spectrum was automatically baseline corrected and normalized to obtain the corresponding standardized spectrum.

[0010] Step (2): infrared data pre-processing

[0011] The full wave number spectrum obtained in the above step (1) is processed by using the spectrum preprocessing method of Norris smoothing (Smooth) and first-order derivative (D1) to process the infrared spectrum.

[0012] Step (3): Selection and weighting of moisture characteristic bands

[0013] By analyzing the near infrared spectral characteristics of flavors and fragrances, the main characteristic peak bands of water are: 4015-4167, 5263-5685 and 7198-7692cm -1 The infrared data of these characteristic bands are multiplied by the amplification factor for weighted amplification, and then added to the full-band spectrum of flavors and fragrances to obtain the weighted full-wavenumber spectrum of the characteristic bands.

[0014] Step (4), gas chromatography detection of water content

[0015] Weigh 0.8 g of the sample and place it in a centrifuge tube, record it as m, accurately add 10 mL of 0.02 mg / mL acetone-methanol extract, seal it with a cap, and place it in an oscillator ( for extraction for 10 to 30 min. Take about 2 mL of the extract and filter it with an organic phase filter membrane (5.5), and transfer it to a chromatographic analysis bottle for analysis.

[0016] Chromatographic conditions:

[0017] Inlet temperature: 250°C, detector temperature: 250°C; column box temperature: 170°C (isotherm), analysis time: 6 minutes; carrier gas: helium, 8.0 mL / min; tail gas: helium, 10 mL / min; reference flow: 25 mL / min; injection volume: 1 μL, split injection, split ratio 20:1.

[0018] Step (5), segmented infrared spectrum modeling of different moisture contents;

[0019] The water content measured in step (4) was grouped into groups, and the contents [0, 0.1), [0.1, 0.2), and [0.2, 0.3) were selected for grouping. After the weighted full-wave number spectrum of the characteristic band obtained in the above step (3) was matched with the water content data obtained in step (4), PLS fitting was performed. During the model establishment process, outliers were continuously eliminated, and then the test fitting was performed until the determination coefficient R 2 Maximum, respectively establish moisture prediction models.

[0020] Step (6), preliminary infrared prediction of moisture;

[0021] For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), they are respectively imported into all the groups in step (5) for PLS-DA analysis to obtain the two-dimensional score (a, b) of the tested sample and the two-dimensional score (xi, yi) of each sample in the grouped sample, and the distance from the sample (a, b) point to the harmonic mean center point H (x0, y0) of the grouped sample is calculated:

[0022]

[0023] in:

[0024]

[0025] Where n is the number of samples in this group;

[0026] Compare the distances from the sample points to the harmonic mean center points of each group of samples, and select the group with the smallest distance, which is the preliminary prediction interval of the moisture content of the sample.

[0027] Step (7) Infrared accurate prediction of moisture

[0028] For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), the samples are imported into the model constructed for the interval grouping selected in step (6), and the accurate infrared prediction value of moisture content can be obtained.

[0029] Furthermore, in step (3), the multiplication factor is 1.1-4.5, preferably 1.5-2.5.

[0030] Furthermore, in step (5), the number of samples in each grouped sample cannot be less than 3.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] (1) The present invention has a short operation time, and the moisture content of flavors and fragrances can be quickly obtained in just a few minutes;

[0033] (2) The present invention adopts the full wave number spectrum with characteristic band weighting, and the accuracy and precision are significantly improved;

[0034] (3) The present invention realizes rapid and accurate prediction of the water content in flavor and fragrance samples through a two-step method of preliminary segmentation-accurate prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 It is the predicted value-true value diagram of model 1 in Example 1;

[0036] Figure 2 It is the predicted value-true value diagram of the control model a in Example 1;

[0037] Figure 3 It is the predicted value-true value diagram of the control model 1 in Example 1;

[0038] Figure 4 The PCA graph of the infrared spectrum of sample 1 and the infrared spectrum of the samples grouped in [0, 0.1) in Example 2;

[0039] Figure 5 The PCA graph of the infrared spectrum of sample 1 and the infrared spectrum of the samples grouped in [0.1, 0.2) in Example 2;

[0040] Figure 6 This is the PCA graph of the infrared spectrum of sample 1 in Example 2 and the infrared spectrum of the samples grouped in [0.2, 0.3). DETAILED DESCRIPTION

[0041] Example 1

[0042] The two-step method of near infrared spectroscopy technology based on characteristic band weighting in this embodiment quickly predicts the moisture content in flavors and fragrances, including the following steps:

[0043] Step (1): Near infrared scanning

[0044] 46 flavor and fragrance sample solutions were scanned using near-infrared diffuse reflectance spectroscopy. The sample solutions were stirred evenly and poured into a sample cup. The sample solution depth was ≥10 mm. The number of scans was 64, and the resolution was 1 cm. -1 The scanning range was 4 000-10 000 cm-1, and the average value of three tests was used to eliminate the error. The original spectrum was automatically baseline corrected and normalized to obtain the corresponding standardized spectrum.

[0045] Step (2): infrared data pre-processing

[0046] The full wave number spectrum obtained in the above step (1) was processed by using Norris smoothing (Segment length: 13, Gap between segments 5) and first-order derivative (D1) spectrum preprocessing methods.

[0047] Step (3): Selection and weighting of moisture characteristic bands

[0048] By analyzing the near infrared spectral characteristics of flavors and fragrances, the main characteristic peak bands of water are: 4015-4167, 5263-5685 and 7198-7692cm -1 The infrared data of these characteristic bands are multiplied by 2.0 for weighted amplification, and then added to the full-band spectrum of flavors and fragrances to obtain the weighted full-wavenumber spectrum of the characteristic bands.

[0049] Step (4), gas chromatography detection of water content

[0050] Weigh 0.8 g of the sample and place it in a centrifuge tube, record it as m, accurately add 10 mL of 0.02 mg / mL acetone-methanol extract, seal it with a cap, and place it in an oscillator ( for extraction for 10 to 30 min. Take about 2 mL of the extract and filter it with an organic phase filter membrane (5.5), and transfer it to a chromatographic analysis bottle for analysis.

[0051] Chromatographic conditions:

[0052] Inlet temperature: 250°C, detector temperature: 250°C; column box temperature: 170°C (isotherm), analysis time: 6 minutes; carrier gas: helium, 8.0 mL / min; tail gas: helium, 10 mL / min; reference flow: 25 mL / min; injection volume: 1 μL, split injection, split ratio 20:1.

[0053] Step (5): Establishment of infrared prediction model

[0054] After matching the weighted full wave number spectrum of the characteristic band obtained in step (3) with the moisture content data obtained in step (4), PLS fitting is performed. During the model establishment process, outliers are continuously eliminated, and then the test fitting is performed until the determination coefficient R 2 Maximum, establish a moisture prediction model.

[0055] Step (6) Infrared prediction of moisture content of flavor and fragrance samples

[0056] For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), the samples are introduced into the model established in step (5) to obtain the infrared prediction value of moisture.

[0057] See also Figure 1-Figure 3As shown, in order to verify the superiority of the characteristic band weighting, a full-wave number control model 1 and a control model a established only with the characteristic band were established according to all the above steps but excluding step (3).

[0058] Table 1 shows the evaluation parameters of the two models. It can be seen from Table 1 that the prediction ability and prediction results of the full wave number model with characteristic band weighting are better than those of the full wave number model without weighting, while the precision and accuracy of the single characteristic band modeling model are very low and cannot meet the requirements of rapid prediction. The corresponding modeling effect diagram is shown in Figure 1-Figure 3 .

[0059] Table 1

[0060] <![CDATA[R 2 ]]> Control Model 1 0.8492 Control Modela 0.7141 Model 1 0.9815

[0061] In order to further verify the effectiveness of the model, six flavor and fragrance samples were analyzed by near infrared, and the moisture content was predicted using the above model. The results are shown in Table 2. Table 2 shows the predicted moisture content and true value (gas chromatography) data of flavors and fragrances. It can be seen from Table 2 that the data obtained by model 1 has very high accuracy and precision. Therefore, it is feasible to use the two-step method based on characteristic band weighted near infrared spectroscopy technology to quickly predict the moisture content in flavors and fragrances in actual operation.

[0062] Table 2

[0063] ID True value Model 1 predicted value Prediction value of control model 1 1 0.145 0.135 0.174 2 0.397 0.379 0.368 3 0.205 0.234 0.244 4 0.326 0.31 0.371 5 0.185 0.169 0.153 6 0.301 0.311 0.342

[0064] Example 2

[0065] In order to further verify the accuracy of the two-step infrared characteristic band weighted prediction model, the same sample as in Example 1 was used, and the first half of the test steps were consistent with steps (1) to (4) in Example 1, and the subsequent steps were as follows:

[0066] Step (5), segmented infrared spectrum modeling of different moisture contents;

[0067] The water content measured in step (4) was grouped into groups, and the contents [0, 0.1), [0.1, 0.2), and [0.2, 0.3) were selected for grouping. After the weighted full-wave number spectrum of the characteristic band obtained in the above step (3) was matched with the water content data obtained in step (4), PLS fitting was performed. During the model establishment process, outliers were continuously eliminated, and then the test fitting was performed until the determination coefficient R 2 Maximum, respectively establish moisture prediction models.

[0068] Step (6), preliminary infrared prediction of moisture;

[0069] For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), they are respectively imported into all the groups in step (5) for PLS-DA analysis to obtain the two-dimensional score (a, b) of the tested sample and the two-dimensional score (xi, yi) of each sample in the grouped sample, and the distance from the sample (a, b) point to the harmonic mean center point H (x0, y0) of the grouped sample is calculated:

[0070]

[0071] in:

[0072]

[0073] Where n is the number of samples in the group

[0074] Compare the distances from the sample points to the harmonic mean center points of each group of samples, and select the group with the smallest distance, which is the preliminary prediction interval of the moisture content of the sample.

[0075] Step (7) Infrared accurate prediction of moisture

[0076] For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), the samples are imported into the model constructed for the interval grouping selected in step (6), and the accurate infrared prediction value of moisture content can be obtained.

[0077] Taking the test sample 1 (YP1) as an example, first substitute it into the [0, 0.1) sample group, and the score chart of the PLS-DA analysis result is as follows: Figure 4As shown, the two-dimensional scores of the modeled data are T1-1 (-13.2547, -4.29033), T1-2 (-4.97117, 5.70065), T1-3 (24.6079, -3.29366), T1-4 (0.372776, 6.31597), T1-5 (58.1696, 1.8241), T1-6 (-32.8625, 10.1629), T1-7 ( -37.9242, 16.5459), T1-8 (-20.1149, 6.23838), T1-9 (-4.64523, -4.06941), T1-10 (-12.9363 , -1.93243), T1-11 (-31.5358, 7.32121), T1-12 (-6.78516, -0.818343), T1-13 (23.7345, -20. 2914),T1-14(25.7923,-7.93844),T1-15(46.5678,21.788),T1-16(-22.4111,-4.08032),T 1-17 (-52.6428, 2.673), T1-18 (6.85013, 7.03396), T1-19 (4.73225, 10.4345), T1-20 (-13.72 35, 3.13888), T1-21 (-5.35557, 8.86566), T1-22 (-12.6338, 0.392281), T1-23 (-4.56446, -33.8072), T1-24 (7.55269, -37.9695), T1-25 (18.2228, -11.1212), and sample YP1 (59.7559, 21.1939). The harmonic mean center point 1H1 (13.08124, 13.24006) was calculated by the harmonic mean center point equation, and L1 was calculated by the distance formula = 47.35.

[0078] In addition, the test sample 1 (YP1) was substituted into the [0.1, 0.2) sample group, and the PLS-DA analysis results were scored as shown in the figure below: Figure 5As shown, the two-dimensional scores of the modeled data are T2-1 (-4.59613, -5.88169), T2-2 (11.9512, 17.151), T2-3 (5.56968, -4.26931), T2-4 (35.4271, 9.01292), T2-5 (20.7884, -0.386544), T2-6 (-33.1966, 61.6232), T2-7 (-5.87205, -9.06264 ), T2-8 (-23.5818, -28.774), T2-9 (6.06604, -0.283141), T2-10 (-5.27466, 1.26954), T2-11 (-3.1262, -12.7283), T2-12 (-0.0860193, -7.43404), T2-13 (18.4098, 7.52058), and sample 1YP1 (-22.4788, -27.7576). The harmonic mean center point 2H2 (-1.08005, -2.25103) was calculated by the harmonic mean center point equation, and L2 = 33.29 was calculated using the distance formula.

[0079] In addition, the test sample 1 (YP1) was substituted into the [0.2, 0.3) sample group, and the PLS-DA analysis results were scored as shown in the figure below: Figure 6 As shown, the two-dimensional scores of the modeled data are T3-1 (-27.5326, 1.06097), T3-2 (-30.4032, -1.40289), T3-3 (-23.5699, 0.259747), T3-4 (41.3644, -8.68076), and sample 1YP1 (40.1413, 8.76294). The harmonic mean center point 3H3 (-45.73338, 1.008974) was calculated by the harmonic mean center point equation, and L2 was calculated by the distance formula = 86.22.

[0080] Through the above calculation and comparison, it can be seen that the distance between sample 1YP1 and the centroid 2 is the smallest, and it is preliminarily predicted that the content range of sample 1 belongs to the interval of 0.1-0.2.

[0081] Therefore, the infrared data of sample 1 is processed according to step (1), step (2), and step (3) and then introduced into the model constructed by grouping the interval [0.1, 0.2) selected in step (6), and its content is predicted to be 0.141. The content of all 6 samples to be tested is predicted according to the above method, and the results are shown in Table 3.

[0082] Table 3

[0083]

[0084]

[0085] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the moisture content in flavors and fragrances based on near-infrared spectroscopy with weighted characteristic bands, characterized in that: The steps include: Step (1), scanning the sample solution by near-infrared diffuse reflectance spectroscopy to obtain an original spectrum, and automatically performing baseline correction and normalization on the original spectrum to obtain a corresponding full-wave number standard spectrum; Step (2), preprocessing the full wave number standard spectrum obtained in step (1); Step (3) Selection and weighting of characteristic bands Analyze the near infrared spectrum characteristics of flavors and fragrances to obtain several characteristic peak bands of water; multiply the infrared data of the characteristic peak bands by the amplification coefficient for weighted amplification, and then add the full-band spectrum of the flavors and fragrances to obtain a weighted full-wave number spectrum of the characteristic bands; Step (4), measuring the moisture content by gas chromatography; Step (5), segment modeling for different upper contents The water content measured in step (4) was grouped into groups, and the contents [0, 0.1), [0.1, 0.2), and [0.2, 0.3) were selected for grouping. After the weighted full-wave number spectrum of the characteristic band obtained in the above step (3) was matched with the water content data obtained in step (4), PLS fitting was performed. During the model establishment process, outliers were continuously eliminated, and then the test fitting was performed until the determination coefficient R 2 Maximum, respectively, establish moisture prediction models; Step (6), infrared preliminary prediction of sample moisture content range For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), they are respectively imported into all the groups in step (5) for PLS-DA analysis to obtain the two-dimensional score (a, b) of the tested sample and the two-dimensional score (xi, yi) of each sample in the grouped sample, and the distance from the sample (a, b) point to the harmonic mean center point H (x0, y0) of the grouped sample is calculated: in: Where n is the number of samples in this group; Compare the distances from the sample point to the harmonic mean center point of each group of samples, and select the group with the smallest distance, which is the preliminary prediction interval of the sample moisture content; Step (7), accurate infrared prediction of moisture content of flavor and fragrance samples For the flavor and fragrance samples to be tested, after being processed according to steps (1), (2), and (3), the samples are imported into the model constructed for the interval grouping selected in step (6), and the accurate infrared prediction value of moisture content can be obtained.

2. The method according to claim 1, characterized in that Step (1) also includes: Stir the sample solution evenly and pour it into the sample cup. The sample solution depth should be ≥10mm. The number of scans should be 64 and the resolution should be 1cm. -1 , scanning range is 4 000~10 000cm -1 , the average value of 3 tests was used to eliminate the error.

3. The method according to claim 1, characterized in that Step (2) further includes: The full wave number spectrum obtained in step (1) is processed by using the spectral preprocessing method of Norris smoothing and first-order derivative to process the infrared spectrum.

4. The method according to claim 1, characterized in that: Step (3) also includes: The characteristic peak bands of the water content are: 4015-4167, 5263-5685 and 7198-7692 cm -1 .

5. The method according to claim 1, characterized in that Step (4) further includes: Weigh 0.8 g of the sample and place it in a centrifuge tube, record it as m, accurately add 10 mL of 0.02 mg / mL acetone-methanol extract, seal it with a cap and place it in an oscillator to extract for 10 to 30 minutes; take 2 mL of the extract and filter it with an organic phase filter membrane, and transfer it to a chromatographic analysis bottle for analysis.

6. The method according to claim 5, characterized in that The conditions for chromatographic analysis are: Inlet temperature: 250°C, detector temperature: 250°C; column box temperature: 170°C (isotherm), analysis time: 6 minutes; carrier gas: helium, 8.0 mL / min; tail gas: helium, 10 mL / min; reference flow: 25 mL / min; injection volume: 1 μL, split injection, split ratio 20:

1.

7. The method according to claim 1, characterized in that In step (5), the number of samples in each grouped sample is more than 3.

8. The method according to any one of claims 1 to 7, characterized in that: In step (3), the amplification factor is 1.1-4.

5.

9. The method according to claim 8, characterized in that The magnification factor is 1.5-2.5.

Images