A rapid non-destructive detection method for the peroxide value of Torreya grandis

Abstract

The invention discloses a rapid non-destructive detection method for the peroxide value of Torreya grandis. Torreya grandis with similar peroxide value spectra are screened on the basis of the substantially consistent shapes and sizes of Torreya grandis by constructing a similar expression of the peroxide value spectrum of Torreya grandis, so as to establish a prediction model suitable for the peroxide value detection of a single Torreya grandis, thereby solving the problems of the existing method that the pre-processing is cumbersome, time-consuming, and the single Torreya grandis detection cannot be realized. In addition, a two-layer model method is adopted to improve the prediction accuracy of the peroxide value of a single Torreya grandis.

Description

Technical Field

[0001] The invention belongs to the field of nut quality detection and relates to a rapid non-destructive detection method for the peroxide value of Torreya grandis. Background Art

[0002] Torreya grandis is known as the "king of nuts". It is as big as a date and shaped like an olive. The kernel is yellow-white, crispy and delicious, and is deeply loved by people. Peroxide value is an important indicator of Torreya grandis rancidity and needs to be tested. At present, the commonly used method for detecting the peroxide value of Torreya grandis mainly uses petroleum ether to extract the oil of Torreya grandis, and then detects it according to the national standard method GB / T5009.37-2003. It has the disadvantages of cumbersome pre-treatment and long detection time. In addition, the oil extraction of Torreya grandis generally requires about 10g of kernels, while a single Torreya grandis kernel is only about 2g, so it is impossible to detect the peroxide value of a single Torreya grandis. In some cases, it is necessary to detect the peroxide value of a single Torreya grandis, such as the detection of old Torreya grandis seeds mixed in new Torreya grandis seeds, and the detection of Torreya grandis rancidity one by one on the assembly line. Therefore, a rapid and non-destructive detection method for the peroxide value of a single Torreya grandis is very needed to meet the above needs.

[0003] Near infrared spectroscopy is a fast, non-destructive, green optical detection method. Its principle is to quantitatively detect the content of organic components based on the frequency doubling and sum frequency absorption of hydrogen-containing groups. The spectrum of Torreya grandis is obtained by near infrared spectroscopy, and the peroxide value of Torreya grandis is determined by standard method. Then, the relationship between the spectrum of Torreya grandis and its peroxide value is established by chemometric method, which is called prediction model. Finally, the spectrum of Torreya grandis sample whose peroxide value needs to be detected is collected, and the peroxide value of Torreya grandis can be quickly and non-destructively detected by using the prediction model. At present, the standard method requires about 10g of kernels to detect the peroxide value of Torreya grandis, while a single Torreya grandis kernel is only about 2g. Therefore, the standard method can only obtain the average peroxide value of multiple Torreya grandis. Since the standard method obtains the average peroxide value of multiple Torreya grandis, when establishing the prediction model of Torreya grandis peroxide value using near infrared spectroscopy, it is also necessary to obtain the corresponding spectra of multiple Torreya grandis, and take the average spectrum to establish the prediction model. When the prediction model established by using the average spectra of multiple Torreya grandis is used to predict the peroxide value of a single Torreya grandis, its prediction accuracy will be greatly reduced, affecting the actual application of the prediction model. The present invention proposes a spectral similarity method for finding multiple Torreya grandis with similar peroxide value spectra; when the size and shape of multiple Torreya grandis are basically the same and the peroxide value spectra are similar, the multiple Torreya grandis can be regarded as one Torreya grandis; at this time, the prediction model established by using the average spectra of multiple Torreya grandis can better predict the peroxide value of a single Torreya grandis. At the same time, the present invention adopts a two-layer model method to improve the prediction accuracy of the peroxide value of a single Torreya grandis. Summary of the invention

[0004] The present invention aims at the problems that the existing method for detecting the peroxide value of Torreya grandis has complicated pre-processing, long time consumption, and cannot realize the detection of a single Torreya grandis. The present invention proposes a spectrum similarity method for finding multiple Torreya grandis with similar peroxide value spectra, and regards multiple Torreya grandis with basically the same size and shape and similar peroxide value spectra as one Torreya grandis. The average spectrum of the multiple Torreya grandis is used to establish a peroxide value prediction model, and the prediction model is used to detect the peroxide value of a single Torreya grandis, which can better solve the problem that the prediction model established by the average spectrum of multiple Torreya grandis has low prediction accuracy for the peroxide value of a single Torreya grandis. At the same time, the present invention adopts a two-layer model method to further improve the prediction accuracy of the peroxide value of a single Torreya grandis.

[0005] The technical solution adopted by the present invention to solve the technical problem is: a rapid non-destructive detection method for the peroxide value of Torreya grandis, characterized in that the method comprises the following steps:

[0006] Step 1: Construction of similarity expression of peroxide value spectrum.

[0007] Step 1.1, collect n Torreya grandis samples harvested in the same year and with basically the same size and shape, divide the n Torreya grandis samples into 10 parts, and accelerate oxidation in an oven at 100°C for 0, 6, 12, 18, 24, 30, 36, 42, 48, and 54 hours, respectively, to obtain Torreya grandis samples with peroxide value differentiation;

[0008] Step 1.2, then using near infrared spectroscopy technology to collect the transmission spectrum of the Torreya grandis sample, taking the accelerated oxidation time of the Torreya grandis sample as the Y value, and calculating the correlation between each wavelength of the Torreya grandis sample spectrum and the Y value;

[0009] Step 1.3, select the first five wavelengths with the highest correlation as the characteristic wavelengths of the peroxide value of Torreya grandis, denoted as λ1, λ2, λ3, λ4, λ5, and the correlation coefficients between the above five characteristic wavelengths and the Y value are denoted as a1, a2, a3, a4, a5 respectively;

[0010] Step 1.4, for the characteristic wavelength λ1, calculate the average spectral energy of all Torreya grandis samples at the characteristic wavelength λ1 under different accelerated oxidation times, and record them as Step 1.5: When the accelerated oxidation time is between 0 and 6 hours, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 6 and 12 hours, it is also believed that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 12 and 18, 18 and 24, 24 and 30, 30 and 36, 36 and 42, 42 and 48, and 48 and 54 hours, respectively, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time within the corresponding accelerated oxidation time range, and the unit time change of the average spectral energy is recorded as Step 1.6, construct the spectral similarity expression of the characteristic wavelength λ1 as in is the spectral energy value of a Torreya grandis sample spectrum at the characteristic wavelength λ1, a1 is the correlation coefficient calculated in step 1.3, To accelerate the oxidation time, the value is 0, 6, 12, ..., 48, is the average spectral energy of the characteristic wavelength λ1 of the Torreya grandis sample under a certain accelerated oxidation time, and the values ​​are is the unit time variation of the average spectral energy, and its values ​​are when Value is and In between, The value is 0. The value is The value is when Value is and In between, The value is 6. The value is The value is The same applies to other situations;

[0011] Step 1.7, for characteristic wavelengths λ2, λ3, λ4, λ5, calculate the average spectral energy of the corresponding wavelengths according to the method in step 1.4, and record them as According to the method in step 1.5, the corresponding unit time change of average spectral energy is calculated and recorded as According to step 1.6, the spectral similarity expressions of the corresponding characteristic wavelengths are constructed, which are

[0012] Step 1.8, construct the similarity expression of the peroxide value spectrum of Torreya grandis

[0013]

[0014] When the size and shape of Torreya grandis are basically the same, the closer the C values ​​of Torreya grandis are, the more similar their peroxide value spectra are; the C value is used to screen Torreya grandis samples with similar peroxide value spectra, so that multiple Torreya grandis can be regarded as one Torreya grandis.

[0015] Step 2: Establishment of peroxide value prediction model;

[0016] Step 2.1, collect Torreya grandis samples of different storage time, and first divide them into 10 groups according to the similarity of Torreya grandis size and shape, with k samples in each group;

[0017] Step 2.2, for the first group of k samples, collect the near-infrared transmission spectra of k Torreya grandis samples, calculate the C value of each Torreya grandis sample according to the similarity expression of the peroxide value spectrum of Torreya grandis, and match them according to the proximity of the C values. Every 8 Torreya grandis samples with similar C values ​​are taken as a new sample, and j1 new samples can be obtained, j1<(k / 8); the near-infrared spectra of 8 samples in the new sample are averaged as the spectrum of the new sample, the diameters of the 8 samples are measured and averaged as the diameter of the new sample, and then the 8 samples are crushed to take out their kernels, and the oil is extracted with petroleum ether, and then the peroxide value of the new sample is measured according to the national standard method GB / T5009.37-2003; the spectrum, diameter and peroxide value of the new sample are respectively recorded as R 11 ,R 12 ,R 13 ,…, d 11 ,d 12 ,d 13 ,…, and V 11 ,V 12 ,V 13 ,…, Step 2.3: To eliminate the influence of Torreya grandis diameter on the spectrum, the new sample spectrum R 11 ,R 12 ,R 13 ,…, Divide by the diameter d of the corresponding new sample 11 ,d 12 ,d 13 ,…, The corrected spectrum is recorded as R' 11 ,R' 12 ,R' 13 ,…, Step 2.4: For the 2nd to 10th groups of Torreya grandis samples, collect the sample spectra and calculate the C values ​​according to step 2.2, and obtain j2, j3, j4, …, j 10 new samples; calculate the spectrum of the new samples according to step 2.2, and determine the diameter and peroxide value of the new samples; the spectrum, diameter and peroxide value of the second group of new samples are recorded as R21 ,R 22 ,R 23 ,…, d 21 ,d 22 ,d 23 ,…, and V 21 ,V 22 ,V 23 ,…, The spectrum, diameter and peroxide value of the third group of new samples are denoted as R 31 ,R 32 ,R 33 ,…, d 31 ,d 32 ,d 33 ,…, and V 31 ,V 32 ,V 33 ,…, The labels of spectra, diameters, and peroxide values ​​of the 4th to 10th groups of new samples are similar;

[0018] Step 2.5: For the 2nd to 10th groups of new sample spectra, perform diameter correction according to step 2.3. The corrected spectra are recorded as R' 21 ,R' 22 ,R' 23 ,…, R' 31 ,R' 32 ,R' 33 ,…, R' 41 ,R' 42 ,R' 43 ,…, R' 51 ,R' 52 ,R' 53 ,…, R' 61 ,R' 62 ,R' 63 ,…, R' 71 ,R' 72 ,R' 73 ,…, R' 81 ,R' 82 ,R' 83 ,…, R' 91 ,R' 92 ,R' 93 ,…, R' 101 ,R'102 ,R' 103 ,…,

[0019] Step 2.6, in order to establish a high-precision prediction model for the peroxide value of Torreya grandis, three methods are used to screen the characteristic wavelengths, and three corresponding prediction models are established as the first-layer prediction model; then, the prediction results of the three prediction models for new samples are used as the input of the artificial neural network, and the true value of the peroxide value of the new sample is used as the output to establish an artificial neural network model as the second-layer prediction model; the prediction accuracy of the peroxide value of Torreya grandis is improved by the two-layer prediction model;

[0020] Step 2.7, for the above-corrected spectrum, extract the spectral energy values ​​at the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5 of the peroxide value respectively, and use the multivariate linear regression method to associate the spectral energy values ​​of the five characteristic wavelengths of the new Torreya grandis sample with the corresponding peroxide value, and establish a linear regression model, which is recorded as Model_F1;

[0021] Step 2.8, using a genetic algorithm to screen the characteristic wavelengths of the above-corrected spectrum, obtain f1 characteristic wavelengths, extract the spectral energy values ​​at the f1 characteristic wavelengths respectively, use the PLS method to associate the spectral energy values ​​of the new Torreya grandis sample at the f1 characteristic wavelengths with the corresponding peroxide value, and establish a PLS prediction model, recorded as Model_F2;

[0022] Step 2.9, using the competitive adaptive reweighted sampling method to screen the characteristic wavelengths of the above-corrected spectrum, obtain f2 characteristic wavelengths, extract the spectral energy values ​​at the f2 characteristic wavelengths respectively, and use the LSSVM method to associate the spectral energy values ​​of the new Torreya grandis sample at the f2 characteristic wavelengths with the corresponding peroxide value, and establish an LSSVM prediction model, recorded as Model_F3; Model_F1, Model_F2 and Model_F3 are used as the first-level models for predicting the peroxide value of Torreya grandis;

[0023] Step 2.10, using Model_F1, Model_F2 and Model_F3 to predict the corrected spectra of the new samples respectively, and obtaining the peroxide value prediction results of the three groups of new samples; taking the three groups of peroxide value prediction results as the input of the artificial neural network, and the real peroxide value of the new samples as the output, an artificial neural network model is established, denoted as Model_S, which is used as the second-layer model of the peroxide value of Torreya grandis; the two-layer model can effectively improve the prediction accuracy of the peroxide value of Torreya grandis;

[0024] Step 3: prediction of peroxide value of a single Torreya grandis;

[0025] Step 3.1, obtain a Torreya grandis sample to be tested, collect the near infrared spectrum of the Torreya grandis sample, and measure the diameter of the Torreya grandis sample, which are recorded as R p and d p ;

[0026] Step 3.2: Perform diameter correction according to step 2.3. The corrected spectrum is recorded as R' p ;

[0027] Step 3.3, for R' p , extract the spectral energy values ​​under the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5, substitute them into Model_F1, and obtain the peroxide value prediction result V p1 ;

[0028] Step 3.4, for R' p , extract the spectral energy value of the f1 characteristic wavelengths screened by the genetic algorithm, substitute it into Model_F2, and obtain the peroxide value prediction result V p2 ;

[0029] Step 3.5, for R' p , extract the spectral energy values ​​of f2 characteristic wavelengths screened by the competitive adaptive reweighted sampling method, substitute them into Model_F3, and obtain the peroxide value prediction result V p3 ;

[0030] Step 3.6: The three predicted results of the peroxide value of Torreya grandis samples V p1 、V p2 and V p3 Input into Model_S to obtain the final peroxide value prediction result of Torreya grandis sample, recorded as V p , thereby achieving rapid, non-destructive and high-precision detection of the peroxide value of a single Torreya grandis.

[0031] The beneficial effects of the present invention are as follows: a method for detecting the peroxide value of Torreya grandis of the present invention can quickly and non-destructively detect the peroxide value of a single Torreya grandis, which can solve the problems of the existing method for detecting the peroxide value of Torreya grandis, such as complicated pre-processing, long time consumption, and inability to detect the peroxide value of a single Torreya grandis; in addition, the prediction accuracy of the peroxide value of Torreya grandis can be effectively improved through the two-layer model. DETAILED DESCRIPTION

[0032] The present invention is further described below by means of specific examples.

[0033] Embodiment: A rapid nondestructive detection method for the peroxide value of Torreya grandis, characterized by comprising the following steps:

[0034] Step 1: Construction of similar expression of peroxide value spectrum;

[0035] Step 1.1, collect n Torreya grandis samples harvested in the same year and with basically the same size and shape, divide the n Torreya grandis samples into 10 parts, and accelerate oxidation in an oven at 100°C for 0, 6, 12, 18, 24, 30, 36, 42, 48, and 54 hours, respectively, to obtain Torreya grandis samples with peroxide value differentiation;

[0036] Step 1.2, then using near infrared spectroscopy technology to collect the transmission spectrum of the Torreya grandis sample, taking the accelerated oxidation time of the Torreya grandis sample as the Y value, and calculating the correlation between each wavelength of the Torreya grandis sample spectrum and the Y value;

[0037] Step 1.3, select the first five wavelengths with the highest correlation as the characteristic wavelengths of the peroxide value of Torreya grandis, denoted as λ1, λ2, λ3, λ4, λ5, and the correlation coefficients between the above five characteristic wavelengths and the Y value are denoted as a1, a2, a3, a4, a5 respectively;

[0038] Step 1.4, for the characteristic wavelength λ1, calculate the average spectral energy of all Torreya grandis samples at the characteristic wavelength λ1 under different accelerated oxidation times, and record them as Step 1.5: When the accelerated oxidation time is between 0 and 6 hours, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 6 and 12 hours, it is also believed that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 12 and 18, 18 and 24, 24 and 30, 30 and 36, 36 and 42, 42 and 48, and 48 and 54 hours, respectively, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time within the corresponding accelerated oxidation time range, and the unit time change of the average spectral energy is recorded as Step 1.6, construct the spectral similarity expression of the characteristic wavelength λ1 as in is the spectral energy value of a Torreya grandis sample spectrum at the characteristic wavelength λ1, a1 is the correlation coefficient calculated in step 1.3, To accelerate the oxidation time, the value is 0, 6, 12, ..., 48, is the average spectral energy of the characteristic wavelength λ1 of the Torreya grandis sample under a certain accelerated oxidation time, and the values ​​are is the unit time variation of the average spectral energy, and its values ​​are when Value is and In between, The value is 0. The value is The value is when Value is and In between, The value is 6. The value is The value is The same applies to other situations;

[0039] Step 1.7, for characteristic wavelengths λ2, λ3, λ4, λ5, calculate the average spectral energy of the corresponding wavelengths according to the method in step 1.4, and record them as According to the method in step 1.5, the corresponding unit time change of average spectral energy is calculated and recorded as According to step 1.6, the spectral similarity expressions of the corresponding characteristic wavelengths are constructed, which are

[0040] Step 1.8, construct the similarity expression of the peroxide value spectrum of Torreya grandis

[0041]

[0042] When the size and shape of Torreya grandis are basically the same, the closer the C values ​​of Torreya grandis are, the more similar their peroxide value spectra are; the C value is used to screen Torreya grandis samples with similar peroxide value spectra, so that multiple Torreya grandis can be regarded as one Torreya grandis.

[0043] Step 2: Establishment of peroxide value prediction model;

[0044] Step 2.1, collect Torreya grandis samples of different storage time, and first divide them into 10 groups according to the similarity of Torreya grandis size and shape, with k samples in each group;

[0045] Step 2.2, for the first group of k samples, collect the near-infrared transmission spectra of k Torreya grandis samples, calculate the C value of each Torreya grandis sample according to the similarity expression of the peroxide value spectrum of Torreya grandis, and match them according to the proximity of the C values. Every 8 Torreya grandis samples with similar C values ​​are taken as a new sample, and j1 new samples can be obtained, j1<(k / 8); the near-infrared spectra of 8 samples in the new sample are averaged as the spectrum of the new sample, the diameters of the 8 samples are measured and averaged as the diameter of the new sample, and then the 8 samples are crushed to take out their kernels, and the oil is extracted with petroleum ether, and then the peroxide value of the new sample is measured according to the national standard method GB / T5009.37-2003; the spectrum, diameter and peroxide value of the new sample are respectively recorded as R 11 ,R 12 ,R 13 ,…, d 11 ,d 12 ,d 13 ,…, and V 11 ,V 12 ,V 13 ,…, Step 2.3: To eliminate the influence of Torreya grandis diameter on the spectrum, the new sample spectrum R 11 ,R 12 ,R 13 ,…, Divide by the diameter d of the corresponding new sample 11 ,d 12 ,d 13 ,…, The corrected spectrum is recorded as R' 11 ,R' 12 ,R' 13 ,…, Step 2.4: For the 2nd to 10th groups of Torreya grandis samples, collect the sample spectra and calculate the C values ​​according to step 2.2, and obtain j2, j3, j4, …, j 10 new samples; calculate the spectrum of the new samples according to step 2.2, and determine the diameter and peroxide value of the new samples; the spectrum, diameter and peroxide value of the second group of new samples are recorded as R 21 ,R 22 ,R 23 ,…, d 21 ,d 22 ,d 23 ,…, and V 21 ,V 22 ,V 23 ,…, The spectrum, diameter and peroxide value of the third group of new samples are denoted as R 31 ,R32 ,R 33 ,…, d 31 ,d 32 ,d 33 ,…, and V 31 ,V 32 ,V 33 ,…, The labels of spectra, diameters, and peroxide values ​​of the 4th to 10th groups of new samples are similar;

[0046] Step 2.5: For the 2nd to 10th groups of new sample spectra, perform diameter correction according to step 2.3. The corrected spectra are recorded as R' 21 ,R' 22 ,R' 23 ,…, R' 31 ,R' 32 ,R' 33 ,…, R' 41 ,R' 42 ,R' 43 ,…, R' 51 ,R' 52 ,R' 53 ,…, R' 61 ,R' 62 ,R' 63 ,…, R' 71 ,R' 72 ,R' 73 ,…, R' 81 ,R' 82 ,R' 83 ,…, R' 91 ,R' 92 ,R' 93 ,…, R' 101 ,R' 102 ,R' 103 ,…,

[0047] Step 2.6, in order to establish a high-precision prediction model for the peroxide value of Torreya grandis, three methods are used to screen the characteristic wavelengths, and three corresponding prediction models are established as the first-layer prediction model; then, the prediction results of the three prediction models for new samples are used as the input of the artificial neural network, and the true value of the peroxide value of the new sample is used as the output to establish an artificial neural network model as the second-layer prediction model; the prediction accuracy of the peroxide value of Torreya grandis is improved by the two-layer prediction model;

[0048] Step 2.7, for the above-corrected spectrum, extract the spectral energy values ​​at the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5 of the peroxide value respectively, and use the multivariate linear regression method to associate the spectral energy values ​​of the five characteristic wavelengths of the new Torreya grandis sample with the corresponding peroxide value, and establish a linear regression model, which is recorded as Model_F1;

[0049] Step 2.8, using a genetic algorithm to screen the characteristic wavelengths of the above-corrected spectrum, obtain f1 characteristic wavelengths, extract the spectral energy values ​​at the f1 characteristic wavelengths respectively, use the PLS method to associate the spectral energy values ​​of the new Torreya grandis sample at the f1 characteristic wavelengths with the corresponding peroxide value, and establish a PLS prediction model, recorded as Model_F2;

[0050] Step 2.9, using the competitive adaptive reweighted sampling method to screen the characteristic wavelengths of the above-corrected spectrum, obtain f2 characteristic wavelengths, extract the spectral energy values ​​at the f2 characteristic wavelengths respectively, and use the LSSVM method to associate the spectral energy values ​​of the new Torreya grandis sample at the f2 characteristic wavelengths with the corresponding peroxide value, and establish an LSSVM prediction model, recorded as Model_F3; Model_F1, Model_F2 and Model_F3 are used as the first-level models for predicting the peroxide value of Torreya grandis;

[0051] Step 2.10, using Model_F1, Model_F2 and Model_F3 to predict the corrected spectra of the new samples respectively, and obtaining the peroxide value prediction results of the three groups of new samples; taking the three groups of peroxide value prediction results as the input of the artificial neural network, and the real peroxide value of the new samples as the output, an artificial neural network model is established, denoted as Model_S, which is used as the second-layer model of the peroxide value of Torreya grandis; the two-layer model can effectively improve the prediction accuracy of the peroxide value of Torreya grandis;

[0052] Step 3: prediction of peroxide value of a single Torreya grandis;

[0053] Step 3.1, obtain a Torreya grandis sample to be tested, collect the near infrared spectrum of the Torreya grandis sample, and measure the diameter of the Torreya grandis sample, which are recorded as R p and d p ;

[0054] Step 3.2: Perform diameter correction according to step 2.3. The corrected spectrum is recorded as R' p ;

[0055] Step 3.3, for R' p , extract the spectral energy values ​​under the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5, substitute them into Model_F1, and obtain the peroxide value prediction result V p1 ;

[0056] Step 3.4, for R' p , extract the spectral energy value of the f1 characteristic wavelengths screened by the genetic algorithm, substitute it into Model_F2, and obtain the peroxide value prediction result V p2 ;

[0057] Step 3.5, for R' p , extract the spectral energy values ​​of f2 characteristic wavelengths screened by the competitive adaptive reweighted sampling method, substitute them into Model_F3, and obtain the peroxide value prediction result V p3 ;

[0058] Step 3.6: The three predicted results of the peroxide value of Torreya grandis samples V p1 、V p2 and V p3 Input into Model_S to obtain the final peroxide value prediction result of Torreya grandis sample, recorded as V p , thereby achieving rapid, non-destructive and high-precision detection of the peroxide value of a single Torreya grandis.

Claims

1. A rapid non-destructive detection method for the peroxide value of Torreya grandis, characterized in that The steps of this method are as follows: Step 1: Construction of similar expression of peroxide value spectrum; Step 1.1, collect n Torreya grandis samples harvested in the same year and with basically the same size and shape, divide the n Torreya grandis samples into 10 parts, and accelerate oxidation in an oven at 100°C for 0, 6, 12, 18, 24, 30, 36, 42, 48, and 54 hours, respectively, to obtain Torreya grandis samples with peroxide value differentiation; Step 1.2, then using near infrared spectroscopy technology to collect the transmission spectrum of the Torreya grandis sample, taking the accelerated oxidation time of the Torreya grandis sample as the Y value, and calculating the correlation between each wavelength of the Torreya grandis sample spectrum and the Y value; Step 1.3, select the first five wavelengths with the highest correlation as the characteristic wavelengths of the peroxide value of Torreya grandis, denoted as λ1, λ2, λ3, λ4, λ5, and the correlation coefficients between the above five characteristic wavelengths and the Y value are denoted as a1, a2, a3, a4, a5 respectively; Step 1.4, for the characteristic wavelength λ1, calculate the average spectral energy of all Torreya grandis samples at the characteristic wavelength λ1 under different accelerated oxidation times, and record them as Step 1.5: When the accelerated oxidation time is between 0 and 6 hours, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 6 and 12 hours, it is also believed that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time, and its unit time change is Recorded as When the accelerated oxidation time is between 12 and 18, 18 and 24, 24 and 30, 30 and 36, 36 and 42, 42 and 48, and 48 and 54 hours, respectively, it is considered that the average spectral energy of the Torreya grandis sample at the characteristic wavelength λ1 changes linearly with the accelerated oxidation time within the corresponding accelerated oxidation time range, and the unit time change of the average spectral energy is recorded as Step 1.6, construct the spectral similarity expression of the characteristic wavelength λ1 as in is the spectral energy value of a Torreya grandis sample spectrum at the characteristic wavelength λ1, a1 is the correlation coefficient calculated in step 1.3, To accelerate the oxidation time, the value is 0, 6, 12, ..., 48, is the average spectral energy of the characteristic wavelength λ1 of the Torreya grandis sample under a certain accelerated oxidation time, and the values ​​are is the unit time variation of the average spectral energy, and its values ​​are when Value is and In between, The value is 0. The value is The value is when Value is and In between, The value is 6. The value is The value is The same applies to other situations; Step 1.7, for characteristic wavelengths λ2, λ3, λ4, λ5, calculate the average spectral energy of the corresponding wavelengths according to the method in step 1.4, and record them as According to the method in step 1.5, the corresponding unit time change of average spectral energy is calculated and recorded as According to step 1.6, the spectral similarity expressions of the corresponding characteristic wavelengths are constructed, which are Step 1.8, construct the similarity expression of the peroxide value spectrum of Torreya grandis When the size and shape of Torreya grandis are basically the same, the closer the C values ​​of Torreya grandis are, the more similar their peroxide value spectra are; The C value is used to screen Torreya grandis samples with similar peroxide value spectra, so that multiple Torreya grandis can be regarded as one Torreya grandis. Step 2: Establishment of peroxide value prediction model; Step 2.1, collect Torreya grandis samples of different storage time, and first divide them into 10 groups according to the similarity of Torreya grandis size and shape, with k samples in each group; Step 2.2, for the first group of k samples, collect the near-infrared transmission spectra of k Torreya grandis samples, calculate the C value of each Torreya grandis sample according to the similarity expression of the peroxide value spectrum of Torreya grandis, and match them according to the proximity of the C values. Every 8 Torreya grandis samples with similar C values ​​are taken as a new sample, and j1 new samples can be obtained, j1<(k / 8); the near-infrared spectra of 8 samples in the new sample are averaged as the spectrum of the new sample, the diameters of the 8 samples are measured and averaged as the diameter of the new sample, and then the 8 samples are crushed to take out their kernels, and the oil is extracted with petroleum ether, and the peroxide value of the new sample is measured according to the national standard method GB / T5009.37-2003; the spectrum, diameter and peroxide value of the new sample are respectively recorded as and Step 2.3: To eliminate the influence of Torreya grandis diameter on the spectrum, the new sample spectrum of group 1 was Divide by the diameter of the corresponding new sample The corrected spectrum is recorded as Step 2.4: For the 2nd to 10th groups of Torreya grandis samples, collect the sample spectra and calculate the C values ​​according to step 2.2, and obtain j2, j3, j4, …, j 10 new samples; calculate the spectrum of the new samples according to step 2.2, and determine the diameter and peroxide value of the new samples; the spectrum, diameter and peroxide value of the second group of new samples are recorded as and The spectra, diameters and peroxide values ​​of the third group of new samples are recorded as and The marking of spectra, diameters, and peroxide values ​​of the 4th to 10th groups of new samples is similar; Step 2.5: For the 2nd to 10th groups of new sample spectra, diameter correction is performed according to step 2.

3. The corrected spectra are recorded as Step 2.6, in order to establish a high-precision prediction model for the peroxide value of Torreya grandis, three methods are used to screen the characteristic wavelengths, and three corresponding prediction models are established as the first-layer prediction model; then, the prediction results of the three prediction models for new samples are used as the input of the artificial neural network, and the true value of the peroxide value of the new sample is used as the output to establish an artificial neural network model as the second-layer prediction model; the prediction accuracy of the peroxide value of Torreya grandis is improved by the two-layer prediction model; Step 2.7, for the above-corrected spectrum, extract the spectral energy values ​​at the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5 of the peroxide value respectively, and use the multivariate linear regression method to associate the spectral energy values ​​of the five characteristic wavelengths of the new Torreya grandis sample with the corresponding peroxide value, and establish a linear regression model, which is recorded as Model_F1; Step 2.8, using a genetic algorithm to screen the characteristic wavelengths of the above-corrected spectrum, obtain f1 characteristic wavelengths, extract the spectral energy values ​​at the f1 characteristic wavelengths respectively, use the PLS method to associate the spectral energy values ​​of the new Torreya grandis sample at the f1 characteristic wavelengths with the corresponding peroxide value, and establish a PLS prediction model, recorded as Model_F2; Step 2.9, using the competitive adaptive reweighted sampling method to screen the characteristic wavelengths of the above-corrected spectrum, obtain f2 characteristic wavelengths, extract the spectral energy values ​​at the f2 characteristic wavelengths respectively, and use the LSSVM method to associate the spectral energy values ​​of the new Torreya grandis sample at the f2 characteristic wavelengths with the corresponding peroxide value, and establish an LSSVM prediction model, recorded as Model_F3; Model_F1, Model_F2 and Model_F3 are used as the first-level models for predicting the peroxide value of Torreya grandis; Step 2.10, using Model_F1, Model_F2 and Model_F3 to predict the corrected spectra of the new samples respectively, and obtaining the peroxide value prediction results of the three groups of new samples; taking the three groups of peroxide value prediction results as the input of the artificial neural network, and the real peroxide value of the new samples as the output, an artificial neural network model is established, denoted as Model_S, which is used as the second-layer model of the peroxide value of Torreya grandis; the two-layer model can effectively improve the prediction accuracy of the peroxide value of Torreya grandis; Step 3: prediction of peroxide value of a single Torreya grandis; Step 3.1, obtain a Torreya grandis sample to be tested, collect the near infrared spectrum of the Torreya grandis sample, and measure the diameter of the Torreya grandis sample, which are recorded as R p and d p ; Step 3.2: Perform diameter correction according to step 2.

3. The corrected spectrum is recorded as R' p ; Step 3.3, for R' p , extract the spectral energy values ​​under the characteristic wavelengths λ1, λ2, λ3, λ4, and λ5, substitute them into Model_F1, and obtain the peroxide value prediction result V p1 ; Step 3.4, for R' p , extract the spectral energy value of the f1 characteristic wavelengths screened by the genetic algorithm, substitute it into Model_F2, and obtain the peroxide value prediction result V p2 ; Step 3.5, for R' p , extract the spectral energy values ​​of f2 characteristic wavelengths screened by the competitive adaptive reweighted sampling method, substitute them into Model_F3, and obtain the peroxide value prediction result V p3 ; Step 3.6: The three predicted results of the peroxide value of Torreya grandis samples V p1 、V p2 and V p3 Input into Model_S to obtain the final peroxide value prediction result of Torreya grandis sample, recorded as V p , thereby achieving rapid, non-destructive and high-precision detection of the peroxide value of a single Torreya grandis.

Images