Near-infrared qualitative and quantitative detection method for adulteration of cooked donkey meat

Abstract

The invention relates to a near-infrared qualitative and quantitative detection method for adulteration of cooked donkey meat, which comprises the following steps: carrying out near-infrared spectrum scanning on a cooked minced donkey meat sample, a cooked minced horse meat sample, a cooked minced mule meat sample, donkey-horse adulterated cooked minced meat samples with different adulteration proportions and donkey-mule adulterated cooked minced meat samples with different adulteration proportions; and establishing a qualitative classification model and a quantitative detection model according to the obtained near infrared diffuse reflection spectrum. Detecting a to-be-detected sample by using the established detection model, detecting the category of cooked meat by using the qualitative classification model, and if the cooked meat belongs to donkey and mule adulteration or donkey and horse adulteration, performing quantitative detection by using the quantitative detection model so as to determine the adulteration proportion. The method is simple, rapid, efficient and accurate in operation and low in cost, does not damage a sample in the scanning process, does not consume chemical reagents, does not cause environmental pollution, can quantitatively detect the adulteration proportion, and is suitable for adulteration identification of cooked minced donkey meat.

Description

Technical Field

[0001] The invention relates to a meat adulteration detection technology, in particular to a near infrared qualitative and quantitative detection method for cooked donkey meat adulteration. Background Art

[0002] Donkey meat has high nutritional value and has the advantages of "two highs and two lows", high protein, low fat, high amino acids, and low cholesterol. It has a good health care effect on arteriosclerosis, coronary heart disease, and hypertension. Donkey meat and donkey meat products have excellent flavor and are widely favored by everyone. However, donkey meat production efficiency is low and supply and demand are mismatched. Since horse meat and mule meat have similar appearance characteristics to donkey meat and are cheaper than donkey meat, some unscrupulous vendors use horse meat, mule meat, etc. to adulterate cooked donkey meat products such as donkey meat and sell fake products. This problem has emerged in an endless stream, seriously damaging the rights and interests of consumers and even threatening their health, which is not conducive to market fairness. Therefore, the problem of adulteration of cooked donkey meat should be solved as soon as possible.

[0003] In order to solve the problem of meat adulteration, the commonly used detection methods now include polymerase chain reaction (PCR) technology, electronic nose detection technology, liquid chromatography tandem mass spectrometry and other methods.

[0004] Among them, polymerase chain reaction (PCR) technology is a molecular biology technology used to amplify specific DNA (or RNA) fragments. It is fast, simple and highly specific, but it has the disadvantages of high cost, complex operation, difficulty in obtaining results and damage to meat samples. Electronic nose detection technology is an intelligent bionic device that uses gas sensors to establish response curves based on the volatile components of the target to quickly identify the odor of donkey meat samples. It is easy to operate and can achieve non-destructive rapid detection, but it is expensive and has low detection efficiency. Liquid chromatography tandem mass spectrometry uses the separation ability of chromatography and the qualitative function of mass spectrometry to achieve qualitative and quantitative analysis of donkey meat adulteration samples. It has fast detection and high sensitivity, but it has a complex structure and high maintenance costs.

[0005] Existing detection technologies have the problems of destroying meat samples and being costly, and existing detection technologies are unable to effectively and quantitatively detect the proportion of adulteration. Summary of the invention

[0006] The purpose of the present invention is to provide a near-infrared qualitative and quantitative detection method for adulteration of cooked donkey meat, so as to solve the problems that the existing donkey meat adulteration detection technology destroys meat samples, has high cost, and cannot effectively and quantitatively detect the proportion of adulteration.

[0007] The present invention is achieved in this way: a near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat comprises the following steps.

[0008] S1. Establish a detection model.

[0009] S11. Sampling and pre-processing the samples to obtain cooked minced donkey meat samples, cooked minced horse meat samples, cooked minced mule meat samples, cooked minced donkey and horse meat samples with different adulteration ratios, and cooked minced donkey and mule meat samples with different adulteration ratios.

[0010] S12, performing near infrared spectrum scanning on each sample to obtain a near infrared spectrum graph of each sample.

[0011] S13, dividing the near infrared spectra of each sample into a calibration set and a prediction set.

[0012] S14. Perform spectral preprocessing on the near-infrared spectra of each sample to eliminate interference factors.

[0013] S15. Establish a qualitative classification model.

[0014] S16. Establish and optimize the quantitative detection model.

[0015] S2. Take the sample to be tested and perform pre-treatment.

[0016] S3, performing near infrared spectroscopy scanning on the sample to be tested to obtain a near infrared diffuse reflectance spectrum of the sample to be tested.

[0017] S4. Substitute the near-infrared diffuse reflectance spectrum of the sample to be tested into the qualitative classification model to distinguish the category of the sample to be tested.

[0018] S5. Substituting the near-infrared diffuse reflectance spectrum of the sample to be tested into the quantitative detection model to detect the content of donkey meat in the sample to be tested.

[0019] Among them, in step S11, the meat block is cooled to room temperature and then chopped into small pieces of 3 to 5 cubic millimeters as cooked minced meat samples, and cooked minced meat samples of donkey meat, horse meat and mule meat are used to prepare donkey meat content of 0% to 100% and donkey and horse adulterated cooked minced meat samples and donkey and mule adulterated cooked minced meat samples according to a certain gradient, and the number of samples for each gradient is no less than 10.

[0020] In step S11, donkey and horse adulterated cooked minced meat samples and donkey and mule adulterated cooked minced meat samples with donkey meat content ranging from 0% to 100% are prepared with a gradient of 10%.

[0021] In step S13, before dividing the calibration set and the prediction set, abnormal samples are eliminated. Abnormal samples include samples with large errors in spectral data.

[0022] In step S14, the preprocessing includes 5-point smoothing, 15-point smoothing, 25-point smoothing, smoothed first-order derivative, first-order and second-order differentials, multivariate scatter correction, standard normal variable transformation, normalization, baseline correction, and detrending algorithm.

[0023] In step S15, three supervised pattern recognition algorithms, namely linear discriminant analysis, support vector machine and generalized regression neural network model, are used to establish a qualitative classification model for the cooked minced meat sample.

[0024] In step S16, the first 20 principal components of the near-infrared spectrum are selected for dimensionality reduction, and a partial least squares regression quantitative detection model is established in combination with cross-validation; the first 10 and first 15 latent variable scores are selected for partial least squares dimensionality reduction, respectively, and the latent variables extracted after partial least squares dimensionality reduction are used as input variables for cooked minced meat to establish a back propagation neural network quantitative detection model, compare the effects of different preprocessing methods on the modeling results, and select the preprocessing method with the best detection results.

[0025] Near infrared spectroscopy technology has the advantages of being non-destructive, environmentally friendly, efficient, and easy to operate. Its essence is to reflect that there is a certain correspondence between the spectral data information of the hydrogen-containing groups of organic molecules in the sample and the composition content or characteristic data measured by standard detection methods or recognized methods. Based on this relationship, a mathematical model is established using chemometric methods, and then the established model is used to achieve quantitative or qualitative analysis of unknown samples. Near-infrared light refers to electromagnetic waves between visible light and mid-infrared light. The spectral wavelength is in the range of 780~2526 nm, which is divided into two regions: near-infrared short wave (780~1100 nm) and near-infrared long wave (1100~2526 nm). It is mainly due to the non-resonance of molecular vibration, which makes the molecule transition from the ground state to the high energy level. It is the frequency doubling and combination spectrum band of the molecular vibration spectrum. It mainly refers to the absorption of hydrogen-containing groups XH (X=C, N, O), which contains rich information on the composition and molecular structure of most types of organic matter. Because different organic matter contains different groups, different groups have different energy levels, different groups and the same group have obvious differences in the absorption wavelength of near-infrared light in different physical and chemical environments, and the absorption coefficient is small, and the heat is less. The inorganic water or other components contained in the donkey meat sample contain OH bonds, fat contains CH bonds, and protein contains NH bonds, so they can all be detected and analyzed by near-infrared spectroscopy.

[0026] There are two main technologies used to obtain near-infrared spectra: transmission spectroscopy and reflection spectroscopy. Diffuse reflection is the reflection of light on the surface or inside of an object (usually powder or other granular objects) with uncertain directions after it is projected onto the object. The analysis using diffuse reflected light is called diffuse reflectance spectroscopy. The sample tested in the present invention is meat, so the technology used is diffuse reflectance spectroscopy.

[0027] The present invention adopts near-infrared spectroscopy analysis technology, and only detects by scanning the near-infrared spectrum of the sample, does not destroy the sample to be tested, can non-destructively and greenly realize the identification of donkey meat adulteration, performs near-infrared spectroscopy scanning on donkey meat cooked minced meat samples, horse meat cooked minced meat samples, mule meat cooked minced meat samples, donkey and horse adulterated cooked minced meat samples with different adulteration ratios, and donkey and mule adulterated cooked minced meat samples with different adulteration ratios, and establishes a qualitative classification model and a quantitative detection model according to the obtained near-infrared diffuse reflectance spectrum. After the detection model is established, the sample to be tested can be detected, and the qualitative classification model is first used to detect the category of cooked meat. If it belongs to donkey and mule adulteration or donkey and horse adulteration, the quantitative detection model is used to perform quantitative detection to determine the adulteration ratio.

[0028] The present invention can identify whether donkey meat is adulterated in a non-destructive and green manner, and has the advantages of simple, rapid, efficient, accurate and low cost operation. The sample is not destroyed during the scanning process, chemical reagents are not consumed, and environmental pollution is not caused. At the same time, the adulteration ratio can be quantitatively detected, and the method is suitable for identifying adulteration of cooked minced donkey meat. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a flow chart of the present invention.

[0030] Figure 2 It is the average near-infrared spectrum of cooked minced meat sample.

[0031] Figure 3 It is the BP quantitative prediction result of donkey meat content in donkey mule adulteration in the cooked minced meat sample set.

[0032] Figure 4 It is the BP quantitative prediction result of donkey meat content in donkey and horse adulteration in the cooked minced meat sample set.

[0033] Figure 5 It is the original near infrared spectrum of the sample to be tested in the embodiment.

[0034] Figure 6 It is a near infrared spectrum of the sample to be tested after SNV pretreatment in the embodiment.

[0035] Figure 7 It is the near infrared spectrum of the sample to be tested after Detrend pretreatment in the embodiment.

[0036] Figure 8 It is the near infrared spectrum of the sample to be tested after normalization pretreatment in the embodiment.

[0037] Fig. 9 It is the BP quantitative prediction result of the donkey and mule adulterated sample to be tested in the embodiment.

[0038] Fig.10 It is the BP quantitative prediction result of the donkey and horse adulterated sample to be tested in the embodiment. DETAILED DESCRIPTION

[0039] like Figure 1 As shown, the near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat of the present invention comprises the following steps.

[0040] S1. Establish a detection model.

[0041] S2. Take the sample to be tested and perform pre-treatment.

[0042] S3, performing near infrared spectroscopy scanning on the sample to be tested to obtain a near infrared diffuse reflectance spectrum of the sample to be tested.

[0043] S4. Substitute the near-infrared diffuse reflectance spectrum of the sample to be tested into the qualitative classification model to distinguish the category of the sample to be tested.

[0044] S5. Substituting the near-infrared diffuse reflectance spectrum of the sample to be tested into the quantitative detection model to detect the content of donkey meat in the sample to be tested.

[0045] Among them, step S1 includes establishing a qualitative classification model and a quantitative classification model, and the specific steps are as follows.

[0046] S11. Sampling and pre-processing the samples to obtain cooked minced donkey meat samples, cooked minced horse meat samples, cooked minced mule meat samples, cooked minced donkey and horse meat samples with different adulteration ratios, and cooked minced donkey and mule meat samples with different adulteration ratios.

[0047] S12, performing near infrared spectrum scanning on each sample to obtain a near infrared spectrum graph of each sample.

[0048] S13, dividing the near infrared spectra of each sample into a calibration set and a prediction set.

[0049] S14. Perform spectral preprocessing on the near-infrared spectra of each sample to eliminate interference factors.

[0050] S15. Establish a qualitative classification model.

[0051] S16. Establish and optimize the quantitative detection model.

[0052] When establishing the detection model, it is necessary to prepare samples first, and then pre-treat the samples after sampling. The pre-treatment process is to cool the meat to room temperature, remove fascia and fat, and chop the meat into small pieces of 3 to 5 cubic millimeters as cooked meat samples. Use cooked meat samples of donkey meat, horse meat, and mule meat to prepare donkey and horse adulterated cooked meat samples and donkey and mule adulterated cooked meat samples with donkey meat content of 0% to 100% according to a certain gradient, and there are no less than 10 samples in each gradient.

[0053] Specifically, donkey and horse adulterated cooked minced meat samples and donkey and mule adulterated cooked minced meat samples with donkey meat content ranging from 0% to 100% can be prepared with a gradient of 10%. In order to enhance the applicability of the model, the sample can also be prepared with a lower gradient.

[0054] In the present invention, 1035 cooked minced meat samples (179 donkey meat, 147 mule meat, 172 horse meat, 267 donkey and mule adulterated meat, and 270 donkey and horse adulterated meat) were prepared according to the above method and principle.

[0055] For the obtained cooked minced donkey meat samples, cooked minced horse meat samples, cooked minced mule meat samples, cooked minced donkey and horse meat samples with different adulteration ratios, and cooked minced donkey and mule meat samples with different adulteration ratios, 50 g of the sample was accurately weighed and placed in a cylindrical quartz dish with a diameter of 5 cm and compacted in preparation for the collection of near-infrared spectra.

[0056] Before scanning the near-infrared spectrum, turn on the spectrometer and preheat it for 30 minutes to check the instrument. Set the spectrometer parameters as follows: band 4000~12500 cm -1 , the background is the gold background of the instrument, the resolution is 2 cm -1 , 64 scans. The spectrum was collected at 3 points and the average value was taken as the sample spectrum. The laboratory temperature was kept at room temperature during the spectrum collection, and the background spectrum was measured every 30 minutes to reduce the error caused by room temperature changes. After the spectrum was collected, the absorbance corresponding to each sample spectrum point was saved.

[0057] like Figure 2 Shown are the average spectra of donkey meat, horse meat, mule meat, donkey and horse adulteration, and donkey and mule adulteration cooked minced meat samples. The absorption peaks and contours of the adulterated sample spectra are very similar to those of pure donkey meat samples and overlap with each other, so modeling analysis is needed.

[0058] In step S13, before dividing the calibration set and the prediction set, abnormal samples are eliminated. Abnormal samples include samples with large errors in spectral data.

[0059] The error of spectral data is mainly due to the error of the spectrometer itself, improper sample pretreatment, the influence of ambient temperature and humidity, etc. Therefore, before building the model, it is necessary to first calculate the Mahalanobis distance of the sample set and sort the size, and use the Chauville criterion to remove abnormal samples to improve the prediction accuracy of the model. After removing the abnormal samples, the sample set is divided into the calibration set and the prediction set in a ratio of 2:1, 3:1 or 4:1.

[0060] The present invention divides the sample set into a ratio of approximately 3:1 to obtain a calibration set and a prediction set. The specific division results are shown in Table 1.

[0061] Table 1 Calibration set and prediction set of cooked minced meat adulteration samples

[0062] In step S14, the preprocessing includes 5-point smoothing, 15-point smoothing, 25-point smoothing, smoothed first-order derivative, first-order and second-order differentials, multivariate scatter correction, standard normal variable transformation, normalization, baseline correction, and detrending algorithm.

[0063] Near infrared spectroscopy analysis utilizes the chemical information of the sample, while physical information is often an interference factor that is not conducive to modeling and should be eliminated or reduced, so spectral preprocessing is required. The preprocessing used in the present invention includes 11 methods: smoothing (5 points, 15 points and 25 points), smooth first-order derivative, first-order and second-order differentials, multivariate scatter correction (MSC), standard normal variable transformation (SNV), normalization, baseline correction and detrend algorithm.

[0064] Smoothing algorithm: It is the most commonly used method to eliminate noise, and is particularly effective in filtering out high-frequency noise. The smoothing algorithms used in this invention include 5-point, 15-point, 25-point smoothing processing and smoothing first-order derivative algorithm.

[0065] First-order and second-order differentials: Derivative differentials are commonly used preprocessing methods in near-infrared spectroscopy analysis, which can effectively eliminate baseline and background interference and improve resolution and sensitivity. However, it will also amplify noise, resulting in a decrease in the signal-to-noise ratio. Therefore, smoothing should be performed before differentiation. For a discrete spectrum, calculate the first-order derivative and second-order derivative spectra at its wavelength point.

[0066] MSC algorithm: It is mainly used to reduce the impact of scattering caused by uneven particle size and distribution on the near-infrared spectrum. MSC is based on the spectral array of a group of samples and is essentially the same as spectral standardization.

[0067] SNV and Detrend algorithms: SNV is used to reduce the impact of uneven particle size and nonspecific scattering on the particle surface. It is based on the rows of the spectral array, that is, it processes a piece of spectral data. The Detrend algorithm generally processes the spectrum after SNV, fits a trend line between the absorbance and wavelength of the original spectrum, and then subtracts the trend line from the original spectrum, thereby eliminating the baseline drift of the diffuse reflectance spectrum.

[0068] Normalization algorithm: The purpose of spectrum normalization is to eliminate the influence of optical path change on the spectrum, which is divided into minimum / maximum normalization, vector normalization, and zero correction. In near-infrared spectroscopy analysis, vector normalization is commonly used, which first calculates the average absorbance of the spectrum, then subtracts the average from the spectrum, and then divides it by the sum of the squares of all spectra.

[0069] In step S15, three supervised pattern recognition algorithms, namely linear discriminant analysis (LDA), support vector machine (SVM) and generalized regression neural network (GRNN) model, are used to establish a qualitative classification model for the cooked minced meat sample.

[0070] Select the appropriate algorithm based on the prediction results.

[0071] LDA includes three methods: Linear, Quadratic and Mahalanobis. SVM uses the support vector classification algorithm (C-SVC). The larger the penalty parameter C is, the less tolerant it is to errors. The kernel function type is Linear, and the default value of the penalty parameter C is 1. Both LDA and GRNN algorithms select the first 20 principal component scores of the spectrum as input variables, and use smoothing (5 points, 15 points and 25 points), smoothing first-order derivative, multivariate scatter correction (MSC), standard normal variable transformation (SNV), normalization, baseline correction and detrend algorithm to preprocess the model.

[0072] The effects of different pretreatment methods on modeling results were compared. The better results of different algorithms combined with pretreatment to establish qualitative models for cooked minced meat sample sets are shown in Table 2. The prediction accuracy of the models established by the three LDA methods were 91.79%, 85.82%, and 85.07%, respectively, and the number of sample misjudgments was 22, 38, and 40, respectively. When the first three principal components were selected, the GRNN model combined with normalization had better results than other pretreatment methods, and the accuracy of the calibration set reached 100%, but the accuracy of the prediction set was only 65.30%, and the prediction results were poor. The SVM model after SNV pretreatment achieved better results than the other models above, and the accuracy of the calibration set and prediction set reached 98.70% and 94.78%, respectively.

[0073] The better results of different algorithms combined with pretreatment to establish qualitative models for cooked minced meat sample sets are shown in Table 2. Table 2 shows the better results of different algorithms combined with pretreatment to establish models for cooked minced meat adulteration samples. The SVM model after SNV pretreatment achieved better results than other models, with the accuracy of the calibration set and prediction set reaching 98.70% and 94.78% respectively. The prediction results of the three methods of LDA were second. Although the GRNN model had a higher accuracy rate in the calibration set, its prediction performance was poor. It can be seen from the prediction results of the adulterated samples that the prediction accuracy of adulterated samples is higher, and pure meat samples are more likely to be misjudged as adulterated samples.

[0074] Table 2 Qualitative prediction results of the cooked minced meat adulteration sample set

[0075] In step S16, the establishment process of the quantitative detection model is similar to that of the qualitative detection model. The first 20 principal components of the near-infrared spectrum are selected for dimensionality reduction, and a partial least squares regression (PLSR) quantitative detection model is established in combination with cross-validation. The first 10 and first 15 latent variable scores are selected for partial least squares dimensionality reduction, and the latent variables extracted after partial least squares dimensionality reduction are used as input variables for cooked minced meat to establish a back propagation neural network (BP) quantitative detection model.

[0076] The cumulative percentage of variance of the first 10 latent variable scores of donkey and mule adulteration in cooked minced meat reached 95.44%, and the cumulative percentage of the first 15 latent variable scores of donkey and horse adulteration was 98.75%. Therefore, the first 10 and first 15 latent variable scores were selected for PLS dimensionality reduction, and the training parameters of the BP model were set as follows: the number of training times was 1000, the learning rate was 0.001, and the minimum error of the training target was 0.001.

[0077] Compare the effects of different preprocessing methods on modeling results, select the preprocessing method with the best detection results, and optimize the detection model.

[0078] The quantitative detection models of cooked meat minced donkey and donkey adulteration and donkey and horse adulteration established by PLSR both have the best prediction results under 15-point smoothing pretreatment. The specific results are shown in Table 3. Compared with the PLSR model, the results of the BP model for sample quantitative detection have been improved. The prediction results of donkey and mule adulteration are better under Detrend pretreatment, and the results of donkey and horse adulteration are better under normalization pretreatment. The specific results are shown in Table 4. Figure 3 , Figure 4 shown.

[0079] Table 3 PLSR quantitative detection results of adulterated cooked minced meat samples

[0080] Table 3 shows the results of the PLSR quantitative detection model for adulterated cooked minced meat samples. In the detection of donkey and mule adulteration, the prediction accuracy of the model calibration set and cross-validation set after various pretreatments was improved compared with the original spectrum, among which the model after 15-point smoothing had the best result. For the detection of donkey and horse adulteration, the accuracy of the prediction set was slightly improved after 15-point smoothing pretreatment, and the overfitting phenomenon of the calibration set was alleviated, and the stability of the model was enhanced.

[0081] Table 4 BP quantitative detection results of adulterated cooked minced meat samples

[0082] Table 4 shows the BP quantitative detection results of the donkey and donkey adulterated cooked meat sample sets. The prediction accuracy of the donkey and donkey adulterated sample sets was improved after 5-point, 15-point, and 25-point smoothing, smoothed first-order derivative, MSC, SNV, Detrend, and normalization preprocessing. After Detrend preprocessing, the regression evaluation index R of the cross-validation set and the prediction set was 2 The coefficient of determination, RMSE (root mean square error), and RPD (residual prediction deviation) were 0.971, 0.067, 5.844, 0.980, 0.086, and 6.984, respectively. The accuracy of the model was improved, the overfitting phenomenon of the calibration set was alleviated, and the stability of the model was improved. The accuracy of the cross-validation set and the prediction set of the donkey and horse adulteration sample set was improved after MSC, SNV, Detrend, Baseline, and normalization. The R of the cross-validation set and the prediction set after normalization was 0.971, 0.067, 5.844, 0.980, 0.086, and 6.984, respectively. The accuracy of the model was improved, the overfitting phenomenon of the calibration set was alleviated, and the stability of the model was improved. 2 , RMSE, and RPD were 0.997, 0.032, 18.026, 0.982, 0.089, and 7.454, respectively, achieving better results.

[0083] The following is explained by examples.

[0084] Step 1: Sample preparation Prepare cooked meat samples of donkey meat, mule meat, and horse meat. Remove fat and fascia in advance, and then cut into small pieces of 3 to 5 cubic millimeters. Use the cooked meat samples to prepare the cooked meat samples of donkey meat, mule meat, horse meat, and donkey, mule, and donkey and horse adulterated meat. Randomly select 3 cooked meat samples of donkey meat, mule meat, horse meat, and donkey, mule, and donkey and horse adulterated meat, for a total of 15 samples. Accurately weigh 50 g of the sample, put it in a quartz dish with a diameter of 5 cm and compact it for testing.

[0085] Step 2: Collect the spectrum Perform near infrared spectroscopy scans on the 15 samples selected in step 1 to obtain 15 spectra, record the absorbance corresponding to the wavelength points of 4000~12500cm-1, and draw the spectral curve as shown in Figure 5 shown.

[0086] Step 3: Sample testing 1) Identification The spectral data of 15 samples to be tested were preprocessed by SNV to obtain the spectral graph as shown in Figure 6 As shown, the SVM model is then substituted for classification and identification. The results are shown in Table 5. The prediction accuracy is 100%, which can accurately distinguish.

[0087] 2) Detection The samples of the donkey and mule adulteration and donkey and horse adulteration examples to be tested were subjected to detrend and normalization pretreatment respectively, and the spectra were as follows: Figure 7 , Figure 8 As shown, the measured data results are shown in Fig. 9 , Fig.10 .

[0088] Table 5 Prediction results of the SNV qualitative classification model in Example

[0089] Fig. 9 and Fig.10 The quantitative detection results of the samples to be tested in the embodiment show that good prediction results are obtained for the quantitative detection of donkey and mule adulteration and donkey and horse adulteration.

Claims

1. A near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat, characterized in that: The following steps are involved: S1. Establish a detection model; S11, sampling and pre-processing the samples to obtain cooked minced donkey meat samples, cooked minced horse meat samples, cooked minced mule meat samples, cooked minced donkey and horse meat samples with different adulteration ratios, and cooked minced donkey and mule meat samples with different adulteration ratios; S12, performing near infrared spectrum scanning on each sample to obtain a near infrared spectrum graph of each sample; S13, dividing the near infrared spectra of each sample into a calibration set and a prediction set; S14, performing spectral preprocessing on the near infrared spectra of each sample to eliminate interference factors; S15. Establish a qualitative classification model; S16. Establish and optimize the quantitative detection model; S2, taking the sample to be tested and performing pre-treatment; S3, performing near infrared spectroscopy scanning on the sample to be tested to obtain a near infrared diffuse reflectance spectrum of the sample to be tested; S4, substituting the near-infrared diffuse reflectance spectrum of the sample to be tested into the qualitative classification model to distinguish the category of the sample to be tested; S5. Substituting the near-infrared diffuse reflectance spectrum of the sample to be tested into the quantitative detection model to detect the content of donkey meat in the sample to be tested.

2. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 1, characterized in that: In step S11, the meat block is cooled to room temperature and then chopped into small cubic pieces with a side length of 3 to 5 mm as cooked minced meat samples. Cooked minced meat samples of donkey meat, horse meat, and mule meat are used to prepare donkey and horse adulterated cooked minced meat samples and donkey and mule adulterated cooked minced meat samples with a donkey meat content of 0% to 100% according to a certain gradient, and there are no less than 10 samples in each gradient.

3. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 2, characterized in that: In step S11, donkey and horse adulterated cooked minced meat samples and donkey and mule adulterated cooked minced meat samples with donkey meat content ranging from 0% to 100% are prepared with a gradient of 10%.

4. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 1, characterized in that: In step S13, before dividing the calibration set and the prediction set, abnormal samples are eliminated. Abnormal samples include samples with large errors in spectral data.

5. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 1, characterized in that: In step S14, the preprocessing includes 5-point smoothing, 15-point smoothing, 25-point smoothing, smoothed first-order derivative, first-order and second-order differentials, multivariate scatter correction, standard normal variable transformation, normalization, baseline correction, and detrending algorithm.

6. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 1, characterized in that: In step S15, three supervised pattern recognition algorithms, namely linear discriminant analysis, support vector machine and generalized regression neural network model, are used to establish a qualitative classification model for the cooked minced meat sample.

7. The near infrared qualitative and quantitative detection method for adulteration of cooked donkey meat according to claim 1, characterized in that: In step S16, the first 20 principal components of the near-infrared spectrum are selected for dimensionality reduction, and a partial least squares regression quantitative detection model is established in combination with cross-validation; the first 10 and first 15 latent variable scores are selected for partial least squares dimensionality reduction, respectively, and the latent variables extracted after partial least squares dimensionality reduction are used as input variables for cooked minced meat to establish a back propagation neural network quantitative detection model, compare the effects of different preprocessing methods on the modeling results, and select the preprocessing method with the best detection results.

Images