Method for predicting water-retaining property of pork based on low-field nuclear magnetic resonance technology

Abstract

The invention relates to a method for constructing a pork water-retaining property prediction model based on a low-field nuclear magnetic resonance technology. The method comprises the following steps: (1) preparing a sample; (2) sample low-field nuclear magnetic resonance signal acquisition and inversion; (3) measuring water-retaining property parameters of the sample; and (4) model construction and evaluation. According to the method, an LF-NMR instrument is used for analyzing and optimizing acquisition parameters of a pork sample, proper characteristic parameters are selected as input parameters, proper pork water-retaining property parameters are selected as output parameters, and a pork water-retaining property prediction model based on a neural network, support vector regression and a random forest is constructed. The prediction method provided by the invention provides a scientific basis for the water-retaining property change of the pork in the processing, storage and transportation processes, and also provides more efficient and accurate detection service for food quality and safety monitoring.

Description

Technical Field

[0001] The invention relates to the field of food detection, and in particular to a method for predicting the water retention capacity of pork based on low-field nuclear magnetic resonance technology. Background Art

[0004] The water retention of frozen pork is one of the key factors affecting meat quality, and is also an important quality indicator for measuring frozen meat during meat processing and storage. Therefore, in-depth research on the impact of repeated freezing and thawing on the water retention of pork and exploring effective detection and evaluation methods are of great significance for improving the quality and efficiency of the pig industry system.

[0005] Low-field Nuclear Magnetic Resonance (LF-NMR) is an emerging technology used to study water distribution, lipid oxidation status, protein conformation, etc. Due to the excellent characteristics of LF-NMR being non-destructive and non-invasive, it is widely used in many fields such as food, biology, and physical chemistry, bringing a new perspective and application possibilities to the detection field.

[0006] LF-NMR technology has the advantages of low cost, fast speed, and high detection accuracy. It has its unique advantages in the determination of moisture content in meat. It can not only accurately determine the overall moisture content in meat, but also detect the distribution and state changes of moisture in meat (non-mobile water, bound water, and free water). However, the multivariate statistical method of LF-NMR technology for water retention of meat is relatively simple, and it mainly establishes a simple correlation analysis between its characteristic data and water retention, or uses the nuclear magnetic signal attenuation curve to establish a prediction model based on a small sample size. The prediction model established in this case has poor generalization ability, is prone to systematic errors in practical applications, and is not widely practical, and cannot achieve accurate prediction of pork water retention. At present, there are relatively few studies on the application of LF-NMR technology to the accurate prediction of water retention, and there is still a lot of room for exploration in-depth research. Summary of the invention

[0007] To solve the above technical problems, the present invention includes the following aspects: A first aspect of the present invention provides a method for constructing a prediction model for water retention of pork based on low-field nuclear magnetic resonance technology, the method comprising the following steps: (1) Sample preparation: Fresh pork was selected as the test sample, and a portion of the fresh pork was subjected to freeze-thaw cycle treatment to obtain fresh pork and pork test samples with different freeze-thaw cycle times; (2) Sample low-field nuclear magnetic resonance signal acquisition and inversion: Place the pork test sample obtained in step (1) in a nuclear magnetic resonance sample tube, set the instrument detection parameters, start collecting the sample low-field nuclear magnetic resonance (LF-NMR) signal, and use software to perform multi-component inversion fitting on the CPMG sequence collected sample to obtain the sample's transverse relaxation time T 2 Curve map, extract the data of characteristic parameters in the map; (3) Determination of water retention parameters of samples: After the pork test sample in step (2) is scanned by LF-NMR, water retention parameter data of the pork sample is immediately taken out and measured, wherein the water retention parameter is the pressure loss rate; (4) Model construction and evaluation: The transverse relaxation time T of the test sample obtained in step (2) is used. 2 The characteristic parameter data of the curve graph is used as the modeling input parameter, and the water retention parameter data of the test sample obtained in step (3) is used as the modeling output parameter. The input parameters and the output parameters are substituted into MATLAB software to establish a regression model for predicting the water retention of pork, and the prediction effect of the regression model is evaluated.

[0008] Preferably, the fresh pork sample in step (1) is stored at -18°C and subjected to freeze-thaw cycle treatment.

[0009] Preferably, the freeze-thaw cycle treatment in step (1) comprises storing the pork at -18°C for 12 hours and then completely thawing it at 4°C.

[0010] Preferably, in step (1), the fresh pork sample is subjected to freeze-thaw cycle treatment for 3, 5 and 10 times respectively.

[0011] Preferably, the instrument detection parameters set in step (2) include waiting time, echo time and number of repeated scans.

[0012] Preferably, the waiting time is set to 1000-5000ms. More preferably, the waiting time is set to 2000-4000ms. Further preferably, the waiting time is set to 2000ms.

[0013] Preferably, the echo time is set to 0.10-0.40 ms. More preferably, the echo time is set to 0.30-0.35 ms. Further preferably, the echo time is set to 0.30 ms.

[0014] Preferably, the number of repeated scans is set to 8-128 times. More preferably, the number of repeated scans is set to 16-32 times. Further preferably, the number of repeated scans is set to 32 times.

[0015] Preferably, in step (2), three samples with different freeze-thaw cycle times are selected. 2 Plot the decay curve.

[0016] Preferably, in step (2), T 2 The characteristic parameters of the curve spectrum are selected from one or more of peak start time, peak apex time, peak end time, peak area and peak area ratio.

[0017] Preferably, the peak start time parameters include T 21S 、T 22S or T 23S The peak time parameters include T 21P 、T 22P or T 23P , the peak end time parameters include T 21E 、T 22E or T 23E , the peak area parameters include S 21 , S 22 , S 23 or S T , the peak area ratio parameters include P 21 , P 22 or P 23 .

[0018] More preferably, in step (2), T 2 The characteristic parameter of the curve spectrum is the peak area.

[0019] Preferably, the peak area parameter includes the bound water peak area S 21 , non-mobile water peak area S 22 , free water peak area S 23 Or total peak area S T .

[0020] Further preferably, in step (2), T 2 The characteristic parameter of the curve spectrum is the bound water peak area S 21 .

[0021] Preferably, the pressure loss rate in step (3) is determined by the following method: a sampler with a diameter of 2.5 cm is used to take a 1.0 cm thick round meat column sample along the vertical direction of the muscle fiber, and the sample is accurately weighed. The upper and lower surfaces of the meat sample are wrapped with 16 layers of filter paper respectively. The meat is pressurized with a hydraulic tester at a constant pressure of 35.0 kg for 5 minutes, and then the filter paper is removed and weighed immediately. The pressure loss rate of the meat sample is calculated according to the following formula: ; Where PL: pressurization loss rate, unit is %; m 1: The mass of the meat sample before pressurization, in g; m 2 : The mass of the meat sample after pressurization, in g.

[0022] Preferably, the regression model in step (4) is an artificial neural network regression model, a support vector regression model or a random forest regression model. More preferably, the regression model is a support vector regression model or a random forest regression model. Further preferably, the regression model is a support vector regression model.

[0023] Preferably, the artificial neural network regression model is implemented using the Levenberg-Marquardt method in the neural network toolbox of MATLAB software. More preferably, the number of hidden layers of the Levenberg-Marquardt method is 10.

[0024] Preferably, the support vector regression model is implemented using the fitrsvm function in the machine learning toolbox of MATLAB software. More preferably, the core parameters of the support vector regression model are penalty factor c=4.0, radial basis function parameter g=0.8, cmd=['-t 2','-c ',num2str(c),'-g ',num2str(g),'-s 3 -p 0.01'], model=svmtrain(t_train, p_train, cmd).

[0025] Preferably, the random forest regression model is implemented using Statistics and Machine Learning Toolbox in MATLAB. More preferably, it is implemented using the TreeBagger class tool in Statistics and Machine Learning Toolbox, with the number of decision trees trees=50 and the minimum number of leaves leaf=5.

[0026] Preferably, the regression prediction model in step (4) uses 2 / 3 of the samples to construct the calibration set data, and 1 / 3 of the sample data is used as external validation to establish the validation set model, and the hold-out method is used for cross-validation.

[0027] Preferably, the regression prediction model in step (4) is evaluated for its prediction effect by the coefficient of determination, root mean square error and mean square error of the validation set.

[0028] Preferably, the R of the regression prediction model in step (4) is c 2 The value is greater than 0.70, and the RMSEC value is less than 5.0. More preferably, the R c 2The value is greater than 0.80, and the RMSEC value is less than 4.0. Further preferably, the R c 2 The value was greater than 0.90 and the RMSEC value was less than 3.0.

[0029] A second aspect of the present invention provides a method for predicting the water retention capacity of pork based on low-field nuclear magnetic resonance technology, the method comprising the following steps: (i) Sample low-field nuclear magnetic resonance signal acquisition: Take the pork sample to be tested, place it in the nuclear magnetic resonance sample tube, set the instrument detection parameters, start collecting the sample low-field nuclear magnetic resonance (LF-NMR) signal, use the software to perform multi-component inversion fitting on the CPMG sequence collected sample, and obtain the sample's transverse relaxation time T 2 Curve map, extract the data of characteristic parameters in the map; (ii) Calculate the water retention of the sample: Substitute the transverse relaxation time T obtained in step (i) into 2 The characteristic parameter input data in the curve graph is substituted into the pork water retention prediction model to calculate the water retention parameter output data of the pork sample, and the water retention of the pork sample is evaluated based on the water retention parameter data, wherein the water retention parameter is the pressure loss rate.

[0030] Preferably, the pork sample to be tested in step (i) is fresh pork or pork that has been frozen and thawed multiple times.

[0031] Preferably, the instrument parameters of the low-field nuclear magnetic resonance in step (i) are as follows: the waiting time is set to 2000 ms, the echo time is set to 0.30 ms, and the number of repeated scans is set to 32 times.

[0032] Preferably, in step (ii), T 2 The characteristic parameter of the curve spectrum is the peak area. More preferably, the peak area parameter includes the bound water peak area S 21 , non-mobile water peak area S 22 , free water peak area S 23 Or total peak area S T . Further preferably, T 2 The characteristic parameter of the curve spectrum is the bound water peak area S 21 .

[0033] Preferably, the pork water retention capacity prediction model in step (ii) is constructed by the aforementioned method.

[0034] More preferably, the pork water retention capacity prediction model in step (ii) is a support vector regression model.

[0035] Preferably, the pork water retention prediction model in step (ii) is pre-loaded into the MATLAB workspace before use, and the load function is used to load the .mat file that saves the model parameters.

[0036] Preferably, in step (ii), after the pork water retention prediction model is loaded and the characteristic parameter input data is prepared, the predict function is used to perform the prediction.

[0037] Preferably, the pressure loss rate used as a water retention parameter in step (ii) is in the range of 20-50%, and the higher the value, the fewer freeze-thaw cycles the pork sample has experienced.

[0038] More preferably, the pressure loss rate as a water retention parameter in step (ii) is in the range of 25-45%, and the higher the value, the fewer freeze-thaw cycles the pork sample has experienced.

[0039] The technical effects produced by the present invention are: 1. The present invention uses an LF-NMR instrument to analyze and optimize the collection parameters of pork samples, determines a method for detecting the meat block used for the pressurized loss rate of pork based on low-field nuclear magnetic resonance technology, and then selects the most suitable data as input parameters according to the different characteristics of the three types of LF-NMR data, constructs a pork water retention prediction model based on neural network, support vector regression and random forest, and evaluates the effectiveness of the model through indicators such as determination coefficient, root mean square error and mean absolute error, and finally obtains a method for accurately predicting the water retention of pork based on LF-NMR technology.

[0040] 2. Using the results obtained by low-field nuclear magnetic resonance technology to make accurate predictions is a difficult problem faced by the industry. The key to accurate model prediction is to select appropriate modeling data, which not only guarantees model performance, but also reduces the workload of data processing and improves computational efficiency. The model for predicting the water retention of pork samples established by the present invention using LF-NMR technology to invert the peak area in the characteristic parameters has high accuracy in predicting water retention. In addition, the three models of neural network, support vector regression and random forest established by the present invention have better prediction capabilities, among which the support vector regression prediction model R c 2 The highest and the lowest RMSEC value confirmed that the support vector regression model had the best predictive ability.

[0041] 3. The present invention studies the effect of repeated freezing and thawing on the water retention of pork, and uses three indicators, namely juice loss rate, cooking loss rate and pressurized loss rate, to evaluate the water retention of pork. It is finally unexpectedly discovered that the pressurized loss method is the most suitable water retention detection method based on low-field nuclear magnetic resonance.

[0042] 4. The pork water retention prediction method based on LF-NMR of the present invention provides a scientific basis for the changes in the water retention of pork during processing, storage and transportation. The method can also be extended to the water retention evaluation of other foods, as well as the broader field of food quality and safety monitoring, providing the food industry with more efficient and accurate testing services. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is the effect of repeated sampling waiting time on the LF-NMR signal amplitude; Figure 2 is the effect of echo time on the multi-component fitted transverse relaxation spectrum; Figure 3 is the effect of repeated sampling on LF-NMR T 2 The influence of relaxation spectra; Figure 4 is the CPMG decay curve of pork; Figure 5 It's pork T 2 Inversion curve; Figure 6 These are some characteristic parameters of the pork relaxation spectrum; Figure 7 This is the correlation diagram between the calibration set (a), validation set (b) and predicted value of the artificial neural network regression model predicting the water retention of pork; Figure 8 This is the correlation diagram between the calibration set (a), validation set (b) and predicted value of the support vector regression model for predicting pork water retention; Fig. 9 This is the correlation diagram between the calibration set (a), validation set (b) and predicted values ​​of the random forest model predicting the water retention of pork; Fig.10 is the mean square error diagram of the actual value of pork water retention and its predicted value in different models (a is the calibration set, b is the validation set); Fig.11 This is a correlation analysis chart between pork water retention parameters and LF-NMR parameters. DETAILED DESCRIPTION

[0044] The present invention is further described below in conjunction with examples, but the embodiments of the present invention are not limited thereto. The experimental methods used in the following examples are conventional methods unless otherwise specified.

[0045] Experimental Example 1: Construction of a prediction model for pork water retention based on low-field nuclear magnetic resonance technology 1. Test methods 1.1 Sample preparation Three-way pigs with the same feeding conditions were selected from Beijing Ershang Meat Food Group Co., Ltd., and the longissimus dorsi muscle of the left half carcass was collected. The fat and connective tissue on the surface of the pork were removed, and the sample was evenly divided into four parts perpendicular to the direction of the myofibrils, with a weight of about 500g each. One of the samples was not subjected to repeated freezing and thawing, marked as 0F-T, and the various indicators were directly measured. The other three samples were stored in a low-temperature environment of -18℃ and subjected to 3, 5 and 10 freeze-thaw cycles, respectively. Each cycle included storing the pork at -18℃ for 12h and then completely thawing at 4℃, marked as 0 FT, 3 FT, 5 FT and 10 FT, respectively. A total of 218 fresh pork samples and pork samples with different freeze-thaw cycles were prepared (0 FT: 54, 3 FT: 66, 5 FT: 49, 10 FT: 49), and the above pork samples were subjected to LF-NMR measurement.

[0046] 1.2. Pork low-field NMR signal acquisition and inversion Place the test sample prepared in Section 1.1 in the NMR sample tube, set the instrument detection parameters and start collecting the sample LF-NMR signal. Scan each sample three times in succession, and use the low-field NMR analyzer application software Ver4.0 to perform multi-component inversion fitting on the CPMG sequence collected samples to obtain the transverse relaxation spectrum of the sample. Its attenuation curve echo signal and T 2 The data of the inversion curve are selected from the same time for representation, and the characteristic parameters of each sample after inversion are represented by the average value of three inversions. The characteristic parameters include peak start time (T 21S 、T 22S and T 23S ), peak time (T 21P 、T 22P and T 23P ), peak end time (T 21E 、T 22E and T 23E ), peak area (S 21 , S 22 , S 23 and S T ), peak area ratio (P 21 , P 22 and P 23 ). Among them, T 21S 、T 21P 、T 21E , S 21 and P 21 Indicates the corresponding parameter of bound water, T 22S 、T 22P 、T 22E , S 22 and P 22 Indicates the parameter corresponding to the non-flowing water, T23S 、T 23P 、T 23E , S 23 and P 23 Represents the free water corresponding parameter, S T represents the total peak area. All data are expressed as mean ± standard deviation, p <0.05 indicated significant differences.

[0047] 1.3. Optimization of acquisition parameters (1) Waiting time (TW) The waiting time is the period from the end of the previous sampling to the start of the next sampling, in milliseconds. If the waiting time is set too long, the sampling time will be longer. This test was carried out under the FID sequence, and the TW values ​​were set to 1000, 2000, 3000, 4000 and 5000ms respectively. The best condition was to compare the signal modulus and make it relatively stable.

[0048] (2) Echo time (TE) After the RF pulse, the time interval from the initial generation of transverse magnetization to the reception of the signal is called the echo time. The time interval between a 180° pulse and the next 180° pulse, in ms, was set to 0.10, 0.15, 0.20, 0.25, 0.30, 0.35, and 0.40 ms in this experiment to compare the attenuation and relaxation information of the samples under different TE.

[0049] (3) Number of repeated scans (NS) The number of repeated scans is the number of times the instrument automatically repeats sampling. The number of repeated scans will affect the signal strength and sampling time of the sample, so it is necessary to optimize the selection according to the actual signal of the sample. Therefore, the number of repeated scans of the instrument is set to 8, 16, 32, 64, and 128 times, and T is compared and analyzed respectively. 2 Relaxation properties of the spectrum.

[0050] 1.4. Determination of pressure loss rate After the LF-NMR scan is completed, the sample is immediately taken out to measure the pressure loss rate. The specific test method is as follows: fresh pork and pork with different freeze-thaw cycles are used as test samples. A sampler with a diameter of 2.5 cm is selected to take a 1.0 cm thick round meat column sample along the vertical direction of the muscle fiber, and accurately weigh it (accurate to 0.01g). The upper and lower surfaces of the meat sample are wrapped with 16 layers of filter paper respectively. The meat is pressurized with a hydraulic tester at a constant pressure of 35.0kg for 5 minutes, then the filter paper is removed and weighed immediately (accurate to 0.01g). The pressure loss rate of the meat sample is calculated according to the following formula: ; Where PL: pressurization loss rate, unit is %; m 1 : The mass of the meat sample before pressurization, in g; m 2 : The mass of the meat sample after pressurization, in g.

[0051] 1.5 Statistical Model and Data Processing (1) Artificial neural network regression model Artificial Neural Network (ANN) is a complex nonlinear mathematical method designed to simulate the complex structure and function of the human brain neural network. It is mainly composed of input layer, hidden layer and output layer, and each layer carries different information processing and conversion tasks. In order to achieve accurate classification or regression tasks, artificial neural network continuously adjusts the weights between different neural layers to ensure that the input feature vector can be mapped to the correct output category.

[0052] (2) Support Vector Regression Model Support vector regression (SVR) mainly maps the data in the input space to a high-dimensional feature space through a nonlinear mapping, and searches for an optimal hyperplane in a specific feature space to ensure that the distance between all sample points and this hyperplane is as small as possible. The optimal hyperplane represents the regression function used to predict new data points.

[0053] (3) Random Forest Model The basic principle of Random Forest (RF) is to use bootstrap self-sampling technology to train and create multiple decision trees, which together constitute a random forest model. This model improves its robustness and generalization ability by voting or weighted averaging the outputs of all decision trees, thereby achieving classification or regression analysis of data.

[0054] (4) Regression model construction and model evaluation MATLAB R2022a software was used to establish the prediction models of ANN, SVR and SF for pork water retention. The quantitative prediction models used 2 / 3 of the samples to construct the calibration set data, and 1 / 3 of the sample data as external validation to establish the validation set model. The hold-out method was used for cross-validation. The regression prediction model was tested by the coefficient of determination of the validation set (R p 2), root mean square error (RMSEP) and mean square error (MSEP) to evaluate the prediction effect. The model performance evaluation indicators of the model are as follows: (a) Coefficient of determination including calibration set (R c 2 ) and the correlation coefficient of the validation set (R p 2 ), which is used to measure the degree of fit of the model to the data, namely R 2 The closer it is to 1, the higher the degree of fit and the better the model accuracy. The calculation formula is: ; (b) The root mean square error includes the root mean square error of the calibration set (RMSEC) and the validation set (RMSEP). It is the square root of the average of the squares of the deviations between the observed values ​​and the predicted values. It is used to evaluate the accuracy of the model. The smaller the RMSE, the better the performance of the model. Its calculation formula is: ; (c) Mean square error includes the mean square error of the calibration set (MSEC) and the validation set (MSEP), which is the average of the squared difference between the predicted value and the true value. The smaller the value of MSE, the closer the model's prediction result is to the actual value, and the better the model's performance is. It is used to evaluate the accuracy of the model's prediction. Its calculation formula is: ; Note: N is the number of samples. is the actual value of the sample, is the predicted value of the sample, is the average value of all samples in the calibration set or validation set.

[0055] 2. Test results 2.1. Effects of acquisition parameters on pork NMR signals (1) Waiting time (TW) like Figure 1 As shown in the figure, the modulus values ​​obtained for each sampling are not the same. When TW is set too small, the error between modulus values ​​is large. As the TW value increases, the signal amplitude between sampling times tends to be stable, but when TW is set to a larger value, the sampling time will be extended, which is not conducive to rapid detection. Therefore, TW is set to 2000ms for the test of pork samples.

[0056] (2) Echo time (TE) The change of echo time will affect the effect of molecular self-diffusion and molecular exchange on relaxation characteristics, and the shape of CPMG signal will also change. The longer the echo time, the stronger the molecular self-diffusion, which leads to the transverse relaxation time T 2 extension. Figure 2 Table 1 shows the T of meat pieces at different echo times. 2 Changes in the area of ​​each peak after spectrum inversion. 21 , S 22 The peaks show an irregular trend as the echo time increases. This is speculated to be because when the sample is scanned, the moisture in pork is constantly migrating, causing the peak areas of bound water and immobile water to change in real time, so the selection of TE value is judged by the stability of its peak area. When the echo time is 0.30 and 0.35, the change in echo time does not have a significant effect on the stability of each peak area of ​​sample moisture, and the RSD of its peak area is relatively stable. However, when TE is 0.35, the peak shapes of immobile water and free water in the transverse relaxation spectrum of multi-component inversion fitting change, and cannot be well separated. Therefore, setting the TE value to 0.30ms can meet the experimental requirements.

[0057] Table 1 Effect of echo time on peak area of ​​multi-component pseudo-transverse relaxation spectrum: .

[0058] (3) Number of repeated scans (NS) As NS increases, the signal amplitude of the spectrum will gradually increase. Figure 3 The influence of different NS on the relaxation spectrum is shown in Table 2. Then, the optimal NS is determined by analyzing the influence of NS on the peak area of ​​characteristic parameters. The results are shown in Table 2. As shown in Table 2, when NS is 32, S 21 and S 22 The minimum RSDs are 1.92% and 0.04% respectively. 23 The stability also meets the requirement of less than 5%, so it can be determined that when NS is 32, the characteristic parameter peak area has good stability. Although the increase in the number of repeated scans will significantly improve the signal-to-noise ratio, too many times will cause the magnet to heat up, causing the sample temperature to rise, thereby changing the sample relaxation characteristics and taking longer time. Therefore, this experiment chose to repeat the scan 32 times, at which time the stable relaxation characteristics can meet the needs of the sample measurement.

[0059] Table 2 Effect of repeated sampling times (NS) on relaxation spectrum peak area: .

[0060] 2.2. Selection of LF-NMR modeling input parameters Under the same testing conditions, 218 pork samples with different freeze-thaw cycles were tested by LF-NMR. 2 The purpose of drawing the attenuation curve is to intuitively show the difference in NMR signals of samples with different freeze-thaw cycles. Figure 4 As shown in the figure, although the decay rates of the samples are different, it is difficult to observe with the naked eye whether there is a clear difference in the decay curves due to the large number of samples and the subtle differences. Therefore, other technologies are needed to deeply analyze the spectrum and amplify the differences to extract key information from it in order to more accurately quantify the water retention of pork.

[0061] The inversion curve T is obtained by inverting the CPMG signal in the attenuation curve 2 The results are as follows Figure 5 The inversion curves of fresh meat and samples with different freeze-thaw cycles all show three peaks (T 21 、T 22 and T 23 ), although the sample curves with the same number of freeze-thaw cycles have a relatively high overlap, the inversion curve data is a 200×218 data matrix with a large amount of data. The calculation is slow when performing data statistics, so it is not selected as the modeling data.

[0062] The characteristic parameters of the relaxation spectrum in the pork sample were further extracted, and the T S 、T P 、T E And the result of P is as follows Figure 6 The significance analysis of the characteristic parameters of the relaxation spectrum of pork samples showed that with the increase of the number of freeze-thaw cycles, the characteristic parameters T of 10 FT and fresh meat (0 FT), 3 FT and 5 FT samples increased. 21S 、T 21P 、T 21E , S 21 and P 21 There is a significant difference ( p <0.05), T 22S 、T 22P 、T 22E and P 22 There were no significant differences.

[0063] Although the characteristic parameters T of samples with different freeze-thaw cycles are S 、T P 、T EThere are some differences between the values ​​of T and P, but each pork sample has its own unique tissue structure, and pork is a "live" sample, the water inside is always in a state of migration, so the peak start time, peak peak time, and peak end time of the water peak in different samples are also different. Therefore, when selecting data for water retention modeling, we cannot just rely on T S 、T P and T E The peak area reflects the total amount of water in different states in the sample, so the peak area (S 21 , S 22 , S 23 and S T ) as a modeling input parameter indicator can more stably and accurately reflect the water retention capacity of samples, thereby better predicting the changes in water retention capacity of pork samples.

[0064] Table 3 Changes of some characteristic parameters of pork relaxation spectrum: ; Note: Different letters indicate significant differences ( p <0.05).

[0065] 2.3 Analysis of artificial neural network regression prediction model The peak area (S 21 , S 22 , S 23 and S T ) characteristic parameters as input variables, and pressurization loss rate as output variables, the neural network model was constructed using the neural network toolbox (Levenberg-Marquardt method, hidden layer number is 10) in Matlab R2022a software (Mathworks, Inc. Natick, USA), and the network weights were continuously optimized through the training process. After the model training was completed, the model based on the neural network was obtained, and the pork water retention prediction model was saved by the save method. When used later, the prediction model was loaded, and the input variables of the sample to be tested were substituted into the model to achieve accurate prediction of the water retention of the pork sample to be tested. The specific steps are as follows: (1) Data verification: Verify the peak area of ​​the low-field NMR inversion of pork and remove the entire spectrum curve data of abnormal spectra. The "abnormal spectrum" refers to the spectrum with excessive background noise, signal peaks masked by noise or difficult to distinguish between signal peaks and noise, and signal peaks that should appear are completely missing or the signal intensity is far lower than expected.

[0066] (2) Data preprocessing: The verified pork low-field NMR inversion peak area and pork pressurization loss rate data were preprocessed, including removing paired pork low-field NMR inversion peak area and pork pressurization loss rate data containing missing values ​​and null values.

[0067] (3) Dataset division: To ensure reasonable evaluation of model training and inversion results, the entire data set (218 groups) was divided into two parts using a random method. 75% (164 groups) of the data were used for model training, and 25% (54 groups) of the data were used for post-training effect evaluation.

[0068] (4) Division of training data set: To ensure the effectiveness of model training, the hold-out method is used for cross-validation to train the model in each model training iteration.

[0069] (5) Loading the model: If you need to use the established model, load the trained model into the MATLAB workspace. Use the load function to load the .mat file that saves the model parameters.

[0070] (6) Prepare input data: Before calling the model, prepare input data of the same format and type as those used during model training, i.e., the peak area obtained after NMR scanning.

[0071] (7) Use the model for prediction: After the model is loaded and the input data is prepared, use the predict function to make predictions.

[0072] In order to quantitatively evaluate the predictive performance of the model, R p 2 The quantitative prediction model was evaluated by RMSEP to comprehensively evaluate the model performance. The correlation between the actual water retention value of the calibration set and the validation set of the artificial neural network prediction water retention model and its predicted value is shown in the figure. Figure 7 As shown, the calibration set R of the model c 2 The validation set R p 2 The average speed of the test sample is 0.72, and the RMSEP is 3.39.

[0073] 2.4 Support Vector Regression Prediction Model Analysis The peak area (S 21 , S 22 , S 23 and S T) characteristic parameters as input variables, and the pressure loss rate as the output variable. Referring to the method in Section 2.3, support vector regression was used to build a prediction model (implemented using the fitrsvm function in the MATLAB machine learning toolbox, with penalty factor c=4.0, radial basis function parameter g=0.8, cmd=['-t 2','-c ',num2str(c),'-g ',num2str(g),'-s 3 -p 0.01'], model=svmtrain(t_train, p_train,cmd)) to achieve accurate prediction of the water retention of pork samples. The correlation diagram of the actual water retention values ​​of the calibration set and the validation set of the support vector regression model for predicting water retention and its predicted value is shown in the figure below. Figure 8 As shown, the calibration set R of the model c 2 The validation set R p 2 The average sensitivity of the two tests was 0.82, the RMSEP was 2.90, and the MSEP was 8.40.

[0074] 2.5 Random Forest Model Analysis The peak area (S 21 , S 22 , S 23 and S T ) characteristic parameters as input variables, and the pressure loss rate as output variable. The prediction model was constructed by random forest according to the method in Section 2.3 (using Statistics and Machine Learning Toolbox (TreeBagger class) in MATLAB containing the random forest algorithm function, with the number of decision trees = 50 and the minimum number of leaves = 5) to achieve accurate prediction of the water retention of pork samples. The correlation diagram of the actual water retention value of the calibration set and the validation set of the random forest regression prediction water retention model and its predicted value is shown in the figure below. Fig. 9 As shown, the calibration set R of the model c 2 The validation set R p 2 The average sensitivity of the two samples was 0.83, the RMSEP was 3.05, and the MSEP was 9.28.

[0075] 2.6 Comparison of different prediction models In this test case, ANN, SVR and RF were used to establish prediction models for pork water retention, and calibration set prediction models and validation set prediction models were established respectively. The evaluation index results of different models are shown in Table 4, and the mean square error of the prediction results of calibration set and validation set of different models is shown in Table 4. Fig.10The performance of ANN, SVR and SF models was evaluated, and it was found that compared with ANN and SF, SVR had a better performance in predicting water retention in both the calibration set and the validation set. c 2 The validation set R p 2 The average water retention rate of fresh pork and frozen-thaw cycle pork was 0.82, the RMSEP was 2.90, and the MSEP was 8.40. The experimental results show that it is feasible to use support vector regression (SVR) to predict the water retention of fresh pork and frozen-thaw cycle pork.

[0076] Table 4 Comparison of quantitative model results of pork water retention training set and validation set: .

[0077] Test Example 2: LF-NMR Modeling Output Parameter Screening Test 1. Test methods 1.1 Sample preparation Three-way pigs with the same feeding conditions were selected from Beijing Ershang Meat Food Group Co., Ltd., and the longissimus dorsi muscle of the left half of the carcass was collected. The fat and connective tissue on the surface of the pork were removed, and the sample was evenly divided into four parts perpendicular to the myofibril direction, with a weight of about 500g each. One of the samples was not repeatedly frozen and thawed, marked as 0 FT, and the various indicators were directly measured. The remaining three samples were stored in a low-temperature environment of -18℃ and subjected to 3, 5 and 10 freeze-thaw cycles, respectively. Each cycle included storing the pork at -18℃ for 12h and then completely thawing at 4℃, marked as 0 FT, 3 FT, 5 FT and 10 FT, respectively. The samples after each treatment were then measured for various indicators.

[0078] 1.2 Determination of juice loss rate Use fresh pork prepared in Section 1.1 and pork with different freeze-thaw cycles as test samples, cut the meat samples into 2cm×3cm×5cm strips along the direction of the muscle fibers, accurately weigh the mass (accurate to 0.01g) and hang them in a bag. After placing them in a 4℃ environment for 24h, absorb the surface moisture of the meat and accurately weigh the mass (accurate to 0.01g). Calculate the juice loss rate of the meat sample according to the following formula: ; Where: DL: juice loss rate, unit is %; m 1 : The mass of the meat sample before hanging, in g; m 2 : The mass of the meat sample after hanging, in g.

[0079] 1.3 Determination of cooking loss rate Use fresh pork prepared in Section 1.1 and pork with different freeze-thaw cycles as test samples, take 2.5cm thick meat samples and weigh them accurately (accurate to 0.01g), place the meat samples in a 72℃ constant temperature water bath and heat them until the center temperature reaches 70℃, take the meat samples out of the water bath, cool them in running water for 30min, store them at 4℃ for 24h, dry the surface water with absorbent paper, and weigh the mass accurately (accurate to 0.01g). Calculate the cooking loss rate of the meat samples according to the following formula: ; Where: CL: cooking loss rate, unit: %; m 1 : Mass of meat sample before cooking, in g; m 2 : The mass of the meat sample after steaming, in g.

[0080] 1.4 Determination of pressure loss rate Fresh pork prepared in Section 1.1 and pork with different freeze-thaw cycles were used as test samples. A sampler with a diameter of 2.5 cm was used to take a 1.0 cm thick round meat column sample along the vertical direction of the muscle fiber, and weighed accurately (accurate to 0.01 g). The upper and lower surfaces of the meat sample were wrapped with 16 layers of filter paper respectively. The meat was pressurized with a hydraulic tester at a constant pressure of 35.0 kg for 5 minutes, then the filter paper was removed and weighed immediately (accurate to 0.01 g). The pressurization loss rate of the meat sample was calculated according to the following formula: ; Where: PL: pressurization loss rate, unit: %; m 1 : The mass of the meat sample before pressurization, in g; m 2 : The mass of the meat sample after pressurization, in g.

[0081] 1.5 Low-field NMR measurement According to the method in Section 1.2 of Experimental Example 1, low-field NMR measurements were performed on fresh pork and pork samples with different freeze-thaw cycles (TW was set to 2000 ms, echo time was set to 0.30 ms, and the number of repeated scans was set to 32 times). The spin-spin transverse relaxation (T) was measured using the Carr-Purcell-Meiboom-Gill (CPMG) sequence. 2 ), the echo signal of the sample collected by the CPMG sequence was processed by multi-component inversion using the low-field nuclear magnetic resonance analyzer application software Ver4.0 to obtain the transverse relaxation spectrum of the sample.

[0082] 1.6 Data Statistics and Analysis Each experiment was repeated at least three times, and each repetition included three parallel determination experiments. SPSS Statistics 26.0 software was used for data analysis, and one-way analysis of variance (ANOVA), Duncan's multiple range test (Duncan) and Pearson's correlation analysis (p < 0.05) were performed. The experimental data were presented in the form of mean ± standard deviation, and p < 0.05 indicated a significant difference. Origin 8.0 was used for graphing. Partial Least Squares-Discriminant Analysis (PLS-DA) was performed using SIMCA 14.1 (Umetrics, Swiss).

[0083] 2. Test results Water retention is one of the key indicators for evaluating the quality of meat. It not only affects the edible quality of meat, but also has certain economic value. In order to further study the effect of repeated freezing and thawing on the water retention of pork, the three indicators of juice loss rate, cooking loss rate and pressure loss rate were used to evaluate the water retention of pork. As shown in Table 5, with the increase of repeated freezing and thawing times, the juice loss rate of pork increased continuously and the difference was significant (p<0.05). The cooking loss rate increased first and then decreased. Compared with the sample with 0 FT, the cooking loss rate of 3 FT and 5 FT samples increased continuously; while the cooking loss rate of 10 FT sample decreased. Different from the changing trend of juice loss and cooking loss rate, the pressure loss showed a significant downward trend with the increase of the number of freeze-thaw cycles (p<0.05). The pressure loss rate decreased from 42.23% of 0 FT to 28.99% after 10 freeze-thaw cycles, and the decrease rate of 10 FT accounted for about 31% of 0 FT.

[0084] Table 5 Changes in pork juice loss rate, cooking loss rate and pressure loss rate: ; Note: Different letters indicate significant differences (p<0.05).

[0085] Through correlation analysis, the intrinsic relationship between pork quality indicators and water retention parameters can be further clarified. Water retention parameters mainly include cooking loss rate, pressure loss rate, and juice loss rate. Low-field NMR characteristic parameters mainly involve peak onset time (T S ), peak time (T P ), peak end time (T E ), peak area (S) and peak area ratio (P).

[0086] like Fig.11 The low-field NMR characteristic parameter T S 、T P 、TE It is significantly negatively correlated with the cooking loss rate and the pressure loss rate ( p <0.05). S 22 , S T There is a significant positive correlation with the pressure loss rate ( p <0.05), and the corresponding correlation coefficients were as high as 0.90 and 0.98. 22S 、T 22E and T 22P There is a strong negative correlation between T 23S 、T 23E and T 23P The loss rate of steaming and boiling is negatively correlated with the pressure loss rate, and the correlation is above -0.80. 22 The ratio was significantly positively correlated with the correlation coefficient of 0.94. 22 There is a positive correlation between the cooking loss rate and T 22S 、T 22E and T 22P There is a negative correlation between the juice loss rate and the low-field nuclear magnetic resonance characteristic parameters, and the correlation is above -0.80. In summary, the pressure loss method is more suitable as a water retention detection method based on low-field nuclear magnetic resonance.

[0087] Although specific embodiments of the present invention have been described, it will be appreciated by those skilled in the art that various changes and modifications may be made to the present invention without departing from the scope or spirit of the present invention. Therefore, the present invention is intended to cover all such changes and modifications that fall within the scope of the appended claims and their equivalents.

Claims

1. A method for constructing a prediction model for pork water retention based on low-field nuclear magnetic resonance technology. It is characterized in that The method comprises the following steps: (1) Sample preparation: Fresh pork was selected as the test sample, and a portion of the fresh pork was subjected to freeze-thaw cycle treatment to obtain fresh pork and pork test samples with different freeze-thaw cycle times; (2) Sample low-field nuclear magnetic resonance signal acquisition and inversion: Place the pork test sample obtained in step (1) in a nuclear magnetic resonance sample tube, set the instrument detection parameters, start collecting the sample low-field nuclear magnetic resonance (LF-NMR) signal, and use software to perform multi-component inversion fitting on the CPMG sequence collected sample to obtain the sample's transverse relaxation time T 2 Curve map, extract the data of characteristic parameters in the map; (3) Determination of water retention parameters of samples: After the pork test sample in step (2) is scanned by LF-NMR, the water retention parameter data of the pork sample is immediately taken out and measured, wherein the water retention parameter is the pressure loss rate; (4) Model construction and evaluation: The transverse relaxation time T of the test sample obtained in step (2) is used. 2 The characteristic parameter data of the curve graph is used as the modeling input parameter, and the water retention parameter data of the test sample obtained in step (3) is used as the modeling output parameter. The input parameters and the output parameters are substituted into MATLAB software to establish a regression model for predicting the water retention of pork, and the prediction effect of the regression model is evaluated. The regression model is a support vector regression model.

2. The method according to claim 1, It is characterized in that The instrument detection parameters set in step (2) include waiting time, echo time and number of repeated scans.

3. The method according to claim 2, It is characterized in that The waiting time is set to 1000-5000ms.

4. The method according to claim 2, It is characterized in that The echo time is set to 0.10-0.40 ms.

5. The method according to claim 2, It is characterized in that The number of repeated scans is set to 8-128 times.

6. The method according to claim 1, It is characterized in that In step (2), T 2 The characteristic parameters of the curve spectrum are selected from one or more of peak start time, peak apex time, peak end time, peak area and peak area ratio.

7. A method for predicting the water retention of pork based on low-field nuclear magnetic resonance technology. It is characterized in that The method comprises the following steps: (i) Sample low-field nuclear magnetic resonance signal acquisition: Take the pork sample to be tested, place it in the nuclear magnetic resonance sample tube, set the instrument detection parameters, start collecting the sample low-field nuclear magnetic resonance (LF-NMR) signal, use the software to perform multi-component inversion fitting on the CPMG sequence collected sample, and obtain the sample's transverse relaxation time T 2 Curve map, extract the data of characteristic parameters in the map; (ii) Calculate the water retention of the sample: Substitute the transverse relaxation time T obtained in step (i) into 2 The characteristic parameter input data in the curve graph is substituted into the pork water retention prediction model constructed by the method described in any one of claims 1 to 6, and the water retention parameter output data of the pork sample is calculated. The water retention of the pork sample is evaluated based on the water retention parameter data, and the water retention parameter is the pressurization loss rate.

Images