CN114354670A - Water and glue injection meat detection method based on MLP neural network - Google Patents
Water and glue injection meat detection method based on MLP neural network Download PDFInfo
- Publication number
- CN114354670A CN114354670A CN202210014216.8A CN202210014216A CN114354670A CN 114354670 A CN114354670 A CN 114354670A CN 202210014216 A CN202210014216 A CN 202210014216A CN 114354670 A CN114354670 A CN 114354670A
- Authority
- CN
- China
- Prior art keywords
- neural network
- glue
- peak
- sample
- water
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 59
- 239000003292 glue Substances 0.000 title claims abstract description 56
- 238000001514 detection method Methods 0.000 title claims abstract description 45
- 238000002347 injection Methods 0.000 title claims abstract description 42
- 239000007924 injection Substances 0.000 title claims abstract description 42
- 235000013372 meat Nutrition 0.000 title claims abstract description 38
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 27
- 235000015277 pork Nutrition 0.000 claims abstract description 61
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000005481 NMR spectroscopy Methods 0.000 claims abstract description 22
- 238000007689 inspection Methods 0.000 claims abstract description 10
- 239000000523 sample Substances 0.000 claims description 54
- 238000003062 neural network model Methods 0.000 claims description 27
- 238000012549 training Methods 0.000 claims description 15
- 229920001285 xanthan gum Polymers 0.000 claims description 14
- 235000010493 xanthan gum Nutrition 0.000 claims description 13
- 239000000230 xanthan gum Substances 0.000 claims description 13
- 229940082509 xanthan gum Drugs 0.000 claims description 13
- 229920001817 Agar Polymers 0.000 claims description 11
- 108010010803 Gelatin Proteins 0.000 claims description 11
- 239000008272 agar Substances 0.000 claims description 11
- 235000010419 agar Nutrition 0.000 claims description 11
- 239000008273 gelatin Substances 0.000 claims description 11
- 229920000159 gelatin Polymers 0.000 claims description 11
- 235000019322 gelatine Nutrition 0.000 claims description 11
- 235000011852 gelatine desserts Nutrition 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 11
- 235000010418 carrageenan Nutrition 0.000 claims description 10
- 229920001525 carrageenan Polymers 0.000 claims description 10
- 230000004913 activation Effects 0.000 claims description 8
- 238000005070 sampling Methods 0.000 claims description 8
- 238000001228 spectrum Methods 0.000 claims description 8
- 239000013068 control sample Substances 0.000 claims description 5
- 230000001419 dependent effect Effects 0.000 claims description 5
- 102000011990 Sirtuin Human genes 0.000 claims description 4
- 108050002485 Sirtuin Proteins 0.000 claims description 4
- 238000002592 echocardiography Methods 0.000 claims description 4
- 238000002360 preparation method Methods 0.000 claims description 4
- 238000003672 processing method Methods 0.000 claims description 3
- 239000013074 reference sample Substances 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 239000002552 dosage form Substances 0.000 claims description 2
- 238000012360 testing method Methods 0.000 abstract description 10
- 238000005516 engineering process Methods 0.000 abstract description 5
- 238000004364 calculation method Methods 0.000 abstract description 2
- 238000003908 quality control method Methods 0.000 abstract description 2
- 239000002994 raw material Substances 0.000 abstract description 2
- 235000020995 raw meat Nutrition 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 9
- 235000013305 food Nutrition 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 238000007619 statistical method Methods 0.000 description 4
- 238000005303 weighing Methods 0.000 description 3
- 238000010521 absorption reaction Methods 0.000 description 2
- 239000000679 carrageenan Substances 0.000 description 2
- 229940113118 carrageenan Drugs 0.000 description 2
- 238000007796 conventional method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000010997 low field NMR spectroscopy Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000001953 sensory effect Effects 0.000 description 2
- UHVMMEOXYDMDKI-JKYCWFKZSA-L zinc;1-(5-cyanopyridin-2-yl)-3-[(1s,2s)-2-(6-fluoro-2-hydroxy-3-propanoylphenyl)cyclopropyl]urea;diacetate Chemical compound [Zn+2].CC([O-])=O.CC([O-])=O.CCC(=O)C1=CC=C(F)C([C@H]2[C@H](C2)NC(=O)NC=2N=CC(=CC=2)C#N)=C1O UHVMMEOXYDMDKI-JKYCWFKZSA-L 0.000 description 2
- GJCOSYZMQJWQCA-UHFFFAOYSA-N 9H-xanthene Chemical compound C1=CC=C2CC3=CC=CC=C3OC2=C1 GJCOSYZMQJWQCA-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 229940023476 agar Drugs 0.000 description 1
- WQZGKKKJIJFFOK-PHYPRBDBSA-N alpha-D-galactose Chemical compound OC[C@H]1O[C@H](O)[C@H](O)[C@@H](O)[C@H]1O WQZGKKKJIJFFOK-PHYPRBDBSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 238000007865 diluting Methods 0.000 description 1
- 238000001035 drying Methods 0.000 description 1
- 239000000839 emulsion Substances 0.000 description 1
- 229930182830 galactose Natural products 0.000 description 1
- 150000002256 galaktoses Chemical class 0.000 description 1
- 239000000499 gel Substances 0.000 description 1
- 229940014259 gelatin Drugs 0.000 description 1
- 150000004676 glycans Chemical class 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 229910052739 hydrogen Inorganic materials 0.000 description 1
- 239000001257 hydrogen Substances 0.000 description 1
- 230000006698 induction Effects 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 229920001282 polysaccharide Polymers 0.000 description 1
- 239000005017 polysaccharide Substances 0.000 description 1
- 230000008054 signal transmission Effects 0.000 description 1
- 238000003307 slaughter Methods 0.000 description 1
- 239000000243 solution Substances 0.000 description 1
- 238000003756 stirring Methods 0.000 description 1
- 238000010998 test method Methods 0.000 description 1
- 230000008719 thickening Effects 0.000 description 1
Images
Landscapes
- Meat, Egg Or Seafood Products (AREA)
Abstract
The invention relates to the field of inspection and quarantine, in particular to the field of inspection and quarantine of pork products, and more particularly relates to a water and glue injection meat detection method based on an MLP neural network. The method creatively combines the low-field nuclear magnetic technology with the MLP neural network technology to be used for identifying the water-injected and glue-injected pork. The method comprises the steps of taking low-field nuclear magnetic resonance detection information of 17 low-field nuclear magnetic resonance detection parameters representing samples as input data of an MLP neural network, taking classification of raw meat and water injection and glue injection meat as output, adopting different modeling parameters, and finally establishing a three-layer MLP neural network recognition model with a structure of 17-6-3 through multiple calculations. Tests prove that the correct recognition rate of the model disclosed by the invention reaches 97.8%. The identification method disclosed by the invention provides a new idea for the rapid detection of water-injected and glue-injected meat, and has important significance for improving the quality control level of the raw material meat in China.
Description
Technical Field
The invention relates to the field of inspection and quarantine, in particular to the field of inspection and quarantine of pork products, and more particularly relates to a water and glue injection meat detection method based on an MLP neural network.
Background
Pork, as a high-protein and high-nutrition meat food, has always taken a leading position in the meat consumption structure of residents in China. However, the frequent events of injecting water and injecting glue into pork make people worry about the quality and safety of the pork more.
The water-injected pork refers to a pork fresh product formed by injecting water into the pig before or after slaughter. Illegal vendors gain weight to gain violence by this flooding approach. The conventional methods for detecting water-injected meat are relatively mature, such as sensory test, microscopic test, cooked food rate, test paper method, drying method and the like.
With the continuous updating of the manipulation of making fake, glue-injected meat is present in the market on the basis of water-injected meat, and the traditional detection methods for water-injected meat, such as sensory test methods, test paper methods and the like, are difficult to apply. The reason is that various food gums such as xanthan gum, carrageenan, gelatin, agar and the like are mainly filled in the process of making fake glue-injected meat, but people know that the main components of the food gums are galactose and dehydrated galactose which are easy to form polysaccharide gel, contain a large amount of hydrogen bonds, have high water absorption rate, can well keep the moisture in a food system, and the common detection method is difficult to identify the glue-injected meat. But simultaneously, the water absorption capacity of the pork after the glue injection can be increased by more than 20 percent, so that the meat quality is greatly reduced, and more serious, the problem of food safety can be caused.
Artificial Neural Networks (ANNs) are a technology for simulating human neural network behavior characteristics to perform distributed parallel information processing, and have good nonlinear characteristics, strong robustness and good learning and induction capabilities. The artificial neural network adopts a supervised learning mode and comprises two processes of forward transmission of signals and back propagation of errors, wherein the two processes are repeatedly and alternately executed, and the weight and the threshold of the network are continuously adjusted until the error function of the network is minimum. ANNs have the advantages that the requirement on data distribution is not strict, the functional relation between independent variables and dependent variables is not required to be expressed in detail, and the problems of abnormal distribution and nonlinear data processing can be effectively solved. However, neural networks have their own drawbacks, namely, the training time is too long and the relative importance of the input variables is difficult to distinguish in order to obtain an optimal network. Therefore, in the process of solving the problem by adopting the artificial neural network, the type of the training set data input as the input layer is crucial, and whether the model can obtain a satisfactory prediction result or not is determined, so that an effective classification effect is realized.
Disclosure of Invention
The invention aims to solve the technical problem of how to quickly, effectively and accurately identify water-injected and glue-injected meat.
In order to solve the technical problem, the invention discloses a water and glue injection meat detection method based on an MLP neural network, which is a method for inputting low-field nuclear magnetic resonance detection data of a sample to be detected into a model obtained according to the steps A-D according to the method disclosed by E so as to obtain a judgment result;
wherein the method comprises the following steps of,
a, respectively preparing a blank control sample, water-injected pork samples with different water injection amounts, glue-injected pork samples with different amounts of carrageenin, glue-injected pork samples with different amounts of gelatin, glue-injected pork samples with different amounts of agar and glue-injected pork samples with different amounts of xanthan gum as samples;
b, performing low-field nuclear magnetic resonance detection on each standard sample obtained in the step A, and selecting peak 2 area, peak 2 area ratio, peak 3 initial time, peak area sum, peak 1 area ratio, peak 1 area, peak 1 initial time, peak 2 vertex time, peak 3 ending time, peak 3 vertex time, peak 1 vertex time, peak 3 area ratio, peak 1 ending time, peak 2 ending time and single-component relaxation time obtained by data processing and analysis as input layer original variables of the three-layer MLP neural network model;
it should be noted that the low-field nuclear magnetic resonance detection is performed according to the operation instructions, and the sample to be detected should be pretreated according to the detection requirements of the low-field nuclear magnetic resonance, for example, a 32 ℃ (the working temperature of the instrument) water bath is performed for 15min before the sample tube is placed into the instrument.
C, taking the input layer as 17 original variables; the number of hidden layer units is 6, and the activation function is hyperbolic tangent; the output layer is 3 dependent variables of contrast (normal), water injection and glue injection, and a three-layer MLP neural network model with a structure of 17-6-3 is established;
d, randomly dividing the sample data in the A into a training set and an inspection set, inputting the sample data in the training set into an MLP neural network model to train the neural network, inputting the sample data in the inspection set into the MLP neural network model after the training is finished, and comparing an output value with an actual value, so that the MLP neural network model constructed in the C is further inspected and corrected, and the MLP neural network model for detecting the water-injected and glue-injected pork is obtained;
and E, processing the sample to be detected according to the processing method of the reference sample in the step A, obtaining 17 original variable values of the sample to be detected according to the method B, substituting the 17 original variable values into the MLP neural network model established in the step D, and obtaining the prediction classification of the sample to be detected by the output layer.
Preferably, the activation function is Softmax.
Meanwhile, the invention further discloses that the water injection amount of the water injection pork sample in the sample preparation A is 0.05ml/g, 0.1ml/g, 0.15ml/g and 0.2 ml/g.
Meanwhile, the invention also discloses that the mass percentages of carrageenin, gelatin and agar in the glue injection pork samples with different amounts of carrageenin, the glue injection pork samples with different amounts of gelatin and the glue injection pork samples with different amounts of agar in the sample preparation A are 5%, 12.5%, 25% and 37.5%; the mass percentages of the xanthan gum in the glue-injected pork samples added with different amounts of xanthan gum are 2.5%, 3.75%, 5% and 6.25%.
Further preferably, the B low-field nuclear magnetic resonance detection adopts CPMG (Carr-Purcell-Meiboom-Gill sequence) sequence detection, and each detection parameter is as follows: radio Frequency Delay (RFD) is 0.08ms, pre-shift stage (PRG) is 1, analog gain (RG) is 20.0db, digital gain (DRG) is 3, main frequency (SF) is 21MHz, sampling frequency (SW) is 100kHz, the number of sampling points (TD) is 1000062, latency (TW) is 20000ms, echo Time (TE) is 0.5ms, the Number of Echoes (NECH) is 10000, and the Number of Summations (NS) is 4.
Further preferably, the low-field nuclear magnetic resonance detection data is processed by a data dosage form in a multi-component inversion and single-component inversion mode to obtain a transverse relaxation spectrum and corresponding values of relaxation spectrum parameters; the inversion parameters are SIRT inversion method, the number of relaxation time points is 1000, the minimum value of relaxation time is 0.01ms, the maximum value of relaxation time is 10000ms, the number of selected data is 200, and the iteration number is 100000.
The method creatively combines the low-field nuclear magnetic technology with the MLP neural network technology to be used for identifying the water-injected and glue-injected pork. The method comprises the steps of taking low-field nuclear magnetic resonance detection information of 17 low-field nuclear magnetic resonance detection parameters representing samples as input data of an MLP neural network, taking classification of raw meat and water injection and glue injection meat as output, adopting different modeling parameters, and finally establishing a three-layer MLP neural network recognition model with a structure of 17-6-3 through multiple calculations. The test proves that the total correct recognition rate of the model disclosed by the invention to 179 samples in the test set reaches 97.8%. The identification method disclosed by the invention provides a new idea for the rapid detection of water-injected and glue-injected meat, and has important significance for improving the quality control level of the raw material meat in China.
Drawings
FIG. 1 is a low-field NMR data chart of a pork sample.
FIG. 2 is a graph of mean comparison lines of 17 LF-NMR measured parameters of pork samples after normalization.
Detailed Description
In order that the invention may be better understood, we now provide further explanation of the invention with reference to specific examples.
Firstly, preparing blank control samples, water-injected pork samples with different water injection amounts, glue-injected pork samples with different amounts of carrageenin, glue-injected pork samples with different amounts of gelatin, glue-injected pork samples with different amounts of agar and glue-injected pork samples with different amounts of xanthan gum as samples respectively;
the method specifically comprises the following steps:
(1) blank control sample: a total of 47 parts of 4.00 g of meat emulsion sample was weighed into a sample bottle as a control sample.
(2) Water injected pork samples of different water injection rates: respectively weighing 4.00 g of minced meat samples into 108 sample bottles, averagely dividing the minced meat samples into four groups, respectively injecting 0.2ml, 0.4ml, 0.6ml and 0.8ml of water into a 1ml injector, standing the mixture in a refrigerator at 4 ℃ after uniformly stirring, and measuring within 24 hours to obtain 108 samples, wherein 27 samples are totally detected in each water injection amount sample.
(2) Processing the frozen fresh pork plum slices into meat paste by a meat grinder. Weighing 4.00 g of meat paste sample in each sample, and then respectively weighing 0.2 g, 0.5 g, 1.0 g and 1.5 g of carragheenan according to different glue injection varieties; 0.2 g, 0.5 g, 1.0 g, 1.5 g gelatin; 0.2 g, 0.5 g, 1.0 g, 1.5 g agar; 0.1 g, 0.15 g, 0.2 g, 0.25 g xanthan gum; dissolving with water, diluting to 20mL (dissolving can be promoted by heating in water bath when the solution can not be dissolved at room temperature), adding into each sample bottle, and respectively preparing into rubber-injected pork samples with different amounts of carrageenin, rubber-injected pork samples with different amounts of gelatin, rubber-injected pork samples with different amounts of agar, and rubber-injected pork samples with different amounts of xanthan gum. The group was tested for 384 samples in total, with 96 xanthan, gelatin, agar and carrageenan samples, each at 28 concentrations. The addition amount of the xanthan gum is mainly considered to be strong in thickening effect of the xanthan gum, so that the xanthan gum is added in a decrement manner after the test.
B, performing low-field nuclear magnetic resonance detection on each standard sample obtained in the step A, wherein the low-field nuclear magnetic resonance detection adopts CPMG (Carr-Purcell-Meiboom-Gill sequence) sequence detection, and each detection parameter is as follows: radio Frequency Delay (RFD) is 0.08ms, pre-shift stage (PRG) is 1, analog gain (RG) is 20.0db, digital gain (DRG) is 3, main frequency (SF) is 21MHz, sampling frequency (SW) is 100kHz, the number of sampling points (TD) is 1000062, latency (TW) is 20000ms, echo Time (TE) is 0.5ms, the Number of Echoes (NECH) is 10000, and the Number of Summations (NS) is 4.
It should be noted that the low-field nuclear magnetic resonance detection is performed according to the operation instructions, and the sample to be detected should be pretreated according to the detection requirements of the low-field nuclear magnetic resonance, for example, a 32 ℃ (the working temperature of the instrument) water bath is performed for 15min before the sample tube is placed into the instrument.
The data obtained by detection needs to be processed, and in the invention, the measured data is subjected to multi-component inversion and single-component inversion by adopting a data processing function of the instrument, so that a transverse relaxation spectrum of each meat sample and a value of a corresponding relaxation spectrum parameter are obtained.
Wherein the inversion parameters are: the SIRT inversion method is selected, the number of relaxation time points is 1000, the minimum value of the relaxation time is 0.01ms, the maximum value of the relaxation time is 10000ms, the number of selected data is 200, and the iteration number is 100000.
Statistical analysis of data was performed using the IBM SPSS Statistics (Version 26, IBM) and the statistical analysis data was screened and applied in the modeling process.
It can be seen that the low-field nmr data map of the pork sample is shown in fig. 1, and the data overlap of the samples is high, and cannot be distinguished by the conventional method. The inventor of the invention uses peak 2 area, peak 2 area ratio, peak 3 initial time, peak area sum, peak 1 area ratio, peak 1 area, peak 1 initial time, peak 2 vertex time, peak 3 end time, peak 3 vertex time, peak 1 vertex time, peak 3 area ratio, peak 1 end time, peak 2 end time and single component relaxation time as the input layer original variables of the three-layer MLP neural network model through data processing and screening; the mean-versus-line plot of the 17 LF-NMR measurement parameters after normalization is shown in fig. 2, and it can be seen that the data after screening has a high significant difference among the sample groups.
C, taking the input layer as 17 original variables; the number of hidden layer units is 6, and the activation function is hyperbolic tangent; the output layer is 3 dependent variables of contrast (normal), water injection and glue injection, and a three-layer MLP neural network model with a structure of 17-6-3 is established;
d, randomly dividing the sample data in the A into a training set and an inspection set, inputting the sample data in the training set into an MLP neural network model to train the neural network, inputting the sample data in the inspection set into the MLP neural network model after the training is finished, and comparing an output value with an actual value, so that the MLP neural network model constructed in the C is further inspected and corrected, and the MLP neural network model for detecting the water-injected and glue-injected pork is obtained;
preferably, the activation function is Softmax.
In this example, 539 samples were randomly divided into a training set (70%) and a test set (30%), where 360 samples were training set samples and 179 samples were prediction set samples.
A three-layer MLP neural network model with a structure of 17-6-3 is finally established through training of a training set, namely, an input layer has 17 original variables; the number of hidden layer units is 6, and the activation function is hyperbolic tangent; the output layer is 3 dependent variables such as contrast (normal), water injection, glue injection and the like, and the activation function is Softmax.
Then, the check set is input into the model, and the reliability of the model is checked. The test results are shown in table 1:
table 1: MLP neural network model classification results
Wherein 1 represents a control sample, 2 represents water-injected pork, and 3 represents injected pork.
As can be seen from table 1, for 19 samples of the control group without water injection and glue injection, 18 samples are predicted by the model to be consistent with the actual result of water injection and glue injection, and 1 sample is misjudged to be the glue injection pork with the accuracy of 94.7%; 31 water-injected pork samples are predicted by a model to be completely consistent with the reality, and the accuracy is 100%; for 129 injected pork samples, 126 injected pork samples are predicted to be actually matched through a model, 2 injected pork samples are misjudged to be injected pork, 1 injected pork sample is misjudged to be injected pork, and the accuracy rate is 97.7%. The total accuracy reaches 97.8%.
The MLP neural network model obtained by the method is reliable in identifying the water-injected and glue-injected pork.
EXAMPLE 2 identification of samples to be tested
And (3) processing the sample to be detected according to the processing method of the reference sample in the step A, obtaining 17 original variable values of the sample to be detected according to the method B, substituting the 17 original variable values into the MLP neural network model established in the step D, and obtaining the prediction classification of the sample to be detected by the output layer.
Specifically, about 500 g of chilled pork plum slice meat is taken and processed into meat paste by a meat grinder to be used as a sample to be detected.
Then, a 32 deg.C (instrument operating temperature) water bath was applied for 15min before placing the sample tube into the instrument.
Carrying out low-field nuclear magnetic resonance detection on the processed sample to be detected, wherein the low-field nuclear magnetic resonance detection adopts CPMG (Carr-Purcell-Meiboom-Gill sequence) sequence detection, and the detection parameters are as follows: radio Frequency Delay (RFD) is 0.08ms, pre-shift stage (PRG) is 1, analog gain (RG) is 20.0db, digital gain (DRG) is 3, main frequency (SF) is 21MHz, sampling frequency (SW) is 100kHz, the number of sampling points (TD) is 1000062, latency (TW) is 20000ms, echo Time (TE) is 0.5ms, the Number of Echoes (NECH) is 10000, and the Number of Summations (NS) is 4.
And then, performing multi-component inversion and single-component inversion on the measured data by adopting a data processing function of the instrument to obtain a transverse relaxation spectrum of the sample to be detected and a value of a corresponding relaxation spectrum parameter.
Wherein the inversion parameters are: the SIRT inversion method is selected, the number of relaxation time points is 1000, the minimum value of the relaxation time is 0.01ms, the maximum value of the relaxation time is 10000ms, the number of selected data is 200, and the iteration number is 100000.
Statistical analysis of data was performed using the IBM SPSS Statistics (Version 26, IBM) and the statistical analysis data was screened and applied in the modeling process.
Inputting peak 2 area, peak 2 area ratio, peak 3 start time, peak area sum, peak 1 area ratio, peak 1 area, peak 1 start time, peak 2 vertex time, peak 3 end time, peak 3 vertex time, peak 1 vertex time, peak 3 area ratio, peak 1 end time, peak 2 end time and single component relaxation time in sample data to be detected into the MLP neural network model obtained in example 1 as input layer original variables of the three-layer MLP neural network model;
the classification result output by the output layer is as follows: normal pork.
And (3) detecting the sample to be detected according to the standard, and determining that the water content of the sample is lower than 76% and the sample is normal pork.
What has been described above is a specific embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and such improvements and modifications are also considered to be within the scope of the present invention.
Claims (6)
1. The water and glue injection meat detection method based on the MLP neural network is characterized in that: the detection method is a method for inputting low-field nuclear magnetic resonance detection data of a sample to be detected into the model obtained according to the steps A-D according to the method disclosed by E so as to obtain a judgment result;
wherein each step is as follows.
A, respectively preparing a blank control sample, water-injected pork samples with different water injection amounts, glue-injected pork samples with different amounts of carrageenin, glue-injected pork samples with different amounts of gelatin, glue-injected pork samples with different amounts of agar and glue-injected pork samples with different amounts of xanthan gum as samples;
b, performing low-field nuclear magnetic resonance detection on each standard sample obtained in the step A, and selecting peak 2 area, peak 2 area ratio, peak 3 initial time, peak area sum, peak 1 area ratio, peak 1 area, peak 1 initial time, peak 2 vertex time, peak 3 ending time, peak 3 vertex time, peak 1 vertex time, peak 3 area ratio, peak 1 ending time, peak 2 ending time and single-component relaxation time obtained by data processing and analysis as input layer original variables of the three-layer MLP neural network model;
c, taking the input layer as 17 original variables; the number of hidden layer units is 6, and the activation function is hyperbolic tangent; the output layer is 3 dependent variables of contrast, water injection and glue injection, and a three-layer MLP neural network model with a structure of 17-6-3 is established;
d, randomly dividing the sample data in the A into a training set and an inspection set, inputting the sample data in the training set into an MLP neural network model to train the neural network, inputting the sample data in the inspection set into the MLP neural network model after the training is finished, and comparing an output value with an actual value, so that the MLP neural network model constructed in the C is further inspected and corrected, and the MLP neural network model for detecting the water-injected and glue-injected pork is obtained;
and E, processing the sample to be detected according to the processing method of the reference sample in the step A, obtaining 17 original variable values of the sample to be detected according to the method B, substituting the 17 original variable values into the MLP neural network model established in the step D, and obtaining the prediction classification of the sample to be detected by the output layer.
2. The MLP neural network-based water and glue injection meat detection method as claimed in claim 1, wherein: the activation function is Softmax.
3. The MLP neural network-based water and glue injection meat detection method as claimed in claim 1, wherein: the water injection amount of the water injection pork sample in the sample preparation A is 0.05ml/g, 0.1ml/g, 0.15ml/g and 0.2 ml/g.
4. The MLP neural network-based water and glue injection meat detection method as claimed in claim 1, wherein: in the preparation of the sample A, the mass percentages of carrageenin, gelatin and agar in the glue-injected pork samples with different amounts of carrageenin, the glue-injected pork samples with different amounts of gelatin and the glue-injected pork samples with different amounts of agar are respectively 5%, 12.5%, 25% and 37.5%; the mass percentages of the xanthan gum in the glue-injected pork samples added with different amounts of xanthan gum are 2.5%, 3.75%, 5% and 6.25%.
5. The MLP neural network-based water and glue injection meat detection method as claimed in claim 1, wherein: the B low-field nuclear magnetic resonance detection adopts CPMG (Carr-Purcell-Meiboom-Gill sequence) sequence detection, and all detection parameters are as follows: radio Frequency Delay (RFD) is 0.08ms, pre-shift stage (PRG) is 1, analog gain (RG) is 20.0db, digital gain (DRG) is 3, main frequency (SF) is 21MHz, sampling frequency (SW) is 100kHz, the number of sampling points (TD) is 1000062, latency (TW) is 20000ms, echo Time (TE) is 0.5ms, the Number of Echoes (NECH) is 10000, and the Number of Summations (NS) is 4.
6. The MLP neural network-based water and glue injection meat detection method as claimed in claim 1, wherein: processing the data dosage form by adopting multi-component inversion and single-component inversion modes in low-field nuclear magnetic resonance detection data to obtain a transverse relaxation spectrum and a value of a corresponding relaxation spectrum parameter; the inversion parameters are SIRT inversion method, the number of relaxation time points is 1000, the minimum value of relaxation time is 0.01ms, the maximum value of relaxation time is 10000ms, the number of selected data is 200, and the iteration number is 100000.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210014216.8A CN114354670A (en) | 2022-01-06 | 2022-01-06 | Water and glue injection meat detection method based on MLP neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210014216.8A CN114354670A (en) | 2022-01-06 | 2022-01-06 | Water and glue injection meat detection method based on MLP neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114354670A true CN114354670A (en) | 2022-04-15 |
Family
ID=81107780
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210014216.8A Pending CN114354670A (en) | 2022-01-06 | 2022-01-06 | Water and glue injection meat detection method based on MLP neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114354670A (en) |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968438A (en) * | 2010-09-25 | 2011-02-09 | 西北农林科技大学 | Method for distinguishing water injection of raw material muscles quickly |
| CN102621172A (en) * | 2012-04-16 | 2012-08-01 | 南京农业大学 | Screening method of water-injected pork rapid detecting indexes based on nuclear magnetic resonance |
| CN106324011A (en) * | 2016-08-25 | 2017-01-11 | 江南大学 | United detection method for determinming freshness of prepared aquatic product at low temperature shelf life |
| WO2020177423A1 (en) * | 2019-03-01 | 2020-09-10 | 江南大学 | Low-field nuclear magnetism-based device and method for intelligent detection of microwave-dried spicy vegetable flavor |
| WO2020209566A1 (en) * | 2019-04-08 | 2020-10-15 | 서울대학교산학협력단 | Artificial neural network-based nuclear magnetic resonance and magnetic resonance spectroscopy data processing method and apparatus thereof |
| CN113125488A (en) * | 2021-04-21 | 2021-07-16 | 南京农业大学 | Method for quickly identifying fat-filled artificial snowflake beef |
-
2022
- 2022-01-06 CN CN202210014216.8A patent/CN114354670A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101968438A (en) * | 2010-09-25 | 2011-02-09 | 西北农林科技大学 | Method for distinguishing water injection of raw material muscles quickly |
| CN102621172A (en) * | 2012-04-16 | 2012-08-01 | 南京农业大学 | Screening method of water-injected pork rapid detecting indexes based on nuclear magnetic resonance |
| CN106324011A (en) * | 2016-08-25 | 2017-01-11 | 江南大学 | United detection method for determinming freshness of prepared aquatic product at low temperature shelf life |
| WO2020177423A1 (en) * | 2019-03-01 | 2020-09-10 | 江南大学 | Low-field nuclear magnetism-based device and method for intelligent detection of microwave-dried spicy vegetable flavor |
| WO2020209566A1 (en) * | 2019-04-08 | 2020-10-15 | 서울대학교산학협력단 | Artificial neural network-based nuclear magnetic resonance and magnetic resonance spectroscopy data processing method and apparatus thereof |
| CN113125488A (en) * | 2021-04-21 | 2021-07-16 | 南京农业大学 | Method for quickly identifying fat-filled artificial snowflake beef |
Non-Patent Citations (3)
| Title |
|---|
| 夏阿林;夏霞明;吉琳琳;赵良忠;: "低场核磁共振结合化学模式识别方法判别休闲豆干品牌", 农业工程学报, no. 10, 23 May 2018 (2018-05-23) * |
| 杨志敏 等: "应用近红外技术快速鉴别原料肉注水的研究", 食品研究与开发, vol. 33, no. 5, 31 May 2012 (2012-05-31), pages 118 - 120 * |
| 盖圣美;张中会;游佳伟;邹玉峰;刘登勇;: "低场核磁共振技术结合化学计量学方法定性、定量检测注水猪肉", 食品科学, no. 04, 25 February 2020 (2020-02-25) * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111722199B (en) | Radar signal detection method based on convolutional neural network | |
| CN109725016B (en) | Nuclear magnetic resonance laboratory measurement method for rock core containing heavy oil and asphaltene | |
| CN109033616B (en) | Method for predicting quality change of frozen pork in storage process | |
| CN103543123A (en) | Infrared spectrum recognition method for adulterated milk | |
| CN112730499B (en) | Method for detecting fresh meat moisture quality and/or moisture content through low-field nuclear magnetic resonance and preparation method and application of standard sample | |
| CN101968438B (en) | Method for distinguishing water injection of raw material muscles quickly | |
| CN113743682B (en) | NMR (nuclear magnetic resonance) relaxation time inversion method based on supervised deep neural network | |
| Xia et al. | Simultaneous, rapid and nondestructive determination of moisture, fat content and storage time in leisure dried tofu using LF-NMR | |
| CN107632010A (en) | A kind of quantitative approach of combination LIBS to steel samples | |
| CN110596176A (en) | Method for detecting water content in aquatic product by using low-field nuclear magnetic resonance technology | |
| CN112731377A (en) | Dielectric constant inversion method, roadbed detection method and detection system | |
| Bizzani et al. | Non-invasive detection of internal flesh breakdown in intact Palmer mangoes using time-domain nuclear magnetic resonance relaxometry | |
| CN112525942A (en) | Method for rapidly and nondestructively detecting moisture content and drying degree of green tea in drying process based on low-field nuclear magnetic resonance technology | |
| Hou et al. | Mechanical properties of mulberry fruit under compression and nuclear magnetic resonance tests | |
| CN105445307B (en) | The low-field nuclear magnetic resonance detection method of texture quality in a kind of sea cucumber salting process | |
| Tan et al. | Actual time determination of egg freshness: A centroid rate based approach | |
| WO2021089001A1 (en) | Method, device, and system for detecting sweetness of fruit, and storage medium | |
| CN109322655A (en) | A kind of microwave hydro rate detection device and method based on neural network Yu double frequency difference model | |
| CN114354670A (en) | Water and glue injection meat detection method based on MLP neural network | |
| CN116593373A (en) | Method and system for detecting porosity and pore distribution of cigarettes | |
| CN106338526B (en) | A kind of correction model and detection method based on microwave moisture instrument | |
| CN115330150A (en) | Method for predicting sensory quality of crystal balls based on crystal ball indexes and application | |
| Zompanti et al. | A Sensor System for Non-Destructive Monitoring of Food Ripening Processes | |
| DE102004014338A1 (en) | Use of electromagnetic field at microwave frequencies, to detect food freshness, e.g. of fish, involves comparing measured and reference energy variations for statistical evaluation | |
| CN112067577A (en) | Method, device and equipment for identifying overproof cream pigment based on support vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |