Method for realizing meat freshness prediction based on neural network in combination with quality index

Abstract

The invention discloses a method for realizing meat freshness prediction based on a neural network in combination with quality indexes, which comprises the following steps: S1, data acquisition: acquiring quality indexes, freshness indexes, flavor indexes and biomarker indexes of protein and amine degradation of different freshness samples; s2, performing data processing to obtain a data set; s3, constructing a neural network model; and S4, predicting the freshness of the meat product by using the neural network model constructed in the step S3. According to the method, the prediction accuracy and efficiency are greatly improved. Traditional prediction often depends on sensory evaluation or chemical analysis, and the methods are time-consuming and labor-consuming and are easily influenced by human factors. Compared with the prior art, the neural network can automatically learn and extract key quality indexes in meat products, and establish a direct relation with freshness. Through the training of a large amount of data, the neural network model can accurately predict the freshness of the meat product, the prediction speed is extremely high, and the requirements of large-scale and rapid detection are met.

Description

Technical Field

[0001] The invention belongs to the field of food detection, and specifically is a method for predicting the freshness of meat based on a neural network combined with quality indicators. Background Art

[0002] As people pay more and more attention to food safety and quality, meat freshness identification technology has become a hot topic of research. Although traditional methods for meat freshness identification, such as sensory evaluation, chemical analysis and microbiological testing, are effective, they have limitations such as long testing cycle, high cost and complex operation. Disadvantages of existing technologies:

[0003] ① High testing costs: Traditional methods such as sensory evaluation, chemical analysis and microbiological testing may require expensive equipment and reagents, as well as professional technicians to operate, resulting in high testing costs. However, neural networks can be combined with quality indicators for prediction, and can achieve automated testing through software, reducing equipment costs and labor costs.

[0004] ② Low detection efficiency: Traditional methods often require testing each sample one by one, which is time-consuming and cannot meet the needs of large-scale and rapid testing. However, the neural network model can realize parallel processing of multiple samples, significantly improving detection efficiency.

[0005] ③ Strong subjectivity: Sensory evaluation and other methods are easily affected by the subjective factors of the assessors, resulting in inconsistent results. However, neural networks combine quality indicators for prediction, based on objective data and algorithms, which can reduce the interference of human factors and improve the objectivity and consistency of the results.

[0006] ④ Limited generalization ability: Traditional methods may only be applicable to the detection of meat freshness under specific types or specific conditions. For meat of different types or different storage conditions, it is necessary to re-establish detection methods and standards. However, the neural network model has strong adaptive learning ability and can adapt to the prediction needs of meat freshness of different types and different storage conditions, thus improving the generalization ability.

[0007] Therefore, it is particularly important to develop a fast, accurate and efficient meat freshness prediction technology.

[0008] Among the existing technologies, there are technical solutions that combine neural networks with quality indicators to predict the freshness of meat. Neural networks, especially deep learning technology, have achieved certain results in the fields of image recognition and classification, providing new ideas for predicting the freshness of meat. Among them:

[0009] ① Convolutional neural network model for meat freshness prediction:

[0010] Convolutional neural network (CNN) is a neural network model that is particularly suitable for image processing. It can extract feature information from images through structures such as convolutional layers and pooling layers. In the prediction of meat freshness, CNN can receive meat image input, extract image features through layer-by-layer convolution and pooling operations, and finally output the meat freshness prediction result.

[0011] ② Radial basis function model for meat freshness prediction:

[0012] The radial basis function (RBF) model is a neural network model based on the radial basis function, which usually includes an input layer, a hidden layer, and an output layer. In the prediction of meat freshness, the RBF model can receive the quality index data of meat (such as temperature, humidity, odor, etc.), undergo nonlinear transformation of the hidden layer, and finally output the prediction result of meat freshness.

[0013] The above models have certain deficiencies in dealing with complex nonlinear relationships, prediction accuracy and generalization ability. How to select accurate input features, build a suitable model to deal with complex nonlinear problems, and improve prediction accuracy and generalization ability are technical problems that need to be solved urgently.

[0014] The purpose of the present invention is to provide a more economical, efficient, objective and generalizable method for predicting the freshness of meat. By combining neural networks with quality indicators for prediction, the present invention aims to achieve the following goals:

[0015] Reduce testing costs: Utilize the software implementation and automated detection features of the neural network model to reduce the need for expensive equipment and reagents and reduce testing costs.

[0016] Improve detection efficiency: Through the parallel processing capabilities of the neural network model, rapid detection of multiple meat samples can be achieved, thereby improving detection efficiency.

[0017] Enhance the objectivity of results: Make predictions based on objective data and algorithms, reduce interference from human factors, and improve the objectivity and consistency of results.

[0018] Improve generalization ability: Through the adaptive learning ability of the neural network model, it can adapt to the freshness prediction needs of meat of different types and storage conditions and improve the generalization ability.

[0019] Technical solution:

[0020] A method for predicting meat freshness based on a neural network combined with quality indicators, comprising:

[0021] S1. Data collection: collecting quality indicators, freshness indicators, flavor indicators, and biomarker indicators of protein and amine degradation of samples of different freshness;

[0022] S2, data processing, processing the data collected in S1 to obtain a data set;

[0023] S3, neural network model construction, using back propagation artificial neural network BP-ANN as the model framework, quality indicators, freshness indicators, and flavor indicators as the input layer; using biomarker indicators of protein and amine degradation as the output layer; dividing the data set obtained in S2 into training set and test set training model to obtain the neural network model;

[0024] S4. Use the neural network model built in S3 to predict the freshness of meat.

[0025] Preferably, the quality index is color difference and texture; the freshness index is pH value and thiobarbituric acid value TBARS; the flavor index is the value measured by electronic nose; the biomarker index of protein and amine degradation is total volatile basic nitrogen TVB-N.

[0026] Preferably, a colorimeter is used to measure the color difference of the samples.

[0027] Preferably, the sample texture is measured using a texture analyzer.

[0028] Preferably, the pH value is measured by the following steps:

[0029] Take 5 g of sample in a beaker, add 50 mL of distilled water, homogenize and centrifuge for 10 min at 7000 r / min, take the supernatant and use a pH meter to measure the pH value, and repeat the measurement 3 times for each sample.

[0030] Preferably, the experimental value of total volatile basic nitrogen TVB-N is obtained by microdiffusion method.

[0031] Preferably, thiobarbituric acid value TBARS is obtained by the following formula:

[0032]

[0033] Wherein, A is the absorbance at 532 nm; V is the volume of the sample solution, in mL; M is the molar mass of MDA; ε is the molar absorption coefficient; L is the optical path; and m is the mass of the meat sample, in g.

[0034] Preferably, the electronic nose obtains the measured value by the following steps:

[0035] Using a cNose electronic nose, 5 g of pork patty sample was placed in a gas collection headspace bottle and balanced for 15 minutes to allow sufficient odor entry. Each group of samples was measured three times. The measurement parameters were as follows: pre-sampling time 5 s, sampling time interval 1 s, self-cleaning time 90 s, feed flow rate 300 mL / min, and sample measurement time 90 s.

[0036] Preferably, in step S2, the data processing refers to: obtaining the original initial data from the response signal detected by the electronic nose, using origin to perform smoothing preprocessing on the collected sensor response signal through the Savitzky-Golay smoothing method; then using the K-means clustering method of SPSS to extract the eigenvalues ​​of the electronic nose response signal; using cluster analysis to divide the 150s detection data of each sample into 6 clusters, and extracting the final cluster center of each cluster as the eigenvalue as the input layer data of the model.

[0037] Specifically, during the construction of the neural network model, the parameters are set as follows:

[0038] Create a network: newff function, 6 hidden layers;

[0039] Set training parameters: number of iterations = 1000, error threshold = 10-6, learning rate = 0.01.

[0040] Beneficial Effects of the Invention

[0041] The use of neural networks combined with quality indicators to predict meat freshness has significant technical advantages and effectively solves many technical problems in traditional prediction methods.

[0042] First, this method greatly improves the accuracy and efficiency of predictions. Traditional predictions often rely on sensory evaluation or chemical analysis, which are not only time-consuming and labor-intensive, but also easily affected by human factors. In contrast, neural networks can automatically learn and extract key quality indicators in meat, such as texture, color, pH, total volatile basic nitrogen, flavor, thiobarbituric acid, etc., and establish a direct connection with freshness. Through training with a large amount of data, the neural network model can accurately predict the freshness of meat, and the prediction speed is extremely fast, meeting the needs of large-scale and rapid testing.

[0043] Secondly, the prediction method of neural network combined with quality indicators is highly objective and generalizable. The sensory evaluation in traditional prediction methods is often affected by the subjective experience of the evaluators, resulting in inconsistent results. The neural network prediction is based on objective data and algorithms, avoiding the interference of human factors, making the prediction results more reliable and consistent. At the same time, the neural network model has a strong adaptive learning ability, and can be optimized and updated according to the data of meat products of different types and storage conditions, which improves the generalization ability of the model and enables it to be widely used in various meat freshness prediction scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 The neural network topology diagram constructed for the present invention

[0045] Figure 2Constructing a flow chart for the neural network model of the present invention Figure 3 Schematic diagram of the model training set training process in the embodiment Figure 4 Schematic diagram of the model test set verification process in the embodiment DETAILED DESCRIPTION

[0046] The present invention will be further described below in conjunction with embodiments, but the protection scope of the present invention is not limited thereto:

[0047] Combination Figure 2 , a method for predicting meat freshness based on neural network combined with quality indicators, comprising:

[0048] S1. Data collection: collecting quality indicators, freshness indicators, flavor indicators, and biomarker indicators of protein and amine degradation of samples of different freshness;

[0049] S2, data processing, processing the data collected in S1 to obtain a data set;

[0050] S3, neural network model construction, using back propagation artificial neural network BP-ANN as the model framework, quality indicators, freshness indicators, and flavor indicators as the input layer; using biomarker indicators of protein and amine degradation as the output layer; dividing the data set obtained in S2 into training set and test set training model to obtain the neural network model;

[0051] S4. Use the neural network model built in S3 to predict the freshness of meat.

[0052] In this embodiment, the quality indicators are: color difference (using a colorimeter to measure the color difference of the sample) and texture (using a texture analyzer to measure the texture of the sample); the freshness indicators are: pH value (slightly modified based on GB5009.237-2016. Take 5g of the sample in a beaker, add 50mL of distilled water, homogenize and centrifuge for 10min at a speed of 7000r / min, take the supernatant and use a pH meter to measure the pH value. Each sample is repeated 3 times), thiobarbituric acid value TBARS (weigh 5g of pork patty sample, add 15mL 20g / 100mL trichloroacetic acid was homogenized with a homogenizer for 35s, centrifuged at 2000r / min for 10min, 2.5mL of supernatant was drawn and added with 2.5mL of 0.02mol / L thiobarbituric acid solution, and the mixture was taken out after being placed in a water bath at 91°C for 40min. The mixture was quickly cooled in a water bath, and its absorbance was measured at a wavelength of 532nm using an ultraviolet spectrophotometer (UV-2800, Unico Instrument Co., Ltd.), and the TBARS value was expressed as the mass of malondialdehyde contained in each kilogram of meat sample); the flavor index was the value measured by an electronic nose; the biomarker index for protein and amine degradation was the total volatile basic nitrogen TVB-N (obtained by microdiffusion method).

[0053] Preferably, thiobarbituric acid value TBARS is obtained by the following formula:

[0054]

[0055] Wherein, A is the absorbance at 532 nm; V is the volume of the sample solution, in mL; M is the molar mass of MDA (72.063 g / mol); ε is the molar absorption coefficient (156000 L / (mol / cm)); L is the optical path (1 cm); and m is the mass of the meat sample, in g.

[0056] Preferably, the electronic nose obtains the measured value by the following steps:

[0057] A cNose electronic nose (Shanghai Baosheng Industrial Development Co., Ltd.) was used to take 5 g of pork patty sample and put it in a gas collection headspace bottle. The sample was balanced for 15 minutes to allow sufficient odor entry. Each group of samples was measured three times. The measurement parameters were as follows: pre-sampling time 5 s, sampling time interval 1 s, self-cleaning time 90 s, feed flow rate 300 mL / min, and sample measurement time 90 s.

[0058] Preferably, in step S2, the data processing refers to: obtaining the original initial data from the response signal detected by the electronic nose, using origin to perform smoothing preprocessing on the collected sensor response signal through the Savitzky-Golay smoothing method; then using the K-means clustering method of SPSS to extract the eigenvalues ​​of the electronic nose response signal; using cluster analysis to divide the 150s detection data of each sample into 6 clusters, and extracting the final cluster center of each cluster as the eigenvalue as the input layer data of the model.

[0059] Specifically, in the process of building the neural network model, combined with Figure 1 , using Matlab software, including:

[0060] (1) Dataset Division

[0061] Input layer: electronic nose data, texture, color difference, pH, TBARS

[0062] Output layer: TVB-N

[0063] 80% for training set, 20% for testing set

[0064] (2) Data normalization

[0065] (3) Create a network: newff function, 6 hidden layers

[0066] (4) Set training parameters: number of iterations = 1000, error threshold = 10-6, learning rate = 0.01

[0067] (5) Training the network

[0068] (6) Simulation test: sim function

[0069] (7) Data denormalization: mapminmax function

[0070] (8) Calculate the root mean square error: sqrt function

[0071] The model training set training process is as follows Figure 3 As shown, the root mean square error is 0.2092, the mean absolute error is 0.1653, and the accuracy is 96.09%. The model test set verification process is as follows Figure 4 As shown, the root mean square error is 0.2637, the mean absolute error is 0.2225, and the accuracy is 93.50%.

[0072] Advantages of this solution:

[0073] The 6-layer BP neural network constructed with the newff function is used to predict the freshness of pork patties. Compared with other neural network prediction models, it has the advantages of high flexibility, improved prediction accuracy, and strong generalization ability.

[0074] (1) High flexibility: BP neural network, especially the complex network structure (such as 6 layers) constructed by newff function, can simulate complex nonlinear relationships. This flexibility enables BP neural network to handle multiple influencing factors and variables in pork patty freshness prediction.

[0075] (2) Prediction accuracy: By optimizing weights and bias values ​​during the training process, the BP neural network can obtain accurate prediction results. During the training process, network parameters (such as learning rate, number of iterations, etc.) can be adjusted to further improve prediction accuracy.

[0076] (3) Generalization ability: The BP neural network can not only make predictions on the training data, but also make effective predictions on unknown data. This means that even when faced with pork pie samples that have never appeared in the training set, the 6-layer BP neural network can still give reasonable prediction results. Compared with other models:

[0077] Compared with traditional statistical models, BP neural network has significant advantages in dealing with nonlinear relationships and can better simulate the complex relationship between pork patty freshness and influencing factors.

[0078] (1) Convolutional Neural Network

[0079] Compared with other neural network models (such as convolutional neural network (CNN), although BP neural network may not be efficient enough in processing image data, it may be more flexible and applicable when processing non-image data (such as sensor data, chemical indicators, etc.). In addition, BP neural network also has greater flexibility in model structure and parameter adjustment. In summary, the 6-layer BP neural network constructed with newff function has the advantages of high flexibility, prediction accuracy, generalization ability, and convenience of visualization and debugging in the prediction of pork patty freshness. These advantages make BP neural network a powerful tool in the field of pork patty freshness prediction.

[0080] (2) Radial Basis Function Neural Network

[0081] Sensitive parameter selection: The performance of RBF neural network depends largely on its parameter settings, such as the number of hidden layer neurons, the width of radial basis function, etc. The selection of these parameters needs to be determined by experience or experiments, which increases the complexity of model construction.

[0082] Limited generalization ability: Compared with BP neural network, RBF neural network may have weaker generalization ability. It may be more dependent on specific features in the training data, so its prediction performance may not be as good as BP neural network when facing unknown or new data.

[0083] RBF neural network is more suitable for processing situations with obvious local features and relatively simple problems. However, 6-layer BP neural network is more suitable for processing complex problems with strong nonlinear relationships, such as the complex relationship between multiple influencing factors and variables in the prediction of pork pie freshness.

[0084] (3) Support Vector Machine Model

[0085] Sensitive parameter selection: The performance of SVM depends largely on its parameter settings, such as penalty parameter C and kernel function parameters, etc. The selection of these parameters needs to be determined by experience or experiments, which increases the complexity of model construction.

[0086] High computational complexity: When the data set is large, the training process of SVM may become very time-consuming because a large-scale quadratic programming problem needs to be solved.

[0087] Not suitable for regression problems: Although SVM can handle regression problems (i.e., SVR) by introducing an ε-insensitive loss function, its performance may be inferior to that of models specifically designed for regression tasks in some cases.

[0088] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for predicting meat freshness based on a neural network combined with quality indicators, characterized in that It includes: S1. Data collection: collecting quality indicators, freshness indicators, flavor indicators, and biomarker indicators of protein and amine degradation of samples of different freshness; S2, data processing, processing the data collected in S1 to obtain a data set; S3, neural network model construction, using back propagation artificial neural network BP-ANN as the model framework, quality indicators, freshness indicators, and flavor indicators as the input layer; using biomarker indicators of protein and amine degradation as the output layer; dividing the data set obtained in S2 into training set and test set training model to obtain the neural network model; S4. Use the neural network model built in S3 to predict the freshness of meat.

2. The method according to claim 1, characterized in that The quality index is color difference and texture; the freshness index is pH value and thiobarbituric acid value TBARS; the flavor index is the value measured by electronic nose; the biomarker index of protein and amine degradation is total volatile basic nitrogen TVB-N.

3. The method according to claim 2, characterized in that Use a colorimeter to measure the color difference of the samples.

4. The method according to claim 2, characterized in that The texture of the samples was measured using a texture analyzer.

5. The method according to claim 2, characterized in that The pH value is obtained by the following steps: Take 5 g of sample in a beaker, add 50 mL of distilled water, homogenize and centrifuge for 10 min at 7000 r / min, take the supernatant and use a pH meter to measure the pH value, and repeat the measurement 3 times for each sample.

6. The method according to claim 2, characterized in that The experimental value of total volatile basic nitrogen TVB-N was obtained by microdiffusion method.

7. The method according to claim 2, characterized in that Thiobarbituric acid value TBARS is obtained by the following formula: Wherein, A is the absorbance at 532 nm; V is the volume of the sample solution, in mL; M is the molar mass of MDA; ε is the molar absorption coefficient; L is the optical path; and m is the mass of the meat sample, in g.

8. The method according to claim 2, characterized in that The electronic nose obtains the measured value through the following steps: Using a cNose electronic nose, 5 g of pork patty sample was placed in a gas collection headspace bottle and balanced for 15 minutes to allow sufficient odor entry. Each group of samples was measured three times. The measurement parameters were as follows: pre-sampling time 5 s, sampling time interval 1 s, self-cleaning time 90 s, feed flow rate 300 mL / min, and sample measurement time 90 s.

9. The method according to claim 1, characterized in that In step S2, the data processing refers to: detecting the response signal obtained by the electronic nose to obtain the original initial data, using origin to perform smoothing preprocessing on the collected sensor response signal through the Savitzky-Golay smoothing method; and then using the K-means clustering method of SPSS to extract the characteristic value of the electronic nose response signal; The 150s detection data of each sample were divided into 6 clusters using cluster analysis, and the final cluster center of each cluster was extracted as the eigenvalue as the input layer data of the model.

10. The method according to claim 1, characterized in that During the construction of the neural network model, the parameters are set as follows: Create a network: newff function, 6 hidden layers; Set training parameters: number of iterations = 1000, error threshold = 10-6, learning rate = 0.01.

Images