Food processing technology optimization method based on fuzzy mathematics and deep learning prediction

Abstract

The embodiment of the invention discloses a food processing technology optimization method based on fuzzy mathematics and deep learning prediction. A specific embodiment of the method comprises the following steps: acquiring N groups of response surface test data of a response surface test; constructing a response surface equation according to the N groups of response surface test data; according to the constructed response surface equation, expansion is carried out on the basis of the N groups of response surface test data, and M groups of virtual test data are generated; determining the N groups of response surface test data and the M groups of virtual test data as a sample set; executing a training step based on the sample set; obtaining a deep learning model based on training, and executing a prediction step; and in the second prediction sensory score output by the deep learning model, based on the target second prediction sensory score, determining a selected processing factor parameter value of the target meat. Therefore, in the scene of industrial processing of the target meat, higher reliability and accuracy are realized in the aspects of sensory prediction and accurate control of the processing technology.

Description

Technical Field

[0001] The present disclosure relates to the field of food processing technologies, and in particular to an optimization method for food processing technologies based on fuzzy mathematics and deep learning prediction. Background Art

[0002] In the industrial production process of complex foods such as prefabricated dishes, due to numerous cooking process parameters and complex interaction relationships, and the lack of a unified standardized control system, problems such as flavor imbalance and quality fluctuations often occur, seriously affecting product consistency and production efficiency. At the same time, the traditional method relying on manual sensory evaluation, although regarded as the "gold standard" for flavor quality judgment, has limitations such as strong subjectivity, long cycle, and high cost, and it is difficult to meet the requirements of real-time, stability, and reproducibility in modern production, becoming an obstacle restricting the large-scale development of the prefabricated food industry. Summary of the Invention

[0003] This part of the disclosure is provided to introduce concepts in a brief form, and these concepts will be described in detail in the following detailed implementation part. This part of the disclosure is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0004] First aspect, embodiments of the present disclosure provide a method for optimizing food processing technology based on fuzzy mathematics and deep learning prediction. The method includes: obtaining N groups of response surface test data of response surface tests, where the response surface test includes processing target meat based on multiple target processing factors to obtain target meat products, and each group of test data in the N groups of response surface test data includes multiple target processing factor parameter values and a first sensory score, where the first sensory score is the sensory score of the target meat product processed based on the multiple target processing factor parameter values; constructing a response surface equation according to the N groups of response surface test data; expanding based on the constructed response surface equation on the basis of the N groups of response surface test data to generate M groups of virtual test data; determining the N groups of response surface test data and the M groups of virtual test data as a sample set; based on the sample set, performing a training step, where the training step includes: importing the multiple target processing factor parameter values in the sample into an untrained deep learning network to obtain a first predicted sensory score output by the deep learning network; generating a loss value based on the first predicted sensory score and the first sensory score in the sample; updating the weight values in the deep learning network based on the loss value until a preset stop training condition is reached to obtain a deep learning model; based on the trained deep learning model, performing a prediction step, where the prediction step includes: obtaining parameter values of processing factors to be verified, and importing the parameter values of processing factors to be verified into the trained deep learning model to obtain a second predicted sensory score output by the deep learning model; determining a target second predicted sensory score that meets the conditions from the second predicted sensory scores output by the deep learning model; determining the selected processing factor parameter values of the target meat based on the parameter values of processing factors to be verified corresponding to the target second predicted sensory score.

[0005] Second aspect, embodiments of the present disclosure provide a method for optimizing food processing technology, including: designing and implementing a response surface test for target processing factors of target meat, where the response surface test includes processing target meat based on multiple target processing factors to obtain target meat products; obtaining the test results of the response surface test; based on N groups of response surface test data of the response surface test, performing the method according to any item in the first aspect to obtain a selected processing factor parameter group of the target meat; where each group of test data in the N groups of response surface test data includes multiple target processing factor parameter values and a first sensory score, where the first sensory score is the sensory score of the target meat product processed based on the multiple target processing factor parameter values. Description of the Drawings

[0006] In combination with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages, and aspects of the embodiments of the present disclosure will become more apparent. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic and that the original elements and elements are not necessarily drawn to scale.

[0007] Figure 1 It is a flowchart of an embodiment of a method for optimizing food processing technology based on fuzzy mathematics and deep learning prediction according to the present disclosure;

[0008] Figure 2 It is a flowchart of an embodiment of a method for optimizing food processing technology according to the present disclosure;

[0009] Figure 3 It is the sensory score curves of making mother wine chicken by frying, precooking, and stewing for different times, where Figure 3 (a) is the sensory score curve of making mother wine chicken by frying for different times, Figure 3 (b) is the sensory score curve of making mother wine chicken by precooking for different times, Figure 3 (c) is the sensory score curve of making mother wine chicken by stewing for different times;

[0010] Figure 4 It is the quality change and yield rate curves of mother wine chicken before and after frying, precooking, and stewing, where Figure 4 (a) is the quality change and yield rate curve of mother wine chicken before and after frying, Figure 4 (b) is the quality change and yield rate curve of mother wine chicken before and after precooking, Figure 4 (c) is the quality change and yield rate curve of mother wine chicken before and after stewing;

[0011] Figure 5 It is the change curve of the moisture content of mother wine chicken by frying, precooking, and stewing for different times;

[0012] Figure 6 It is the pH change curves of mother wine chicken soup and chicken by frying, precooking, and stewing for different times. Among them, Figure 6 (a) is the pH change curve of mother wine chicken soup by frying, precooking, and stewing for different times, Figure 6 (b) is the pH change curve of mother wine chicken by frying, precooking, and stewing for different times;

[0013] Figure 7 It is the change curve of the color difference of chicken skin by frying, precooking, and stewing for different times. Among them, Figure 7 (a) is the change curve of the color difference of chicken skin by frying for different times, Figure 7 (b) is the change curve of the color difference of chicken skin by precooking for different times, Figure 7 (c) is the change curve of the color difference of chicken skin by stewing for different times;

[0014] Figure 8 The change curves of the color difference values of chicken for different stir-frying, precooking, and stewing times, where Figure 8 (a) is the change curve of the color difference value of chicken for different stir-frying times, Figure 8 (b) is the change curve of the color difference value of chicken for different precooking times, Figure 8 (c) is the change curve of the color difference value of chicken for different stewing times;

[0015] Figure 9 are the loss curves of the training set and the validation set;

[0016] Figure 10 is the sensory score response surface model of the mother rice wine chicken;

[0017] Figure 11 is the comparison curve between the true value and the predicted value. Specific embodiments

[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Instead, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0019] It should be understood that the various steps recited in the method embodiments of the present disclosure can be executed in a different order and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this regard.

[0020] As used herein, the term "including" and its variants are open-ended, i.e., "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the following description.

[0021] It should be noted that the concepts such as "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules, or units, and are not used to limit the order of the functions executed by these devices, modules, or units or their interdependent relationships.

[0022] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive. Those skilled in the art should understand that unless clearly stated otherwise in the context, it should be understood as "one or more".

[0023] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are for illustrative purposes only and are not used to limit the scope of these messages or information.

[0024] In one or more embodiments of the present disclosure, through single-factor experiments on key process parameters combined with the Box-Behnken response surface method to design experiments, real data samples are fused with the constructed virtual data samples, and mathematical model adaptive learning and fitting training are carried out to construct a fuzzy sensory evaluation - deep neural network (DNN) integrated model, which breakthroughly solves the problems that traditional sensory evaluation is difficult to quantify and predict in real time.

[0025] In one or more embodiments of the present disclosure, this model is successfully applied to the optimization of the processing technology of Hakka mother rice wine chicken, effectively identifying the non-linear mapping relationship between key process parameters and sensory quality. The optimal processing parameters obtained after optimization are stir-frying time of 4 minutes, pre-boiling time of 6 minutes, and stewing time of 60 minutes. Based on the established DNN model, the relative error between the predicted optimal sensory score and the actual score is only 2.25%, which is significantly better than the prediction accuracy of the traditional response surface model (the relative error with the actual score is 4.29%), achieving higher reliability and accuracy in flavor quality prediction and precise control of processing technology. This technical solution not only realizes the standardization and optimization of the processing technology of Hakka mother rice wine chicken, significantly improves the sensory quality and production stability of the product, but also successfully promotes its progress towards large-scale and process-based industrial production. While ensuring the consistency of product quality, it greatly improves the food processing efficiency and meets the comprehensive requirements of the modern food industry for high quality, high efficiency, and low cost.

[0026] In addition, the technical path proposed in the present disclosure has good expandability and generality, and can be further extended to scenarios such as multi-objective optimization, multi-cooking methods, and cross-category transfer learning in the future to construct an intelligent processing optimization platform for multi-category foods. This solution provides forward-looking technical support for intelligent manufacturing and industrial upgrading in the food processing field.

[0027] Please refer to Figure 1 , which shows the flow of an embodiment of the food processing technology optimization method based on fuzzy mathematics and deep learning prediction according to the present disclosure. As Figure 1 shown, the food processing technology optimization method based on fuzzy mathematics and deep learning prediction includes the following steps:

[0028] Step 101, obtain N groups of response surface test data of the response surface test.

[0029] In this embodiment, the execution subject (such as a server and / or a terminal device) of the food processing technology optimization method based on fuzzy mathematics and deep learning prediction can obtain N groups of response surface test data of the response surface test.

[0030] In this embodiment, the response surface test includes processing target meat based on multi-objective processing factors to obtain a target meat product.

[0031] In this embodiment, each group of test data in the N groups of response surface test data includes multi-objective processing factor parameter values and a first sensory score.

[0032] In this embodiment, the first sensory score is the sensory score of the target meat product processed based on the multi-objective processing factor parameter values.

[0033] Step 102, construct a response surface equation according to the N groups of response surface test data.

[0034] Step 103, based on the constructed response surface equation, expand on the basis of the N groups of response surface test data to generate M groups of virtual test data.

[0035] Step 104, determine the N groups of response surface test data and the M groups of virtual test data as a sample set.

[0036] Step 105, perform a training step based on the sample set.

[0037] In this embodiment, the training step includes: importing the multi-objective processing factor parameter values in the sample into an untrained deep learning network to obtain a first predicted sensory score output by the deep learning network; generating a loss value based on the first predicted sensory score and the first sensory score in the sample; updating the weight values in the deep learning network based on the loss value until a preset stop training condition is reached to obtain a deep learning model.

[0038] Step 106, perform a prediction step based on the trained deep learning model.

[0039] In this embodiment, the above prediction step includes: obtaining the to-be-verified processing factor parameter values, and importing the to-be-verified processing factor parameter values into the trained deep learning model to obtain a second predicted sensory score output by the deep learning model.

[0040] Step 107, determine the target second predicted sensory score that meets the conditions from the second predicted sensory scores output by the deep learning model; determine the to-be-verified processing factor parameter values corresponding to the target second predicted sensory score as the selected processing factor parameter values of the target meat.

[0041] The target meat can be any kind of meat, for example, beef, pork, chicken, etc. Optionally, the target meat can be chicken.

[0042] The target meat is processed based on the parameter values of multiple target processing factors to obtain a target meat product. Evaluators evaluate the sensory aspects of the target meat product to obtain a sensory score.

[0043] The specific value of N in the N groups of response surface test data can be selected according to the actual application scenario and is not limited here.

[0044] The target processing factors in each of the N groups are the same, but the parameter values of the target processing factors are different. Each set of target processing factor parameter values corresponds to a sensory score.

[0045] As an example, taking the processing of chicken as an example, the target processing factors can include frying time, pre-boiling time, and stewing time. The first set of response surface test data in the N groups of response surface test data can include a frying time of 2 minutes, a pre-boiling time of 4 minutes, and a stewing time of 60 minutes; the first set of response surface test data can also include a first sensory score of 7.0. The first set of response surface test data in the N groups of response surface test data can include a frying time of 4 minutes, a pre-boiling time of 4 minutes, and a stewing time of 60 minutes; the first set of response surface test data can also include a first sensory score of 7.2.

[0046] Taking multiple target processing factors as independent variables and the first sensory score as the response value, a response surface equation is constructed. The response surface equation can also be called a response surface model.

[0047] The response surface method (RSM) is an optimization method that combines experimental design and mathematical modeling, which can effectively reduce the number of experiments and can also examine the interaction between influencing factors. The response surface equation can be a mathematical model obtained based on experimental design and experimental results. The response surface equation can fit the relationship between independent variables and response values.

[0048] The specific structure of the deep learning network can be set according to the actual application scenario and is not limited here.

[0049] As an example, the deep learning network can adopt a neural network with an RNN or DNN structure. The neural network can include three hidden layers, and the number of nodes in the three hidden layers is 256, 128, and 64 in sequence. The ReLU is selected as the activation function, the learning rate is set to 0.01, and the number of iterations is 1200 times.

[0050] The input of the deep learning network can be the parameter values of multiple target processing factors, and the output can be the first predicted sensory score.

[0051] As an example, the input of the deep learning network can be the frying time, pre-cooking time, and stewing time, and the output can be the first predicted sensory score.

[0052] The eligible target second predicted sensory score can be one or more scores. The conditions can be pre-set, such as the maximum value or greater than a preset threshold.

[0053] The trained deep learning model can be used to predict the sensory score. Optionally, the prediction result of the deep learning model can be utilized to determine the highest second predicted sensory score, and then the processing factor parameter group corresponding to the highest second predicted sensory score can be determined as the selected processing factor parameter value.

[0054] Optionally, the selected processing factor parameter value can be processed (such as taking a value convenient for actual operation) to obtain a processing factor parameter value suitable for industrial processing. For example, if the selected processing factor parameter value is 5.012, 5 can be used as the processing factor parameter value for industrial processing.

[0055] During the processing of the target meat, especially during the industrial processing of meat, it is difficult to simulate the changes of multiple factors through real experiments, and in the industrial production process, it is necessary to precisely control the processing process of meat to obtain meat products with better sensory properties; this makes it difficult to obtain the processing factor parameter values for precise control through real experiments.

[0056] It should be noted that the food processing technology optimization based on fuzzy mathematics and deep learning prediction provided in this embodiment constructs a response surface equation according to the response surface test data; expands on the basis of N groups of response surface test data according to the constructed response surface equation to generate M groups of virtual test data; determines the N groups of response surface test data and the M groups of virtual test data as the sample set; trains a deep learning model based on the sample set; executes a prediction step based on the trained deep learning model; determines the optimal second predicted sensory score from the prediction result of the prediction step, and determines the processing factor parameter value corresponding to the optimal second predicted sensory score as the selected processing factor parameter value of the target meat; thus, a deep neural network prediction model of processing factor parameters and fuzzy sensory scores can be constructed by introducing virtual samples based on the response surface test design, so as to reduce the trial-and-error cost of industrial production, quickly determine accurate optimal processing factor parameters, and improve the possibility and process of putting the target meat into industrial production.

[0057] It should be noted that the training process of a deep learning model usually requires a large number of samples. In the food processing field, if there are already a large number of samples, it seems that the selected processing factor parameter values can be obtained without introducing a deep learning model. Based on N groups of response surface test data, virtual samples (i.e., virtual test data) are expanded based on the response surface model; based on the expanded and more samples, a deep learning model is trained, and the trained deep learning model is used for prediction; thus, the number of real tests can be greatly reduced by conducting response surface tests, and virtual samples are expanded based on the response surface model, making it possible to train an available deep learning model; the prediction of the deep learning model can avoid conducting real tests with all variables, that is, accurate selected processing factor parameter values can be obtained with fewer tests. Moreover, after testing, the prediction effect of the deep learning model is better than that of the response surface model.

[0058] In some embodiments, the above method may further include: obtaining the meat processing steps of the target meat; determining candidate processing factors and candidate processing factor selected standard values according to the meat processing steps; generating a single-factor test plan according to the candidate processing factors and candidate processing factor selected standard values; obtaining and displaying the single-factor test results.

[0059] The single-factor test plan may include single-factor sub-plans for each single factor, and the single-factor sub-plan includes the first change range of the single factor and the candidate processing factor selected standard values of other candidate processing factors except the single factor.

[0060] The single-factor test results include multiple evaluation indicators and scores corresponding to the evaluation indicators.

[0061] The single-factor test results are used to determine the target processing factor and the second change range of the target processing factor.

[0062] As an example, obtain the processing steps of Hakka mother wine chicken; determine candidate processing factors (such as oil consumption, ginger consumption, frying time, pre-boiling time, and stewing time) and candidate processing factor selected standard values (such as oil consumption of 24 g, ginger consumption of 12 g, frying time of 4 min, pre-boiling time of 4 min, and stewing time of 60 min) according to the obtained meat processing steps. The single-factor test plan includes a single-factor sub-plan for oil consumption, a single-factor sub-plan for ginger consumption, a single-factor sub-plan for frying time, a single-factor sub-plan for pre-boiling time, and a single-factor sub-plan for stewing time.

[0063] Taking the single-factor sub-scheme of frying time as an example, the first variation range of frying time is 0 min, 2 min, 4 min, 6 min, and 8 min; for other candidate processing factors except frying time, the standard values of candidate processing factors are adopted, i.e., the oil consumption is 24 g, the ginger consumption is 12 g, the pre-cooking time is 4 min, and the stewing time is 60 min.

[0064] Implement the single-factor test scheme to obtain the single-factor test results. The single-factor test results can include the respective results corresponding to the single-factor sub-schemes. Taking the single-factor sub-scheme of frying time as an example, for the single-factor test with a frying time of 0 min, an oil consumption of 24 g, a ginger consumption of 12 g, a pre-cooking time of 4 min, and a stewing time of 60 min, a finished chicken product is processed, and this finished chicken product can be evaluated from multiple evaluation indexes to obtain the scores corresponding to multiple evaluation indexes; for the single-factor test with a frying time of 2 min, an oil consumption of 24 g, a ginger consumption of 12 g, a pre-cooking time of 4 min, and a stewing time of 60 min, a finished chicken product is processed, and this finished chicken product can be evaluated from multiple evaluation indexes to obtain the scores corresponding to multiple evaluation indexes. And so on.

[0065] Analyzing the single-factor test results, the processing factors that have a greater impact on the target meat product can be obtained, as well as the better parameter variation range (i.e., the second variation range) corresponding to the processing factors. As an example, in the processing technology of Hakka mother rice wine chicken, the target processing factors can include frying time, pre-cooking time, and stewing time. As the frying time of the target processing factor, the second variation range is 2 - 6 min.

[0066] The multiple evaluation indexes of the single-factor test can include but are not limited to at least one of the following: sensory, yield rate, chicken pH value, soup pH value, color difference, and texture.

[0067] Thus, before industrial processing, the processing factors that have a greater impact on the target meat for industrial processing are determined, as well as the corresponding second variation range.

[0068] In some embodiments, the above method may further include: determining a response surface test scheme according to the target processing factors and the second variation range determined by the single-factor test results, wherein the test factors of the response surface test scheme include the target processing factors, and the parameter values corresponding to the factor levels of the response surface test are within the second transformation range; obtaining the test results of the response surface test, wherein the test results of the response surface test are the sensory evaluation indexes and the first sensory scores corresponding to the sensory evaluation indexes.

[0069] The factors of the response surface test are the target processing factors, and the factor levels of the response surface test are determined from the above-mentioned second transformation range. As an example, the test factors of the response surface test are a frying time of 4 min, a pre-cooking time of 6 min, and a stewing time of 60 min. The values corresponding to the test levels of the frying time are 2 min (-1 level), 4 min (0 level), and 6 min (1 level).

[0070] Thus, based on the results of the single-factor test, the processing factors that have a greater impact on the target meat product can be determined, the range of parameter values of the processing factors to be tested can be narrowed, the test difficulty of the response surface test and the fitting difficulty of the response surface test can be reduced, and the test accuracy of the response surface test can be improved.

[0071] In some embodiments, the response surface test is based on the BBD response surface test method.

[0072] In some embodiments, the above step 103 may include: for each set of response surface test data in the N sets of response surface test data, constructing several sets of virtual test data.

[0073] Optionally, several sets may be M / N sets. As an example, N is 17 and M is 204.

[0074] Here, the sample mean of each set of virtual test data is 0, and the standard deviation of the virtual test data is less than or equal to the standard deviation of the corresponding set of response surface test data in the response surface repeated test.

[0075] As an example, for the set of response surface test data with a frying time of 4 min, a pre-cooking time of 6 min, and a stewing time of 60 min, during the test process, it was done multiple times, that is, the response surface repeated test; the standard deviation of the results of the set of response surface repeated tests can be calculated; it can be understood that the levels can be set for the data used as independent variables in the response surface test. Based on the response surface model, 11 sets of virtual test data can be expanded. For each set of test data in these 11 sets of virtual test data, the sample mean is 0, and the standard deviation of the virtual test data is less than or equal to the standard deviation of the corresponding set of response surface test data in the response surface repeated test.

[0076] Thus, reliable virtual test data can be expanded based on each set of data in the response surface test data, increasing the reliable samples and making it possible to train the deep learning model using multiple samples.

[0077] In some embodiments, the generation process of the sensory score includes: for the meat products processed by each set of processing factors, obtaining the original evaluation data for the sensory evaluation indicators; generating a fuzzy matrix R1 according to the original evaluation data; generating a fuzzy comprehensive evaluation set Y1 according to the weight X matrix corresponding to each evaluation indicator and the fuzzy matrix R1; and obtaining the sensory score corresponding to each set of processing factors.

[0078] As an example, please refer to Table 3 in the following text. Analyze according to the results of Table 3. Taking the test group numbered 1 as an example, the voting results of the sensory scores of 10 participating personnel for the first group of samples (i.e., the original average data) are: A appearance = (3, 5, 2, 0); A smell = (2, 4, 3, 1); A taste = (1, 5, 4, 0); A texture = (3, 7, 0, 0). After normalization, the fuzzy matrix R1 is obtained. According to the principle of fuzzy matrix transformation, the weight set X is combined with the fuzzy matrix R, that is, the fuzzy comprehensive judgment set Y = X·R. Take the median H = (h1, h2, h3, h4) = (8.5, 6.5, 4.5, 2.0) of the corresponding scores of each comment set. According to , the fuzzy comprehensive sensory scores S1~S 15 .

[0079] Calculate the fuzzy comprehensive evaluation set Y1:

[0080]

[0081] Calculate the fuzzy comprehensive sensory score S1 (i.e., the sensory score):

[0082] S1 = Y1*H = 0.213*8.5 + 0.523*6.5 + 0.239*4.5 + 0.025*2.0 = 6.34

[0083] It can be understood that the above calculation process of the sensory score can be applied to the evaluation process of single-factor experiments and also to the evaluation process of response surface experiments.

[0084] Thus, the relatively subjective sensory evaluation can be converted into a relatively objective sensory score, presenting a new evaluation method for the results of industrial meat processing, providing objective samples for the establishment of deep learning models, and providing an objective basis for the precise control of industrially processed meat.

[0085] In some embodiments, the method includes: determining the high point of the response value in the response surface equation; determining the corresponding high point to-be-verified processing factor parameter group according to the high point of the response value in the response surface equation; and determining the to-be-verified processing factor parameter group according to the high point to-be-verified processing factor parameter group and the third variation range.

[0086] As an example, based on the response surface equation, determine the high point of the response value. Determine several to-be-verified parameter value groups according to the third variation range around the high point. Use the to-be-verified parameter value groups as the to-be-verified processing factor parameter groups and import them into the deep learning model to obtain the predicted values (i.e., the second sensory predicted values).

[0087] Thus, the computational load of using the deep model for exploration and prediction can be reduced, and the speed of obtaining the selected processing factor parameter values using the deep model can be increased.

[0088] In some embodiments, the target meat is chicken; the candidate processing factors include one or more of the following: raw material ratio, ginger addition amount, peanut oil addition amount, frying time, pre-cooking time, stewing time, frying temperature, and stewing temperature; the test factors of the response surface test plan include frying time, pre-cooking time, and stewing time.

[0089] The parameter values corresponding to the frying time in the response surface test plan include 2 min, 4 min, and 6 min;

[0090] The parameter values corresponding to the pre-cooking time in the response surface test plan include 4 min, 6 min, and 8 min

[0091] The parameter values corresponding to the stewing time in the response surface test plan include 30 min, 60 min, and 90 min;

[0092] In some embodiments, the established response surface equation is:

[0093] Y = H1 - H2A - H3B - H4C - H5AB - H6AC - H7BC - H8A² - H9B² - H 10 C², where Y represents the response value, A represents the frying time value, B represents the pre-cooking time value, C represents the stewing time value, and Hx represents the coefficient obtained by fitting, and x takes values from 1 to 10 in the response surface equation.

[0094] In some embodiments, according to the flow of an embodiment of the food processing technology optimization method based on fuzzy mathematics and deep learning prediction of the present disclosure, the method includes the following steps:

[0095] First step, obtain the meat processing steps of the target meat.

[0096] Second step, according to the meat processing steps, determine the candidate processing factors and the selected standard values of the candidate processing factors;

[0097] Third step, according to the candidate processing factors and the selected standard values of the candidate processing factors, generate a single-factor test plan.

[0098] Among them, the single-factor test plan includes single-factor sub-plans for each single factor, and the single-factor sub-plan includes the first change range of the single factor and the selected standard values of other candidate processing factors outside the single factor;

[0099] Fourth step, obtain and display the single-factor test results.

[0100] Determine the response surface test plan according to the target processing factors and the second change range determined by the single-factor test results

[0101] Step 5: Obtain N groups of response surface test data for the response surface test.

[0102] In this embodiment, the execution subject (such as a server and / or a terminal device) of the food processing technology optimization method based on fuzzy mathematics and deep learning prediction can obtain N groups of response surface test data for the response surface test.

[0103] In this embodiment, each group of test data in the N groups of response surface test data includes multi-item target processing factor parameter values and a first sensory score.

[0104] In this embodiment, the first sensory score is the sensory score of the target meat product processed based on the multi-item target processing factor parameter values.

[0105] Step 6: Construct a response surface equation according to the N groups of response surface test data.

[0106] Step 7: Based on the constructed response surface equation, expand on the basis of the N groups of response surface test data to generate M groups of virtual test data.

[0107] Step 8: Determine the N groups of response surface test data and the M groups of virtual test data as the sample set.

[0108] Step 9: Execute the training step based on the sample set.

[0109] In this embodiment, the training step includes: importing the multi-item target processing factor parameter values in the sample into an untrained deep learning network to obtain a first predicted sensory score output by the deep learning network; generating a loss value based on the first predicted sensory score and the first sensory score in the sample; updating the weight values in the deep learning network based on the loss value until a preset stop training condition is reached to obtain a deep learning model.

[0110] Step 10: Execute the prediction step based on the trained deep learning model.

[0111] In this embodiment, the above prediction step includes: obtaining the processing factor parameter values to be verified, and importing the processing factor parameter values to be verified into the trained deep learning model to obtain a second predicted sensory score output by the deep learning model.

[0112] Step 11: Determine the target second predicted sensory score that meets the conditions from the second predicted sensory scores output by the deep learning model; determine the processing factor parameter values corresponding to the target second predicted sensory score as the selected processing factor parameter values of the target meat.

[0113] Therefore, through the single-factor experiment of optimized key process parameters combined with the Box-Behnken response surface method to design the experiment, integrating real data samples and constructed virtual data samples, conducting adaptive learning and fitting training of the mathematical model, a fuzzy sensory evaluation - deep neural network (DNN) integrated model was constructed, which breakthroughly solved the problems that traditional sensory evaluation is difficult to quantify and predict in real time.

[0114] Please refer to Figure 2 , which shows the flow of an embodiment of the food processing technology optimization method according to the present disclosure. As Figure 1 shown, the food processing technology optimization method includes the following steps:

[0115] Step 201, for the target processing factors of the target meat, design and implement a response surface experiment, where the response surface experiment includes processing the target meat based on multiple target processing factors to obtain a target meat product.

[0116] Step 202, obtain the experimental results of the response surface experiment.

[0117] Step 203, based on the N groups of response surface experiment data of the response surface experiment, execute the method described in any one of claims 1-7 to obtain a selected processing factor parameter group of the target meat.

[0118] Among them, each group of experimental data in the N groups of response surface experiment data includes multiple target processing factor parameter values and a first sensory score, where the first sensory score is the sensory score of the target meat product processed based on the multiple target processing factor parameter values.

[0119] In some embodiments, obtain a single-factor experiment plan, where the single-factor experiment plan includes single-factor sub-plans for each single factor, and the single-factor sub-plan includes a first change range of the single factor and a selected standard value of other candidate processing factors other than the single factor; implement the single-factor experiment; obtain the experimental results of the single-factor experiment, where the single-factor experiment results include multiple evaluation indicators and scores corresponding to the evaluation indicators.

[0120] In some embodiments, the target meat is chicken; the candidate processing factors include one or more of the following: raw material ratio, ginger addition amount, peanut oil addition amount, frying time, precooking time, stewing time, frying temperature, stewing temperature; the experimental factors of the response surface experiment plan include frying time, precooking time, and stewing time.

[0121] The parameter values corresponding to the frying time in the response surface experiment plan include 2 min, 4 min, and 6 min;

[0122] The parameter values corresponding to the precooking time in the response surface test plan include 4 min, 6 min, and 8 min.

[0123] The parameter values corresponding to the stewing time in the response surface test plan include 30 min, 60 min, and 90 min.

[0124] In some embodiments, the established response surface equation is:

[0125] Y = H1 - H2A - H3B - H4C - H5AB - H6AC - H7BC - H8A² - H9B² - H 10 C², where Y represents the response value, A represents the frying time value, B represents the precooking time value, C represents the stewing time value, and Hx represents the coefficient obtained by fitting, and x in the response surface equation takes values from 1 to 10.

[0126] Thus, through the single-factor test of the optimized key process parameters combined with the Box-Behnken response surface method to design the experiment, integrating the real data samples and the constructed virtual data samples, carrying out the adaptive learning and fitting training of the mathematical model, a fuzzy sensory evaluation - deep neural network (DNN) integrated model is constructed, which breakthroughly solves the problems that traditional sensory evaluation is difficult to quantify and predict in real time.

[0127] In the exemplary application scenarios of one or more embodiments of the present disclosure, a series of experiments and the construction of a deep neural network model are proposed, as follows:

[0128] Single-factor test

[0129] Pretreatment of the raw materials for Hakka mother rice wine chicken: Select fresh chicken legs, debone and wash, dry the surface moisture, cut them into pieces with a side length of 2 - 3 cm, weigh them, and store them in a 4°C windproof environment for later use.

[0130] There are multiple process parameters involved in the processing technology of Hakka mother rice wine chicken, at least including raw material ratio, ginger addition amount, peanut oil addition amount, frying time, precooking time, stewing time, frying temperature, stewing temperature, etc. Single-factor tests are carried out on the above candidate process parameters based on multiple evaluation indexes (including sensory, yield rate, chicken meat pH value, soup pH value, color difference, and texture), and the mutual mapping relationship and influence degree are analyzed. Based on the results of the single-factor test, the process parameters with small influence and poor regularity are eliminated, and finally the frying time, precooking time, and stewing time are selected as variable factors, as the test factors for the Box-Behnken response surface optimization test, and the fuzzy total sensory score is used as the response value of the Box-Behnken response surface optimization test for subsequent research.

[0131] The influencing factors of frying time (0, 2, 4, 6, 8 minutes), pre-cooking time (0, 2, 4, 6, 8 minutes) and stewing time (0, 30, 60, 90, 120 minutes) were selected to study their effects on the edible quality indicators (sensory quality, yield, chicken pH, soup pH, color difference and texture) of Hakka Niangjiu Chicken.

[0132] (1) Effects of stir-frying time, pre-cooking time, and stewing time on sensory quality

[0133] Four factors, namely appearance, aroma, flavor, and texture, were selected to construct the factor set U = {u1,u2,u3,u4} = {appearance, aroma, flavor, texture}, and the evaluation set V = {v1,v2,v3,v4} = {excellent, good, fair, poor}. A weight set X = {x1,x2,x3,x4} was determined using a 0-4 scale. A sensory evaluation panel consisting of 10 evaluators with food professional backgrounds (5 males and 5 females, aged 22-25 years) scored each of the four factors according to the evaluation criteria discussed and established in Table 1. The 10 sensory assessors then assigned weights to the four evaluation factors: appearance, aroma, flavor, and texture. The weighted results are shown in Table 2, resulting in the weight set X = {0.20, 0.25, 0.31, 0.24}. The results of the fuzzy mathematical sensory evaluation of the 15 test samples using single-factor fuzzy analysis are shown in Table 3.

[0134] Table 1 Sensory scoring criteria

[0135]

[0136] Table 2 Weights of fuzzy sensory evaluation indicators

[0137]

[0138] Table 3 Statistical results of voting for sensory evaluation of Niangjiu Chicken

[0139]

[0140] According to the results in Table 3, taking the test group numbered 1 as an example, the sensory scoring results of the 10 participants for the samples in Group 1 are as follows: A appearance = (3, 5, 2, 0); A smell = (2, 4, 3, 1); A taste = (1, 5, 4, 0); A texture = (3, 7, 0, 0). After normalization, the fuzzy matrix R1 is obtained. According to the fuzzy matrix transformation principle, the weight set X is synthesized with the fuzzy relationship matrix R, that is, the fuzzy comprehensive judgment set Y = X·R. Take the median value H = (h1, h2, h3, h4) = (8.5, 6.5, 4.5, 2.0) of the corresponding scores of each comment set, according to , we can get the fuzzy comprehensive sensory score S1~S 15 .

[0141] Calculate the fuzzy comprehensive evaluation set Y1:

[0142]

[0143] Calculate the fuzzy comprehensive sensory score S1:

[0144] S1 = Y1 ∙ H = 0.213 * 8.5 + 0.523 * 6.5 + 0.239 * 4.5 + 0.025 * 2.0 = 6.34

[0145] Calculating according to the above steps, the fuzzy comprehensive sensory scores S of other experimental groups can be obtained, and the results are as Figure 3 shown.

[0146] From Figure 3 it can be seen that the changing trends of the effects of frying time, precooking time, and stewing time on the sensory score of chicken with fermented glutinous rice are the same, all showing a changing law of first increasing and then decreasing. The flavor of cooked meat mainly comes from a series of complex chemical processes such as lipid oxidation degradation, Maillard reaction, and lipid-Maillard interaction. Different cooking methods will significantly change the contents of fatty acids, small molecule metabolites, and volatile flavor substances in chicken, thus causing changes in the product flavor.

[0147] As shown in Figure 3 Figure (a), a frying duration of 2 - 6 minutes is beneficial to the shaping of chicken and gives the chicken skin and chicken an attractive golden color. This color change is an intuitive presentation of the Maillard reaction between reducing sugar and amino acids. This reaction has a significant impact on the color of the chicken skin and chicken, and at the same time greatly stimulates the meaty aroma, reaching the best at 4 minutes. As shown in Figure 3 Figure (b), during precooking, when adding Hakka fermented glutinous rice wine, the alcohol volatilizes with the water vapor, the soup gradually concentrates, and the bitter taste of the wine gradually disappears while the sweetness increases. Combining the data analysis of Figure 3 Figure (b) and Table 3, it can be seen that the first four minutes have a greater impact on the formation of the taste of chicken with fermented glutinous rice. When precooking for 6 minutes, the comprehensive sensory quality of the chicken reaches the best. If continuously heated at a high temperature until 8 minutes, overheating will cause quality defects such as the chicken showing a charred black appearance and a bitter taste in the mouth, resulting in a significant reduction in the sensory score.

[0148] From Figure 3As can be seen from (c), the influence of stewing time on the sensory score shows strong symmetry. As the stewing time increases, the color of the Hakka rice wine chicken gradually changes from light yellowish-brown to reddish-brown and finally to brownish-black, and the texture of the chicken becomes significantly softer. When the stewing time is about 30 - 90 minutes, the Hakka rice wine chicken forms a reddish-brown appearance, emits a uniform aroma, presents a sweet and fresh taste, and has a soft and tender texture, with the best effect at 60 minutes of stewing. Therefore, the stewing time has a great influence on the sensory quality of the Hakka rice wine chicken. While reasonably controlling the stewing time, adding ginger can, to a certain extent, inhibit lipid oxidation, reduce the formation of saturated aldehydes, and improve the overall flavor of the Hakka rice wine chicken. Based on the influence on sensory quality, the basic processing factors for the Hakka rice wine chicken are a frying time of 4 minutes, a pre-boiling time of 6 minutes, and a stewing time of 60 minutes.

[0149] (2)Effect of frying time, pre-boiling time and stewing time on the yield rate of Hakka rice wine chicken

[0150] Determination of the yield rate: Weigh the mass m1 (g) of the chicken skin and chicken before processing, and weigh again m2 (g) after processing.

[0151] Yield rate = (m₂ / m₁) * 100%

[0152] The yield rate is usually closely related to the water retention and juiciness of meat products. As can be seen from Figure 4 it, the yield rate of cooked meat samples is significantly affected by cooking methods and processing times. Since there is no outer shell on the surface of the samples, water evaporation and melted fat can easily escape from the meat.

[0153] From Figure 4 (a), it can be seen that within the range of 0 - 8 minutes of frying, the yield rate significantly decreases from 64.77% to 57.20% as the time prolongs. For the sample fried for 8 minutes, the mass of the chicken skin decreases by 7.09 g and the mass of the chicken decreases by 44.55 g after processing, with the most mass loss compared to other cooked meat samples. This is because fat melts and is lost during long-term high-temperature cooking, and the coiling of collagen helix changes, resulting in tissue softening. High-temperature frying causes the water in the muscle tissue to decrease due to evaporation, and the rapid loss of water leads to a significant decrease in the mass of the chicken skin and chicken. Protein heating causes muscle fiber contraction, and actin and myosin undergo irreversible denaturation, further squeezing out water. As Figure 4 shown in (b), within the range of 0 - 8 minutes of pre-boiling, the rates of water loss, protein denaturation, and fat dissolution are in dynamic changes, so the yield rate fluctuates irregularly but is basically in a stable state. As Figure 4As shown in (c), during the stewing process, the yield decreased due to initial water loss and protein denaturation. As the stewing time exceeded 60 minutes, connective tissue decomposed, collagen was converted into gelatin, and water was reabsorbed, resulting in a brief increase in the yield. Therefore, the effect of stewing time on the yield showed a trend of first decreasing and then increasing. According to the above experimental results, the three different processing methods had a significant impact on the chicken yield. Processing times of frying for 0 - 6 minutes, precooking for 2 - 8 minutes, and stewing for 0 - 120 minutes were beneficial to improving the chicken yield.

[0154] (3)Effect of frying time, precooking time, and stewing time on the water content of chicken

[0155] The water content of the finished chicken was determined by the direct drying method in GB 5009.3—2016 "National Food Safety Standard - Determination of Moisture in Foods".

[0156] As Figure 5 shown, as the processing time increased, the water content of the chicken continuously decreased. For samples after frying for 4 minutes and precooking for 6 minutes, the downward trend was particularly significant. The continuous heat treatment process destroyed the gel network structure of the protein, resulting in an impact on the water retention of the chicken. The water content of the precooked meat sample after precooking for 8 minutes was the lowest, at 48.27 g / 100 g. The above experimental results showed that the three processing steps and the heat treatment time had a significant impact on the water retention of the chicken, and as the heat treatment time increased, the water content of the chicken gradually decreased. There was an inverse change trend between the processing time and the water retention. Selecting the critical time period for cooking the chicken could better ensure good water retention of the chicken.

[0157] (4)Effect of frying time, precooking time, and stewing time on the pH value of the soup and chicken

[0158] The pH values of the homogenized chicken sample and the soup of the finished chicken were determined according to GB 5009.237—2016 "National Food Safety Standard - Determination of Moisture in Foods".

[0159] As Figure 6 shown in (a), the pH value of the Hakka mother rice wine was approximately 3.82, showing strong acidity. After processing, the pH value of the soup increased significantly compared to the raw wine. During the heating process, the water in the chicken and ginger flowed out, causing the pH value of the soup to rise. When the frying time was 6 minutes, a large amount of water and other substances in the chicken transferred to the soup, and the pH value of the sample soup increased. At the same time, the pH value of the chicken decreased significantly. For samples with a precooking time exceeding 2 minutes, the pH value of the soup decreased significantly, indicating that the precooking process was the main factor affecting the pH value of the soup.

[0160] During the frying and precooking process at 170°C, the pH value of the soup showed a gradually decreasing trend; while during the stewing process at 100°C, the pH value of the soup gradually increased, and this change was related to the Maillard reaction. Experimental studies have shown that when the temperature is 100°C, the pH value of the Maillard reaction system only decreases slightly; but when the temperature rises to 110°C and 120°C, the pH value decreases significantly. This is because the carbonyl-amine reaction continuously consumes NH4+ and organic acids are generated during the reaction process. In addition, when heated, the fat in chicken will hydrolyze to produce fatty acids and glycerol, and the release of fatty acids will enhance the acidity of the soup, resulting in a decrease in pH value. On the other hand, chicken contains some alkaline components such as certain amino acids and nitrogen-containing compounds. After reacting with the components in the wine, they may release alkaline substances, prompting the pH value of the soup to increase. For the samples without the stewing process, the pH value of their soup is lower than that of the raw wine; for the samples stewed for 30 minutes, the pH value of the soup increases significantly; for the samples stewed for 60 minutes, the pH value decreases; for the samples stewed for more than 60 minutes, the pH value gradually rises with the extension of time.

[0161] As the processing time extends, proteins degrade and free amino acids dissolve out. As can be seen from Figure 6 Figure (b), the pH value of the cooked meat samples is significantly lower than that of the raw meat. During the frying process, when the frying time is less than 4 minutes, the pH value of the chicken changes little overall, remaining in the range of 5.95 - 5.97. When the frying time reaches 6 minutes, the pH value drops significantly. For the chicken samples precooked for 4 minutes, their pH value significantly increases to 5.81 and then continues to decrease. The difference in stewing time has a more significant impact on the decrease in the pH value of the chicken, because as the stewing time continuously extends, the degree of protein denaturation increases and its network structure becomes more porous. This porous structure enables more sufficient contact between the chicken and the soup, not only promoting the dissolution of acidic substances in the chicken but also making the soup more easily penetrate into the interior of the chicken, resulting in the accumulation of acidic substances such as lactic acid and succinic acid in the chicken, ultimately causing the pH value of the chicken to decrease significantly and gradually approach the pH value of the soup. Compared with the unprocessed samples, the pH values of the cooked meat soup and cooked meat samples have been greatly improved. And based on the fact that the water-holding capacity of cooked meat products increases with the increase in the pH value of the chicken, especially when the pH value is close to the neutral range (6.0 - 7.0), the water-holding capacity reaches the best state. The above experimental results show that the three processing methods and times can significantly change the pH values of the mother rice wine chicken soup and chicken, causing changes in product quality. Especially in the processing methods of frying for 2 - 8 minutes, precooking for 2 - 8 minutes, and stewing for 30 - 120 minutes, the product quality has been effectively improved.

[0162] (5)Effects of frying time, precooking time, and stewing time on the color difference of chicken skin and chicken

[0163] The color of the samples was measured using a portable color difference meter. After calibrating the color difference meter, the L* (lightness value), a* (redness value), and b* (yellowness value) of the chicken skin, the skin-attached surface, and the bone-attached surface were measured.

[0164] The color of chicken skin and chicken meat is an important factor affecting consumers' purchase. The change in the color of chicken skin during the cooking process of Hakka mother wine chicken is as Figure 7 shown. Compared with raw chicken skin, the L* value of the processed chicken skin gradually decreases, while the a* value and b* value increase significantly. The undercooked samples showed higher L* values because more water was impregnated on the surface of the samples. As the cooking time increases, the cooking loss rate increases and L* gradually decreases, that is, the reduction of sample moisture leads to the reduction of the surface gloss of the sample. During the high-temperature cooking process, substances such as mother wine color the chicken skin, making it gradually turn into reddish-brown.

[0165] The b* value of the processed chicken skin all becomes positive, because heat treatment makes the color of the chicken skin significantly yellow. As the pre-cooking time extends, the b* value of the chicken skin significantly decreases at 8 minutes; while as the stewing time extends, the b* value of the chicken skin reaches the highest at 90 minutes and then significantly decreases. The increase in the b* value is related to the increase in lipid oxidation, but as the cooking time further extends, the b* value decreases because the fat content of the chicken skin itself is less, and long-term cooking leads to partial protein degradation and leaching, thus reducing the yellowness.

[0166] Hakka mother wine chicken with bright red color is more popular among consumers. As Figure 8 shown, compared with raw meat, the L* value of the skin-attached surface and bone-attached surface of the processed chicken meat gradually decreases, while the a* value and b* value increase significantly. After processing, the L* value of the bone-attached surface first rebounds and then decreases with the increase of the set time, because the water extruded from the chicken meat during the heat treatment adheres to the meat surface, resulting in a temporary increase in the L* value. After stewing, the L* value of the chicken meat significantly decreases (P<0.05). After heating and cooking, the a* value of the chicken meat significantly increases (P<0.05), and the b* value also gradually increases with the cooking time. This may be due to the reaction of myoglobin with the residual oxygen in the meat at the initial stage of heating, thus enhancing the redness value of the meat color. At the same time, non-enzymatic browning reactions such as the Maillard reaction occur between the components in the chicken meat and the reducing sugars, amino acids, and phenolic substances in the mother wine, further enhancing the redness and yellowness of the meat. Therefore, different processing methods and times have the effect of improving the color quality of Hakka mother wine chicken, especially when the frying time is 2-8 minutes, the pre-cooking time is 2-8 minutes, and the stewing time is 30-90 minutes, the color effect is more obvious.

[0167] (6) Effects of frying time, pre-cooking time, and stewing time on the texture of chicken

[0168] Cut the chicken into pieces of 1 - 2 cm. Select the P36 / R probe in the TPA mode, set the test speed before and after at 2 mm / s, the pressing speed at 1 mm / s, the deformation at 50%, the pressing interval at 5 s, and the trigger force at 5 g, and measure the hardness, elasticity, chewiness and other indexes of the cooked samples under different conditions.

[0169] Table 4 Determination of the texture of chicken

[0170]

[0171] Note: Separate comparisons were made for the five treatment levels of stir-frying, precooking, and stewing. a, b, c are significant annotations. Different annotated letters represent significant differences (P < 0.05).

[0172] The texture of cooked meat products mainly includes factors such as hardness, adhesiveness, elasticity, cohesiveness, gumminess, chewiness and resilience. During the cooking process, the tenderness of meat is mainly affected by the thermal dissolution of connective tissue and the thermal denaturation of myofibrillar proteins, and cooking time and temperature have significant effects on both cooking loss and tenderness. As can be seen from Table 4, the hardness of the cooked samples is in the range of 668.26 - 2015.29 g. And the hardness of the samples gradually decreases with the increase of stir-frying time and stewing time, while increases with the increase of precooking time. Within 0 - 8 minutes of stir-frying, the hardness decreases with the increase of time. When precooking for 6 minutes, due to the denaturation of proteins under heat, the gel network structure is enhanced, and the hardness of the chicken increases significantly. Within 0 - 120 minutes of stewing, the hardness, cohesiveness, gumminess, chewiness and resilience of the chicken change significantly compared with the stir-frying and precooking stages. From the above results, it can be seen that all three-stage processing methods have significant effects on the texture characteristics of the chicken in fermented glutinous rice chicken. Selecting cooking times of 2 - 8 minutes for stir-frying, 4 - 8 minutes for precooking, and 30 - 90 minutes for stewing can make the cooked meat products have low hardness, small adhesiveness, high elasticity, medium cohesiveness, chewiness, resilience and cohesiveness, and achieve the improvement of the quality of fermented glutinous rice chicken.

[0173] In summary, in the processing of fermented glutinous rice chicken, cooking methods and cooking time play crucial roles. On the premise that other factors are constant, the effects and functions of stir-frying, pre-boiling, and stewing durations on the quality of chicken are mainly investigated. High-temperature stir-frying causes the rapid evaporation and loss of moisture in the chicken tissue, and the water-holding capacity of the chicken drops rapidly. On the one hand, it promotes the contraction and increased elasticity of the meat tissue, and promotes the Maillard reaction to improve the color, appearance, and flavor of the chicken. However, the reduction of moisture also leads to a decrease in the yield of the chicken, and excessive stir-frying results in too low water retention of the chicken, causing the pH value of the chicken to be too low, affecting the color difference between the chicken and the chicken skin, that is, resulting in a decrease in the surface gloss of the chicken. Moreover, too low a pH value will also inhibit the Maillard reaction of the chicken, thus affecting the flavor, color, etc. of the chicken. During the pre-boiling stage, with the addition of water, the water-holding capacity of the chicken gradually recovers, the protein network structure gradually loosens, and the pH value of the chicken rises. If the pre-boiling time is too short, the protein tissue structure of the chicken will not be in sufficient contact with the soup, affecting the opening of the chicken tissue and the dissolution of acidic substances in the chicken. Also, the pre-boiling stage is a key process affecting the pH value of the soup. In the early stage of pre-boiling, due to the outflow of moisture from the chicken and ginger, the chicken has not started to reverse water absorption, resulting in a rapid increase in the pH value of the soup and a decrease in the pH value of the chicken. After 2 to 4 minutes, the pH value of the soup drops significantly and shows a tendency to approach the pH value of the chicken. The length of the stewing time is also crucial for the quality of the chicken. According to the single-factor experiment, too long a stewing time is not conducive to the yield, color difference, and texture of the chicken. Long-time stewing will cause further degradation and leaching of proteins in the chicken and the chicken skin, resulting in a loss of the yellowness of the chicken product, affecting the color of the chicken, and causing the chicken to lose nutrients and taste. Therefore, combining the single-factor experiment and the coupling and correlation of various evaluation indicators of chicken quality, and selecting the Box-Behnken response surface optimization experiment, the factors and levels of the response surface experiment are determined, as shown in Table 5.

[0174] Table 5 Factors and Levels of Response Surface Experiment

[0175]

[0176] 1. Response Surface Experiment and Construction of Deep Neural Network Model

[0177] As can be seen from Table 6, the response surface test model is extremely significant (P<0.0001), and at the same time, the lack-of-fit term is not significant (P = 0.3457>0.05), indicating that the established model has a good fitting effect in the entire regression region. The determination coefficient of the model R² = 0.9766, indicating good correlation of the test data; the adjusted determination coefficient Radj² = 0.9464, indicating that the regression model can explain 94.64% of the response value changes. The primary and secondary order of the influence of each factor on the sensory score of the chicken cooked with fermented glutinous rice is frying time > pre-boiling time > stewing time, and the obtained regression equation of the sensory score is: Y = 7.50 - 0.5125A - 0.3812B - 0.1237C - 0.3550AB - 0.5800AC - 0.5675BC - 0.6028A² - 0.4403B² - 0.4702C².

[0178] Table 6 Response Surface Analysis Design and Results

[0179]

[0180] Table 7 Analysis of Variance of the Regression Model

[0181]

[0182] Note: Among them, "**" indicates extremely significant difference (P<0.01); "-" indicates no significant difference.

[0183] Based on the response surface model, a deep neural network model was further constructed. The data set was divided into a training set, a validation set and a test set according to the ratios of 64%, 16% and 20% respectively, which were used to verify the model training and generalization ability. The number of nodes in the hidden layer of the model is 256, 128, and 64 in sequence. The ReLU is selected as the activation function, the learning rate is set to 0.01, and the number of iterations is 1200 times.

[0184] From Figure 9 it can be seen that in the initial stage of training (about the first 10 Epochs), both the training loss and the validation loss decreased sharply, which means that the model is learning efficiently. As the training process progresses, although the loss continues to decrease, the rate of decrease gradually slows down. Finally, the two curves tend to a stable state, indicating that the model may have approached its best performance. The mean absolute error (MAE) of this model is 0.3735, indicating that the model has good prediction accuracy; the mean square error (MSE) is 0.3603, and the root mean square error (RMSE) is 0.6003, further confirming that the prediction error of the model is small; the prediction success rate is 90.00%.

[0185] 3. Analysis, Verification and Comparison of Model Prediction Results

[0186] The response surface and contour line constructed based on the regression equation are presented inFigure 10 From this figure, it can be clearly judged that the pairwise interactions among the three factors have significant effects on the sensory score of Hakka mother wine chicken. Among them, the interactions between frying time and stewing time, and between pre-boiling time and stewing time have particularly prominent effects on the sensory quality. The DesignExpert software was used to optimize the experimental results, and the process parameters that can maximize the sensory score were obtained: frying time of 2.81 minutes, pre-boiling time of 4.81 minutes, and stewing time of 79.39 minutes. Considering the feasibility of actual operation, the parameters were adjusted to: frying time of 3 minutes, pre-boiling time of 5 minutes, and stewing time of 79 minutes. The sensory score predicted by the response surface model is 7.69. Three parallel experiments were carried out according to the optimized process parameters for verification, and the average value of the actual sensory score obtained was 7.36, which is within the range of the 95% confidence interval (7.34 - 8.05), and the relative error compared with the theoretical value is 4.29%.

[0187] The comparison between the true values and predicted values of the DNN model constructed through the response surface experiment is as Figure 11 shown. The maximum predicted value of the DNN model is 7.55, and the corresponding process parameters at this time are: frying time of 4 minutes, pre-boiling time of 6 minutes, and stewing time of 60 minutes. To verify the accuracy of the model, three parallel experiments were carried out, and the actual sensory score obtained was 7.38 ± 0.12, and the relative error was only 2.25%. As can be seen from Table 8, the sensory score of the DNN model after actual verification has slightly improved compared with the response surface model, and the relative error is significantly lower. This result indicates that in terms of predicting sensory quality, the DNN model shows higher accuracy and reliability, and its prediction effect is better than that of the response surface model. With the continuous improvement and enrichment of actual production data, the advantages of the DNN model in prediction will become more prominent.

[0188] Table 8 Comparison of verification test results of two models

[0189]

Claims

1. A food processing technology optimization method based on fuzzy mathematics and deep learning prediction, characterized in that, Including: Obtaining N groups of response surface test data of a response surface test, where the response surface test includes processing target meat into a target meat product based on multiple target processing factors, and each group of test data in the N groups of response surface test data includes multiple target processing factor parameter values and a first sensory score, where the first sensory score is a sensory score of the target meat product processed based on the multiple target processing factor parameter values; Constructing a response surface equation according to the N groups of response surface test data; Based on the constructed response surface equation, expanding on the basis of the N groups of response surface test data to generate M groups of virtual test data; Determining the N groups of response surface test data and the M groups of virtual test data as a sample set; Performing a training step based on the sample set, where the training step includes: importing the multiple target processing factor parameter values in the sample into an untrained deep learning network to obtain a first predicted sensory score output by the deep learning network; generating a loss value based on the first predicted sensory score and the first sensory score in the sample; updating the weight values in the deep learning network based on the loss value until a preset training stop condition is reached to obtain a deep learning model; Performing a prediction step based on the trained deep learning model, where the prediction step includes: obtaining parameter values of processing factors to be verified, and importing the parameter values of processing factors to be verified into the trained deep learning model to obtain a second predicted sensory score output by the deep learning model; Determining a target second predicted sensory score that meets the conditions from the second predicted sensory scores output by the deep learning model; determining selected processing factor parameter values of the target meat based on the parameter values of the processing factors to be verified corresponding to the target second predicted sensory score.

2. The method according to claim 1, wherein The method further includes: Obtaining the meat processing steps of the target meat; Determining candidate processing factors and candidate processing factor selected standard values according to the meat processing steps; Generating a single-factor test plan according to the candidate processing factors and the candidate processing factor selected standard values, where the single-factor test plan includes single-factor sub-plans for each single factor, and the single-factor sub-plan includes a first change range of the single factor and the candidate processing factor selected standard values of other candidate processing factors except the single factor; Obtaining and displaying single-factor test results, where the single-factor test results include multiple evaluation indicators and scores corresponding to the evaluation indicators; Wherein, the single-factor test results are used to determine the target processing factors and the second change range of the target processing factors.

3. The method according to claim 2, characterized in that According to the method, it further includes: Determining a response surface test plan according to the target processing factors and the second change range determined by the single-factor test results, where the test factors of the response surface test plan include the target processing factors, and the parameter values corresponding to the factor levels of the response surface test are within the second transformation range; Obtaining the test results of the response surface test, where the test results of the response surface test are sensory evaluation indicators and the first sensory scores corresponding to the sensory evaluation indicators.

4. The method according to claim 3, characterized in that, The response surface test is based on the BBD response surface test method; And The determining the N groups of response surface test data and the M groups of virtual test data as a sample set includes: For each set of response surface test data among the N sets of response surface test data, a number of sets of virtual test data are constructed; wherein, the sample mean of each set of virtual test data is 0, and the standard deviation of the virtual test data is less than or equal to the standard deviation of the corresponding set of response surface test data in the response surface repeated tests.

5. The method according to claim 1, wherein The generation process of the sensory score includes: For the meat products processed by each set of processing factors, obtain the original evaluation data for the sensory evaluation index; Generate the fuzzy matrix R1 according to the original evaluation data; Generate the fuzzy comprehensive evaluation set Y1 according to the weight X matrix corresponding to each evaluation index and the fuzzy matrix R1; According to the sensory score corresponding to each set of processing factors.

6. The method according to claim 1, wherein The method includes: Determine the high point of the response value in the response surface equation; According to the high point of the response value in the response surface equation, determine the corresponding high point processing factor parameter group to be verified; According to the high point processing factor parameter group to be verified and the third change range, determine the processing factor parameter group to be verified.

7. The method according to any one of claims 1-6, wherein The target meat is chicken; The candidate processing factors include one or more of the following: raw material ratio, ginger addition amount, peanut oil addition amount, frying time, precooking time, stewing time, frying temperature, stewing temperature; The test factors of the response surface test plan include frying time, precooking time and stewing time; The parameter values corresponding to the frying time in the response surface test plan include 2 min, 4 min, and 6 min; The parameter values corresponding to the precooking time in the response surface test plan include 4 min, 6 min, and 8 min The parameter values corresponding to the stewing time in the response surface test plan include 30 min, 60 min, and 90 min; The established response surface equation is: Y = H1 - H2A - H3B - H4C - H5AB - H6AC - H7BC - H8A² - H9B² - H 10 C², where Y represents the response value, A represents the frying time value, B represents the precooking time value, C represents the stewing time value, Hx represents the coefficient obtained by fitting, and the value of x in the response surface equation ranges from 1 to 10.

8. An optimization method for a food processing technology, characterized in that, Including: For the target processing factors of the target meat, design and implement a response surface test, wherein the response surface test includes processing the target meat based on multiple target processing factors to obtain target meat products; Obtain the test results of the response surface test; Based on the N sets of response surface test data of the response surface test, execute the method according to any one of claims 1-7 to obtain the selected processing factor parameter group of the target meat; Wherein, each set of test data in the N sets of response surface test data includes multiple target processing factor parameter values and the first sensory score, wherein the first sensory score is the sensory score of the target meat product processed based on the multiple target processing factor parameter values.

9. The method according to claim 8, characterized in that The method further includes: Obtain a single factor test plan, wherein the single factor test plan includes single factor sub-plans for each single factor, and the single factor sub-plan includes the first change range of the single factor and the selected standard values of the candidate processing factors other than the single factor; Implement a single factor test; Obtain the test results of the single factor test, wherein the single factor test results include multiple evaluation indexes and the scores corresponding to the evaluation indexes.

10. The method according to claim 9, wherein The target meat is chicken; The candidate processing factors include one or more of the following: raw material ratio, ginger addition amount, peanut oil addition amount, frying time, precooking time, stewing time, frying temperature, stewing temperature; The test factors of the response surface test plan include frying time, precooking time, and stewing time; The parameter values corresponding to the frying time in the response surface test plan include 2 min, 4 min, and 6 min; The parameter values corresponding to the precooking time in the response surface test plan include 4 min, 6 min, and 8 min The parameter values corresponding to the stewing time in the response surface test plan include 30 min, 60 min, and 90 min; The established response surface equation is: Y = H1 - H2A - H3B - H4C - H5AB - H6AC - H7BC - H8A² - H9B² - H 10 C², where Y represents the response value, A represents the frying time value, B represents the precooking time value, C represents the stewing time value, Hx represents the coefficient obtained by fitting, and the value of x in the response surface equation ranges from 1 to 10.

Images