Method and system for judging fermentation maturity of bean products

Abstract

The invention provides a bean product fermentation maturity determination method and system, and relates to the technical field of bean product fermentation maturity determination. Temperature, humidity, pH value and ammonia gas concentration of a bean product fermentation environment and concentration data of amino acid and short-chain fatty acid in a bean product are monitored in real time, and data preprocessing is performed; constructing a deep learning network, taking the concentrations of the internal amino acids and the short-chain fatty acids as labels for training, obtaining self-influence parameters of the bean products, performing linear regression by utilizing the preprocessed environmental data, calculating environmental influence parameters, and finally combining the self-influence parameters with the environmental influence parameters to obtain the bean products. And comparing the comprehensive judgment value with a set maturity threshold value so as to judge the fermentation maturity condition of the bean product.

Description

Technical Field

[0001] The invention relates to the technical field of fermentation maturity of bean products, and in particular to a method and system for determining the fermentation maturity of bean products. Background Art

[0002] The flavor and texture of soy products, such as tofu, bean curd, soy milk, etc., will change significantly during the fermentation process, affecting the quality of the final product and the acceptance of consumers. Accurate determination of fermentation maturity is a key step in ensuring the quality of soy products. It is not only related to the flavor and taste of the product, but also directly related to production efficiency and economic benefits. In the traditional soy product fermentation process, the evaluation of fermentation maturity usually relies on experienced technicians through sensory evaluation, chemical analysis, etc. Although these methods can reflect the fermentation state to a certain extent, due to the lack of objective quantitative standards, there are often large subjective and non-standard problems. At the same time, in industrial production, dynamic monitoring and real-time adjustment of the fermentation process are of great significance to improving product quality and large-scale production. Therefore, the application of accurate, real-time and automated fermentation maturity determination methods in soy product fermentation technology has important practical significance.

[0003] At present, the evaluation of the maturity of fermented soy products mostly uses single parameter monitoring methods, such as temperature, humidity or pH monitoring. These parameters are measured by simple sensors in traditional processes, and then judged by the operator based on experience and measured data. This method can reflect some characteristics of the fermentation process to a certain extent, but it cannot fully characterize the complex biochemical changes in the fermentation process. In order to enhance the accuracy of the judgment, some studies and applications have begun to introduce chemical analysis methods, such as high performance liquid chromatography (HPLC) to determine amino acid content and gas chromatography to determine short-chain fatty acids. These methods can provide more accurate data, but due to the complexity of the instruments and slow analysis speed, it is difficult to meet the needs of real-time monitoring. In addition, these technologies are mainly used in laboratory environments, and their application in actual production is limited.

[0004] Existing technologies have some obvious shortcomings in monitoring fermentation maturity. First, technologies based on single parameter monitoring cannot fully reflect the combined effects of multiple biochemical reactions in the fermentation process of soy products, resulting in inaccurate and unstable evaluation results. Secondly, although chemical analysis methods can provide detailed component data, their processes are complicated and cannot achieve real-time online monitoring, making it difficult to adapt to the efficient and continuous production needs in modern industrial production. Furthermore, these traditional technologies usually rely on manual judgment and operation, making it difficult to achieve automation and standardized management, resulting in a greater impact of the operator's experience and skills on the control of the fermentation process, thereby increasing uncertainty and risk in production. In summary, existing technologies cannot simultaneously take into account the three key requirements of real-time, comprehensive and automated, which to a certain extent limits the optimization and quality control of the fermentation process of soy products.

[0005] The present invention proposes a method for determining the fermentation maturity of soy products based on multi-sensor data fusion and neural network algorithm, which effectively solves the above-mentioned shortcomings of the prior art. By integrating multiple sensors such as temperature, humidity, pH value, ammonia concentration, and amino acid and short-chain fatty acid concentrations, the present invention can comprehensively and real-time monitor the fermentation environment and fermentation status. The data of these sensors are transmitted to the data processing module wirelessly, and after cleaning and standardization, they are input into the trained neural network model to output the fermentation maturity results. This method can realize fully automated monitoring of the fermentation process, significantly improving the accuracy and robustness of maturity judgment. Therefore, the present invention is significantly superior to traditional technologies in realizing real-time, comprehensive and automated determination of fermentation maturity, and solves the various limitations and shortcomings of traditional methods in the production process.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for determining the fermentation maturity of a soy product to solve the problems raised in the above-mentioned background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A method for determining fermentation maturity of a bean product, the specific steps comprising:

[0010] Step 1: Real-time monitoring of environmental data and fermentation data of the fermentation environment of soybean products, wherein the environmental data include temperature, humidity and pH value, and the fermentation data include ammonia concentration data and corresponding fermentation days, collecting ammonia concentration data and corresponding fermentation days in historical data, and corresponding amino acid and short-chain fatty acid concentration data in fermented soybean products, and preprocessing the collected data;

[0011] Step 2: Construct a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the soy product fermentation into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy product, and generate the soy product's own influencing parameters;

[0012] Step 3: Analyze and calculate the corresponding environmental impact parameters using the pre-treated temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product;

[0013] Step 4: Compare the comprehensive judgment value of the fermentation maturity of the soy product with the set maturity threshold, and judge the fermentation maturity of the soy product according to the comparison result.

[0014] Furthermore, temperature sensors, humidity sensors and pH sensors are installed in the fermentation room respectively, and biomarker sensors are installed on the surface or inside of the soy products, which include enzyme sensors and optical fiber chemical sensors. These sensors monitor the temperature T, humidity H and pH value pH of the fermentation environment in real time; fermentation data include ammonia concentration data NH3 and the corresponding fermentation days D, recording the number of days from the start of fermentation to the current moment; ammonia concentration data in historical data is collected And the corresponding fermentation days D h , and the corresponding amino acid AA in fermented soy products h and short-chain fatty acid concentration data SCFA h .

[0015] Furthermore, data preprocessing includes filtering out noise data and filling in missing data. The moving average method calculates the average value of a set of data within a certain window period and smoothes the data to filter out noise and process the temperature sensor data:

[0016]

[0017] Among them, T s，i represents the smoothed temperature measurement value at the i-th moment, T i+j represents the original temperature measurement value at the i+jth moment, and n is the size of the moving average window;

[0018] The median filter method filters out noise by taking the median of the values ​​in a window and processes the humidity sensor data:

[0019] H s，i =Median(H (i-m / 2) , H(i-m / 2+1) , …, H i , H (i+m / 2) )

[0020] Among them, H s，i represents the humidity measurement value after processing at the i-th moment, H i represents the original humidity measurement value at the i-th moment, and m is the size of the median filter window;

[0021] For missing data, linear interpolation is used to fill the gaps for pH sensor data:

[0022]

[0023] Among them, pH s，i represents the interpolation of missing data points, pH i-1 , pH i+1 represents the known pH measurements before and after the missing data point;

[0024] Processing fermentation days data:

[0025]

[0026] Among them, D s，i Interpolation for missing data points;

[0027] The moving average method, median filtering method and linear interpolation method are used to process the concentration of ammonia NH3, the concentration of amino acid AA and short-chain fatty acid SCFA in the same way to obtain the processed data.

[0028] Furthermore, feature data and label data are input to construct the input layer of the deep learning network. The ammonia concentration data and the corresponding fermentation days in the historical data are used as the training set. The input feature vector combination is:

[0029]

[0030] in, The data after data cleaning, feature dimension n in =2;

[0031] Use the amino acid concentration and short-chain fatty acid concentration in the historical data as labels:

[0032] Y=[AA h , SCFA h ]

[0033] Among them, the label dimension of Y is 2, corresponding to AA h and SCFA h , which are historical amino acid concentration data and exercise fatty acid concentration data:

[0034] The input layer passes X to the hidden layer of the deep learning network, with n nodes. hi , the output is:

[0035] H (1) =f(W (1) X+b (1) )

[0036] Among them, W (1) is the weight matrix from the input layer to the first hidden layer, with a dimension of n hi ×n in , b (1) is the bias term of the first hidden layer, with dimension n hi , f is a nonlinear activation function;

[0037] Assume that the network contains L hidden layers, and the output of the lth layer is:

[0038] H (l) =f(W (l) H (l-1) +b (l) )

[0039] Among them, H (l-1) is the output of the previous layer, W (l) and b (l) are the weight and bias of the lth layer respectively;

[0040] The output of the hidden layer H (l) It is passed to the output layer, forming the output of the connected output layer:

[0041] Y p =W (o) H (L) +b (o)

[0042] Among them, H (L) is the output of the last hidden layer, W (o) The weight matrix of the output layer has a dimension of 2×n hi , b (o) is the bias term of the output layer, with dimension 2.

[0043] Furthermore, the model performance is evaluated by calculating the mean square error between the predicted value and the true value. Suppose there are a samples, AA t，i and AA p，i are the amino acid concentrations of the true value and predicted value of the i-th sample; SCFA t，i and SCFA p，i is the true value and predicted value of short-chain fatty acids, then the mean square error loss function can be expressed as:

[0044]

[0045] Where a is the sample size, AA p，i and SCFA p，i is the value predicted by the model, AA t，i and SCFA t，i is the actual value;

[0046] The model is optimized by iteratively optimizing the model parameters to minimize the loss function. The gradient descent update formula is:

[0047]

[0048] Among them, η represents the learning rate, Represents the gradient of the loss function with respect to weights and biases;

[0049] After the model training is completed, the real-time monitored fermentation data is input into the model for prediction:

[0050]

[0051] in, D s，i The ammonia concentration data and the corresponding fermentation days after data cleaning for real-time monitoring;

[0052] Use the trained deep learning model to predict the input real-time data, pass the input data in to get the output:

[0053] Y pr =W (o) H (L) +b (o)

[0054] According to the input X r Calculate the hidden layer output H (L) :

[0055]

[0056]

[0057] The prediction result given by the final output layer is:

[0058] Y pr =[AA p , SCFA p ]

[0059] Among them, AA p is the predicted amino acid concentration, SCFA p is the predicted short-chain fatty acid concentration;

[0060] After the training is completed, the self-influence parameters of soy products are predicted and generated:

[0061] P p =[sum(AA p ), sum(SCFA p )]

[0062] Among them, P p It represents the influencing parameters of soy products themselves, and sum() is a function used to calculate the sum of the amino acid concentration and short-chain fatty acid concentration predicted by the model in the array.

[0063] Furthermore, the pre-processed environmental data is used for analysis to calculate the environmental impact parameter P e :

[0064]

[0065] Among them, T s，i , H s，i , pH s，i All are data after data processing;

[0066] Combining the soy product's own impact parameters and environmental impact parameters, the final judgment value C is generated:

[0067] C=α·P p +β·P e

[0068] Among them, α reflects the importance of the impact of soy products themselves in the comprehensive judgment, and β reflects the importance of environmental impact in the comprehensive judgment.

[0069] Furthermore, a maturity threshold is set, and the comprehensive judgment value C is compared with the set bean product maturity threshold C T , make a maturity judgment:

[0070]

[0071] Among them, IM stands for immature and M stands for mature.

[0072] The present invention further provides a soybean product fermentation maturity determination system, which is used to execute the above-mentioned soybean product fermentation maturity determination method, comprising:

[0073] A data acquisition module is used to monitor the environmental data and fermentation data of the fermentation environment of the soy product in real time, wherein the environmental data includes temperature, humidity and pH value, and the fermentation data includes ammonia concentration data and corresponding fermentation days, collects the ammonia concentration data and corresponding fermentation days in the historical data, and the corresponding amino acid and short-chain fatty acid concentration data in the fermented soy product, and pre-processes the collected data;

[0074] The model building module is used to build a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the fermentation of soy products into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy products, and generate the self-influencing parameters of the soy products;

[0075] The data processing module is used to analyze and calculate the corresponding environmental impact parameters using the pre-processed temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product;

[0076] The data judgment module is used to compare the comprehensive judgment value of the maturity of the bean product with the set maturity threshold, and judge the fermentation maturity of the bean product according to the comparison result.

[0077] Compared with the prior art, the present invention has the following beneficial effects:

[0078] The present invention integrates multiple sensors such as temperature, humidity, pH value, ammonia concentration, and amino acid and short-chain fatty acid concentrations to monitor the fermentation environment and fermentation status comprehensively and in real time. These multi-parameter data are transmitted wirelessly to the data processing module, so that the system has the ability to comprehensively reflect the fermentation process. Compared with the single parameter monitoring method, the comprehensiveness and accuracy of the data are significantly improved.

[0079] The present invention designs a special data cleaning and processing module to denoise and standardize data from a variety of sensors. Through mathematical algorithms such as moving average filtering, the sensor data is smoothed and random noise is removed, thereby obtaining more stable and reliable measurement data. In addition, through analog-to-digital conversion and linear calibration, the different dimensional data output by the sensor is converted into comparable consistent dimensional data, which enhances the availability and consistency of the data. These data cleaning and processing steps effectively improve the data quality and provide reliable basic data guarantee for subsequent intelligent judgment.

[0080] The present invention adopts an advanced neural network model to perform fusion analysis on the cleaned multi-parameter data. By training a large amount of historical fermentation data, the neural network model can learn and identify the complex nonlinear relationship between different parameters, so as to more accurately determine the fermentation maturity. Specifically, the neural network inputs multi-sensor data, performs feature extraction and nonlinear mapping through the neural network model, and finally outputs the fermentation maturity result. This process can give full play to the powerful function of machine learning, dig out data patterns and rules that cannot be effectively utilized by traditional methods, and realize accurate evaluation of fermentation maturity.

[0081] The present invention effectively solves the shortcomings of the existing technology in determining the maturity of fermented soy products. Through the comprehensive application of technical features such as multi-parameter fusion monitoring, data cleaning and processing, neural network analysis and intelligent real-time monitoring, the automation and intelligence level of the fermentation process is significantly improved, the quality consistency and production efficiency of the fermented products are improved, and it has important economic and social value. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 It is a schematic diagram of the overall method flow of the present invention;

[0083] Figure 2 It is a schematic diagram of the overall system module of the present invention. DETAILED DESCRIPTION

[0084] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0085] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0086] Example:

[0087] See also Figure 1 , the present invention provides a technical solution:

[0088] A method for determining fermentation maturity of a bean product, the specific steps comprising:

[0089] Step 1: Real-time monitoring of environmental data and fermentation data of the fermentation environment of soybean products, wherein the environmental data include temperature, humidity and pH value, and the fermentation data include ammonia concentration data and corresponding fermentation days, collecting ammonia concentration data and corresponding fermentation days in historical data, and corresponding amino acid and short-chain fatty acid concentration data in fermented soybean products, and preprocessing the collected data;

[0090] Temperature sensors, humidity sensors and pH sensors are installed in the fermentation room respectively, and biomarker sensors are installed on the surface or inside of the soy products, which include enzyme sensors and optical fiber chemical sensors. These sensors monitor the temperature T, humidity H and pH value pH of the fermentation environment in real time; fermentation data includes ammonia concentration data NH3 and the corresponding fermentation days D, recording the number of days from the start of fermentation to the current moment; ammonia concentration data in historical data is collected And the corresponding fermentation days D h , and the corresponding amino acid AA in fermented soy products h and short-chain fatty acid concentration data SCFA h ;

[0091] The temperature sensor is used to monitor the temperature of the fermentation environment. The temperature sensor is connected to the integrated data acquisition terminal and transmits the data to the data processing module in real time through the wireless communication module:

[0092]

[0093] Where T is temperature, V out is the sensor output voltage, V max is the maximum output voltage of the sensor, R is the proportional constant;

[0094] Humidity sensor, used to monitor the humidity of the fermentation environment, the humidity sensor is connected to the integrated data acquisition terminal and transmits data through the wireless communication module:

[0095]

[0096] Where, H is relative humidity (percentage), V out is the sensor output voltage, V max is the maximum output voltage of the sensor;

[0097] pH sensor, used to monitor the pH value of soy products during fermentation. The pH sensor is connected to the integrated data acquisition terminal and transmits data through the wireless communication module:

[0098]

[0099] Where pH is the pH value, P is the sensor constant, V out is the sensor output voltage, V ref is the reference voltage;

[0100] Ammonia sensor, used to monitor ammonia concentration in the fermentation environment, is connected to the integrated data acquisition terminal and transmits data through the wireless communication module:

[0101]

[0102] Among them, NH3 is the ammonia concentration, K is the sensor constant, V out is the sensor output voltage, V max is the maximum output voltage of the sensor;

[0103] Biosensors detect the concentration of amino acids AA and short-chain fatty acids SCFA in soy products in real time. Multiple biosensors are integrated on the surface or inside of soy products, and data is transmitted through wireless communication modules. For amino acids AA:

[0104] AA=C A ·exp(-τ A ·t)

[0105] Where AA is the amino acid concentration, C A is the initial concentration of amino acids, τ A is the time constant, t is the time;

[0106] For short-chain fatty acids (SCFA):

[0107] SCFA=C S ·exp(-τ S ·t)

[0108] Among them, SCFA is the concentration of short-chain fatty acids, C S is the initial concentration of short-chain fatty acids, τ S is the time constant, t is the time;

[0109] The constants R, P, K, C of each sensor can be calibrated by combining the characteristics of the sensor and actual experimental data A , C S , τ A and τ S Etc., to ensure the accuracy of the conversion formula in practical applications;

[0110] The data processing module mainly includes three steps: data cleaning, format conversion and data storage to ensure that the sensor data can be processed in a high-quality and standardized manner, providing a reliable data basis for subsequent analysis; data cleaning is the first step in processing sensor data, mainly including filtering out noise data and filling in missing data. Commonly used techniques include moving average method and median filtering;

[0111] Moving Average is a simple filtering method that processes temperature sensor data by calculating the average value of a set of data within a certain window period and smoothing the data to filter out noise:

[0112]

[0113] Among them, T s，irepresents the temperature measurement value after smoothing at the i-th moment. By taking the average of the original measurement values ​​within the window period, the influence of mutation and random noise is reduced. i+j is the original temperature measurement value at the i+jth moment, indicating the actual temperature data of the sensor at each moment, and n is the size of the moving average window;

[0114] If the original measurement value T i+j A sudden increase in a value in the s，i Slightly increased, but due to the averaging process, this effect is smoothed. When abnormal points (noise) appear in the original data, the moving average method can effectively reduce the impact of such abnormalities on the smoothed value by smoothing multiple data points.

[0115] Median filtering is a nonlinear signal processing technology that filters out noise by taking the median of the values ​​in a window. It is particularly suitable for removing spike noise and processing humidity sensor data:

[0116] H s，i =Median(H i-m / 2 , H i-m / 2+1， …, H i , H i+m / 2 )

[0117] Among them, H s，i represents the humidity measurement value after processing at the i-th moment, H i represents the original humidity measurement value at the i-th moment, and m is the size of the median filter window;

[0118] Independent variable (H i-m / 2 , …, H i+m / 2 ) is the raw humidity measurement value in a series of windows, and the dependent variable H s，i It is the median of these values. If a value in the window is particularly large (noise), the median filter will ignore this deviation value and select the value in the middle. The median filter is suitable for removing occasional large noise and can smooth the data sequence to make it closer to the actual situation of continuous change.

[0119] For missing data, we can use interpolation method, linear interpolation to fill the gaps and process pH sensor data:

[0120]

[0121] Among them, pH s，i represents the interpolation of missing data points, pH i-1 , pH i+1 represents the known pH measurements before and after the missing data point;

[0122] Independent variable pH i-1, pH i+1 is the known pH measurement value before and after the missing point. The dependent variable is the estimated value obtained by linear interpolation. The larger the value before and after the missing point, the larger the interpolation result. Interpolation will smoothly fill the missing point to ensure the continuity and integrity of the data series.

[0123] Fermentation days data usually do not need to be processed. If there is missing data, linear interpolation is used to supplement it:

[0124]

[0125] Among them, D s，i Interpolation for missing data points;

[0126] The moving average method, median filtering method and linear interpolation method are used to process the concentration of ammonia NH3, the concentration of amino acid AA and short-chain fatty acid SCFA in the same way to obtain the processed data.

[0127] Step 2: Construct a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the soy product fermentation into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy product, and generate the soy product's own influencing parameters;

[0128] Input feature data and label data to construct the input layer of the deep learning network. Use the ammonia concentration data and the corresponding fermentation days in the historical data as the training set. The input feature vector combination is:

[0129]

[0130] in, The data after data cleaning, feature dimension n in =2, the number of nodes in the input layer, indicating the number of input features;

[0131] X is the input feature vector, which contains the cleaned and processed data; The smoothed historical ammonia concentration affects the growth and fermentation process of microorganisms; The number of days of fermentation after smoothing affects the flavor and preservation of soy products;

[0132] Use the amino acid concentration and short-chain fatty acid concentration in the historical data as labels:

[0133] Y=[AA h , SCFA h ]

[0134] Among them, the label dimension of Y is 2, which indicates the number of predicted target variables, corresponding to AA h and SCFA h , which are historical amino acid concentration data and exercise fatty acid concentration data;

[0135] The input layer passes X to the hidden layer of the deep learning network, with n nodes. hi , the output is:

[0136] H (1) =f(W (1) X+b (1) )

[0137] Among them, W (1) is the weight matrix from the input layer to the first hidden layer, with a dimension of n hi ×n in , each weight affects the output degree of the input feature in the first hidden layer, b (1) is the bias term of the first hidden layer, with dimension n hi , used to adjust the output value and enhance the flexibility of the model. f is a nonlinear activation function (such as ReLU and Sigmoid), which introduces nonlinear characteristics to help the model learn complex patterns. (1) is the output of the first hidden layer, after being processed by the activation function;

[0138] Assume that the network contains L hidden layers, and the output of the lth layer is:

[0139] H (l) =f(W (l) H (l-1) +b (l) )

[0140] Among them, H (l-1) is the output of the previous layer, W (l) and b (l) are the weights of the lth layer, which control the contribution and bias of the previous layer output to the current layer result and adjust the output of this layer; H (l) is the output of the lth layer, indicating the processing result of this layer on the output of the previous layer;

[0141] The output of the hidden layer H (l) It is passed to the output layer, forming the output of the connected output layer:

[0142] Y p =W (o) H (L) +b (o)

[0143] Among them, H (L) is the output of the last hidden layer, providing input for the output layer, W(o) The weight matrix of the output layer has a dimension of 2×n hi , control the influence of hidden layer features on the final prediction results, b (o) Y is the bias term of the output layer, which adjusts the output value to better match the target label, with a dimension of 2; p The final output of the model includes the predicted amino acid and short-chain fatty acid concentrations, which transforms the hidden layer features into the final prediction results for evaluating the maturity state of soy products;

[0144] The model performance is evaluated by calculating the mean square error between the predicted value and the true value. Suppose there are a samples, AA t，i and AA p，i are the amino acid concentrations of the true value and predicted value of the i-th sample; SCFA t，i and SCFA p，i is the true value and predicted value of short-chain fatty acids, then the mean square error loss function can be expressed as:

[0145]

[0146] Where a is the sample size, AA p，i and SCFA p，i is the value predicted by the model, AA t，i and SCFA t，i is the actual value; the loss function L reflects the average difference between the model prediction value and the actual value. The smaller its value is, the closer the model prediction is to the true value, indicating that the model performs better;

[0147] AA t，i represents the true amino acid concentration of the i-th sample, AA p，i represents the amino acid concentration predicted by the model for the i-th sample, SCFA t，i Represents the true short-chain fatty acid concentration of the i-th sample, SCFA p，i represents the model-predicted short-chain fatty acid concentration of the i-th sample;

[0148] The independent variable AA in the formula t，i , A.A. p，i , SCFA t，i , SCFA p，i It reflects the difference between the actual value and the predicted value, and these differences directly affect the value of the loss function. The change of the independent variable will affect the output of the loss function, reflecting the influence of the independent variable on the dependent variable.

[0149] The model is optimized by iteratively optimizing the model parameters to minimize the loss function. The loss function (mean square error) is used to measure the gap between the model's predicted value and the true value: the smaller the loss value, the more accurate the model's prediction. The gradient descent update formula is:

[0150]

[0151] Among them, η represents the learning rate, which is a hyperparameter that controls the update amplitude of weights and biases in each step of optimization. It represents the gradient of the loss function with respect to the weight and bias. It is the gradient of the loss function with respect to the weight W and the bias b, that is, the partial derivative of the loss function with respect to the parameter, reflecting the current parameter adjustment direction and size. The gradient descent method iteratively optimizes the weight and bias of the model by calculating the gradient of the loss function for each parameter. With repeated iterations, the gradient descent method gradually reduces the value of the loss function, thereby optimizing the model parameters and improving the accuracy of the prediction to minimize the loss function.

[0152] The larger the learning rate η, the larger the model parameter update amplitude, which may lead to unstable model training; on the contrary, a small learning rate may lead to slow convergence. Through these gradients, we can determine how to adjust W (l) and b (l) To reduce the prediction error;

[0153] After the model training is completed, the real-time monitored fermentation data is input into the model for prediction:

[0154]

[0155] in, D s，i The ammonia concentration data and the corresponding fermentation days after data cleaning for real-time monitoring;

[0156] Use the trained deep learning model to predict the input real-time data, pass the input data in to get the output:

[0157] Y pr =W (o) H (L) +b (o)

[0158] According to the input X r Calculate the hidden layer output H (L) :

[0159]

[0160] The prediction result given by the final output layer is:

[0161] Y pr =[AA p , SCFA p ]

[0162] Among them, AA pis the predicted amino acid concentration, SCFA p is the predicted short-chain fatty acid concentration;

[0163] After the training is completed, the self-influence parameters of soy products are predicted and generated:

[0164] P p =[sum(AA p ), sum(SCFA p )]

[0165] Among them, P p It represents the influencing parameters of soy products themselves, and sum() is a function used to calculate the sum of the amino acid concentration and short-chain fatty acid concentration predicted by the model in the array.

[0166] Among them, P p Indicates the influencing parameters of soy products themselves, including the amino acid concentration and short-chain fatty acid concentration predicted by the model, and the predicted value AA p , SCFA p The larger the size, the more mature the bean products are; the predicted value is AA p and SCFA p is obtained by inputting feature X r This means that changes in input features directly affect the output prediction results.

[0167] Step 3: Analyze and calculate the corresponding environmental impact parameters using the pre-treated temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product;

[0168] Analyze the preprocessed environmental data and calculate the environmental impact parameter P e :

[0169]

[0170] Among them, T s，i , H s，i , pH s，i All are data after data processing;

[0171] P e It is a comprehensive indicator used to reflect the influence of environmental factors on the fermentation maturity of soy products. It comprehensively considers three key environmental factors: temperature, humidity and pH value, and can indicate whether the environmental conditions during the fermentation of soy products are suitable. e The larger the value of P, the higher the degree of support for the fermentation of soy products, which usually means higher maturity and thus may produce better product quality. eSmaller may indicate less than ideal environmental conditions, which may result in inadequate fermentation or undesirable flavor and texture;

[0172] Temperature is one of the key factors affecting the fermentation process. Suitable temperature can accelerate the metabolism and growth of microorganisms, thereby promoting fermentation. In the formula, the temperature is squared and multiplied by the humidity, indicating that the effect of temperature on fermentation is nonlinear. The higher the temperature, the faster the fermentation. s，i The larger the value of e will also increase, indicating more ideal fermentation conditions;

[0173] Humidity affects the activity of microorganisms and the transfer of substances. If the humidity is too low, the fermentation process may be inhibited. In the formula, humidity appears in square form and becomes part of the denominator, which means that an increase in humidity will lead to an improvement in environmental impact parameters. The higher the humidity, the better the overall P e The values ​​tend to increase, reflecting conditions that are more favorable for fermentation;

[0174] pH directly affects the growth and metabolic activity of microorganisms. During the fermentation process, changes in pH will significantly affect the flavor and safety of the product. In the formula, pH s，i Appears in the denominator, humidity squared and pH s，i The ratio shows that when pH s，i When it decreases (more acidic), the parameter P is affected e The value of will increase accordingly, which means a more suitable environment for fermentation;

[0175] The final judgment value C generated by combining the self-influencing parameters of soy products with the environmental influencing parameters reflects the comprehensive index of maturity of soy products:

[0176] C=α·P p +β·P e

[0177] Among them, α reflects the importance of the influence of soy products themselves in the comprehensive judgment, β reflects the importance of environmental influences in the comprehensive judgment, and the final judgment value C combines the nutritional components of soy products themselves (amino acid and short-chain fatty acid concentrations) with the influence of environmental factors. The higher the value, the higher the maturity and quality of the soy products.

[0178] The weights α and β are used to measure the importance of the influence of soy products themselves and the environmental influence in the judgment. If α is larger, the influence of soy products themselves on maturity is more significant, and if β is larger, the influence of environmental factors is more important.

[0179] Step 4: comparing the comprehensive judgment value of the fermentation maturity of the bean product with the set maturity threshold, and judging the fermentation maturity of the bean product according to the comparison result;

[0180] Set the maturity threshold, and compare the comprehensive judgment value C with the set bean product maturity threshold C T , make a maturity judgment:

[0181]

[0182] Among them, IM stands for immature, indicating that the maturity of the bean product has not reached the predetermined standard, and may not be ideal in flavor and nutrition. M stands for mature, indicating that the maturity of the bean product has reached the predetermined standard, which usually means that the flavor and nutritional value are high.

[0183] This judgment logic helps production and quality control departments evaluate the status of soy products. Effective maturity judgment can guide production decisions and determine whether to continue fermentation, sales or further processing;

[0184] In this formula, the comprehensive judgment value C is compared with the set threshold value C T , if C is greater than or equal to C T , it means that the maturity of the bean products has reached the standard; if it is less than C T , it means that the bean product is immature, so the larger the C value is, the more mature the bean product is;

[0185] Factors affecting the comprehensive judgment value C: Higher P p and P e will lead to an increase in CC value, which means that soy products perform well in terms of nutritional content and environmental conditions. p or P e ) increases, which usually leads to an increase in the comprehensive judgment value C, making it more likely to meet C ≥ C T The conditions are judged to be mature;

[0186] Maturity Threshold C T It is a preset standard value, usually set through historical data analysis, industry standards or expert experience, and aims to provide a clear benchmark for judging the maturity of soy products.

[0187] See also Figure 2 The present invention further provides a soybean product fermentation maturity determination system, the system is used to execute the above-mentioned soybean product fermentation maturity determination method, comprising:

[0188] A data acquisition module is used to monitor the environmental data and fermentation data of the fermentation environment of the soy product in real time, wherein the environmental data includes temperature, humidity and pH value, and the fermentation data includes ammonia concentration data and corresponding fermentation days, collects ammonia concentration data and corresponding fermentation days in historical data, and corresponding amino acid and short-chain fatty acid concentration data in the fermented soy product, and pre-processes the collected data;

[0189] The model building module is used to build a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the fermentation of soy products into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy products, and generate the self-influencing parameters of the soy products;

[0190] The data processing module is used to analyze and calculate the corresponding environmental impact parameters using the pre-processed temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product;

[0191] The data judgment module is used to compare the comprehensive judgment value of the maturity of the bean product with the set maturity threshold, and judge the fermentation maturity of the bean product according to the comparison result.

[0192] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0193] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0194] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0195] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for determining the fermentation maturity of a bean product, characterized in that: The specific steps include: Step 1: Real-time monitoring of environmental data and fermentation data of the fermentation environment of soybean products, wherein the environmental data include temperature, humidity and pH value, and the fermentation data include ammonia concentration data and corresponding fermentation days, collecting ammonia concentration data and corresponding fermentation days in historical data, and corresponding amino acid and short-chain fatty acid concentration data in fermented soybean products, and preprocessing the collected data; Step 2: Construct a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the soy product fermentation into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy product, and generate the soy product's own influencing parameters; Step 3: Analyze and calculate the corresponding environmental impact parameters using the pre-treated temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product; Step 4: Compare the comprehensive judgment value of the fermentation maturity of the bean product with the set maturity threshold, and judge the fermentation maturity of the bean product according to the comparison result.

2. The method for determining fermentation maturity of a soybean product according to claim 1, characterized in that: In step 1, the temperature sensor, humidity sensor and pH sensor are installed in the fermentation room respectively, and the biomarker sensor is installed on the surface or inside of the soy product, which includes an enzyme sensor and an optical fiber chemical sensor. These sensors monitor the temperature T, humidity H and pH value pH of the fermentation environment in real time; the fermentation data includes ammonia concentration data NH3 and the corresponding fermentation days D, recording the number of days from the start of fermentation to the current moment; the ammonia concentration data in the historical data is collected And the corresponding fermentation days D h , and the corresponding amino acid AA in fermented soy products h and short-chain fatty acid concentration data SCFA h .

3. The method for determining fermentation maturity of a bean product according to claim 2, characterized in that: Data preprocessing includes filtering out noise data and filling in missing data. The moving average method calculates the average value of a set of data within a certain window period and smoothes the data to filter out noise and process temperature sensor data: Among them, T s，i represents the smoothed temperature measurement value at the i-th moment, T i+j represents the original temperature measurement value at the i+jth moment, n is the size of the moving average window; The median filter method filters out noise by taking the median of the values ​​in a window and processes the humidity sensor data: H s，i =Median(H (i-m / 2) ,H (i-m / 2+1) ,…,H i ,H (i+m / 2) ) Among them, H s，i represents the humidity measurement value after processing at the i-th moment, H i represents the original humidity measurement value at the i-th moment, and m is the size of the median filter window; For missing data, linear interpolation is used to fill the gaps for pH sensor data: Among them, pH s，i represents the interpolation of missing data points, pH i-1 , pH i+1 represents the known pH measurements before and after the missing data point; Processing fermentation days data: Among them, D s，i Interpolation for missing data points; The moving average method, median filtering method and linear interpolation method are used to process the concentration of ammonia NH3, the concentration of amino acid AA and short-chain fatty acid SCFA in the same way to obtain the processed data.

4. The method for determining fermentation maturity of a bean product according to claim 3, characterized in that: In step 2, feature data and label data are input to construct the input layer of the deep learning network. The ammonia concentration data and the corresponding fermentation days in the historical data are used as the training set. The input feature vector combination is: in, The data after data cleaning, feature dimension n in =2; Use the amino acid concentration and short-chain fatty acid concentration in the historical data as labels: Y=[AA h ,SCFA h ] Among them, the label dimension of Y is 2, corresponding to AA h and SCFA h , which are historical amino acid concentration data and exercise fatty acid concentration data; The input layer passes X to the hidden layer of the deep learning network, with n nodes. hi , the output is: H (1) =f(W (1) X+b (1) ) Among them, W (1) is the weight matrix from the input layer to the first hidden layer, with a dimension of n hi ×n in , b (1) is the bias term of the first hidden layer, with dimension n hi , f is a nonlinear activation function; Assume that the network contains L hidden layers, and the output of the lth layer is: H (l) =f(W (l) H (l-1) +b (l) ) Among them, H (l-1) is the output of the previous layer, W (l) and b (l) are the weight and bias of the lth layer respectively; The output of the hidden layer H (l) It is passed to the output layer, forming the output of the connected output layer: Y p =W (o) H (L) +b (o) Among them, H (L) is the output of the last hidden layer, W (o) The weight matrix of the output layer has a dimension of 2×n hi , b (o) is the bias term of the output layer, with dimension 2.

5. The method for determining fermentation maturity of a bean product according to claim 4, characterized in that: The model performance is evaluated by calculating the mean square error between the predicted value and the true value. Suppose there are a samples, AA t，i and AA p，i are the amino acid concentrations of the true value and predicted value of the i-th sample; SCFA t，i and SCFA p，i is the true value and predicted value of short-chain fatty acids, then the mean square error loss function can be expressed as: Where a is the sample size, AA p，i and SCFA p，i is the value predicted by the model, AA t，i and SCFA t，i is the actual value; The model is optimized by iteratively optimizing the model parameters to minimize the loss function. The gradient descent update formula is: Among them, η represents the learning rate, Represents the gradient of the loss function with respect to weights and biases; After the model training is completed, the real-time monitored fermentation data is input into the model for prediction: in, D s，i The ammonia concentration data and the corresponding fermentation days after data cleaning for real-time monitoring; Use the trained deep learning model to predict the input real-time data, pass the input data in to get the output: Y pr =W (o) H (L) +b (o) Among them, according to the input X r Calculate the hidden layer output H (L) : The prediction result given by the final output layer is: THE pr =[AA p ,SCFA p ] Among them, AA p is the predicted amino acid concentration, SCFA p is the predicted short-chain fatty acid concentration; After the training is completed, the self-influence parameters of soy products are predicted and generated: P p =[sum(AA p ),sum(SCFA p )] Among them, P p It represents the influencing parameters of soy products themselves, and sum() is a function used to calculate the sum of the amino acid concentration and short-chain fatty acid concentration predicted by the model in the array.

6. The method for determining fermentation maturity of a bean product according to claim 5, characterized in that: In step 3, the pre-processed environmental data is used for analysis to calculate the environmental impact parameter P e : Among them, T s，i , H s，i , pH s，i All are data after data processing; Combining the soy product's own impact parameters and environmental impact parameters, the final judgment value C is generated: C=α·P p +β·P e Among them, α reflects the importance of the impact of soy products themselves in the comprehensive judgment, and β reflects the importance of environmental impact in the comprehensive judgment.

7. A method for determining fermentation maturity of soybean products according to claim 6, characterized in that: In step 4, a maturity threshold is set, and the comprehensive judgment value C is compared with the set bean product maturity threshold C T , make a maturity judgment: Among them, IM stands for immature and M stands for mature.

8. A soybean product fermentation maturity determination system, characterized in that: The system is used to execute the method for determining fermentation maturity of a bean product according to any one of claims 1 to 7, comprising: A data acquisition module is used to monitor the environmental data and fermentation data of the fermentation environment of the soy product in real time, wherein the environmental data includes temperature, humidity and pH value, and the fermentation data includes ammonia concentration data and corresponding fermentation days, collects ammonia concentration data and corresponding fermentation days in historical data, and corresponding amino acid and short-chain fatty acid concentration data in the fermented soy product, and pre-processes the collected data; The model building module is used to build a deep learning network, input the ammonia concentration data and the corresponding fermentation days in the historical data after feature cleaning as the training set, and the corresponding amino acid concentration and short-chain fatty acid concentration as labels, train the deep learning network, and input the fermentation data of the real-time monitoring environment of the fermentation of soy products into the deep learning network model after training, obtain the predicted amino acid and short-chain fatty acid concentration data inside the fermented soy products, and generate the self-influencing parameters of the soy products; The data processing module is used to analyze and calculate the corresponding environmental impact parameters using the pre-processed temperature, humidity and pH value, and combine them with the inherent impact parameters of the soy product to obtain a comprehensive judgment value of the fermentation maturity of the soy product; The data judgment module is used to compare the comprehensive judgment value of the maturity of the bean product with the set maturity threshold, and judge the fermentation maturity of the bean product according to the comparison result.

Images