Perishable meat product freshness state real-time monitoring method based on big data and AI

Abstract

The application discloses a perishable meat product preservation state real-time monitoring method based on big data and AI, and relates to the field of meat product storage, which comprises the following steps: real-time monitoring of the surface microbial metabolic gas concentration, myoglobin oxidation degree, and dynamic data of the transport environment temperature and humidity, gas composition and vibration amplitude of the perishable meat product, so as to establish a multi-source original monitoring data set; the application captures the related dynamic data of the perishable meat product and the multi-dimensional data of the transport environment in real time, dynamically adjusts the collection frequency to adapt to different preservation state requirements, accurately screens effective data, realizes space-time correlation fusion, and thus efficiently outputs the corruption quantization value, the remaining preservation period and the core influence factor sorting, and quickly generates a targeted transport environment regulation scheme, which realizes real-time optimization of the temperature and humidity, the gas ratio and the vibration reduction parameters, effectively delays the corruption process, prolongs the preservation period, reduces the transportation loss, provides reliable guarantee for the preservation and transportation of the perishable meat product, and reduces the resource waste and economic loss.

Description

Technical Field

[0001] This invention relates to the field of meat product storage technology, specifically a method for real-time monitoring of the freshness status of perishable meat products based on big data and AI. Background Technology

[0002] Perishable meat products are rich in nutrients such as protein, but they are susceptible to microbial contamination and enzymatic hydrolysis, which can lead to spoilage. During transportation, factors such as temperature fluctuations, packaging sealing, and transportation time can cause problems such as color deterioration, flavor loss, and reduced safety, resulting not only in significant product losses but also potential harm to consumer health.

[0003] Perishable meat products typically rely on cold chain transportation. However, existing cold chain transport containers often provide a cold chain transportation environment in a preset mode, failing to intelligently adjust their own mode according to the transportation environment. This can lead to spoilage or partial loss of perishable meat products during transportation because the transport container fails to adapt to the transportation environment and adjust its own operating mode in a timely manner.

[0004] To this end, we propose a real-time monitoring method for the preservation status of perishable meat products based on big data and AI. Summary of the Invention

[0005] In view of the above-mentioned shortcomings of the existing technology, the present invention provides a method for real-time monitoring of the preservation status of perishable meat products based on big data and AI, which can effectively solve the problems of the existing technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions; This invention discloses a method for real-time monitoring of the preservation status of perishable meat products based on big data and AI, including: This system monitors dynamic data in real time on the surface of perishable meat products, including the concentration of microbial metabolic gases, the degree of myoglobin oxidation, and the temperature, humidity, gas composition, and vibration amplitude of the transportation environment, to establish a multi-source raw monitoring dataset. It then iterates through the raw monitoring dataset, removing outliers, and spatiotemporally fuses the dynamic data of meat quality characteristics with multi-parameter data of the transportation environment to generate a standardized feature dataset containing microbial metabolic time-series features and environmental coupling features. A deep learning model driven by the synergistic interaction of microbial metabolism and environmental factors is constructed. A spoilage state identification sub-model is generated based on the microbial metabolic time-series features in the standardized feature dataset, and an environmental impact assessment sub-model is generated based on the environmental coupling features. The standardized feature dataset is input into the trained deep learning model in real time, outputting the current quantified spoilage value of the perishable meat products, the remaining shelf life, and the priority ranking results of core influencing factors. Based on the core influencing factors and the remaining shelf life, a transportation environment control scheme is generated. Based on this scheme, key parameters such as temperature, humidity, gas ratio, and vibration reduction intensity within the transportation cabin are adjusted in real time. Finally, complete monitoring data from each transportation, environmental control records, and the final meat product preservation status verification results are summarized to construct a cross-scenario monitoring database, providing users with access to this database simultaneously. Among them, the dynamic data of meat quality characteristics are the concentration of microbial metabolic gases and the degree of myoglobin oxidation on the surface of perishable meat products, and the multi-parameter data of the transportation environment are the dynamic data of temperature, humidity, gas composition and vibration amplitude in the transportation environment. The combination of the two sub-models is a deep learning model driven by the synergy of microbial metabolism and environmental factors.

[0007] Furthermore, the dynamic data of microbial metabolic gas concentration, myoglobin oxidation degree, and temperature, humidity, gas composition, and vibration amplitude of the transport environment on the surface of the perishable meat products are collected inside the transport cabin through a distributed multi-dimensional sensor array. The distributed multi-dimensional sensor array includes several sensor groups, each sensor group integrating a microbial metabolic gas detection unit, a myoglobin oxidation degree detection unit, a temperature and humidity detection unit, a gas composition detection unit, and a vibration detection unit. The sensor groups are deployed in various corners inside the transport cabin, beside the inlet and outlet airflow channels, and in locations far from the inlet and outlet airflow channels. Each sensor group is equipped with a unique spatial identification code and a time synchronization module. The monitoring frequency of the sensor group is changed in real time based on a preset dynamic adjustment logic. That is, the adjustment basis is determined according to the current remaining shelf life and the rate of change of spoilage. When the current remaining shelf life is less than the preset short period threshold and the rate of change of spoilage exceeds the preset high rate threshold, the sampling frequency is increased to the preset high frequency range; when the current remaining shelf life exceeds the preset long period threshold and the rate of change of spoilage is less than the preset low rate threshold, the sampling frequency is decreased to the preset low frequency range; in other cases, the preset default sampling frequency is maintained.

[0008] Furthermore, when removing abnormal data from the original monitoring data, the determination of whether the data is abnormal is made by calculating the multi-parameter spatiotemporal correlation reliability coefficient of the data points: ; In the formula: Let be the confidence coefficient of the data point corresponding to the i-th sampling time and the j-th sampling position; For parameter difference weighting coefficients; This represents the total number of categories of monitored parameters. This refers to the raw data of the k-th type of monitoring parameter corresponding to the i-th sampling time and the j-th sampling location; The mean value of the k-th type of monitoring parameter within the preset time window adjacent to the i-th sampling time and the preset spatial range adjacent to the j-th sampling location; The historical normal fluctuation standard deviation of the k-th type of monitoring parameter; This is the time deviation weighting coefficient; This is the normalized result of the difference between the i-th sampling time and its adjacent sampling times; Spatial deviation weighting coefficient; This is the normalized distance between the j-th sampling position and its adjacent sampling positions.

[0009] Furthermore, when performing spatiotemporal correlation fusion of dynamic data on meat quality characteristics and multi-parameter data of the transportation environment, the following applies: Correlation calculations are performed on meat quality characteristic data and environmental data within the same spatiotemporal dimension to generate standardized feature data: ; In the formula: The fused feature value corresponds to the i-th sampling time and the j-th sampling position; Weights for fusion of meat quality characteristic data; The standardized value of the meat quality characteristics data corresponding to the i-th sampling time and the j-th sampling position; This is the time decay coefficient; This is the difference between the i-th sampling time and the data fusion reference time; Weighting for transportation environment data fusion; This represents the standardized value of the transportation environment data corresponding to the i-th sampling time and the j-th sampling location; This is the spatial attenuation coefficient; is the distance between the j-th sampling position and the data fusion reference position.

[0010] Furthermore, in the deep learning model driven by the synergistic interaction of microbial metabolism and environmental factors, the putrefaction state identification sub-model applies an attention-enhanced temporal convolutional network: taking microbial metabolic time-series data and dynamic data on myoglobin oxidation as input, a 3-head self-attention module is embedded in the temporal convolutional layer. By calculating the attention weights of features at different time steps, it adaptively focuses on key time node information such as abrupt changes in microbial metabolic rate and inflection points in myoglobin oxidation, with an output dimension of [missing information]. The temporal feature vector of meat spoilage, where The feature vector dimension is adaptively set based on the number of input features, model training convergence efficiency, and feature representation accuracy. The environmental impact assessment sub-model applies a graph attention neural network: Temperature, humidity, oxygen concentration, carbon dioxide concentration, and vibration amplitude from the transportation environment parameters are treated as independent nodes in a graph structure. Each node's initial feature vector includes the real-time value, time-series rate of change, and historical fluctuation characteristics of the corresponding parameter. By calculating the bidirectional association weights between nodes, a multi-head attention mechanism is used to update node features, uncovering the nonlinear coupling relationships between environmental parameters. The output dimension is... The environmental coupling feature vector, where The dimension of the feature vector is positively correlated with the number of environmental parameters and the complexity of parameter coupling relationships; In this process, the meat spoilage time-series feature vector and the environmental coupling feature vector are dimensionally matched by zero-padding or dimensional truncation, and then input into a two-layer fully connected network. The first layer performs a nonlinear transformation on the fused features through the ReLU activation function, and the second layer maps the transformed features to a freshness status evaluation value in the 0-1 interval through the Sigmoid activation function, representing the final decision result of the model.

[0011] Furthermore, the logic for obtaining the quantifiable value of corruption is as follows: ; In the formula: This is a quantification of corruption. Weights for the concentration characteristics of microbial metabolic gases; The time-series peak value of microbial metabolic gas concentration within the monitoring period [0,T]; This is a nonlinear index of microbial metabolic gas concentration; Weights for the degree of myoglobin oxidation; This is a standardized value for the degree of myoglobin oxidation. This is a non-linear index representing the degree of myoglobin oxidation. Number of categories of key parameters for the transportation environment; The influence weight of the k-th type of environmental parameter; This represents the standardized value of the k-th type of environmental parameter; For the k-th type of environmental parameter, there is a nonlinear adjustment coefficient. The total weight is determined by the coupling effect of the environment.

[0012] Furthermore, the logic for calculating the remaining shelf life is as follows: First, the real-time rate of change of the putrefaction metric is obtained through linear fitting. ,in They are respectively A quantification of corruption at any given moment; Then, based on the preset spoilage threshold, calculate the remaining shelf life. , Indicates the corruption threshold. This represents the current quantification of corruption. Indicates the corruption inhibition coefficient. This indicates the real-time rate of change of the corruption quantification. When v is a non-positive value, the remaining preservation period is assigned according to the preset longest period.

[0013] Furthermore, the priority ranking of the core influencing factors is based on the calculation results of the factor's contribution to the impact; The formula for calculating the contribution of the influence is as follows: ; In the formula: Contribution to the impact of the k-th type of environmental factor; Let the corruption quantification value D be relative to the k-th type of environmental factor. The first-order partial derivative; Let be the variance of the k-th type of environmental factor; These are the covariance weighting coefficients; Let $\mathbf{k}$ be the covariance between the $k$-th type of environmental factor and the quantified value of corruption. Let be the second mixed partial derivative of the corruption quantification with respect to the k-th environmental factor and the l-th environmental factor; This represents the total number of categories of environmental factors. according to Sort in descending order to obtain the priority ranking of the core impact factors.

[0014] Furthermore, the transportation environment control scheme, when generated, follows the following: Logic1: For the core influencing factors with the highest priority, determine the corresponding control parameters, including temperature and humidity, gas ratio, or vibration reduction intensity. Logic2: Calculate the parameter adjustment range based on the deviation between the current value and the optimal range of each core influencing factor; Logic3: By Impact Contribution Parameters are adjusted sequentially from largest to smallest.

[0015] Furthermore, when constructing the cross-scenario monitoring database, the core classification dimensions are transportation scenario type, perishable meat product category, and transportation duration range. Various types of data generated during the monitoring process are classified and archived. A hybrid storage structure combining time-series database and relational database is adopted. The time-series database is used to store real-time monitoring data, environmental control operation records, and time-series change sequences of various parameters. The relational database is used to store basic attribute information of perishable meat products, freshness status verification results, scenario characteristic parameters, and transportation task association information.

[0016] Compared with the known prior art, the technical solution provided by this invention has the following beneficial effects: This invention provides a real-time monitoring method for the preservation status of perishable meat products based on big data and AI. During execution, this method captures dynamic data related to perishable meat products and multi-dimensional data of the transportation environment in real time, dynamically adjusts the collection frequency to adapt to different preservation requirements, accurately filters effective data and achieves spatiotemporal correlation fusion, thereby efficiently outputting quantifiable values ​​of spoilage, remaining shelf life, and ranking of core influencing factors, and quickly generating targeted transportation environment control schemes. It optimizes temperature and humidity, gas ratio, and vibration reduction parameters in real time, effectively delaying the spoilage process, extending the shelf life, reducing transportation losses, providing reliable protection for the preservation and transportation of perishable meat products, and reducing resource waste and economic losses. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0018] Figure 1 This is a flowchart illustrating a method for real-time monitoring of the preservation status of perishable meat products based on big data and AI. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The present invention will be further described below with reference to embodiments.

[0021] Example: This embodiment presents a method for real-time monitoring of the preservation status of perishable meat products based on big data and AI, such as... Figure 1 As shown, it includes: Real-time monitoring of dynamic data on the concentration of metabolic gases from microorganisms on the surface of perishable meat products, the degree of myoglobin oxidation, and the temperature, humidity, gas composition, and vibration amplitude of the transportation environment, in order to establish a multi-source raw monitoring dataset; Dynamic data on the concentration of microbial metabolic gases on the surface of perishable meat products, the degree of myoglobin oxidation, and the temperature, humidity, gas composition, and vibration amplitude of the transportation environment are collected inside the transport cabin through a distributed multi-dimensional sensor array. The distributed multi-dimensional sensor array includes several sensor groups, each of which integrates a microbial metabolic gas detection unit, a myoglobin oxidation degree detection unit, a temperature and humidity detection unit, a gas composition detection unit, and a vibration detection unit. The sensor groups are deployed in various corners inside the transport cabin, beside the inlet and outlet airflow channels, and in locations far from the inlet and outlet airflow channels. Each sensor group is equipped with a unique spatial identification code and a time synchronization module. The monitoring frequency of the sensor array is changed in real time based on a preset dynamic adjustment logic. That is, the adjustment is determined according to the current remaining shelf life and the rate of change of spoilage. When the current remaining shelf life is less than the preset short-cycle threshold and the rate of change of spoilage exceeds the preset high-rate threshold, the sampling frequency is increased to the preset high-frequency range. When the current remaining shelf life exceeds the preset long-cycle threshold and the rate of change of spoilage is less than the preset low-rate threshold, the sampling frequency is decreased to the preset low-frequency range. In other cases, the preset default sampling frequency is maintained. By traversing the original monitoring dataset and removing abnormal data, dynamic data of meat quality characteristics and multi-parameter data of transportation environment are spatiotemporally correlated and fused to generate a standardized feature dataset containing microbial metabolic time-series features and environmental coupling features. When removing abnormal data from the original monitoring data, the reliability coefficient of the multi-parameter spatiotemporal correlation of data points is calculated to determine whether the data is abnormal: ; In the formula: Let be the confidence coefficient of the data point corresponding to the i-th sampling time and the j-th sampling position; For parameter difference weighting coefficients; This represents the total number of categories of monitored parameters. This refers to the raw data of the k-th type of monitoring parameter corresponding to the i-th sampling time and the j-th sampling location; The mean value of the k-th type of monitoring parameter within the preset time window adjacent to the i-th sampling time and the preset spatial range adjacent to the j-th sampling location; The historical normal fluctuation standard deviation of the k-th type of monitoring parameter; This is the time deviation weighting coefficient; This is the normalized result of the difference between the i-th sampling time and its adjacent sampling times; Spatial deviation weighting coefficient; This is the normalized distance between the j-th sampling position and its adjacent sampling positions; The above formula comprehensively considers three core factors: differences in monitoring parameters, time deviation, and spatial deviation. By introducing weighting coefficients, it achieves flexible adaptation to different influencing dimensions. With the help of normalization processing and standard deviation calibration, it accurately quantifies the spatiotemporal correlation credibility of data points and effectively identifies and eliminates abnormal data caused by sensor errors, sudden environmental changes, etc. when When the data is less than the preset confidence threshold, the corresponding data point is determined to be abnormal data and is removed. in, >0 indicates a larger value when the stability requirements of the monitoring parameter are high and the impact of parameter anomalies on the judgment of the state of decay is significant, and a smaller value when the monitoring parameter itself allows for large normal fluctuations and the impact of parameter differences on the reliability of the data is small. >0 indicates a larger value when the sampling time interval is uneven and the data correlation in the time dimension is required, and a smaller value when the sampling frequency is stable and the time deviation has little impact on the accuracy of the monitoring results. >0 indicates a larger value when the spatial environment inside the transport compartment is highly heterogeneous and the monitoring parameters at different locations differ significantly; conversely, a smaller value indicates a smaller value when the spatial distribution is uniform and the location differences have little impact on the validity of the data. When spatiotemporally fusing dynamic data on meat quality characteristics with multi-parameter data of the transportation environment, the following applies: Correlation calculations are performed on meat quality characteristic data and environmental data within the same spatiotemporal dimension to generate standardized feature data: ; In the formula: The fused feature value corresponds to the i-th sampling time and the j-th sampling position; Weights for fusion of meat quality characteristic data; The standardized value of the meat quality characteristics data corresponding to the i-th sampling time and the j-th sampling position; This is the time decay coefficient; This is the difference between the i-th sampling time and the data fusion reference time; Weighting for transportation environment data fusion; This represents the standardized value of the transportation environment data corresponding to the i-th sampling time and the j-th sampling location; This is the spatial attenuation coefficient; The distance between the j-th sampling position and the data fusion reference position; The above formula is based on the inherent relationship between meat quality characteristic data and transportation environment data. By setting fusion weights, a reasonable ratio of the two types of data is achieved. At the same time, time decay coefficient and spatial decay coefficient are introduced to adapt to the shelf life of perishable meat products, the degree of temperature deviation in the transportation environment, the uniformity of spatial distribution in the vehicle environment, and the stacking density of meat products, respectively. Furthermore, the interference of non-reference spatiotemporal dimension data is weakened by the exponential decay model, so as to achieve efficient fusion of the two types of data under different spatiotemporal conditions and finally generate standardized data that can comprehensively reflect the temporal characteristics of microbial metabolism and environmental coupling characteristics. in, , All are positive numbers and their sum is 1. When the monitoring accuracy, temporal stability, and correlation with spoilage status of meat quality characteristic data are more significant, The larger the value, the more likely the meat quality data is to be affected by fluctuations, reduced reliability, or a greater dominance of the transportation environment in driving spoilage. The smaller the value, the more prominent the real-time performance, spatial consistency, and coupling driving effect of transportation environment data on meat quality changes. The larger the value, the more likely it is that environmental data fluctuations are gradual, their impact on spoilage is weakened, or the characteristics of the meat itself dominate the spoilage process. The smaller the value; >0 indicates that the value is larger when the shelf life of perishable meat products is shorter and the temperature of the transportation environment deviates significantly from the optimal preservation range, and smaller when the shelf life of perishable meat products is longer and the temperature of the transportation environment is closer to the optimal preservation range. >0 indicates that the higher the degree of spatial non-uniformity of the environmental parameters in the transport compartment and the greater the stacking density of perishable meat products, the larger the value; the more uniform the spatial distribution of the environmental parameters in the transport compartment and the smaller the stacking density of perishable meat products, the smaller the value. The value is obtained by performing linear normalization on the original data of microbial metabolic gas concentration and myoglobin oxidation degree corresponding to the i-th sampling time and j-th sampling location, and then weighting and summing them according to the preset meat quality parameter weight ratio. The value range is [0,1]. The value is obtained by performing linear normalization on the original data of temperature, humidity, gas composition and vibration amplitude corresponding to the i-th sampling time and j-th sampling location, and then weighting and summing them according to the preset environmental parameter weight ratio. The value range is [0,1]. A deep learning model driven by the synergistic interaction of microbial metabolism and environmental factors is constructed. A sub-model for identifying putrefaction status is generated based on the time-series features of microbial metabolism in a standardized feature dataset, and a sub-model for assessing environmental impact is generated based on the environmental coupling features. In a deep learning model driven by the synergy of microbial metabolism and environmental factors, the putrefaction state identification sub-model employs an attention-enhanced temporal convolutional network. Taking microbial metabolic time-series data and dynamic data on myoglobin oxidation as input, a three-head self-attention module is embedded in the temporal convolutional layer. By calculating the attention weights of features at different time steps, it adaptively focuses on key time nodes such as abrupt changes in microbial metabolic rate and inflection points in myoglobin oxidation. The output dimension is... The temporal feature vector of meat spoilage, where The feature vector dimension is adaptively set based on the number of input features, model training convergence efficiency, and feature representation accuracy. The core basis is the coupled calculation result of the number of monitoring parameter categories and the temporal window length, satisfying... , The input is the number of categories of meat quality-related parameters, and L is the length of the time series window. This is the feature dimension adjustment coefficient, and its value range is determined by model training and validation. The environmental impact assessment sub-model applies a graph attention neural network: Temperature, humidity, oxygen concentration, carbon dioxide concentration, and vibration amplitude are treated as independent nodes in a graph structure. Each node's initial feature vector includes the real-time value, temporal rate of change, and historical fluctuation characteristics of the corresponding parameter. Bidirectional correlation weights between nodes are calculated, with weights derived from the Pearson correlation coefficient and temporal synchronicity coefficient between parameters. A multi-head attention mechanism is used to update node features, uncovering the nonlinear coupling relationships between environmental parameters. The output dimension is... The environmental coupling feature vector, where The feature vector dimension is positively correlated with the number of environmental parameters and the complexity of parameter coupling, satisfying the following conditions: , Where m is the number of environmental parameter categories, and m is the number of edges associated with the node. This is the adjustment coefficient for the environmental characteristic dimension, and... Independent optimization; In this process, the meat spoilage time-series feature vector and the environmental coupling feature vector are dimensionally matched by zero-padding or dimensional truncation, and then input into a two-layer fully connected network. The first layer performs a non-linear transformation on the fused features through the ReLU activation function, and the second layer maps the transformed features to a freshness status evaluation value in the 0-1 interval through the Sigmoid activation function, which represents the final decision result of the model. The standardized feature dataset is input into the trained deep learning model in real time, and the output is the current spoilage quantification value of perishable meat products, the remaining shelf life, and the priority ranking results of core influencing factors. The logic for obtaining the corruption quantification value is as follows: ; In the formula: This is a quantification of corruption. Weights for the concentration characteristics of microbial metabolic gases; The time-series peak value of microbial metabolic gas concentration within the monitoring period [0,T]; This is a nonlinear index of microbial metabolic gas concentration; Weights for the degree of myoglobin oxidation; This is a standardized value for the degree of myoglobin oxidation. This is a non-linear index representing the degree of myoglobin oxidation. Number of categories of key parameters for the transportation environment; The influence weight of the k-th type of environmental parameter; This represents the standardized value of the k-th type of environmental parameter; For the k-th type of environmental parameter, there is a nonlinear adjustment coefficient. The environmental coupling influences the total weight; The above formula focuses on the comprehensive impact of microbial metabolic gas concentration, myoglobin oxidation degree, and transportation environment parameters on spoilage. It distinguishes the proportion of each factor in the spoilage process through characteristic weights, and uses a nonlinear exponent to fit the difference in the rate of change of different indicators with the spoilage process. The environmental parameter part introduces a nonlinear adjustment coefficient to accurately characterize the spoilage acceleration effect after the parameters exceed the optimal range. At the same time, it integrates the synergistic effect of various environmental factors through the total weight of environmental coupling influence, so as to achieve accurate quantification of the spoilage degree of perishable meat products and provide an effective quantitative indicator for the evaluation of freshness status. in, ∈ (0,1), the higher the proportion of the influence of the concentration of microbial metabolic gas on the degree of spoilage during the spoilage process of this type of perishable meat products, the larger the value; the weaker the driving effect of microbial metabolism on spoilage, the smaller the value. ∈ (1,3), the faster the growth rate of microbial metabolic gas concentration with the putrefaction process, the larger the value; the smaller the value, the more the gas concentration is approximately linearly related to the degree of putrefaction. ∈ (0,1), the higher the sensitivity of the freshness perception of this type of perishable meat products to the degree of myoglobin oxidation, the larger the value; the smaller the impact of myoglobin oxidation on meat freshness, the smaller the value. ∈ (1, 2.5), the faster the rate of change of myoglobin oxidation with transport time or ambient oxygen concentration, the larger the value; the slower the change in oxidation degree, the smaller the value. For each environmental parameter ∈ (0, 1), the more significant the individual effect of the k-th type of environmental parameter on the spoilage of this type of meat product, the larger its value; the smaller the fluctuation of this parameter on spoilage, the smaller its value. For example, if the value of environmental parameter k exceeds the optimal range, the more obvious the accelerated decay effect is, the larger the value is; if the decay effect of environmental parameter is slow to change in the non-optimal range, the smaller the value is. ∈ (0.1, 0.8), the stronger the comprehensive driving effect of transportation environment parameters on the spoilage of this type of meat products, the larger the value; the weaker the influence of environmental factors on spoilage, the smaller the value. The calculation logic for the remaining shelf life is as follows: First, the real-time rate of change of the putrefaction metric is obtained through linear fitting. ,in They are respectively A quantification of corruption at any given moment; Then, based on the preset spoilage threshold, calculate the remaining shelf life. , Indicates the corruption threshold. This represents the current quantification of corruption. Indicates the corruption inhibition coefficient. This indicates the real-time rate of change of the corruption quantification. The above formula obtains the real-time rate of change of spoilage through linear fitting, which intuitively reflects the dynamic trend of the spoilage process. Combined with the preset spoilage threshold and the current quantified value of spoilage, a spoilage inhibition coefficient is introduced to adapt the adjustment range and blocking effect of the core influencing factors. When the spoilage rate is not positive, the preset longest period is assigned a value, which not only ensures the scientificity and timeliness of the period calculation, but also can deal with special scenarios such as the stagnation of the spoilage process, and provides a reference for the preservation decision during transportation. When v is a non-positive value, the remaining preservation period is assigned according to the preset longest period. in, The preset value range is [0.1, 0.9]. The closer the adjustment range of the core influencing factor is to its optimal range, the more significant the blocking effect of the control measures on the key pathway of meat product spoilage. The larger the value, the more the control range deviates from the optimal range and the weaker the blocking effect. The smaller the value; The priority ranking of core impact factors is based on the calculation results of the factor's impact contribution. The formula for calculating the contribution of influence is: ; In the formula: Contribution to the impact of the k-th type of environmental factor; Let the corruption quantification value D be relative to the k-th type of environmental factor. The first-order partial derivative reflects the strength of the direct influence of the factor on the degree of corruption; Let be the variance of the k-th type of environmental factor; These are the covariance weighting coefficients; Let $\mathbf{k}$ be the covariance between the $k$-th type of environmental factor and the quantified value of corruption. The second mixed partial derivative of the corruption quantification with respect to the k-th and l-th environmental factors reflects the intensity of the interaction between the factors. This represents the total number of categories of environmental factors. The above formula quantifies the direct impact of environmental factors on the degree of corruption through first-order partial derivatives, combines factor variance and covariance to reflect their own fluctuations and the stability of their correlation with the degree of corruption, introduces second-order mixed partial derivatives to consider the interaction effects between factors, balances the overall weight of multi-factor interactions through mean-based processing, and adapts the covariance weight coefficient to the covariance stability and the strength of factor interaction effects, comprehensively and accurately calculates the impact contribution of each environmental factor, and provides support for the priority ranking of core impact factors. according to Sort in descending order to obtain the priority ranking results of the core impact factors; in, ∈ (0,1), when the covariance stability of the k-th environmental factor and the quantification of corruption is higher and the interaction between this factor and other environmental factors is weaker. The larger the value, the greater the covariance fluctuation and the stronger the interaction between this factor and other environmental factors. The smaller the value; Based on the core influencing factors and the remaining shelf life, a transportation environment control plan is generated, and key parameters such as temperature and humidity, gas ratio and vibration reduction strength in the transportation cabin are adjusted in real time based on the transportation environment control plan. When a transportation environment control scheme is generated, it follows the following rules: Logic1: For the core influencing factors with the highest priority, determine the corresponding control parameters, including temperature and humidity, gas ratio, or vibration reduction intensity. Logic2: Calculate the parameter adjustment range based on the deviation between the current value and the optimal range of each core influencing factor; Logic3: By Impact Contribution The parameters are adjusted sequentially from largest to smallest, and the adjustment of each subsequent parameter must compensate for the correlation effect of the previous parameter adjustment on other factors. in, >0, and the higher the contribution of the k-th type of environmental factor, the greater the deviation between the current value and the optimal value. The larger the value, the lower the contribution of the influence and the smaller the deviation between the current value and the optimal value. The smaller the value; The complete monitoring data, environmental control records, and final meat product preservation status verification results of each transportation are summarized to build a cross-scenario monitoring database and provide users with access permissions simultaneously. When constructing the cross-scenario monitoring database, the core classification dimensions are transportation scenario type (including land transportation, water transportation, and air transportation), perishable meat product category, and transportation duration range. Various types of data generated during the monitoring process are classified and archived. A hybrid storage structure combining time-series database and relational database is adopted. The time-series database is used to store real-time monitoring data, environmental control operation records, and time-series change sequences of various parameters, while the relational database is used to store basic attribute information of perishable meat products, freshness status verification results, scenario characteristic parameters, and transportation task-related information. Among them, the dynamic data of meat quality characteristics are the concentration of microbial metabolic gases and the degree of myoglobin oxidation on the surface of perishable meat products, and the multi-parameter data of the transportation environment are the dynamic data of temperature, humidity, gas composition and vibration amplitude in the transportation environment. The combination of the two sub-models is a deep learning model driven by the synergy of microbial metabolism and environmental factors.

[0022] The method described in the above embodiments can capture dynamic data related to the characteristics of perishable meat products and the transportation environment in real time, dynamically adapt the collection frequency, accurately filter effective information and perform in-depth fusion analysis, quickly output the degree of spoilage, remaining shelf life and key influencing factors, optimize transportation environment parameters in a timely manner, and significantly reduce spoilage losses.

[0023] The following is an application example of the system described in the above embodiments: XX Food Logistics Company adopted this method to monitor the preservation of a batch of chilled meat during land transportation. Distributed multi-dimensional sensor arrays were deployed in the corners of the transport compartment, near the inlet and outlet airflow channels, and in areas far from these channels. Each sensor group is equipped with a unique spatial identifier and time synchronization module to collect real-time data on the concentration of microbial metabolic gases on the surface of the chilled meat, the degree of myoglobin oxidation, and dynamic data on the temperature, humidity, gas composition (oxygen, carbon dioxide), and vibration amplitude of the transport environment. In the initial stage of transportation, when the remaining shelf life is long and the rate of spoilage change is slow, the sensors collect data at a preset default frequency. After 12 hours of transportation, when the remaining shelf life shortens and the rate of spoilage change accelerates, the collection frequency automatically increases to a preset high-frequency range.

[0024] When processing the raw monitoring data, the reliability coefficient of the multi-parameter spatiotemporal correlation of each data point was calculated, and three abnormal data points whose reliability did not reach the preset threshold were removed. Subsequently, the meat quality characteristic data and the transportation environment data were spatiotemporally correlated and fused to obtain the fusion feature value corresponding to each sampling time and location, with the result ranging from 0.31 to 0.68.

[0025] The standardized feature dataset was input into the trained deep learning model, which consists of a spoilage state identification sub-model and an environmental impact assessment sub-model. The model outputs a current spoilage metric of 0.42 for the chilled meat, with a remaining shelf life of 28 hours. The core influencing factors, prioritized as follows: transportation temperature, vibration amplitude, and carbon dioxide concentration.

[0026] Based on the above results, a control plan was generated: the transportation temperature was first adjusted from 4℃ to the optimal range of 0-2℃, with an adjustment range of 33%; then the vibration reduction intensity was adjusted to compensate for the vibration and reduce the vibration amplitude by 20%; finally, the carbon dioxide concentration was fine-tuned to the optimal range.

[0027] After the transportation task is completed, the real-time monitoring data, environmental control records, and final preservation verification results of the chilled meat will be classified according to land transportation, chilled meat category, and 24-48 hour transportation time range, and archived into a cross-scenario monitoring database. The database will be stored in a hybrid manner using time-series and relational databases for the company to access later.

[0028] In summary, the method described in the above embodiments captures dynamic data related to perishable meat products and multi-dimensional data of the transportation environment in real time, dynamically adjusts the collection frequency to adapt to different preservation requirements, accurately filters effective data and achieves spatiotemporal correlation fusion, thereby efficiently outputting quantified values ​​of spoilage, remaining shelf life, and ranking of core influencing factors, and quickly generating targeted transportation environment control schemes, optimizing temperature and humidity, gas ratio, and vibration reduction parameters in real time, effectively delaying the spoilage process, extending the shelf life, reducing transportation losses, providing reliable protection for the preservation and transportation of perishable meat products, and reducing resource waste and economic losses.

[0029] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for real-time monitoring of the preservation status of perishable meat products based on big data and AI, characterized in that, include: Real-time monitoring of dynamic data on the concentration of metabolic gases from microorganisms on the surface of perishable meat products, the degree of myoglobin oxidation, and the temperature, humidity, gas composition, and vibration amplitude of the transportation environment, in order to establish a multi-source raw monitoring dataset; By traversing the original monitoring dataset and removing abnormal data, dynamic data of meat quality characteristics and multi-parameter data of transportation environment are spatiotemporally correlated and fused to generate a standardized feature dataset containing microbial metabolic time-series features and environmental coupling features. When removing abnormal data from the original monitoring data, the determination of whether the data is abnormal is made by calculating the multi-parameter spatiotemporal correlation reliability coefficient of the data points: ； In the formula: Let be the confidence coefficient of the data point corresponding to the i-th sampling time and the j-th sampling position; For parameter difference weighting coefficients; This represents the total number of categories of monitored parameters. This refers to the raw data of the k-th type of monitoring parameter corresponding to the i-th sampling time and the j-th sampling location; The mean value of the k-th type of monitoring parameter within the preset time window adjacent to the i-th sampling time and the preset spatial range adjacent to the j-th sampling location; The historical normal fluctuation standard deviation of the k-th type of monitoring parameter; This is the time deviation weighting coefficient; This is the normalized result of the difference between the i-th sampling time and its adjacent sampling times; Spatial deviation weighting coefficient; This is the normalized distance between the j-th sampling position and its adjacent sampling positions; A deep learning model driven by the synergistic interaction of microbial metabolism and environmental factors is constructed. A sub-model for identifying putrefaction status is generated based on the time-series features of microbial metabolism in a standardized feature dataset, and a sub-model for assessing environmental impact is generated based on the environmental coupling features. The standardized feature dataset is input into the trained deep learning model in real time, and the output is the current spoilage quantification value of perishable meat products, the remaining shelf life, and the priority ranking results of core influencing factors. Based on the core influencing factors and the remaining shelf life, a transportation environment control plan is generated, and key parameters such as temperature and humidity, gas ratio and vibration reduction strength in the transportation cabin are adjusted in real time based on the transportation environment control plan. The complete monitoring data, environmental control records, and final meat product preservation status verification results of each transportation are summarized to build a cross-scenario monitoring database, and users are given access to the database simultaneously.

2. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 1, characterized in that, The dynamic data of microbial metabolic gas concentration, myoglobin oxidation degree, temperature and humidity, gas composition and vibration amplitude on the surface of the perishable meat products are collected inside the transport cabin by a distributed multi-dimensional sensor array. The distributed multi-dimensional sensor array includes several sensor groups, each sensor group integrating a microbial metabolic gas detection unit, a myoglobin oxidation degree detection unit, a temperature and humidity detection unit, a gas composition detection unit and a vibration detection unit. The sensor groups are deployed in various corners inside the transport cabin, beside the inlet and outlet airflow channels, and in locations far from the inlet and outlet airflow channels. Each sensor group is equipped with a unique spatial identification code and a time synchronization module. The monitoring frequency of the sensor group is changed in real time based on a preset dynamic adjustment logic. That is, the adjustment basis is determined according to the current remaining shelf life and the rate of change of spoilage. When the current remaining shelf life is less than the preset short period threshold and the rate of change of spoilage exceeds the preset high rate threshold, the sampling frequency is increased to the preset high frequency range; when the current remaining shelf life exceeds the preset long period threshold and the rate of change of spoilage is less than the preset low rate threshold, the sampling frequency is decreased to the preset low frequency range; in other cases, the preset default sampling frequency is maintained.

3. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 1, characterized in that, When performing spatiotemporal correlation fusion of dynamic data on meat quality characteristics with multi-parameter data of the transportation environment, the following applies: Correlation calculations are performed on meat quality characteristic data and environmental data within the same spatiotemporal dimension to generate standardized feature data: ； In the formula: The fused feature value corresponds to the i-th sampling time and the j-th sampling position; Weights for fusion of meat quality characteristic data; The standardized value of the meat quality characteristics data corresponding to the i-th sampling time and the j-th sampling position; This is the time decay coefficient; This is the difference between the i-th sampling time and the data fusion reference time; Weighting for transportation environment data fusion; This represents the standardized value of the transportation environment data corresponding to the i-th sampling time and the j-th sampling location; This is the spatial attenuation coefficient; is the distance between the j-th sampling position and the data fusion reference position.

4. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 1, characterized in that, In the deep learning model driven by the synergistic interaction of microbial metabolism and environmental factors, the putrefaction state identification sub-model applies an attention-enhanced temporal convolutional network: taking microbial metabolic time-series data and dynamic data on myoglobin oxidation as input, a 3-head self-attention module is embedded in the temporal convolutional layer. By calculating the attention weights of features at different time steps, it adaptively focuses on key time node information such as abrupt changes in microbial metabolic rate and inflection points in myoglobin oxidation, with an output dimension of [missing information]. The temporal feature vector of meat spoilage, where The feature vector dimension is adaptively set based on the number of input features, model training convergence efficiency, and feature representation accuracy. The environmental impact assessment sub-model applies a graph attention neural network: Temperature, humidity, oxygen concentration, carbon dioxide concentration, and vibration amplitude from the transportation environment parameters are treated as independent nodes in a graph structure. Each node's initial feature vector includes the real-time value, time-series rate of change, and historical fluctuation characteristics of the corresponding parameter. By calculating the bidirectional association weights between nodes, a multi-head attention mechanism is used to update node features, uncovering the nonlinear coupling relationships between environmental parameters. The output dimension is... The environmental coupling feature vector, where The dimension of the feature vector is positively correlated with the number of environmental parameters and the complexity of parameter coupling relationships; In this process, the meat spoilage time-series feature vector and the environmental coupling feature vector are dimensionally matched by zero-padding or dimensional truncation, and then input into a two-layer fully connected network. The first layer performs a nonlinear transformation on the fused features through the ReLU activation function, and the second layer maps the transformed features to a freshness status evaluation value in the 0-1 interval through the Sigmoid activation function, representing the final decision result of the model.

5. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 1, characterized in that, The logic for obtaining the corruption quantification value is as follows: ； In the formula: This is a quantification of corruption. Weights for the concentration characteristics of microbial metabolic gases; The time-series peak value of microbial metabolic gas concentration within the monitoring period [0,T]; This is a nonlinear index of microbial metabolic gas concentration; Weights for the degree of myoglobin oxidation; This is a standardized value for the degree of myoglobin oxidation. This is a non-linear index representing the degree of myoglobin oxidation. Number of categories of key parameters for the transportation environment; The influence weight of the k-th type of environmental parameter; This represents the standardized value of the k-th type of environmental parameter; For the k-th type of environmental parameter, there is a nonlinear adjustment coefficient. The total weight is determined by the coupling effect of the environment.

6. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 5, characterized in that, The logic for calculating the remaining shelf life is as follows: First, the real-time rate of change of the putrefaction metric is obtained through linear fitting. ,in They are respectively A quantification of corruption at any given moment; Then, based on the preset spoilage threshold, calculate the remaining shelf life. , Indicates the corruption threshold. This represents the current quantification of corruption. Indicates the corruption inhibition coefficient. This indicates the real-time rate of change of the corruption quantification. When v is a non-positive value, the remaining preservation period is assigned according to the preset longest period.

7. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 5, characterized in that, The priority ranking of the core influencing factors is based on the calculation results of the factor's contribution to the impact. The formula for calculating the contribution of the influence is as follows: ； In the formula: Contribution to the impact of the k-th type of environmental factor; Let the corruption quantification value D be relative to the k-th type of environmental factor. The first-order partial derivative; Let be the variance of the k-th type of environmental factor; These are the covariance weighting coefficients; Let $\mathbf{k}$ be the covariance between the $k$-th type of environmental factor and the quantified value of corruption. Let be the second mixed partial derivative of the corruption quantification with respect to the k-th environmental factor and the l-th environmental factor; This represents the total number of categories of environmental factors. according to Sort in descending order to obtain the priority ranking of the core impact factors.

8. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 7, characterized in that, The transportation environment control scheme, when generated, follows the following: Logic1: For the core influencing factors with the highest priority, determine the corresponding control parameters, including temperature and humidity, gas ratio, or vibration reduction intensity. Logic2: Calculate the parameter adjustment range based on the deviation between the current value and the optimal range of each core influencing factor; Logic3: By Impact Contribution Parameters are adjusted sequentially from largest to smallest.

9. The method for real-time monitoring of the preservation status of perishable meat products based on big data and AI according to claim 1, characterized in that, When constructing the cross-scenario monitoring database, the core classification dimensions are transportation scenario type, perishable meat product category, and transportation duration range. Various types of data generated during the monitoring process are classified and archived. A hybrid storage structure combining time-series database and relational database is adopted. The time-series database is used to store real-time monitoring data, environmental control operation records, and time-series change sequences of various parameters. The relational database is used to store basic attribute information of perishable meat products, freshness status verification results, scenario characteristic parameters, and transportation task association information.

Images