Intelligent active package optimization method and system based on machine learning

Abstract

The invention discloses an intelligent active packaging optimization method and system based on machine learning. Comprising the steps of performing periodic detection and activity parameter recording on target active food based on a test scheme, constructing a detection parameter matrix, introducing a CNN model to perform matrix feature learning and characterize a parameter change mode, calculating a first qualitative change coefficient in combination with an ideal state matrix, performing fitting prediction on the qualitative change coefficient, and obtaining a second qualitative change coefficient; and introducing an ideal activity parameter to evaluate a second qualitative change coefficient, and optimizing a fitting prediction process. And finally, according to a fitting result, evaluating a packaging effect and predicting food activity change, thereby realizing accurate optimization of a preset packaging test scheme, effectively analyzing and predicting potential activity influence factors, reducing participation of human experience, and improving activity packaging optimization efficiency.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of packaging detection, more particularly, to an intelligent active packaging optimization method and system based on machine learning. BACKGROUND

[0002] Traditional food packaging technology mainly relies on empirical material selection and single test parameter analysis, such as observing the degree of food rot to determine the effect of active packaging, which is difficult to adapt to the dynamic change needs of different foods during storage and make corresponding active packaging optimization. Active foods (such as fresh fruits and vegetables, meat, etc.) will deteriorate in quality during storage due to microbial metabolism, enzymatic reaction and oxidation, etc. Its active parameters (such as oxygen concentration, PH, carbon dioxide concentration, humidity, ethylene content, microbial indicators, etc.) change over time. The existing technology lacks the mining of the change rule of multi-dimensional active parameters, and lacks the joint change degree test evaluation, resulting in that the production optimization process of intelligent active packaging relies on artificial experience adjustment, and it is difficult to have substantial breakthrough. Therefore, at present, an intelligent and informationized intelligent active packaging optimization analysis process is urgently needed. SUMMARY

[0003] The present application overcomes the defects of the prior art and provides an intelligent active packaging optimization method and system based on machine learning.

[0004] The first aspect of the present application provides an intelligent active packaging optimization method based on machine learning, comprising: S11: based on a preset packaging test scheme, packaging and detecting food parameters of a target active food, and obtaining and recording food active parameters based on each detection period; S12: constructing a periodic detection parameter matrix according to the time dimension and the food active parameters, introducing a CNN model to perform cyclic convolution calculation on the detection parameter matrix, and generating a change feature set by fusing multiple calculation results; S13: introducing the ideal parameter matrix to compare the difference between the change feature set, and calculating a first change coefficient of each detection period; S14: comparing the difference between the food active parameters and the expected parameters according to a plurality of preset parameter weights, and evaluating the change degree by weighted average, to obtain a second change coefficient of each detection period; S15: fitting the linear change of the first change coefficient by linear regression, and taking the minimum mean square error of the fitting value and the second change coefficient as the target, adjusting the fitting coefficient by the least square method, to obtain the fitting result; S16: evaluating the packaging effect according to the fitting result, and predicting the food active change according to the preset packaging test scheme, to optimize the packaging scheme.

[0005] In the scheme, the S11 comprises: According to the preset packaging test scheme, the target active food is set for real-time detection by multiple sensors; A plurality of detection cycles are set, and a plurality of recording nodes are set in each detection cycle; For each detection cycle, the food activity parameters of multiple dimensions are obtained.

[0006] In the scheme, the S12 specifically comprises: In one detection cycle, a plurality of recording nodes are included, time dimension is taken as the first dimension, and a plurality of food activity parameters are taken as the second dimension, so as to construct a detection parameter matrix; Each detection cycle corresponds to a detection parameter matrix; A CNN model is constructed, which comprises a convolution layer, an activation layer and a pooling layer, a preset size of convolution kernel is set in the convolution layer, convolution calculation is performed on the detection parameter matrix through the convolution layer, until the convolution calculation covers the entire detection parameter matrix, and the local features obtained are introduced into the nonlinear features through the activation layer; The local features are subjected to maximum pooling operation through the pooling layer, and a change feature set is generated.

[0007] In the scheme, the S13 comprises: According to a plurality of recording nodes of one detection cycle, the food activity parameters in the ideal state are input, and an ideal parameter matrix is constructed according to the idealized parameters; The CNN model is used to perform cyclic convolution calculation on the ideal parameter matrix and generate a comparison feature set; Based on the mean-shift algorithm, a clustering space is constructed, and the change feature set and the comparison feature set are respectively introduced into the clustering space; The change feature set and the comparison feature set are respectively subjected to density clustering, the distance between the feature data is calculated by Manhattan distance, and the clustering is performed cyclically until the center point converges, and two groups of clustering center points are obtained; In the clustering space, the difference between the two groups of clustering center points is analyzed, the average distance from one group of clustering center points to the other group of clustering center points is calculated, and the first qualitative change coefficient is obtained in combination with the difference in the number of the two groups of clustering center points.

[0008] In the scheme, the S14 specifically comprises: According to the preservation property of the target active food, a preset parameter weight is set for each active parameter; For one detection cycle, the food activity parameters and the expected parameters are compared and evaluated, the difference value of each parameter is weighted and averaged, the preset parameter weight is introduced into the weighted calculation, and the second qualitative change coefficient is obtained.

[0009] In the scheme, the S15 comprises: Linear regression fitting is performed on the plurality of first quality change coefficients corresponding to the plurality of detection periods to obtain a fitting equation; The mean value of the fitting data in a detection period is calculated from the fitting equation to obtain a fitting value; For all detection periods, the mean square error of the fitting value and the corresponding second quality change coefficient is calculated, and the least square method is used to adjust the fitting coefficient until the iteration number is reached, and the fitting result is recorded.

[0010] In the scheme, the S16 comprises: Based on the fitting result, the packaging effect of the food active state is evaluated; A prediction period is set, and the food active state is predicted based on the fitting result, and the trend analysis of the packaging effect and the food active state is reflected through the prediction process; According to the prediction process, the time period in which the active change and the quality change degree do not meet the expectation is analyzed, and the packaging optimization is performed according to the preset packaging test scheme.

[0011] The second aspect of the application also provides an intelligent active packaging optimization system based on machine learning, which comprises a memory, a processor and a communication interface, wherein the memory comprises an intelligent active packaging optimization program based on machine learning, and the intelligent active packaging optimization program based on machine learning is executed by the processor to realize the following steps: S11: Based on the preset packaging test scheme, the target active food is packaged and the food parameters are detected, and the food active parameters are obtained and recorded based on each detection period; S12: A periodic detection parameter matrix is constructed according to the time dimension and the food active parameters, a CNN model is introduced to perform cyclic convolution calculation on the detection parameter matrix, and a change feature set is generated by fusing the calculation results; S13: The difference between the change feature set and the contrast feature set of the ideal parameter matrix is compared, and the first quality change coefficient of each detection period is calculated; S14: According to a plurality of preset parameter weights, the difference between the food active parameters and the expected parameters is compared and the quality change is evaluated, and the weighted average of the quality change degree is obtained to obtain the second quality change coefficient of each detection period; S15: The linear change of the first quality change coefficient is fitted by linear regression, and the least square method is used to adjust the fitting coefficient to obtain the fitting result; S16: According to the fitting result, the packaging effect is evaluated, and the food active change is predicted according to the preset packaging test scheme, and the packaging scheme is optimized.

[0012] The third aspect of the present application also provides a machine readable storage medium, which stores instructions, and the instructions make the processor execute the intelligent active packaging optimization method based on machine learning when executed by the processor.

[0013] The present application discloses an intelligent active packaging optimization method and system based on machine learning. The method comprises: periodically detecting a target active food based on a test scheme and recording active parameters, constructing a detection parameter matrix, introducing a CNN model to learn matrix features and represent parameter change patterns, calculating a first quality change coefficient based on an ideal state matrix, fitting and predicting the quality change coefficient, introducing an ideal active parameter to evaluate a second quality change coefficient, and optimizing the fitting and prediction process. Finally, the packaging effect is evaluated according to the fitting result, and the active change of the food is predicted, so as to realize accurate optimization of the preset packaging test scheme, effectively analyze and predict potential active influencing factors, reduce the participation of human experience, and improve the active packaging optimization efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0014] Figure 1 A flowchart of the intelligent active packaging optimization method based on machine learning is shown. Figure 2 A flowchart of the intelligent active packaging prediction optimization is shown. Figure 3 A block diagram of the intelligent active packaging optimization system based on machine learning is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in combination with the drawings and specific embodiments. It can be understood that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0016] In the following description, many specific details are set forth in order to provide a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, therefore, the scope of protection of the present application is not limited by the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of the intelligent active packaging optimization method based on machine learning is shown.

[0018] As Figure 1 shown, the first aspect of the present application provides an intelligent active packaging optimization method based on machine learning, comprising: S11: based on a preset packaging test scheme, packaging and detecting food parameters of a target active food, and obtaining and recording food active parameters based on each detection period; S12: Construct a periodic detection parameter matrix according to the time dimension and the food activity parameter, introduce a CNN model to perform cyclic convolution calculation on the detection parameter matrix, and generate a change feature set by fusing multiple calculation results; S13: Introduce the comparative feature set of the ideal parameter matrix to difference compare the change feature set, and calculate a first qualitative change coefficient of each detection period; S14: According to a plurality of preset parameter weights, difference compare the food activity parameter and the expected parameter, and perform qualitative change evaluation, and weighted average the qualitative change degree to obtain a second qualitative change coefficient of each detection period; S15: Linearly fit the linear change of the first qualitative change coefficient, and take the minimum mean square error of the fitting value and the second qualitative change coefficient as the target, adjust the fitting coefficient through the least square method, and obtain the fitting result; S16: According to the fitting result, evaluate the packaging effect, and predict the food activity change according to the preset packaging test scheme, and optimize the packaging scheme.

[0019] Figure 2 The intelligent activity packaging prediction optimization flowchart of the application is shown.

[0020] As Figure 2 The simplified schematic diagram of steps S11-S16 is shown.

[0021] According to the embodiment of the application, the S11 comprises: According to the preset packaging test scheme, a plurality of sensors are set for real-time detection of the target active food; A plurality of detection periods are set, and a plurality of recording nodes are set in each detection period; For each detection period, a plurality of food activity parameters are obtained.

[0022] It can be understood that in the target active food set of sensors, sensors for detecting PH, concentration of various gases, ethylene content, temperature, humidity, specific compounds related to food qualitative change, microbial indicators and other activity parameters can be used. The activity parameters can be obtained in real time by the sensors and the food change and deterioration degree can be analyzed. The preset packaging test scheme includes test time, detection period setting, activity parameter type and other information.

[0023] According to the embodiment of the application, the S12 specifically comprises: In one detection period, a plurality of recording nodes are included, the time dimension is taken as the first dimension, and a plurality of food activity parameters are taken as the second dimension, and a detection parameter matrix is constructed; Each detection period corresponds to a detection parameter matrix; The CNN model comprises a convolution layer, an activation layer and a pooling layer, a preset size of a convolution kernel is set in the convolution layer, convolution calculation is performed on the detection parameter matrix through the convolution layer, and the local features obtained are introduced into the non-linear features through the activation layer; The local features are subjected to a maximum pooling operation through the pooling layer, and a change feature set is generated.

[0024] It can be understood that the food activity parameters include multiple parameters, which are multidimensional parameter information, and are used to set the second dimension data of the matrix, and the first dimension data is specifically a plurality of record nodes. The preset size of the convolution kernel can be 3*3 or 5*5. The activation layer generally introduces a Sigmoid function to analyze the non-linear features. For each detection period, a change feature set is generated.

[0025] According to the embodiment of the present application, the S13 comprises: According to a plurality of record nodes of one detection period, the food activity parameters in an ideal state are input, and an ideal parameter matrix is constructed according to the idealized parameters; The CNN model is used to perform cyclic convolution calculation on the ideal parameter matrix and generate a comparison feature set; Based on the mean-shift algorithm, a clustering space is constructed, and the change feature set and the comparison feature set are introduced into the clustering space respectively; The change feature set and the comparison feature set are respectively subjected to density clustering, the distance between the feature data is calculated by Manhattan distance, and the clustering is performed cyclically until the center points converge, and two groups of clustering center points are obtained. In the clustering space, the difference between the two groups of clustering center points is analyzed, the average distance from one group of clustering center points to the other group of clustering center points is calculated, and the first qualitative change coefficient is obtained in combination with the difference in the number of the two groups of clustering center points.

[0026] It can be understood that the center points converge, that is, they do not move. The two groups of clustering center points correspond to the clustering results of the two feature sets respectively. In the calculation of the average distance from one group of clustering center points to the other group of clustering center points, the distance value from each center point in the first group of clustering center points to the nearest point in the second group of clustering center points is calculated, and a plurality of distance values are obtained. The average distance is obtained by equalizing the distance values. The first qualitative change coefficient is related to the average distance and the difference in the number of center points, and is specifically proportional to the average distance and the difference in the number of center points.

[0027] For the difference comparison in S13, a traditional data difference analysis process is included.

[0028] It is worth mentioning here that, in the difference comparison of the two feature sets obtained, the mean-shift clustering algorithm is introduced for comparison. Compared with the simple difference analysis of traditional feature data (such as Euclidean distance analysis of data difference), the application can evaluate the difference in the classification state of the data set, and reflect the difference in the classification state with the spatial distance of the cluster center point. The application can effectively analyze the similarity between the data set from the overall and local analysis of the data set, and further calculate the activity change difference to obtain the qualitative change coefficient, and realize the multi-dimensional difference analysis of the activity parameters of the detection period.

[0029] According to the embodiment of the application, the S14 is specifically: According to the preservation property of the target active food, a preset parameter weight is set for each active parameter; For a detection period, the food activity parameters and the expected parameters are compared and evaluated, and the weighted average is calculated according to the difference value of each parameter, the preset parameter weight is introduced in the weighted calculation, and a second qualitative change coefficient is obtained.

[0030] It can be understood that the ideal food activity parameter can be input by the user or calculated according to the preset food property and the continuous test time.

[0031] According to the embodiment of the application, the S15 comprises: Linear regression fitting is performed on the plurality of first qualitative change coefficients corresponding to the plurality of detection periods to obtain a fitting equation; The mean value of the fitting data in a detection period is calculated from the fitting equation to obtain a fitting value; For all detection periods, the mean square error of the fitting value and the corresponding second qualitative change coefficient is calculated, and the least square method is used to adjust the fitting coefficient until the iteration number is reached, and the fitting result is recorded.

[0032] It can be understood that the expected parameter is the ideal food activity parameter. Linear regression fitting can be fitted by using a linear function y=Kx+B, K and B are fitting coefficients. The fitting data is the dependent variable value in the fitting equation, the fitting value is the continuous value extracted from the fitting equation, and the fitting result is the optimized fitting equation.

[0033] In the present application, the ideal parameters and actual parameters are compared, and the second quality change coefficient is set, which is used to reflect the difference of real parameters in a single dimension. In the fitting prediction process of the first quality change coefficient, the second quality change coefficient is introduced to contest the fitting process, which can effectively solve the problem of data overfitting in multi-dimensional parameter analysis, and lead to the deviation of the predicted detection parameters from the real values. Further, according to the fitting result, the activity state of the preset packaging test scheme can be effectively analyzed in multiple cycles, and the activity trend of the packaged food is judged according to the fitting prediction, and the packaging scheme is further optimized, such as material optimization, release compound optimization, packaging optimization based on pH response, etc.

[0034] The mean square error is calculated as follows: ; Wherein, is the fitting value of the i th detection cycle and the second quality change coefficient, N is the total number of detection cycles, and MSE is the mean square error.

[0035] According to the embodiment of the present application, the S16 comprises: Based on the fitting result, the food activity state and the packaging effect are evaluated; Set the prediction period, predict the food activity state based on the fitting result, and reflect the trend analysis of the packaging effect and the food activity state through the prediction process; According to the prediction process, analyze the time period when the activity change and the quality change degree do not meet the expectation, and optimize the packaging according to the preset packaging test scheme.

[0036] In this embodiment, the target active food is packaged and periodically detected for its active parameters by a preset packaging test scheme, thereby constructing a detection parameter matrix capable of reflecting multi-dimensional active state. Subsequently, a CNN model is introduced to perform cyclic convolution calculation on the matrix, thereby generating a change feature set representing the parameter change pattern. By comparing the change feature set with the comparative feature set of the ideal parameter matrix, a first qualitative change coefficient is calculated, which effectively combines multi-dimensional active parameters to analyze the long-term and short-term active parameter characteristic changes (here, the detection period and multiple recording nodes are used to evaluate the long-term and short-term parameters). At the same time, the parameter difference is weighted and averaged based on the preset weight, thereby obtaining a second qualitative change coefficient. Then, the change trend of the first qualitative change coefficient is fitted by linear regression, and the fitting coefficient is optimized to minimize the mean square error between the fitting value and the second qualitative change coefficient, thereby reducing the overfitting of the qualitative change prediction. Finally, the packaging effect is evaluated based on the fitting result, and the food activity change is predicted, thereby realizing the precise optimization of the preset packaging test scheme (such as detection period, packaging material, etc.). The present application solves the problems of long test period, high analysis cost, single analysis dimension, and difficulty in accurately predicting (or testing) the long-term preservation effect of intelligent packaging in the traditional method, realizes the purpose of predicting long-term food activity and packaging effect based on short-term test data, and significantly improves the research and development and optimization efficiency of active packaging.

[0037] According to the embodiments of the present application, the method further comprises: Based on multiple detection periods, linear regression fitting is performed on multiple food activity parameters, and the parameters are used as dependent variables and the time is used as independent variable for fitting, thereby forming multiple fitting curves; In the multiple fitting curves, the mean of the first qualitative change coefficient and the second qualitative change coefficient is minimized as the objective function, and the multiple food activity parameters are used as decision variables. Based on the Bayesian optimization algorithm, the optimal parameter value of each food activity parameter is obtained; The difference between each optimal parameter value and the mean of the corresponding food activity parameter is compared, and the qualitative change influence degree of each food activity parameter is set based on the difference, and the optimization priority of each food activity parameter is set.

[0038] It can be understood that the first qualitative change coefficient is obtained based on the corresponding target detection period, and the second qualitative change coefficient is calculated based on the actual values of multiple decision variables (i.e., multiple food activity parameters). The optimal parameter value includes multiple, and each optimal parameter value corresponds to one food activity parameter.

[0039] In the difference comparison between each optimal parameter value and the mean of the corresponding food activity parameter, first, the parameter mean of each food activity parameter in the entire preset packaging test scheme is calculated, and the difference between each optimal parameter value and the parameter mean is calculated.

[0040] In particular, during the analysis of the packaging preservation effect of the target food and the active parameter, there is a certain correlation between the influence of different active parameters on the quality change. However, there is a lack of effective technical means for comprehensive analysis of multiple active parameters, and the evaluation is often based on a single dimension or manual experience for packaging optimization adjustment.

[0041] To this end, in the present embodiment, by analyzing the change state of multiple active parameters, the Bayesian optimization algorithm is used to search for the active parameter value of the quality change trend in the optimal state within a certain period of time, and the optimal (or preferred) parameter value is compared with the corresponding active parameter mean. The greater the difference, the greater the influence of the parameter on the food quality change, that is, the parameter has a higher correlation and fluctuation relevance for the quality change effect, and the higher the priority of optimizing the active parameter. Furthermore, the packaging is optimized according to the active factors of different priorities. Through the optimization search process, the influence of multiple dimensions of active parameters can be evaluated comprehensively, solving the problem of inaccurate and incomplete single parameter dimension analysis in the prior art.

[0042] Figure 3 A block diagram of an intelligent active packaging optimization system based on machine learning is shown.

[0043] The second aspect of the present application also provides an intelligent active packaging optimization system based on machine learning, which comprises a memory 103, a processor 102 and a communication interface 101. The communication interface is used to connect a user display terminal, and the optimization prediction process is visualized and displayed through the user display terminal. The memory comprises an intelligent active packaging optimization program based on machine learning. When the intelligent active packaging optimization program based on machine learning is executed by the processor, the following steps are implemented: S11: Based on a preset packaging test scheme, the target active food is packaged and the food parameters are detected, and the food active parameters are obtained and recorded based on each detection period; S12: A periodic detection parameter matrix is constructed according to the time dimension and the food active parameters, a CNN model is introduced to perform cyclic convolution calculation on the detection parameter matrix, and a change feature set is generated by fusing multiple calculation results; S13: The change feature set is compared with the ideal parameter matrix, and a first quality change coefficient of each detection period is calculated; S14: According to a plurality of preset parameter weights, the food active parameters and the expected parameters are compared and evaluated, the quality change degree is weighted and averaged, and a second quality change coefficient of each detection period is obtained; S15: fitting the linear change of the first quality change coefficient by linear regression, and adjusting the fitting coefficient by least square method to obtain a fitting result, with the objective of minimizing the mean square error of the fitting value and the second quality change coefficient; S16: evaluating the packaging effect according to the fitting result, and predicting the food activity change for a preset packaging test scheme, and optimizing the packaging scheme.

[0044] According to an embodiment of the present application, the S11 comprises: According to the preset packaging test scheme, a plurality of sensors are set for real-time detection of the target active food; A plurality of detection periods are set, and a plurality of recording nodes are set in each detection period; For each detection period, a plurality of food activity parameters are obtained.

[0045] It can be understood that in the plurality of sensors set for the target active food, sensors for detecting PH, concentration of a plurality of gases, ethylene content, temperature, humidity, specific compounds related to food quality change, microbial indicators and other activity parameters can be used. The activity parameters can be obtained in real time by the sensors, and the degree of food change and deterioration can be analyzed. The preset packaging test scheme includes test time, detection period setting, and information such as types of activity parameters.

[0046] According to an embodiment of the present application, the S12 specifically comprises: In one detection period, a plurality of recording nodes are included, time dimension is used as the first dimension, and a plurality of food activity parameters are used as the second dimension, and a detection parameter matrix is constructed; Each detection period corresponds to a detection parameter matrix; A CNN model is constructed, which includes a convolution layer, an activation layer, and a pooling layer. A preset size of convolution kernel is set in the convolution layer, and convolution calculation is performed on the detection parameter matrix by convolution layer circulation until the convolution calculation covers the entire detection parameter matrix, and the obtained local features are introduced into the activation layer to introduce nonlinear features; The local features are subjected to maximum pooling operation by the pooling layer, and a change feature set is generated.

[0047] It can be understood that the food activity parameters include a plurality of parameters, which are multi-dimensional parameter information, and are used here to set the second dimension data of the matrix. The first dimension data specifically refers to a plurality of recording nodes. The preset size of the convolution kernel can be 3x3 or 5x5. The activation layer generally introduces a Sigmoid function for nonlinear feature analysis. For each detection period, a change feature set is generated.

[0048] According to an embodiment of the present application, the S13 comprises: According to a plurality of recording nodes of a detection cycle, an ideal state food activity parameter is input, and an ideal parameter matrix is constructed according to the idealized parameter; The ideal parameter matrix is subjected to cyclic convolution calculation through the CNN model, and a contrast feature set is generated. Based on the mean-shift algorithm, a clustering space is constructed, and the change feature set and the contrast feature set are respectively imported into the clustering space. The change feature set and the contrast feature set are respectively subjected to density clustering, the distance between the feature data is calculated by Manhattan distance, and clustering is cyclically performed until the center point converges, and two groups of clustering center points are obtained. In the clustering space, the difference between the two groups of clustering center points is analyzed, the average distance from one group of clustering center points to another group of clustering center points is calculated, and the first qualitative change coefficient is obtained in combination with the number difference of the two groups of clustering center points.

[0049] It can be understood that the center point converges, that is, it no longer moves. The two groups of clustering center points correspond to the clustering results of the two feature sets respectively. In the calculation of the average distance from one group of clustering center points to another group of clustering center points, the distance value from each center point in the first group to the nearest point in the second group of clustering center points is calculated, and a plurality of distance values are obtained. The average distance is obtained by homogenizing the distance values. The first qualitative change coefficient is related to the average distance and the number difference of the center points, and specifically proportional to the average distance and the number difference of the center points.

[0050] For the difference comparison in S13, the traditional data difference analysis process is included.

[0051] It is worth mentioning here that in the difference comparison of the two feature sets obtained, the mean-shift clustering algorithm is introduced for comparison. Compared with the simple difference analysis of traditional feature data (such as Euclidean distance analysis of data difference), the present application can evaluate the difference in the classification state of the data set, and reflect the classification state difference by the spatial distance of the clustering center points. The similarity between the data sets is effectively analyzed from the overall and local analysis, and the activity change difference is further calculated, the qualitative change coefficient is obtained, and the multi-dimensional difference analysis of the activity parameters of the detection cycle is realized.

[0052] According to the embodiment of the present application, S14 is specifically: According to the preservation property of the target active food, a preset parameter weight is set for each active parameter; For a detection cycle, the food activity parameter and the expected parameter are compared and evaluated, the difference value of each parameter is weighted and averaged, the preset parameter weight is introduced in the weighted calculation, and a second qualitative change coefficient is obtained.

[0053] It can be understood that the food activity parameter in the ideal state can be input by the user or calculated according to the preset food property and the continuous test time.

[0054] According to an embodiment of the present application, the S15 comprises: Linear regression fitting is performed on the plurality of first quality change coefficients corresponding to the plurality of detection periods to obtain a fitting equation; The mean value of the fitting data in one detection period is calculated from the fitting equation to obtain a fitting value; For all detection periods, the mean square error of the fitting value and the corresponding second quality change coefficient is calculated, and the minimization of the mean square error is taken as the target, and the fitting coefficient is adjusted through the least square method until the iteration number is reached, and the fitting result is recorded.

[0055] It can be understood that the expected parameter is the food activity parameter in the ideal state. Linear regression fitting can be fitted using a linear function y=Kx+B, K and B being fitting coefficients. The fitting data is the dependent variable value in the fitting equation, the fitting value is the continuous value extracted from the fitting equation, and the fitting result is the optimized fitting equation.

[0056] In the present application, the comparison between the ideal parameter and the actual parameter is referenced, and the second quality change coefficient is set. The second quality change coefficient is used to reflect the difference of the real parameter in a single dimension. In the fitting and prediction process of the first quality change coefficient, the second quality change coefficient is introduced to contest the fitting process, which can effectively solve the problem of data overfitting in multi-dimensional parameter analysis, and cause the prediction of the detection parameter to deviate from the real value. Further, according to the fitting result, the activity state of the preset packaging test scheme can be effectively analyzed in multiple cycles, and the activity trend of the packaged food can be judged according to the fitting prediction, and the packaging scheme can be further optimized, such as material optimization, release compound optimization, packaging optimization based on pH response, etc.

[0057] The mean square error is calculated as follows: ; Wherein, is the fitting value of the i-th detection period and the second quality change coefficient, N is the total number of detection periods, and MSE is the mean square error.

[0058] According to an embodiment of the present application, the S16 comprises: Based on the fitting result, the food activity state and the packaging effect are evaluated; A prediction period is set, the food activity state is predicted based on the fitting result, and the trend analysis of the packaging effect and the food activity state is reflected through the prediction process; According to the prediction process, a time period in which the activity change does not conform to an expected degree of quality change is analyzed, and packaging optimization is performed according to a preset packaging test scheme.

[0059] The third aspect of the present application also provides a machine-readable storage medium, which stores instructions, and the instructions, when executed by a processor, cause the processor to execute the intelligent active packaging optimization method based on machine learning.

[0060] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other manners. The above described device embodiments are merely schematic, for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the various components shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, and can be electrical, mechanical or other forms.

[0061] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units; they can be located in one place, or distributed on a plurality of network units; and part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0062] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be realized in the form of hardware, or in the form of hardware plus software functional units.

[0063] Those skilled in the art can understand that all or part of the steps of the above method embodiments can be completed by program instruction related hardware, and the foregoing program can be stored in a computer readable storage medium, and the program, when executed, executes steps including the above method embodiments; and the foregoing storage medium includes mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks or optical disks, and various media that can store program codes.

[0064] Alternatively, the above-mentioned integrated unit of the present application, if realized in the form of a software function module and sold or used as an independent product, can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solutions of the embodiments of the present application can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the methods described in the embodiments of the present application. The aforementioned storage medium includes: mobile storage devices, ROM, RAM, magnetic disks or optical disks, and various media that can store program codes.

[0065] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A machine learning-based intelligent active packaging optimization method, characterized in that, include: S11: Based on the preset packaging test plan, package the target active food and test the food parameters, and acquire and record the food activity parameters based on each test cycle; S12: A periodic detection parameter matrix is ​​constructed based on the time dimension and food activity parameters. A CNN model is introduced to perform cyclic convolution calculation on the detection parameter matrix, and the results of multiple calculations are fused to generate a change feature set. S13: Introduce the contrast feature set and the change feature set of the ideal parameter matrix to compare the differences and calculate the first qualitative change coefficient of each detection cycle; S14: Based on the weights of multiple preset parameters, the difference between the food activity parameters and the expected parameters is compared and qualitative changes are evaluated. The degree of qualitative change is weighted and averaged to obtain the second qualitative change coefficient for each detection cycle. S15: The linear change of the first qualitative change coefficient is fitted by linear regression, and the mean square error between the fitted value and the second qualitative change coefficient is minimized as the objective. The fitting coefficient is adjusted by the least squares method to obtain the fitting result. S16: Evaluate the packaging effect based on the fitting results, predict changes in food activity for the preset packaging test plan, and optimize the packaging plan.

2. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, S11 includes: According to the preset packaging test plan, multiple sensors are set up to detect the target active food in real time. Set multiple detection cycles and set multiple recording nodes in each detection cycle; For each testing cycle, obtain multi-dimensional food activity parameters.

3. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, Specifically, S12 is as follows: In one detection cycle, multiple recording nodes are included. The detection parameter matrix is ​​constructed with time as the first dimension and various food activity parameters as the second dimension. Each detection cycle corresponds to a detection parameter matrix; A CNN model is constructed, which includes convolutional layers, activation layers, and pooling layers. A convolutional kernel of a preset size is set in the convolutional layer. The detection parameter matrix is ​​convolved and calculated repeatedly through the convolutional layer until the convolution calculation covers the entire detection parameter matrix. The obtained local features are then introduced into non-linear features through the activation layer. Local features are max-pooled through pooling layers to generate a set of changing features.

4. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, S13 includes: Based on multiple recording nodes of a detection cycle, the ideal food activity parameters are input, and an ideal parameter matrix is ​​constructed based on the ideal parameters. A CNN model is used to perform cyclic convolution calculations on the ideal parameter matrix and generate a comparative feature set. Based on the mean-shift algorithm, a clustering space is constructed, and a set of variation features and a set of contrast features are imported into the clustering space respectively. Density clustering is performed on the change feature set and the contrast feature set respectively. The distance between the feature data is calculated using Manhattan distance. The clustering is repeated until the center points converge, and two sets of cluster center points are obtained. In the cluster space, the differences between the two groups of cluster centers are analyzed, the average distance from one group of cluster centers to the other group of cluster centers is calculated, and the first qualitative change coefficient is obtained by combining the difference in the number of cluster centers between the two groups.

5. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, Specifically, S14 is: Based on the preservation properties of the target active food, preset parameter weights are set for each active parameter; For a testing cycle, the differences between the food activity parameters and the expected parameters are compared and qualitative changes are assessed. The weighted average of the differences in each parameter is calculated, and the weighted calculation incorporates the preset parameter weights to obtain the second qualitative change coefficient.

6. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, S15 includes: Linear regression fitting is performed on multiple first qualitative change coefficients corresponding to multiple detection cycles to obtain the fitting equation; The mean value of the fitted data within a detection period is calculated from the fitted equation to obtain the fitted value. For all detection cycles, the mean square error between the fitted value and the corresponding second qualitative change coefficient is calculated. The fitted coefficient is adjusted by least squares method with the goal of minimizing the mean square error, until the number of iterations is reached, and the fitted equation is recorded to obtain the fitting result.

7. The intelligent active packaging optimization method based on machine learning according to claim 1, characterized in that, S16 includes: The fitting results are used to evaluate the food's active state and packaging effectiveness. Set a prediction period, predict the food's active state based on the fitting results, and reflect the trend analysis of packaging effect and food's active state through the prediction process. Based on the prediction process, analyze the time periods in which the changes in activity and the degree of qualitative change do not meet expectations, and optimize the packaging for the preset packaging test plan.

8. A machine learning-based intelligent active packaging optimization system, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory includes a machine learning-based intelligent active packaging optimization program. When executed by the processor, the machine learning-based intelligent active packaging optimization program performs the following steps: S11: Based on the preset packaging test plan, package the target active food and test the food parameters, and acquire and record the food activity parameters based on each test cycle; S12: A periodic detection parameter matrix is ​​constructed based on the time dimension and food activity parameters. A CNN model is introduced to perform cyclic convolution calculation on the detection parameter matrix, and the results of multiple calculations are fused to generate a change feature set. S13: Introduce the contrast feature set and the change feature set of the ideal parameter matrix to compare the differences and calculate the first qualitative change coefficient of each detection cycle; S14: Based on the weights of multiple preset parameters, the difference between the food activity parameters and the expected parameters is compared and qualitative changes are evaluated. The degree of qualitative change is weighted and averaged to obtain the second qualitative change coefficient for each detection cycle. S15: The linear change of the first qualitative change coefficient is fitted by linear regression, and the mean square error between the fitted value and the second qualitative change coefficient is minimized as the objective. The fitting coefficient is adjusted by the least squares method to obtain the fitting result. S16: Evaluate the packaging effect based on the fitting results, predict changes in food activity for the preset packaging test plan, and optimize the packaging plan.

9. The intelligent active packaging optimization system based on machine learning according to claim 8, characterized in that, S11 includes: According to the preset packaging test plan, multiple sensors are set up to detect the target active food in real time. Set multiple detection cycles and set multiple recording nodes in each detection cycle; For each testing cycle, obtain multi-dimensional food activity parameters.

10. A machine-readable storage medium storing instructions thereon, characterized in that, When executed by a processor, this instruction causes the processor to perform the machine learning-based intelligent active packaging optimization method according to any one of claims 1 to 7.

Images