Food quality dynamic monitoring system and method based on multi-sensor fusion

Abstract

The application belongs to the technical field of food quality monitoring, and discloses a food quality dynamic monitoring system and method based on multi-sensor fusion, which comprises the following steps: collecting initial quality data of food, storage influence data in the storage stage and food categories; performing feature extraction on the storage influence data to obtain storage feature data; inputting the food categories, the initial quality data and the storage feature data into a shelf life prediction model to obtain a predicted shelf life of finished products; comparing and analyzing the predicted shelf life of finished products with a preset shelf life threshold to determine whether to generate a storage adjustment instruction; and especially in the case that the predicted shelf life is short or the taste is not up to standard, the storage or processing strategy is optimized in advance to maximize the reduction of waste of raw materials and energy and reduce the fluctuation of food quality.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of food quality monitoring, more specifically, the present application relates to a food quality dynamic monitoring system and method based on multi-sensor fusion. BACKGROUND

[0002] In the food industry, the storage and processing quality of food directly affects its shelf life and product taste, however, due to the difference in the quality of food raw materials and the dynamic changes of various factors in the storage and processing process, the traditional quality control method is difficult to meet the strict requirements of modern consumers on food safety and quality; the existing technology usually relies on a single sensor or a static data analysis method, and lacks comprehensive monitoring and intelligent prediction ability for the dynamic changes of food quality in the whole storage-processing chain.

[0003] As in the prior art, the correlation analysis of storage and processing data is limited, and the overall quality performance of food from storage to processing completion cannot be accurately predicted, especially the comprehensive evaluation of product shelf life and taste is insufficient; when the storage environment or processing parameters cause the predicted product shelf life or taste to not meet the expectation, the existing system cannot provide intelligent storage parameter optimization or processing parameter adjustment scheme, resulting in large fluctuations in food quality.

[0004] In view of this, the present application proposes a food quality dynamic monitoring system and method based on multi-sensor fusion to solve the above problems. SUMMARY

[0005] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned purposes, the present application provides the following technical scheme: a food quality dynamic monitoring method based on multi-sensor fusion, comprising:

[0006] Collecting initial quality data of food, storage influence data in the storage stage and food categories;

[0007] Feature extraction is performed on the storage influence data to obtain storage feature data;

[0008] The food categories, initial quality data and storage feature data are input into a shelf life prediction model to obtain a predicted product shelf life;

[0009] The predicted product shelf life is compared and analyzed with a preset shelf life threshold value to determine whether to generate a storage adjustment instruction;

[0010] If the storage adjustment instruction is generated, the food categories, the predicted product shelf life and the shelf life threshold value are input into a storage data adjustment model to obtain optimized storage influence data; it is judged whether the optimized storage influence data is executable; if it is executable, the process is ended, if it is not executable, a processing adjustment instruction is generated;

[0011] According to the processing adjustment instruction, the processing parameters corresponding to the food category, the predicted shelf life of the finished product, and the shelf life threshold are input into the pre-trained processing parameter adjustment model to obtain optimized processing parameter data; the optimized processing parameter data and the food category are input into the pre-trained taste analysis model to obtain a taste score, and according to the taste score, it is determined whether the food category is processed preferentially.

[0012] Further, the training method of the shelf life prediction model comprises:

[0013] A set of pre-processing influence data is collected in advance, A is an integer greater than 1, the pre-processing influence data includes pre-processing feature data and the finished product shelf life corresponding to the pre-processing feature data; the pre-processing feature data includes initial quality data and storage feature data;

[0014] Each set of pre-processing influence data is used as the input of the shelf life prediction model, the shelf life prediction model takes the finished product shelf life corresponding to each set of pre-processing feature data as the output, takes the actual finished product shelf life corresponding to each set of pre-processing feature data as the prediction target, and takes the sum of the loss function values of all pre-processing feature data as the training target; the shelf life prediction model is trained until the sum of the prediction errors reaches convergence, and the training is stopped; the shelf life prediction model is a deep neural network model.

[0015] Further, the optimized storage influence data includes adjusting the temperature, humidity, gas concentration or pressure of the storage equipment; the method for determining whether the optimized storage influence data is executable comprises:

[0016] If any data in the optimized storage influence data exceeds the rated adjustment range of the storage equipment, it is not executable.

[0017] Further, if the predicted shelf life of the finished product does not exceed the preset shelf life threshold, no storage adjustment instruction is generated; if the predicted shelf life of the finished product exceeds the preset shelf life threshold, a storage adjustment instruction is generated.

[0018] Further, the training method of the storage data adjustment model comprises:

[0019] k sets of storage data adjustment data are collected in advance, the storage data adjustment data includes storage feature data and optimized storage influence data corresponding to the storage feature data, and the storage feature data includes food category, predicted shelf life of finished product and shelf life threshold;

[0020] Convert k groups of storage data adjustment data into feature vectors and divide them into training set and test set. Use the storage feature data in the training set as the input of the storage data adjustment model, and use the optimized storage impact data in the training set as the output of the storage data adjustment model. Use the test set to test the storage data adjustment model, and output a storage data adjustment model that meets the prediction accuracy. The storage data adjustment model is a deep neural network or a gradient boosting decision tree model.

[0021] Furthermore, the shelf life of the finished product is the shelf life after the food is processed.

[0022] Furthermore, the storage impact data includes the temperature, humidity, gas concentration and pressure of the storage equipment per unit time.

[0023] Furthermore, the stored characteristic data include the average temperature, the standard deviation of the temperature, the average humidity, the standard deviation of the humidity, the average gas concentration, the standard deviation of the gas concentration, and the average pressure and the standard deviation of the pressure within a unit time.

[0024] Furthermore, the initial quality data includes microbial indicators, physical indicators and nutritional components; microbial indicators include total colony count, E. coli count and mold count; physical indicators include moisture content and water activity; nutritional components include protein content, fat content and carbohydrate content.

[0025] Furthermore, the training method of the taste analysis model includes:

[0026] A taste analysis data set is obtained, wherein the taste analysis data set includes optimized processing parameter data, food types, and taste scores; the taste analysis data set is divided into a sample training set and a sample test set, and a regression network is constructed; the regression network is trained using the optimized processing parameter data and food types in the sample training set as input data of the regression network, and using the taste scores in the sample training set as output data of the regression network to obtain an initial regression network for predicting taste scores; the initial regression network is tested using the sample test set, and the initial regression network with an error value less than a preset value is output as a taste analysis model.

[0027] The food quality dynamic monitoring system based on multi-sensor fusion implements the food quality dynamic monitoring method based on multi-sensor fusion, including:

[0028] The collection module is used to collect the initial quality data of food, storage impact data during the storage stage, and food types;

[0029] A processing module, used for extracting features from storage impact data and obtaining storage feature data;

[0030] The first analysis module is used for inputting the food category, initial mass data and storage characteristic data into a shelf life prediction model to obtain a predicted shelf life of a finished product;

[0031] The second analysis module is used for comparing and analyzing the predicted shelf life of the finished product with a preset shelf life threshold value to determine whether to generate a storage adjustment instruction;

[0032] The third analysis module is used for inputting the food category, predicted shelf life of the finished product and shelf life threshold value into a storage data adjustment model to obtain optimized storage influence data, judging whether the optimized storage influence data is executable, and ending if the optimized storage influence data is executable, and generating a processing adjustment instruction if the optimized storage influence data is not executable;

[0033] The fourth analysis module is used for inputting the food category corresponding processing parameters, predicted shelf life of the finished product and shelf life threshold value into a pre-trained processing parameter adjustment model according to the processing adjustment instruction to obtain optimized processing parameter data, inputting the optimized processing parameter data and food category into a pre-trained taste analysis model to obtain a taste score, and determining whether the food category is preferentially processed according to the taste score.

[0034] The food quality dynamic monitoring system based on multi-sensor fusion has the following beneficial effects:

[0035] The food quality dynamic monitoring system based on multi-sensor fusion can not only realize accurate real-time monitoring during food storage and processing, but also can predict the shelf life and quality performance of the finished product according to the collected initial mass data of the food and storage influence data during the storage stage before the food enters the processing stage. The intelligent prediction and control capability of the system greatly improves the flexibility and efficiency of the food processing and storage process, mainly in the following aspects:

[0036] By fusing the initial mass data of the food and the storage influence data, the shelf life of the finished product can be predicted, the risk of food deterioration or quality decline can be found in advance, the storage environment parameters can be adjusted in time before the risk occurs, and the occurrence of food deterioration can be avoided, so as to ensure food safety; the system automatically generates a storage adjustment instruction according to the food category, predicted shelf life of the finished product and shelf life threshold value, optimizes the storage influence data, prolongs the shelf life of the food, and reduces food loss.

[0037] If the optimized storage impact data cannot be executed, the system further generates processing adjustment instructions to optimize processing parameters, improve product quality, and predict the taste score of the finished product, ensuring that the food meets consumer demand; The system predicts shelf life and taste before the food enters the processing stage, which can avoid food waste caused by unreasonable processing parameters or improper storage environment; Especially in the case of short predicted shelf life or substandard taste, by optimizing storage or processing strategies in advance, the waste of raw materials and energy is minimized, and the fluctuation of food quality is reduced.

[0038] The multi-sensor fusion system provides accurate prediction data and intelligent adjustment suggestions to help food companies make scientific decisions at different stages: when the shelf life prediction result is below the threshold, the company can choose to adjust the storage impact data or processing parameters to ensure the quality of the finished product; If the adjusted taste score of the finished product still does not meet market demand, the company can prioritize processing to avoid food waste caused by delayed processing.

[0039] Through multi-sensor fusion data collection and analysis, the system can monitor food quality in real time, predict the taste score after processing is completed, and provide optimization schemes for the storage and processing stages; This dynamic monitoring and control mechanism makes the entire food production process more intelligent and transparent: from the initial quality of the food to the dynamic monitoring of the finished product quality, achieving closed-loop management of storage, processing, and quality control.

[0040] Food quality and taste directly affect consumers' willingness to buy. By predicting the quality and taste of the finished product before processing and adjusting the processing strategy to meet market demand, companies can significantly improve the market acceptance of the finished product; At the same time, dynamic monitoring systems help to establish a high-quality brand image and enhance market competitiveness. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 A schematic diagram of the food quality dynamic monitoring system based on multi-sensor fusion of the present application;

[0042] Figure 2 A schematic diagram of the food quality dynamic monitoring system method based on multi-sensor fusion of the present application. DETAILED DESCRIPTION

[0043] The technical solutions in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0044] Example 1

[0045] Referring to Figure 1 The food quality dynamic monitoring system based on multi-sensor fusion described in this embodiment includes the following modules:

[0046] The acquisition module is used to acquire initial quality data of the food, storage influence data in the storage stage, and food categories. The storage influence data includes temperature, humidity, gas concentration (such as carbon dioxide and oxygen), and pressure of the storage equipment per unit time.

[0047] The initial quality data includes microbial indicators, physical indicators, and nutritional components. The microbial indicators include total bacterial colony count, Escherichia coli count, and mold count. The total bacterial colony count is expressed in CFU / g (colony-forming units per gram), indicating the number of active microorganisms per gram of food. High bacterial colony count can accelerate food spoilage. Plate counting method is used to culture and count microorganisms in the sample. Escherichia coli count is expressed in MPN / 100g (number per 100 grams). Escherichia coli is one of the food hygiene indicators. Its presence can be detected to judge the hygiene quality of the food. Multiple tube fermentation method or membrane filtration method is used for determination. Mold count is expressed in CFU / g. The growth of mold can cause food to become moldy. Selective medium is used for counting.

[0048] The physical indicators include moisture content and water activity. Moisture content directly affects the reproductive ability of microorganisms. Low-moisture foods (such as dried fruits) have a longer shelf life. Dry weight loss method or Karl Fischer moisture meter is used for determination. Water activity is valued between 0 and 1. Water activity controls the growth of microorganisms. The lower the water activity, the slower the growth rate of microorganisms. Water activity meter is used for measurement.

[0049] Nutritional components include protein content, fat content, and carbohydrate content. Protein decomposition can lead to the production of volatile nitrogen compounds, resulting in spoilage. Kjeldahl method is used to determine protein content. Fat oxidation can produce peroxides, leading to rancidity. Soxhlet extraction method or Fourier transform infrared spectroscopy (FTIR) is used to determine fat content. Sugar decomposition can cause Maillard reaction, affecting food color and taste flavor. Total sugar content determination method is used to determine carbohydrate content.

[0050] The processing module is used to extract features from the storage influence data and obtain storage feature data. The storage feature data includes temperature average value, temperature standard deviation, humidity average value, humidity standard deviation, gas concentration average value, gas concentration standard deviation, pressure average value, and pressure standard deviation per unit time.

[0051] The first analysis module is configured to input the food type, initial quality data and storage characteristic data into a shelf life prediction model to obtain a predicted shelf life of the finished product; the shelf life of the finished product is the shelf life of the finished product after food processing, and the shelf life of the finished product is predicted according to the initial quality data and the storage characteristic data; and the food is processed by default processing parameters such as ingredient ratio, processing temperature, mixing intensity and processing PH.

[0052] The training method of the shelf life prediction model comprises the following steps:

[0053] A set of pre-processing influence data is collected in advance, A is an integer greater than 1, the pre-processing influence data includes pre-processing characteristic data and the shelf life of the finished product corresponding to the pre-processing characteristic data; the pre-processing characteristic data includes initial quality data and storage characteristic data.

[0054] Each set of pre-processing influence data is used as an input of the shelf life prediction model, the shelf life prediction model uses the shelf life of the finished product corresponding to each set of pre-processing characteristic data as an output, uses the actual shelf life of the finished product corresponding to each set of pre-processing characteristic data as a prediction target, and uses the sum of loss function values of all pre-processing characteristic data as a training target; the shelf life prediction model is trained until the sum of prediction errors reaches convergence to stop training; and the shelf life prediction model is specifically a deep neural network model.

[0055] The second analysis module compares and analyzes the predicted shelf life of the finished product with a preset shelf life threshold to determine whether to generate a storage adjustment instruction.

[0056] If the predicted shelf life of the finished product does not exceed the preset shelf life threshold, no storage adjustment instruction is generated; if the predicted shelf life of the finished product exceeds the preset shelf life threshold, a storage adjustment instruction is generated.

[0057] The third analysis module inputs the food type, predicted shelf life of the finished product and shelf life threshold into a storage data adjustment model to obtain optimized storage influence data if the storage adjustment instruction is generated; it is determined whether the optimized storage influence data is executable; if it is executable, the process ends; if it is not executable, a processing adjustment instruction is generated.

[0058] The optimized storage influence data includes data such as temperature, humidity, gas concentration (carbon dioxide, oxygen) or pressure of the storage equipment, and the storage influence data is adjusted to ensure that the food type is within the preset shelf life threshold in the initial quality data and the predicted shelf life of the finished product after a unit of time, that is, the storage influence data is changed to prolong the shelf life of the finished product.

[0059] The method for determining whether the optimized storage influence data is executable comprises the following steps:

[0060] If any of the optimized storage influence data exceeds the rated adjustment range of the storage device, for example, the temperature exceeds the rated adjustment range of the storage device, the storage device cannot adjust the temperature, and at this time, the process cannot be executed. The reason for increasing this process is that the storage data adjustment model learns the relationship between the food category, the predicted shelf life of the finished product, the shelf life threshold, and the optimized storage influence data, without considering the performance of the storage device. At this time, the optimized storage influence data obtained by the storage data adjustment model may cause the above-mentioned situation that the process cannot be executed. Once this situation occurs, the requirement of changing the shelf life of the finished product of the food cannot be met.

[0061] The training method of the storage data adjustment model comprises:

[0062] K sets of storage data adjustment data are collected in advance, k is an integer greater than 1; the storage data adjustment data comprises storage characteristic data and corresponding optimized storage influence data, and the storage characteristic data comprises a food category, a predicted shelf life of a finished product, and a shelf life threshold.

[0063] It should be noted that the predicted shelf life of the finished product and the shelf life threshold can reflect the shelf life gap between the two, and the shelf life threshold also has a limiting effect. Under the limiting effect of the shelf life threshold, optimized storage influence data that meets the limited conditions is collected during the experimental stage. That is, when the predicted shelf life of the finished product obtained by the shelf life prediction model for the first time does not meet the limited conditions of the shelf life threshold, the storage conditions of the food are changed by using the optimized storage influence data, and the predicted shelf life of the finished product obtained by the shelf life prediction model again meets the limited conditions of the shelf life threshold.

[0064] The k sets of storage data adjustment data are converted into feature vectors, k is an integer greater than 1, and are divided into a training set and a test set. The storage characteristic data in the training set is used as the input of the storage data adjustment model, the optimized storage influence data in the training set is used as the output of the storage data adjustment model, the test set is used to test the storage data adjustment model, and the storage data adjustment model that meets the prediction accuracy is output. The storage data adjustment model is a deep neural network or a gradient boosting decision tree model.

[0065] The fourth analysis module inputs the processing parameters corresponding to the food category, the predicted shelf life of the finished product, and the shelf life threshold into the pre-trained processing parameter adjustment model according to the processing adjustment instruction, obtains optimized processing parameter data, and inputs the optimized processing parameter data and the food category into the pre-trained taste analysis model, obtains a taste score, and determines whether the food category is processed preferentially according to the taste score, so as to ensure that the quality of the finished product of the food category meets the demand of the public for taste and guarantees the sales of the food.

[0066] The training method of the processing parameter adjustment model comprises:

[0067] Pre-collecting C group processing impact data, C is an integer greater than 1, the processing impact data includes processing feature data and the optimized processing parameter data corresponding to the processing feature data; the processing feature data includes the processing parameters corresponding to the food category, the predicted shelf life of the finished product and the shelf life threshold.

[0068] It should be noted that the optimized processing parameter data corresponds to the food category after the processing parameter data is optimized, and the actual shelf life is within the shelf life threshold.

[0069] Secondly, the optimization of the processing parameter data will also affect the change of the taste of the finished food product, and the optimized processing parameter data includes the adjustment of the ingredient ratio, processing temperature, processing time, mixing intensity and processing pH; the processing temperature affects the microbial killing effect, enzyme inactivation, physical structure change and taste of the food; the processing time determines the degree of heating or cooling of the food, which affects the texture and taste of the final product; the ingredient ratio includes the addition ratio of various ingredients, such as the addition ratio of preservatives, etc., and the change of the ingredient ratio also determines the taste of the final food product; the mixing intensity is the rotation speed of the mixing equipment paddle per minute, which affects the uniformity of the ingredients and the physical structure of the food, and also affects the taste of the finished food product; the processing pH value affects the enzyme activity and microbial growth, and further affects the taste and texture of the food.

[0070] Each group of processing impact data is used as the input of the processing parameter adjustment model, the processing parameter adjustment model takes the optimized processing parameter data corresponding to each group of processing feature data as the output, takes the actual optimized processing parameter data corresponding to each group of processing feature data as the prediction target, and takes the sum of the loss function values of all processing feature data as the training target; the processing parameter adjustment model is trained until the sum of the prediction errors reaches convergence, and the training is stopped; the processing parameter adjustment model is specifically a deep neural network model.

[0071] The taste score reflects the public acceptance of the finished food product, which is obtained through sensory test of consumers or experts, and the score range is generally 1 to 10, the larger the taste score value, the higher the consumer acceptance, and the better the sales.

[0072] The training method of the taste analysis model includes:

[0073] Obtain a texture analysis dataset, which includes optimized processing parameter data, food category and texture score; divide the texture analysis dataset into a sample training set and a sample test set, and construct a regression network; use the optimized processing parameter data and food category in the sample training set as input data of the regression network, and use the texture score in the sample training set as output data of the regression network, train the regression network, and obtain an initial regression network for predicting the texture score; test the initial regression network using the sample test set, and output the initial regression network with an error value less than a preset error value as a texture analysis model.

[0074] The embodiment is based on a multi-sensor fusion food quality dynamic monitoring system, which can not only realize accurate real-time monitoring during food storage and processing, but also can predict the shelf life and quality performance of finished products according to the collected initial quality data of food and storage influence data during the storage stage before the food enters the processing stage. The intelligent prediction and control capability of the system greatly improves the flexibility and efficiency of the food processing and storage process, mainly reflected in the following aspects:

[0075] By fusing the initial quality data of food and the storage influence data, the shelf life of the finished product can be predicted, the risk of food deterioration or quality decline can be found in advance, and the storage environment parameters can be adjusted in time before the risk occurs, thereby avoiding the occurrence of food deterioration and ensuring food safety; the system automatically generates storage adjustment instructions according to the food category and the predicted shelf life of the finished product, prolongs the shelf life of the food by optimizing the storage influence data, and reduces food waste.

[0076] If the optimized storage influence data cannot be executed, the system will further generate processing adjustment instructions to optimize the processing parameters, improve the quality of the finished product, and predict the texture score of the finished product to ensure that the food meets the needs of consumers; the system predicts the shelf life and texture before the food enters the processing stage, which can avoid food waste caused by unreasonable processing parameters or improper storage environment; especially in the case of short predicted shelf life or unqualified texture, the optimization of storage or processing strategy in advance can minimize the waste of raw materials and energy.

[0077] The multi-sensor fusion system provides accurate prediction data and intelligent adjustment suggestions to help food enterprises make scientific decisions at different stages: when the predicted shelf life is lower than the threshold value, the enterprise can choose to adjust the storage storage influence data or processing parameters to ensure the quality of the finished product; if the adjusted texture score of the finished product still does not meet market demand, the enterprise can prioritize processing to avoid food waste caused by delayed processing.

[0078] Through data acquisition and analysis of multi-sensor fusion, the system can monitor food quality in real time, predict the taste score after processing is completed, and provide optimization scheme for storage and processing stage; such dynamic monitoring and control mechanism makes the whole food production process more intelligent and transparent: from the initial quality of food to the whole process of dynamic monitoring of finished product quality, realizing closed-loop management of storage, processing and quality control.

[0079] Food quality and taste directly affect the willingness of consumers to purchase. By predicting the quality and taste of food products before processing and adjusting the processing strategy to meet market demand, enterprises can significantly improve the market acceptance of finished products; at the same time, the dynamic monitoring system helps to establish a high-quality brand image and enhance market competitiveness.

[0080] Embodiment 2

[0081] Referring to Figure 2 The embodiment provides a food quality dynamic monitoring method based on multi-sensor fusion, comprising:

[0082] Collecting initial quality data of food, storage influence data in storage stage and food categories;

[0083] Feature extraction is performed on the storage influence data to obtain storage feature data;

[0084] The food categories, initial quality data and storage feature data are input into a shelf life prediction model to obtain a predicted shelf life of finished products;

[0085] The predicted shelf life of finished products is compared and analyzed with a preset shelf life threshold value to determine whether to generate a storage adjustment instruction;

[0086] If the storage adjustment instruction is generated, the food categories, the predicted shelf life of finished products and the shelf life threshold value are input into a storage data adjustment model to obtain optimized storage influence data; it is determined whether the optimized storage influence data is executable; if executable, the process is ended, and if not executable, a processing adjustment instruction is generated;

[0087] According to the processing adjustment instruction, the food categories corresponding processing parameters, the predicted shelf life of finished products and the shelf life threshold value are input into a pre-trained processing parameter adjustment model to obtain optimized processing parameter data; the optimized processing parameter data and the food categories are input into a pre-trained taste analysis model to obtain a taste score, and it is determined whether the food categories are processed preferentially according to the taste score.

[0088] The above merely describes specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any modification or substitution within the technical range disclosed by the present application can be easily thought of by those skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0089] Finally, the above merely describes preferred embodiments of the present application, and is not used to limit the present application, and any modification, equivalent substitution, improvement, etc. within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for dynamic monitoring of food quality based on multi-sensor fusion, characterized in that, The method comprises the following steps: Collecting initial quality data of food, storage influence data in storage stage and food category; Extracting features from the storage influence data to obtain storage feature data; Inputting the food category, initial quality data and storage feature data into a shelf life prediction model to obtain a predicted shelf life of finished products; Comparing and analyzing the predicted shelf life of finished products with a preset shelf life threshold to determine whether to generate a storage adjustment instruction; If the storage adjustment instruction is generated, inputting the food category, predicted shelf life of finished products and shelf life threshold into a storage data adjustment model to obtain optimized storage influence data; determining whether the optimized storage influence data is executable; if executable, ending; if not executable, generating a processing adjustment instruction; According to the processing adjustment instruction, inputting the processing parameters corresponding to the food category, predicted shelf life of finished products and shelf life threshold into a pre-trained processing parameter adjustment model to obtain optimized processing parameter data; inputting the optimized processing parameter data and food category into a pre-trained taste analysis model to obtain a taste score, and determining whether the food category is processed preferentially according to the taste score.

2. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, characterized in that, The training method of the shelf life prediction model comprises: Collecting A sets of pre-processing influence data in advance, A being an integer greater than 1, the pre-processing influence data comprising pre-processing feature data and the shelf life of finished products corresponding to the pre-processing feature data; the pre-processing feature data comprising initial quality data and storage feature data; Taking each set of pre-processing influence data as the input of the shelf life prediction model, taking the shelf life of finished products corresponding to each set of pre-processing feature data as the output, taking the actual shelf life of finished products corresponding to each set of pre-processing feature data as the prediction target, and taking the sum of loss function values of all pre-processing feature data as the training target; training the shelf life prediction model until the sum of prediction errors reaches convergence to stop training; the shelf life prediction model is a deep neural network model.

3. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, characterized in that, The optimized storage influence data comprises adjusting the temperature, humidity, gas concentration or pressure of the storage equipment; the method for determining whether the optimized storage influence data is executable comprises: If any data in the optimized storage influence data exceeds the rated adjustment range of the storage equipment, it is not executable.

4. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, wherein, If the predicted shelf life of finished products does not exceed the preset shelf life threshold, no storage adjustment instruction is generated; if the predicted shelf life of finished products exceeds the preset shelf life threshold, a storage adjustment instruction is generated.

5. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, wherein, The training method of the storage data adjustment model comprises: Collecting k sets of storage data adjustment data in advance, the storage data adjustment data comprising storage feature data and optimized storage influence data corresponding to the storage feature data, the storage feature data comprising food category, predicted shelf life of finished products and shelf life threshold; The k groups of storage data adjustment data are converted into feature vectors and divided into a training set and a test set, the storage feature data in the training set is taken as an input of a storage data adjustment model, the optimized storage influence data in the training set is taken as an output of the storage data adjustment model, the test set is used to test the storage data adjustment model, and a storage data adjustment model meeting a prediction accuracy is output.

6. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, wherein, The storage influence data includes temperature, humidity, gas concentration and pressure of the storage device per unit time.

7. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 6, characterized in that, The storage feature data includes temperature average value, temperature standard deviation, humidity average value, humidity standard deviation, gas concentration average value, gas concentration standard deviation, pressure average value and pressure standard deviation per unit time.

8. The multi-sensor fusion based dynamic monitoring method of food quality according to claim 1, wherein, The initial quality data includes microbial indicators, physical indicators and nutritional components; the microbial indicators include total bacterial colony count, Escherichia coli count and mold count; the physical indicators include moisture content and water activity; and the nutritional components include protein content, fat content and carbohydrate content. 9.The multi-sensor fusion based dynamic food quality monitoring method according to claim 1, wherein, The training method of the mouthfeel analysis model includes: An initial quality data set of the food is collected, and the initial quality data set includes initial quality data of the food, a food category and an initial mouthfeel score; the initial quality data set is divided into a sample training set and a sample test set, and a regression network is constructed; the initial quality data of the food and the food category in the sample training set are taken as input data of the regression network, and the initial mouthfeel score in the sample training set is taken as output data of the regression network, the regression network is trained, and an initial regression network for predicting the mouthfeel score is obtained; the initial regression network is tested by using the sample test set, and the initial regression network with an error value less than a preset error value is output as the mouthfeel analysis model.

10. A food quality dynamic monitoring system based on multi-sensor fusion, implementing the food quality dynamic monitoring method based on multi-sensor fusion according to any one of claims 1-9, characterized in that, It includes: A collection module is configured to collect initial quality data of the food, storage influence data in a storage phase and a food category; A processing module is configured to perform feature extraction on the storage influence data to obtain storage feature data; A first analysis module is configured to input the food category, the initial quality data and the storage feature data into a shelf life prediction model to obtain a predicted shelf life of a finished product; A second analysis module is configured to compare and analyze the predicted shelf life of the finished product with a preset shelf life threshold to determine whether to generate a storage adjustment instruction; A third analysis module is configured to input the food category, the predicted shelf life of the finished product and the shelf life threshold into a storage data adjustment model to obtain optimized storage influence data if the storage adjustment instruction is generated; and determine whether the optimized storage influence data is executable; if the optimized storage influence data is executable, the process ends; if the optimized storage influence data is not executable, a processing adjustment instruction is generated; A fourth analysis module is configured to input the food category, the predicted shelf life of the finished product and the shelf life threshold into a pre-trained processing parameter adjustment model according to the processing adjustment instruction to obtain optimized processing parameter data; input the optimized processing parameter data and the food category into a pre-trained mouthfeel analysis model to obtain a mouthfeel score; and determine whether the food category is preferentially processed according to the mouthfeel score.

Images