Food material freshness identification method and device, electronic equipment and storage medium

Abstract

The invention discloses a food material freshness identification method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining historical fresh-keeping data of different food materials, and extracting index features based on different storage times of different food materials during freshness identification; predicting missing index features of the food materials based on a sorting learning algorithm, and screening core features influencing the freshness of the food materials in combination with the extracted index features; and training a freshness identification model according to the core features influencing the freshness of the food materials and the sample data of the food materials. According to the method, the index characteristics of the freshness of the food materials are obtained by using the lambda Mart model realized by using the XGboost, so that the problems that the storage time of different food materials is incomparable and the index characteristics of part of the food materials are missing are solved; the method comprises the following steps: providing an evaluation standard for the importance of index features by using feature importance Weight in XGboost, and screening out core features influencing the freshness of food materials; and the stability of model iteration is ensured by a method of adding an artificial standard in the freshness identification model.

Description

Technical Field

[0001] The present invention relates to the technical field of training data, and in particular to a method, device, electronic device and storage medium for identifying the freshness of food. Background Art

[0002] At present, the refrigeration technology of refrigerators has largely solved the problem of rapid deterioration of food. However, as people's pace of life accelerates, many users have the habit of stocking up a large amount of food in the refrigerator. Since users usually judge the freshness of food by intuitive feeling, they may fail to find and clean up expired food in time. Therefore, it is very important to study the food freshness data to help users identify the freshness of food.

[0003] Existing food freshness recognition technology generally uses machine learning models to predict the freshness of food based on data collected by various sensors in the refrigerator. The training data label (label) required to train such a machine learning model is freshness. Generally, the freshness of food is marked as different levels from fresh to corrupt, and the food is comprehensively judged whether the food is fresh and whether it can meet the edible standard based on different indicators of the food. For example, Chinese patent CN113945039A discloses a method for determining the freshness of food, a control method and system for refrigerator freshness, including obtaining historical freshness data, the historical freshness data includes the freshness of different food at different temperatures and different storage times; extracting the freshness factors that affect the freshness of food from the historical freshness data; training the historical freshness data and the freshness factors based on a machine learning algorithm to obtain a determination model for the freshness of food and the freshness coefficient of the freshness factor of each food at different temperatures and different storage times; the determination model is used to determine the freshness of any food.

[0004] However, no matter which identification method is used, it faces the problem of missing data features due to different initial conditions of food. Summary of the invention

[0005] The purpose of the present invention is to provide a method, device, electronic device and storage medium for identifying the freshness of food, so as to solve the problem of missing data features caused by different initial conditions of food in the prior art.

[0006] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention provides a method for identifying the freshness of food, comprising:

[0007] Obtain historical freshness data of different ingredients, and extract indicator features based on the different storage times of different ingredients when identifying freshness;

[0008] Predict missing indicator features of ingredients based on ranking learning algorithm, and filter out the core features that affect the freshness of ingredients based on the extracted indicator features;

[0009] The freshness recognition model is trained based on the core features that affect the freshness of the food and the sample data of the food.

[0010] As a further improvement of an embodiment of the present invention, the method further includes: the historical freshness-keeping data includes the freshness-keeping degree of different food materials under different storage conditions and different storage times.

[0011] As a further improvement of an embodiment of the present invention, the method further includes: the “indicator features for predicting missing ingredients based on a ranking learning algorithm” includes:

[0012] Define a set of known indicator features x i and the corresponding sorted label y i ;

[0013] For each pair of indicator features x i and x j , calculate the difference value to get d ij =x i -x j , and sort the labels y i and j Convert to comparison label to get c ij =sign(x i -x j );

[0014] The indicator feature x i and x j Input the neural network model and output x i and x j The ranking score difference of ij , then x i The sorting of j The probability before is expressed as:

[0015]

[0016] As a further improvement of an embodiment of the present invention, the method further includes: the “indicator features for predicting missing ingredients based on a ranking learning algorithm” further includes,

[0017] By using the cross entropy loss function, we measure the x i Ranked x j The probability p before ij Compare with label c ij The difference is expressed as:

[0018] L=-sum(c ij×log(p ij )+(1-c ij )×log(1-p ij ))

[0019] By minimizing the loss function L using the gradient descent optimization algorithm, the ranking relationship between indicator features is learned, and the missing indicator features are predicted based on the ranking relationship.

[0020] As a further improvement of an embodiment of the present invention, the method further includes: the “screening of core features that affect the freshness of food” includes:

[0021] The decision tree model is trained based on the gradient boosting tree algorithm to calculate the feature importance corresponding to the indicator features obtained through extraction and prediction;

[0022] According to the value of the feature importance, the indicator features are sorted to screen out the core features that affect the freshness of the food.

[0023] As a further improvement of an embodiment of the present invention, the method further includes: the “training freshness recognition model” includes:

[0024] When the freshness recognition model is training for freshness recognition, samples that cannot be recognized for freshness are manually labeled for recognition;

[0025] The freshness prediction grade is determined by the type of food.

[0026] As a further improvement of an embodiment of the present invention, the method further includes: the manual annotation recognition includes:

[0027] When the freshness recognition model performs freshness recognition, when the prediction probabilities of freshness prediction and grading of different core features are all less than the first threshold, freshness prediction and grading recognition cannot be performed, and manual annotation is then used for recognition;

[0028] When manual annotation recognition is connected, the manually annotated data and the correct data recognized by the freshness recognition model are merged to form a new training data set, and the freshness recognition model is retrained using the data set to iterate a new freshness recognition model.

[0029] To achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention provides a device for identifying the freshness of food materials, including a data acquisition module, an indicator feature processing module and a recognition model training module.

[0030] The data acquisition module is used to acquire historical freshness data of different ingredients and extract index features based on different storage times of different ingredients during freshness identification;

[0031] The indicator feature processing module is used to predict the missing indicator features of the ingredients based on the ranking learning algorithm, and screen the core features that affect the freshness of the ingredients in combination with the extracted indicator features;

[0032] The recognition model training module is used to train a freshness recognition model based on the core features that affect the freshness of the food and the sample data of the food.

[0033] To achieve one of the above-mentioned purposes of the invention, one embodiment of the present invention provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the above-mentioned method for identifying the freshness of food are implemented.

[0034] To achieve one of the above-mentioned purposes of the invention, one embodiment of the present invention provides a storage medium, wherein the storage medium stores a computer program, and is characterized in that when the computer program is executed by a processor, the steps in the above-mentioned method for identifying the freshness of food are implemented.

[0035] Compared with the prior art, the method, device, electronic device and storage medium for identifying the freshness of food provided by the present invention obtain the index characteristics of the freshness of food by using the lambdaMart model implemented by XGboost, thereby solving the problem that the storage time of different food ingredients is incomparable and the problem that some index characteristics of food ingredients are missing; by using the feature importance Weight in XGboost to provide an evaluation standard for the importance of the index characteristics, the core characteristics that affect the freshness of food ingredients are screened out; in the freshness identification model, the stability of the model iteration is ensured by adding an artificial standard method. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is an overall flow chart of the method for identifying the freshness of food materials of the present invention.

[0037] Figure 2 It is a structural schematic diagram of the device for identifying the freshness of food materials of the present invention. DETAILED DESCRIPTION

[0038] The present invention will be described in detail below in conjunction with the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by a person skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0039] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0040] In the first embodiment of the present invention, the present invention provides a method for identifying the freshness of food, such as Figure 1 As shown, the method includes:

[0041] S1: Obtain historical freshness data of different ingredients, and extract indicator features based on the different storage times of different ingredients during freshness identification;

[0042] S2: Predict missing indicator features of ingredients based on ranking learning algorithm, and screen the core features that affect the freshness of ingredients based on the extracted indicator features;

[0043] S3: training a freshness recognition model based on the core features that affect the freshness of the food and the sample data of the food.

[0044] In a specific embodiment of the present invention, the historical freshness-keeping data includes the freshness-keeping degree of different food materials under different storage conditions and different storage times.

[0045] Specifically, historical preservation data refers to data that records and collects the freshness of various ingredients under different storage conditions and different storage times. These data contain the results of observations and measurements of the changes in ingredients under different environments. When collecting historical preservation data, factors such as the type of ingredients, storage temperature, humidity, and gas composition are taken into account. For example, for meat ingredients, the freshness at different temperatures may be recorded, such as changes under refrigeration, freezing, or room temperature. For ingredients such as vegetables and fruits, the freshness under different humidity and atmosphere conditions may be recorded, such as changes under refrigeration, vacuum packaging, or room temperature. By collecting and analyzing these historical preservation data, we can gain a deep understanding of the freshness characteristics of different ingredients under different storage conditions and times. These data can be used to build models and algorithms to predict the freshness and shelf life of ingredients. By analyzing historical preservation data, we can find the impact of different storage conditions on the freshness of ingredients, and then optimize the storage conditions and extend the shelf life of ingredients.

[0046] In a specific embodiment of the present invention, the indicator characteristics for predicting missing ingredients based on the ranking learning algorithm are specifically:

[0047] The present invention uses the lambdaMart model implemented by XGboost to solve the problem of missing indicator features, specifically,

[0048] Define a set of known indicator features x i and the corresponding sorted label y i ;

[0049] For each pair of indicator features x i and x j , calculate the difference value to get d ij =x i -x j , and sort the labels y i and j Convert to comparison label to get c ij =sign(x i -x j );

[0050] The indicator feature x i and x j Input the neural network model and output x i and x j The ranking score difference of ij , then x i The sorting of j The probability before is expressed as:

[0051]

[0052] By using the cross entropy loss function, we measure the x i Ranked x j The probability p before ij Compare with label c ij The difference is expressed as:

[0053] L=-sum(c ij ×log(p ij )+(1-c ij )×log(1-p ij ))

[0054] By minimizing the loss function L using the gradient descent optimization algorithm, the ranking relationship between indicator features is learned, and the missing indicator features are predicted based on the ranking relationship.

[0055] It should be noted that XGBoost is a gradient boosting tree algorithm, and LambdaMART is a sorting learning model in XGBoost. The LambdaMART model trains the sorting model by minimizing the equivalent form of the sorting error, and uses the gradient information of the Lambda value to adjust the model parameters to better optimize the sorting performance.

[0056] The training process of the LambdaMART model includes:

[0057] Data preparation: First, you need to prepare a set of training data, where each training sample consists of the features of the object, the relevance label, and the ranking information of the object. These relevance labels can be manually annotated or calculated based on some evaluation indicators. Feature extraction: For each object, you need to extract meaningful information from its features. These features can include text features, structural features, user behavior features, etc., depending on the requirements of the sorting task. Model training: The LambdaMART model uses the gradient boosting tree as the base model for training. In each round of iteration, the model calculates the feature representation of the object and outputs a score to represent the ranking score of the object. By iteratively training a series of weak sorting models, the LambdaMART model gradually improves the overall sorting performance. Loss function optimization: During the training process, the LambdaMART model uses a gradient information called Lambda value to adjust the model parameters. The Lambda value represents the ranking difference between object pairs. By minimizing the equivalent form of the sorting error, the LambdaMART model can better optimize the sorting performance. Predicted ranking: After the model training is completed, the model can be used to predict the ranking of new objects. The model calculates the feature representation of the object and outputs a score to represent the ranking score of the object. Based on these scores, the objects can be sorted.

[0058] In one embodiment of the present invention, the core features that affect the freshness of food are screened, specifically,

[0059] In the present invention, the "Weight" provided by XGboost is used as the evaluation criterion for the importance of indicator features, specifically,

[0060] A decision tree model is trained by using the gradient boosting tree algorithm. During the training process, the model learns and optimizes based on the indicator features of the sample and the corresponding labels to minimize the prediction error. After the training is completed, the decision tree model can be used to calculate the importance of the indicator features. The feature importance indicates the contribution of the feature to the model prediction. By analyzing the structure of the decision tree model and the node splitting process, we can get the relative importance of each indicator feature and sort the indicator features according to the value of the feature importance. The indicator features with higher importance are placed in front, and the indicator features with lower importance are placed in the back, so as to select the core features that have a greater impact on the freshness of the food.

[0061] It should be noted that in XGBoost, "weight" refers to the number of times or weights of a feature used during model training. Specifically, "weight" indicates the number of times a feature is selected as a split node when building a decision tree. During XGBoost training, each sample has a weight that is used to adjust its importance in model training. These weights can be adjusted based on the distribution of samples to better handle sample imbalance. When calculating feature importance, XGBoost considers the number of times each feature is selected as a split node in the decision tree, as well as the sample weight corresponding to the feature. A higher "Weight" value indicates that the feature is frequently used in model training and is therefore considered a more important feature. By analyzing the "Weight" value of a feature, we can understand which features play a more important role in the model, which helps us understand the contribution and influence of the feature. This can be used for tasks such as feature selection, model interpretation, and optimization feature engineering.

[0062] In a specific embodiment of the present invention, the freshness recognition model is trained, specifically,

[0063] When the freshness recognition model cannot accurately identify the freshness of certain samples, these samples are handed over to manual labeling and identification. Manual labeling can be performed by professionals or domain experts, who can accurately judge the freshness of samples based on their experience and knowledge. When predicting and grading freshness, the prediction and grading standards are determined according to the different types of ingredients. Different ingredients have different characteristics and preservation requirements, so it is necessary to determine appropriate freshness prediction and grading standards based on the type of ingredients. For example, for meat ingredients, they can be graded based on indicators such as the color, smell, and elasticity of the meat; for fruits and vegetables, they can be graded based on changes in the color, texture, and freshness of the appearance.

[0064] In a specific embodiment of the present invention, manual labeling and identification is specifically,

[0065] When the freshness recognition model performs freshness recognition, when the prediction probabilities of freshness prediction and grading of different core features are all less than the first threshold, freshness prediction and grading recognition cannot be performed, and manual annotation is then used for recognition;

[0066] It should be noted that for the same ingredient, its freshness will inevitably deteriorate over time, and this process is irreversible. For example, the freshness of ingredients is divided into seven levels: a, ab, b, bc, c, cd, and d. If the freshness of the ingredient on the second day is b, then the freshness state of the ingredient on the third day cannot become a or ab, but can only be b or the freshness after b.

[0067] In the present invention, the first threshold is 0.7, and the freshness is predicted by using the classification model. If the model can distinguish the freshness of the food with certainty, the current freshness is directly determined. For example, if the model predicts that the probabilities of a, b, c, and d are [0.7, 0.15, 0.1, 0.05], the current freshness of the food is a. If the model cannot clearly distinguish and process, it is manually accessed. For example, if the model predicts that the probabilities of a, b, c, and d are [0.3, 0.27, 0.23, 0.2], the test is manually intervened and labeled.

[0068] It should be noted that the first threshold value can be different according to different prediction accuracies, and this patent does not impose any restrictions.

[0069] Furthermore, the reliability of the model prediction value is automatically determined by calculating the cross entropy between the model prediction value and the possible true value. The cross entropy formula is expressed as:

[0070] H(p,q)=-∑(plog(q))

[0071] For the 4-class classification, the probability distribution of possible true values ​​is expressed as: a: (1, 0, 0, 0), b: (0, 1, 0, 0), c: (0, 0, 1, 0), d: (0, 0, 0, 1);

[0072] Optimizing the above four probability distributions, we get: a:(0.92,0.05,0.02,0.01), b:(0.04,0.9,0.04,0.02), c:(0.02,0.04,0.9,0.04), and d:(0.01,0.02,0.05,0.92).

[0073] By taking the cross entropy of the predicted probability distribution and the probability distribution of the above four possible true values, the maximum value is taken. The smaller the value, the greater the reliability of the model's prediction. In the reliability judgment, the present invention takes 0.7 as the threshold. If the value is greater than 0.7, it means that the reliability of the model prediction is not enough, and it is necessary to introduce manual assistance to evaluate the freshness of the food.

[0074] Furthermore, the manually annotated data is combined with the correct data identified by the freshness recognition model to form a new training dataset. This dataset contains the accurate results of manual annotation and the correct results identified by the model. The freshness recognition model is retrained using this new training dataset to iterate a new and more accurate freshness recognition model.

[0075] In the second embodiment of the present invention, the present invention provides a device for identifying the freshness of food, such as Figure 2 As shown, it includes a data acquisition module, an indicator feature processing module and a recognition model training module.

[0076] Data acquisition module 1, used to acquire historical freshness data of different ingredients, and extract index features based on different storage times of different ingredients during freshness identification;

[0077] The indicator feature processing module 2 is used to predict the missing indicator features of the ingredients based on the ranking learning algorithm, and to screen the core features that affect the freshness of the ingredients based on the extracted indicator features;

[0078] The recognition model training module 3 is used to train the freshness recognition model according to the index characteristics affecting the freshness of the food and the sample data of the food.

[0079] In a third embodiment of the present invention, the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the method for identifying the freshness of food as described above are implemented.

[0080] In a fourth embodiment of the present invention, the present invention provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps in the above-mentioned method for identifying the freshness of food.

[0081] In summary, the method, device, electronic device and storage medium for identifying the freshness of food provided by the present invention obtain the index characteristics of the freshness of food by using the lambdaMart model implemented by XGboost, which solves the problem that the storage time of different food ingredients is incomparable and the problem that some index characteristics of food ingredients are missing; by using the feature importance Weight in XGboost to provide an evaluation standard for the importance of the index characteristics, the core features that affect the freshness of food ingredients are screened out; in the freshness identification model, the stability of the model iteration is ensured by adding an artificial standard method.

[0082] It should be understood that although this specification is described according to implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each implementation mode may also be appropriately combined to form other implementation modes that can be understood by those skilled in the art.

[0083] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the modules described above can refer to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0084] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present embodiment.

[0085] In addition, each functional module in each embodiment of the present application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above integrated modules can be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0086] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for a computer system (which can be a personal computer, a server, or a network system, etc.) or a processor to perform some steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), disk or optical disk and other media that can store program codes.

[0087] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned implementation modes, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned implementation modes, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various implementation modes of the present application.

Claims

1. A method for identifying the freshness of food, characterized in that: The following steps are involved: Obtain historical freshness data of different ingredients, and extract indicator features based on the different storage times of different ingredients when identifying freshness; Predict missing indicator features of ingredients based on ranking learning algorithm, and filter out the core features that affect the freshness of ingredients based on the extracted indicator features; The freshness recognition model is trained based on the core features that affect the freshness of the food and the sample data of the food.

2. The method for identifying the freshness of food according to claim 1, characterized in that: The historical freshness-keeping data includes the freshness-keeping degree of different food materials under different storage conditions and different storage times.

3. The method for identifying the freshness of food according to claim 2, characterized in that: The "indicator features for predicting missing ingredients based on ranking learning algorithm" include: Define a set of known indicator features x i and the corresponding sorted label y i ; For each pair of indicator features x i and x j , calculate the difference value to get d ij =x i -x j , and sort the labels y i and j Convert to comparison label to get c ij =sign(x i -x j ); The indicator feature x i and x j Input the neural network model and output x i and x j The ranking score difference of ij , then x i The sorting of j The probability before is expressed as:

4. The method for identifying the freshness of food according to claim 3, characterized in that: The "indicator features for predicting missing ingredients based on ranking learning algorithm" also include: By using the cross entropy loss function, we measure the x i Ranked x j The probability p before ij Compare with label c ij The difference is expressed as: L=-sum(c ij ×log(p ij )+(1-c ij )×log(1-p ij )) By minimizing the loss function L using the gradient descent optimization algorithm, the ranking relationship between indicator features is learned, and the missing indicator features are predicted based on the ranking relationship.

5. The method for identifying the freshness of food according to claim 4, characterized in that: The "screening of core features that affect the freshness of food ingredients" includes: The decision tree model is trained based on the gradient boosting tree algorithm to calculate the feature importance corresponding to the indicator features obtained through extraction and prediction; According to the value of the feature importance, the indicator features are sorted to screen out the core features that affect the freshness of the food.

6. The method for identifying the freshness of food according to claim 1, characterized in that: The "training freshness recognition model" includes: When the freshness recognition model is training for freshness recognition, samples that cannot be recognized for freshness are manually labeled for recognition; The freshness prediction grade is determined by the type of food.

7. The method for identifying the freshness of food according to claim 6, characterized in that: The manual annotation recognition includes: When the freshness recognition model performs freshness recognition, when the prediction probabilities of freshness prediction and grading of different core features are all less than the first threshold, freshness prediction and grading recognition cannot be performed, and manual annotation is then used for recognition; When manual annotation recognition is connected, the manually annotated data and the correct data recognized by the freshness recognition model are merged to form a new training data set, and the freshness recognition model is retrained using the data set to iterate a new freshness recognition model.

8. A device for identifying the freshness of food, comprising a data acquisition module, an indicator feature processing module and a recognition model training module, characterized in that: The data acquisition module is used to acquire historical freshness data of different ingredients and extract index features based on different storage times of different ingredients during freshness identification; The indicator feature processing module is used to predict the missing indicator features of the ingredients based on the ranking learning algorithm, and screen the core features that affect the freshness of the ingredients in combination with the extracted indicator features; The recognition model training module is used to train a freshness recognition model based on the indicator characteristics that affect the freshness of the food and the sample data of the food.

9. An electronic device, comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the method for identifying the freshness of food as described in any one of claims 1 to 7 are implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for identifying the freshness of food materials described in any one of claims 1 to 7 are implemented.

Images