Method and system for training model for evaluating freshness of object, method and system for evaluating freshness of object, and computer program product

Abstract

The invention relates to a method and a system for training a model for evaluating the freshness of an object. The method comprises the following steps of: respectively labeling a label for representing the freshness state of the object of a corresponding category on VOC (Volatile Organic Compound) detection data of the object of each category acquired by each semiconductor gas sensor; a category recognition model for determining a category of the object and each freshness recognition model for determining a freshness state of the object of the corresponding category are trained based at least on the tagged VOC detection data set. The present application also relates to a method and system for assessing freshness of an object, and a computer program product. According to the method and the device, through a two-stage training mode and an evaluation mode, each trained deep learning model can fully mine information contained in the VOC detection data, and is more focused on the sub-task to be executed, so that the accuracy of predicting the freshness of the object is effectively improved, and the accuracy of predicting the freshness of the object is improved. Therefore, the fresh-keeping adjusting function of the household appliance based on the object freshness is effectively improved.

Description

Technical Field

[0001] This application relates to the field of household appliances, and more particularly to a method for training a model for evaluating the freshness of objects, a system for training a model for evaluating the freshness of objects, a method for evaluating the freshness of objects, a system for evaluating the freshness of objects, and a computer program product for at least assisting in implementing the steps of the method according to this application. Background Technology

[0002] In current household appliances used for food preservation (such as refrigerators, crisper cases, and freezers), VOC (volatile organic compound) gas sensors constructed of metal oxides are typically used to detect the VOC gas composition in the storage space and determine the freshness of the objects placed in the storage space based on the VOC gas composition. However, considering that different types of objects release different types and concentrations of VOC gases as their freshness decreases, the freshness of objects determined directly based on the type and concentration of VOC gases deviates significantly from the actual freshness of the food. This not only affects the object freshness prediction function of household appliances but also the preservation regulation function based on object freshness.

[0003] Therefore, how to accurately and quickly assess the freshness of objects in household appliances has become a technical challenge that needs to be solved. Summary of the Invention

[0004] The purpose of this application is to provide a method for training a model for evaluating the freshness of an object, a system for training a model for evaluating the freshness of an object, a method for evaluating the freshness of an object, a system for evaluating the freshness of an object, and a computer program product to solve the problems in the prior art.

[0005] According to a first aspect of this application, a method for training a model for evaluating the freshness of objects is provided, the method comprising:

[0006] - Step S1: Label the VOC detection data of each category of object acquired by each semiconductor gas sensor with labels indicating the freshness status of the corresponding category of object; and

[0007] - Step S2: Train at least a category recognition model for determining the category of an object and a freshness recognition model for determining the freshness status of an object of the corresponding category, based on the labeled VOC detection dataset.

[0008] The core concept of the method for training a model for assessing object freshness according to this application is to divide the training process of the model into a first training phase for training a category recognition model to determine the category of an object, and a second training phase for training various freshness recognition models to determine the freshness status of objects of corresponding categories. Compared with existing techniques that use VOC detection data to train a single deep model, the two-stage training approach allows each trained deep learning model to fully mine the information contained in the VOC detection data and focus more on its sub-task, effectively improving the accuracy of predicting object freshness and thus effectively improving the freshness regulation function of household appliances based on object freshness.

[0009] According to an optional embodiment of this application, the category recognition model and / or the freshness recognition model can be configured as a deep learning model.

[0010] According to an optional embodiment of this application, step S1 may include:

[0011] - Step S101: Perform cyclic temperature modulation on multiple semiconductor gas sensors within a preset temperature range, and within the cyclic temperature modulation period, sample and acquire VOC detection data of each semiconductor gas sensor for each category of object, for example, using a sliding sampling window technique; and

[0012] - Step S102: Based on the temperature characteristics of each semiconductor gas sensor, label the VOC detection data obtained by the corresponding semiconductor gas sensor in each cyclic temperature modulation cycle with labels to indicate the freshness status of the corresponding category of objects.

[0013] By heating and maintaining each semiconductor gas sensor within different temperature ranges, detection data for different types of VOC gases can be acquired. Furthermore, the sliding sampling window technique can significantly reduce labor and time costs, effectively minimize random errors and human interference with sample data, and ensure a balanced distribution of the acquired VOC detection data samples.

[0014] According to another optional embodiment of this application, the method may further include:

[0015] - Step S11: The VOC detection data acquired by a semiconductor gas sensor in each cycle of temperature modulation are merged into a VOC detection data of a dimension belonging to this semiconductor gas sensor, and labels are assigned to the corresponding VOC detection data based on the labeled labels, thereby constructing a multi-dimensional and labeled VOC detection dataset.

[0016] By constructing a multi-dimensional VOC detection dataset, the richness of information contained in the VOC detection dataset can be improved, laying the foundation for training deep learning models to perform different recognition tasks, thereby effectively improving the recognition ability and prediction accuracy of deep learning models.

[0017] According to another optional embodiment of this application, the method may further include:

[0018] - Step S12: Normalize the VOC detection data of one dimension based on the maximum and minimum VOC detection data values ​​in the corresponding dimension.

[0019] In this way, all elements of the VOC detection data in each dimension are normalized to a value range of 0 to 1, thereby improving the processing efficiency of VOC detection data in each dimension.

[0020] According to another optional embodiment of this application, the method may further include:

[0021] - Step S13: Perform feature importance analysis on the VOC detection data of each dimension, and merge the VOC detection data of each dimension based on the analysis results of the feature importance analysis to reduce the dimensionality of the VOC detection dataset.

[0022] Optionally, based on the analysis results of the feature importance, weight factors belonging to the VOC detection data of the dimension with low feature importance can be assigned to the VOC detection data of the dimension respectively, and the product of the VOC detection data of the dimension with low feature importance and the weight factor belonging to the dimension can be merged into the VOC detection data of the dimension with high feature importance. This can reduce the dimensionality of the VOC detection data and effectively improve the efficiency of deep learning models in processing VOC detection data.

[0023] Optionally, in particular, the weight factor of VOC detection data in dimensions with feature importance below a preset threshold is set to zero, thereby deleting the VOC detection data of inaccurate dimensions as a whole, eliminating their potential negative impact on the detection results, and further improving the processing efficiency and prediction accuracy of the deep learning model.

[0024] According to another optional embodiment of this application, the plurality of semiconductor gas sensors may be constructed from the same or different semiconductor materials, wherein semiconductor gas sensors constructed from different semiconductor materials have different temperature characteristics, such that different semiconductor gas sensors exhibit different sensitivities to different types of VOC gases in different temperature ranges.

[0025] According to a second aspect of this application, a system is provided for training a model for evaluating the freshness of objects, the system being configured to perform a method according to this application for training a model for evaluating the freshness of objects, wherein the system may include the following components:

[0026] - Multiple semiconductor gas sensors are configured to acquire VOC detection data for each category of object;

[0027] - A data labeling module, configured to label VOC detection data of various categories of objects acquired through various semiconductor gas sensors with tags indicating the freshness status of the corresponding category of objects; and

[0028] - The model training module is configured to train, at least based on an annotated VOC detection dataset, a category recognition model for determining the category of an object and a freshness recognition model for determining the freshness status of an object of the corresponding category.

[0029] According to another optional embodiment of this application, the system may further include a data construction module configured to merge VOC detection data acquired by a semiconductor gas sensor in each cyclic temperature modulation cycle into a VOC detection data of a dimension belonging to the semiconductor gas sensor, and assign labels to the corresponding VOC detection data of the dimension based on the labeled labels, thereby constructing a multi-dimensional and labeled VOC detection dataset, laying the foundation for deep learning models to be trained using VOC detection data.

[0030] According to a third aspect of this application, a method for assessing the freshness of an object is provided, the method comprising:

[0031] - Step S1': Determine the category of the object to be identified based on VOC detection data of the object acquired through a semiconductor gas sensor using a category recognition model; and

[0032] - Step S2': Evaluate the freshness status of the object based on the VOC detection dataset of the object to be identified by using a freshness recognition model that is assigned to the category of the determined object.

[0033] The core concept of the method for assessing the freshness of objects according to this application lies in dividing the assessment process into two stages. In the first stage, a category recognition model is used to determine the category of the object to be identified. In the second stage, a freshness recognition model for the corresponding category is used to assess the freshness status of the object. Compared with existing technologies that directly identify the freshness of objects based on VOC detection data, this two-stage assessment method allows each deep learning model to fully mine the information contained in the VOC detection data and focus more on its sub-task, effectively improving the accuracy of predicting the freshness of each category of objects, and thus effectively improving the freshness regulation function of household appliances based on object freshness.

[0034] According to another optional embodiment of this application, the category recognition model and / or the freshness recognition model can be configured as a deep learning model.

[0035] According to another optional embodiment of this application, the method may further include:

[0036] - Step S11': Combine the VOC detection data obtained by one of the multiple semiconductor gas sensors into a single VOC detection data of a dimension belonging to that semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset.

[0037] By constructing a multi-dimensional VOC detection dataset, the richness of information contained in the VOC detection dataset can be improved, laying the foundation for the execution of subtasks of various deep learning models, thereby effectively improving the accuracy of predicting object freshness.

[0038] According to another optional embodiment of this application, step S11' may include:

[0039] -Step S111': Perform cyclic temperature modulation on multiple semiconductor gas sensors within a preset temperature range, and sample and acquire VOC detection data of each semiconductor gas sensor regarding the object to be identified within the cyclic temperature modulation period, for example, using a sliding sampling window technique; and

[0040] - Step S112': The VOC detection data acquired by a semiconductor gas sensor in each cycle of temperature modulation will be merged into a VOC detection data of a dimension belonging to this semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset.

[0041] By heating and maintaining each semiconductor gas sensor within different temperature ranges, detection data for different types of VOC gases can be acquired. Furthermore, the sliding sampling window technique can significantly reduce labor and time costs, effectively minimize random errors and human interference with sample data, and ensure a balanced distribution of the acquired VOC detection data samples.

[0042] According to another optional embodiment of this application, the method may further include:

[0043] - Step S12': Normalize the VOC detection data of one dimension based on the maximum and minimum VOC detection data values ​​in the corresponding dimension.

[0044] In this way, all elements of the VOC detection data in each dimension are normalized to a value range of 0 to 1, thereby improving the processing efficiency of VOC detection data in each dimension.

[0045] According to another optional embodiment of this application, the method may further include:

[0046] - Step S13': Perform feature importance analysis on the VOC detection data of each dimension, and merge the VOC detection data of each dimension based on the analysis results of the feature importance analysis to reduce the dimensionality of the VOC detection dataset.

[0047] Optionally, based on the analysis results of the feature importance, weight factors belonging to the VOC detection data of the dimension with low feature importance can be assigned to the VOC detection data of the dimension respectively, and the product of the VOC detection data of the dimension with low feature importance and the weight factor belonging to the dimension can be merged into the VOC detection data of the dimension with high feature importance. This can reduce the dimensionality of the VOC detection data and effectively improve the efficiency of deep learning models in processing VOC detection data.

[0048] Optionally, in particular, the weight factor of VOC detection data in dimensions with feature importance below a preset threshold is set to zero, thereby deleting the VOC detection data of inaccurate dimensions as a whole, eliminating their potential negative impact on the detection results, and further improving the processing efficiency and prediction accuracy of the deep learning model.

[0049] According to another optional embodiment of this application, the plurality of semiconductor gas sensors may be constructed from the same or different semiconductor materials, wherein semiconductor gas sensors constructed from different semiconductor materials have different temperature characteristics, such that different semiconductor gas sensors exhibit different sensitivities to different types of VOC gases in different temperature ranges.

[0050] According to another optional embodiment of this application, the method may further include:

[0051] -Step S3': Output notification information about the freshness status of the evaluated object.

[0052] The output of the notification information allows users to promptly understand the freshness status of items in the storage space of household appliances.

[0053] According to a fourth aspect of this application, a system for assessing the freshness of an object is provided, the system being used to perform the method for assessing the freshness of an object according to this application, wherein the system may include the following components:

[0054] - Semiconductor gas sensors, configured to acquire VOC detection data about the object to be identified;

[0055] - A category recognition model configured to determine the category of the object to be identified based on VOC detection data about the object acquired through various semiconductor gas sensors; and

[0056] - A freshness recognition model that evaluates the freshness status of an object based on a VOC detection dataset of the object to be identified, by assigning the freshness recognition model to the category of the identified object.

[0057] According to another optional embodiment of this application, the system may further include a data construction module configured to merge VOC detection data acquired by a semiconductor gas sensor into VOC detection data of a dimension belonging to that semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset. By constructing a multi-dimensional VOC detection dataset, the richness of information contained in the VOC detection dataset can be improved, enabling deep learning models performing different recognition tasks to extract more effective information from the multi-dimensional VOC detection dataset, thereby effectively improving the recognition ability and prediction accuracy of the deep learning model.

[0058] According to another optional embodiment of this application, the system may further include a notification module configured to output notification information about the freshness status of the assessed object, enabling the user to be informed in a timely manner about the freshness status of the object in the storage space of the household appliance.

[0059] According to a fifth aspect of this application, a computer program product, such as a computer-readable program carrier, is provided, comprising computer program instructions that, when executed by a processor, at least assist in implementing the steps of the method for evaluating the freshness of an object according to this application and the steps of the method for training a model for evaluating the freshness of an object. Attached Figure Description

[0060] The principles, features, and advantages of this application will be better understood below with reference to the accompanying drawings. The drawings include:

[0061] Figure 1 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to an exemplary embodiment of this application, is shown.

[0062] Figure 2 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to another exemplary embodiment of this application, is shown.

[0063] Figure 3 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to another exemplary embodiment of this application, is shown.

[0064] Figure 4 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to another exemplary embodiment of this application, is shown.

[0065] Figure 5 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to another exemplary embodiment of this application, is shown.

[0066] Figure 6 A block diagram of a system for training a model for evaluating the freshness of objects, according to an exemplary embodiment of this application, is shown.

[0067] Figure 7 A flowchart illustrating a method for assessing the freshness of an object according to an exemplary embodiment of this application is shown.

[0068] Figure 8 A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown;

[0069] Figure 9 A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown;

[0070] Figure 10 A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown;

[0071] Figure 11 A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown;

[0072] Figure 12A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown; and

[0073] Figure 13 A block diagram of a system for evaluating the freshness of an object according to an exemplary embodiment of this application is shown. Detailed Implementation

[0074] To make the technical problems to be solved, the technical solutions, and the beneficial technical effects of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and several exemplary embodiments. It should be understood that the specific embodiments described herein are only for explaining this application and are not intended to limit the scope of protection of this application.

[0075] Figure 1 A flowchart illustrating a method for training a model for evaluating the freshness of objects, according to an exemplary embodiment of this application, is shown. The following exemplary embodiments describe the method according to this application in more detail.

[0076] like Figure 1 As shown, the method may include steps S1 and S2. In step S1, the VOC detection data of each category of object acquired by each semiconductor gas sensor are labeled with tags indicating the freshness status of the corresponding category of object.

[0077] Household appliances used for food preservation (such as refrigerators, freezers, and crisper cases) typically incorporate multiple semiconductor gas sensors configured to acquire VOC detection data for the object to be identified. These multiple semiconductor gas sensors can be constructed from the same semiconductor material, such as zinc oxide gas sensors, which offer excellent gas sensitivity and wide applicability. Due to the temperature characteristics of zinc oxide gas sensors, they exhibit varying sensitivities to different types of VOC gases within different temperature ranges. Therefore, detection data for specific VOC gases can be acquired by heating and maintaining each zinc oxide gas sensor within different temperature ranges.

[0078] Optionally, the plurality of semiconductor gas sensors may also be constructed from different semiconductor materials. Semiconductor gas sensors constructed from different semiconductor materials have different temperature characteristics. That is, different semiconductor gas sensors exhibit different sensitivities to different types of VOC gases in different temperature ranges. For example, a semiconductor gas sensor constructed from Co-doped SnO2 nanosensitive material exhibits high sensitivity to formaldehyde and acetone in a preset temperature range.

[0079] Figure 6The system 1 shown for training a model to assess the freshness of objects includes, for example, a first semiconductor gas sensor 111, a second semiconductor gas sensor 112, a third semiconductor gas sensor 113, and a fourth semiconductor gas sensor 114. These semiconductor gas sensors exhibit different sensitivities to different types of gases within different temperature ranges, thereby acquiring VOC detection data of various types and / or concentrations of VOC gases released by the object to be detected. Considering that deep learning models require labeled VOC detection data for training, and that different categories of objects release different types and concentrations of VOC gases as freshness decreases, the labeling of VOC detection data must include the category and freshness of the object to be detected.

[0080] The following combination Figure 2 The flowchart illustrating step S1 of a method for training a model for evaluating object freshness according to another exemplary embodiment of this application is shown in detail. Figure 2 As shown, step S1 may include steps S101 and S102. In step S101, cyclic temperature modulation is applied to multiple semiconductor gas sensors within a preset temperature range, and VOC detection data of each semiconductor gas sensor for each category of object is sampled and acquired within the cyclic temperature modulation period.

[0081] In the test scenario defined in this application, different types of objects, such as strawberries, lychees, and bayberries, can be placed in the storage space of a household appliance. As their freshness decreases, they release different types and concentrations of VOC gases. Within a cyclic temperature modulation period, for example, from 100°C to 120°C, a modulated waveform—for example, with a modulation period T—can be applied to the semiconductor gas sensor. t Maximum current value I min Maximum current value I max A rectangular wave current with duty cycle D is applied to raise the temperature of a semiconductor gas sensor and maintain it within a first preset temperature range. This semiconductor gas sensor is highly sensitive to, for example, a first VOC gas within this first preset temperature range, thereby acquiring the VOC timing response for the first VOC gas. The VOC timing response is sampled at a fixed sampling frequency to obtain VOC timing response data for the first VOC gas. Next, a current with a modulated waveform is applied to further raise the temperature of the semiconductor gas sensor and maintain it within a second preset temperature range, thereby acquiring VOC timing response data for a second VOC gas. This process is repeated until the temperature of the semiconductor gas sensor traverses the entire 100°C to 120°C temperature modulation cycle. Then, the temperature of the semiconductor gas sensor is lowered to 100°C, and the next temperature modulation cycle is executed.

[0082] Here, each semiconductor gas sensor can be heated and maintained within a different preset temperature range to acquire VOC time-series response data for as many types of VOC gases as possible. Alternatively, multiple semiconductor gas sensors can be heated and maintained within the same preset temperature range, and VOC time-series response data for certain types of VOC gases can be repeatedly detected to achieve the function of predicting the freshness of objects and to verify the reliability of VOC gases of types that are important to the freshness of the food, such as VOC gases produced when food spoils, VOC gases produced by mold in food, etc.

[0083] For example, a sliding sampling window can be set on the time axis of the VOC timing response data, with a sampling window width T. c The value is equal to an integer multiple of the modulation period Tt of the applied waveform, and the sliding step size is one sampling data point. The sampling window slides along the time axis starting from the first sampling data point of the VOC time-series response data. Each time the sliding step size is increased, all the sampling data within the sampling window is extracted into a set of VOC time-series response sample data. This allows multiple sets of VOC time-series response sample data to be obtained for each type of VOC gas. By continuously resampling the VOC time-series response data point by point, the VOC detection data of each semiconductor gas sensor can be obtained. This can significantly reduce manpower and time costs, effectively reduce the interference of random errors and human factors on the sample data, and the obtained VOC detection data sample distribution is balanced.

[0084] In step S102, based on the temperature characteristics of each semiconductor gas sensor, the VOC detection data acquired by the corresponding semiconductor gas sensor within each cyclic temperature modulation cycle are labeled with tags indicating the freshness status of the corresponding category of object. In test scenarios where different categories of objects are placed in the storage space, the VOC detection data acquired in the corresponding test scenarios can be labeled with the category of the object to be detected, and the freshness of the object to be detected can be labeled as the freshness of the corresponding category of object decreases.

[0085] In step S2, at least a category recognition model 22 for determining the category of an object and a freshness recognition model for determining the freshness status of an object of the corresponding category are trained based on the labeled VOC detection dataset.

[0086] Considering that different types of objects release different types and concentrations of VOC gases as their freshness decreases, the freshness of an object determined directly by a single deep learning model based on the type and concentration of VOC gases deviates significantly from the actual freshness of the food. Therefore, the training process of the model used to evaluate the freshness of an object is divided into a first training stage for training a category recognition model 22 to determine the category of the object, and a second training stage for training each freshness recognition model to determine the freshness status of the corresponding category of object. This allows each trained deep learning model to fully explore the information contained in the VOC detection data and focus more on the sub-task it is to perform, effectively improving the prediction accuracy of the freshness status of each category of object.

[0087] For example, a freshness recognition model can be set up for each category of the object to be detected to determine the freshness status of the corresponding category of object. For example, a first freshness recognition model 231 can be set up to determine the freshness status of strawberries, a second freshness recognition model 232 can be set up to determine the freshness status of lychees, a third freshness recognition model 233 can be set up to determine the freshness status of bayberries, and so on. The category recognition model 22 and / or the freshness recognition models 231, 232, and 233 can be configured as deep learning models, such as convolutional neural networks, multilayer autoencoder neural networks, and / or deep belief networks, or they can be configured as other types of mathematical models that can achieve the same function.

[0088] In existing technologies, the time-series response sample data directly acquired by semiconductor gas sensors are all one-dimensional data. Deep learning models can extract limited effective information and have low prediction accuracy from one-dimensional time-series response sample data, which greatly limits the application of VOC gas sensors in complex scenarios of household appliances. The following combines... Figure 3 The flowchart illustrating a method for training a model for evaluating object freshness according to another exemplary embodiment of this application details the process of constructing a multi-dimensional VOC detection dataset. The following only describes the process related to... Figure 1 The differences between the embodiments shown are omitted for brevity, and the same steps will not be repeated.

[0089] like Figure 3As shown, the method may further include step S11. In step S11, the VOC detection data acquired by a semiconductor gas sensor in each cyclic temperature modulation cycle are merged into a VOC detection data of a dimension belonging to that semiconductor gas sensor, and labels are assigned to the corresponding dimension of VOC detection data based on the labeled labels, thereby constructing a multi-dimensional and labeled VOC detection dataset. In the context of this application, the term "merge" can be understood as follows: A row of data elements is constructed from the VOC detection data acquired by a semiconductor gas sensor in one cyclic temperature modulation cycle, and the VOC detection data of the semiconductor gas sensor is processed in this way for each cyclic temperature modulation cycle to construct corresponding rows of data elements. Since the number of VOC detection data acquired in each cyclic temperature modulation cycle is equal, that is, the number of data elements in each row is equal, a matrix with the number of rows equal to the number of cyclic temperature modulation cycles can be constructed from these rows of data elements. This matrix represents the VOC detection data of a dimension belonging to that semiconductor gas sensor. Here, VOC detection data of the same dimension can be processed uniformly, including label assignment, normalization, feature importance analysis, and / or weight factor allocation. For each semiconductor gas sensor, VOC detection data of the corresponding dimension is constructed in this way, resulting in a multi-dimensional VOC detection dataset. For example, the first dimension of VOC detection data is constructed using VOC detection data acquired by the first semiconductor gas sensor 111 over multiple temperature modulation cycles; the second dimension uses VOC detection data acquired by the second semiconductor gas sensor 112 over multiple temperature modulation cycles; the third dimension uses VOC detection data acquired by the third semiconductor gas sensor 113 over multiple temperature modulation cycles; the fourth dimension uses VOC detection data acquired by the fourth semiconductor gas sensor 114 over multiple temperature modulation cycles, and so on. Considering that the number of temperature modulation cycles for VOC gas detection is equal for each semiconductor gas sensor, the amount of VOC detection data in each dimension is also equal.

[0090] In the process of merging VOC detection data from different dimensions, labels can be assigned to the VOC detection data of the corresponding dimensions based on the labels of the original one-dimensional VOC detection data, so that each dimension of VOC detection data obtained by merging has a corresponding label, thereby constructing a multi-dimensional and labeled VOC detection dataset, wherein the labels include the category and freshness of the object to be detected.

[0091] To improve the processing efficiency of VOC detection data from various dimensions, such as Figure 4The flowchart shown is a process diagram of a method for training a model for evaluating the freshness of an object according to another exemplary embodiment of this application. The method may further include step S12. In step S12, the VOC detection data of one dimension are normalized based on the maximum and minimum VOC detection data values ​​in the corresponding dimension, such that all elements of the VOC detection data in each dimension are normalized to a value range of 0 to 1.

[0092] With a large number of semiconductor gas sensors in System 1, the acquired VOC detection data also has many dimensions. Excessive dimensionality of VOC detection data may affect the processing efficiency of deep learning models, and VOC detection data acquired through semiconductor gas sensors with lower detection accuracy will affect the prediction accuracy of deep learning models. Therefore, in some application scenarios, there is a need to reduce the dimensionality of VOC detection data. The following combines... Figure 5 The flowchart shown below, illustrating a method for training a model for evaluating object freshness according to another exemplary embodiment of this application, details the dimensionality reduction process of VOC detection data.

[0093] like Figure 5 As shown, the method may further include step S13. In step S13, feature importance analysis is performed on the VOC detection data of each dimension, and the VOC detection data of each dimension are merged based on the analysis results of the feature importance analysis, so as to reduce the dimensionality of the VOC detection dataset.

[0094] Here, feature importance analysis can be performed on VOC detection data, such as VOC gas type, concentration, and detection accuracy. VOC detection data with a greater variety of VOC gas types (especially those important for predicting food freshness), higher concentrations, and / or higher detection accuracy have higher feature importance, while VOC detection data with fewer VOC gas types, lower concentrations, and / or lower detection accuracy have lower feature importance. Based on the analysis results, weighting factors are assigned to VOC detection data in dimensions with low feature importance, and the product of the weighting factors for each dimension is merged into the VOC detection data in dimensions with high feature importance.

[0095] For example, if the labels of the fourth-dimensional VOC detection data contain fewer types of VOC gases, lower concentrations, or lower detection accuracy, the fourth-dimensional VOC detection data can be merged into the VOC detection data of other dimensions, thereby reducing the dimensionality of the VOC detection data. Especially when the fourth-dimensional VOC detection data reveals that the VOC detection data of the fourth semiconductor gas sensor 114 significantly deviates from the normal detection concentration range, it can be determined that the feature importance of the fourth-dimensional VOC detection data is below a preset threshold. The weight factor of the VOC detection data in dimensions with feature importance below the preset threshold is then set to zero. This effectively removes the inaccurate VOC detection data from these dimensions as a whole, eliminating its potential negative impact on the detection results, thereby improving the processing efficiency and prediction accuracy of the deep learning model.

[0096] According to the current embodiment of this application, the training process of the model used to evaluate the freshness of objects is divided into a first training stage for training a category recognition model to determine the category of an object and a second training stage for training each freshness recognition model to determine the freshness status of objects of the corresponding category. Compared with the prior art of training a single deep model using VOC detection data, the two-stage training method enables each trained deep learning model to fully mine the information contained in the VOC detection data and focus more on the sub-task it is to perform, effectively improving the accuracy of predicting the freshness of objects, and thus effectively improving the freshness regulation function of household appliances based on the freshness of objects.

[0097] Figure 6 A block diagram of a system 1 for training a model for evaluating the freshness of objects, according to an exemplary embodiment of this application, is shown.

[0098] like Figure 6 As shown, the system 1 may include the following components:

[0099] - Multiple semiconductor gas sensors, configured to acquire VOC detection data for various categories of objects, including, for example, a first semiconductor gas sensor 111, a second semiconductor gas sensor 112, a third semiconductor gas sensor 113 and / or a fourth semiconductor gas sensor 114, etc.

[0100] - Data labeling module 12, configured to label VOC detection data of various categories of objects acquired by various semiconductor gas sensors with labels indicating the freshness status of the corresponding category of objects; and

[0101] - Model training module 13 is configured to train, at least based on an annotated VOC detection dataset, a category recognition model 22 for determining the category of an object and various freshness recognition models for determining the freshness status of objects of the corresponding category, wherein the freshness recognition models include, for example, a first freshness recognition model 231 for determining the freshness status of strawberries, a second freshness recognition model 232 for determining the freshness status of lychees and / or a third freshness recognition model 233 for determining the freshness status of bayberries, etc.

[0102] Optionally, the system 1 may further include a data construction module 14, which is configured to merge the VOC detection data acquired by a semiconductor gas sensor in each cyclic temperature modulation cycle into a VOC detection data of a dimension belonging to the semiconductor gas sensor, and assign labels to the corresponding VOC detection data of the dimension based on the labeled labels, thereby constructing a multi-dimensional and labeled VOC detection dataset.

[0103] After the deep learning models—including the category recognition model and the freshness recognition model for objects belonging to each category—are trained using the methods described above, they can be used as follows: Figure 7 The method for assessing the freshness of objects, according to an exemplary embodiment of this application, evaluates the freshness status of various categories of objects in the storage space of household appliances.

[0104] like Figure 7 As shown, the method may include steps S1' and S2'. In step S1', the category of the object to be identified is determined by the category recognition model 22 based on the VOC detection data of the object to be identified obtained by the semiconductor gas sensor. For example, the object in the storage space of a household appliance is determined to include strawberries, lychees and / or bayberries.

[0105] Considering that different types of objects release different types and concentrations of VOC gases as their freshness decreases, in step S2', the freshness status of the object can be evaluated based on the VOC detection dataset for the object to be identified using a freshness recognition model belonging to the determined object category. For example, the first freshness recognition model 231 is used to evaluate the freshness status of strawberries; the second freshness recognition model 232 is used to determine the freshness status of lychees; the third freshness recognition model 233 is used to determine the freshness status of bayberries, and so on.

[0106] According to the current embodiment of this application, the evaluation process of object freshness is divided into two evaluation stages. In the first evaluation stage, the category of the object to be identified is determined by the category recognition model 22. In the second evaluation stage, the freshness recognition model of the corresponding category of object is used to evaluate the freshness status of the object. Compared with the prior art of directly identifying object freshness based on VOC detection data, the two-stage evaluation method allows each deep learning model to fully explore the information contained in the VOC detection data and focus more on the sub-tasks it needs to perform, effectively improving the accuracy of predicting the freshness of each category of object, and thus effectively improving the freshness regulation function of household appliances based on object freshness.

[0107] In existing technologies, the time-series response sample data directly acquired by semiconductor gas sensors are all one-dimensional data. Deep learning models can extract limited effective information and have low prediction accuracy from one-dimensional time-series response sample data, which greatly limits the application of VOC gas sensors in complex scenarios of household appliances. The following combines... Figure 8 The flowchart illustrating another exemplary embodiment of the method for evaluating object freshness according to this application details the process of constructing a multi-dimensional VOC detection dataset. The following only describes the process related to... Figure 7 The differences between the embodiments shown are omitted for brevity, and the same steps will not be repeated.

[0108] like Figure 8 As shown, the method may further include step S11'. In step S11', the VOC detection data acquired by one of the multiple semiconductor gas sensors are merged into a single VOC detection data dimension belonging to that semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset. The construction method of the multi-dimensional VOC detection dataset is the same as that in the model training phase, as described below. Figure 9 The flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application details step S11'.

[0109] like Figure 9As shown, step S11' may include steps S111' and S112'. In step S111', cyclic temperature modulation is applied to multiple semiconductor gas sensors within a preset temperature range, and VOC detection data of each semiconductor gas sensor regarding the object to be identified is sampled within the cyclic temperature modulation cycle. As explained above, similarly, currents with modulated waveforms can be applied to semiconductor gas sensors 211, 212, 213, and 214 within a cyclic temperature modulation cycle of, for example, 100°C to 120°C, to increase the temperature of the corresponding semiconductor gas sensors and maintain it within a first preset temperature range. Within this first preset temperature range, the semiconductor gas sensors are, for example, highly sensitive to a first VOC gas, thereby obtaining the VOC timing response to the first VOC gas. The VOC timing response is sampled at a fixed sampling frequency to obtain VOC timing response data for the first VOC gas. Next, the temperature of the semiconductor gas sensor is further increased by applying a current with a modulated waveform and maintained in the second preset temperature range, thereby obtaining VOC timing response data for the second VOC gas. This process is repeated until the temperature of the semiconductor gas sensor traverses the entire 100°C to 120°C cyclic temperature modulation cycle. Then, the temperature of the semiconductor gas sensor is reduced to 100°C and the next cyclic temperature modulation cycle is repeated.

[0110] Here, each semiconductor gas sensor can be heated and maintained within a different preset temperature range to acquire VOC time-series response data for as many types of VOC gases as possible. Alternatively, multiple semiconductor gas sensors can be heated and maintained within the same preset temperature range, and VOC time-series response data for certain types of VOC gases can be repeatedly detected to achieve the function of predicting the freshness of objects and to verify the reliability of VOC gases of types that are important to the freshness of the food, such as VOC gases produced when food spoils, VOC gases produced by mold in food, etc.

[0111] For example, a sliding sampling window can be set on the time axis of the VOC timing response data, with a sampling window width T. cThe value is equal to an integer multiple of the modulation period Tt of the applied waveform, and the sliding step size is one sampling data point. The sampling window slides along the time axis starting from the first sampling data point of the VOC time-series response data. Each time the sliding step size is increased, all the sampling data within the sampling window is extracted into a set of VOC time-series response sample data. This allows multiple sets of VOC time-series response sample data to be obtained for each type of VOC gas. By continuously resampling the VOC time-series response data point by point, the VOC detection data of each semiconductor gas sensor can be obtained. This can significantly reduce manpower and time costs, effectively reduce the interference of random errors and human factors on the sample data, and the obtained VOC detection data sample distribution is balanced.

[0112] In step S112', the VOC detection data acquired by a semiconductor gas sensor in each of the various temperature modulation cycles are merged into a single dimension of VOC detection data belonging to that semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset. For example, the VOC detection data acquired by the first semiconductor gas sensor 211 in multiple temperature modulation cycles are merged into the first-dimensional VOC detection data; the VOC detection data acquired by the second semiconductor gas sensor 212 in multiple temperature modulation cycles are merged into the second-dimensional VOC detection data; the VOC detection data acquired by the third semiconductor gas sensor 213 in multiple temperature modulation cycles are merged into the third-dimensional VOC detection data; the VOC detection data acquired by the fourth semiconductor gas sensor 214 in multiple temperature modulation cycles are merged into the fourth-dimensional VOC detection data, and so on. Considering that the number of temperature modulation cycles for VOC gas detection performed by each semiconductor gas sensor is equal, the amount of VOC detection data in each dimension is also equal.

[0113] It should be noted that current deep learning models have low accuracy in assessing the freshness of objects based on one-dimensional VOC detection data. According to experimental data, the average accuracy of deep learning sub-models trained on one-dimensional VOC detection data in assessing the freshness of various types of food is only 26%. For example, the accuracy in assessing the freshness of strawberries is 63%, the accuracy in assessing the freshness of lychees is 46%, and the accuracy in assessing the freshness of lychees is 42%. In contrast, the deep learning model trained according to this application has an average accuracy of 85% in assessing the freshness of various types of food based on a multi-dimensional VOC detection dataset. For example, the accuracy of the first freshness recognition model 231 trained on a multi-dimensional VOC detection dataset in assessing the freshness of strawberries is 86%, the accuracy of the second freshness recognition model 232 trained on a multi-dimensional VOC detection dataset in assessing the freshness of lychees is 85%, and the accuracy of the third freshness recognition model 233 trained on a multi-dimensional VOC detection dataset in assessing the freshness of lychees is 75%, and so on.

[0114] According to embodiments of this application, by constructing a multi-dimensional VOC detection dataset, the richness of information contained in the VOC detection dataset can be improved, laying the foundation for the training of deep learning models used to evaluate the freshness state of objects and the execution of related tasks, and further effectively improving the prediction accuracy of deep learning models for the freshness state of objects.

[0115] To improve the processing efficiency of VOC detection data from various dimensions, such as Figure 10 The flowchart shown is a process diagram of a method for assessing the freshness of an object according to another exemplary embodiment of this application. The method may further include step S12'. In step S12', the VOC detection data of one dimension are normalized based on the maximum and minimum VOC detection data values ​​in the corresponding dimension, such that all elements of the VOC detection data of each dimension are normalized to a value range of 0 to 1.

[0116] With a large number of semiconductor gas sensors in System 2, the acquired VOC detection data also has many dimensions. Excessive dimensionality of VOC detection data may affect the processing efficiency of deep learning models, and VOC detection data acquired through semiconductor gas sensors with lower detection accuracy will affect the prediction accuracy of deep learning models. Therefore, in some application scenarios, there is a need to reduce the dimensionality of VOC detection data. The following combines... Figure 11 The flowchart shown below illustrates the process of dimensionality reduction of VOC detection data in detail, according to another exemplary embodiment of this application, of a method for assessing the freshness of an object.

[0117] like Figure 11 As shown, the method may further include step S13'. In step S13', feature importance analysis is performed on the VOC detection data for each dimension, and the VOC detection data for each dimension is merged based on the analysis results to reduce the dimensionality of the VOC detection dataset. Here, feature importance analysis can be performed on VOC detection data, for example, regarding the type, concentration, and detection accuracy of VOC gases. VOC detection data with a greater variety of VOC types (especially those important for predicting object freshness), higher concentrations, and / or higher detection accuracy have higher feature importance, while VOC detection data with fewer VOC types, lower concentrations, and / or lower detection accuracy have lower feature importance. Based on the analysis results, weight factors are assigned to the VOC detection data for dimensions with low feature importance, and the product of the VOC detection data for dimensions with low feature importance and the weight factors assigned to those dimensions is merged into the VOC detection data for dimensions with high feature importance.

[0118] For example, if the labels for the fourth-dimensional VOC detection data contain fewer types of VOC gases, lower concentrations, or lower detection accuracy, the fourth-dimensional VOC detection data can be merged into the VOC detection data of other dimensions, thereby reducing the dimensionality of the VOC detection data. Especially when the fourth-dimensional VOC detection data reveals that the VOC detection data of the fourth semiconductor gas sensor 214 significantly deviates from the normal detection concentration range, it can be determined that the feature importance of the fourth-dimensional VOC detection data is below a preset threshold. The weight factor of the VOC detection data in dimensions with feature importance below the preset threshold is then set to zero. This effectively removes the inaccurate VOC detection data from these dimensions as a whole, eliminating its potential negative impact on the detection results, thereby improving the processing efficiency and prediction accuracy of the deep learning model.

[0119] Figure 12 A flowchart illustrating a method for assessing the freshness of an object according to another exemplary embodiment of this application is shown. The following only describes the method in relation to... Figure 11 The differences between the embodiments shown are omitted for brevity, and the same steps will not be repeated.

[0120] like Figure 12 As shown, the method may further include step S3'. In step S3', notification information about the freshness status of the evaluated object is output, for example, through notification module 25. The notification module 25 is configured, for example, to display the notification information to the user via a display screen and / or voice device, so that the user can promptly understand the freshness status of objects in the storage space of household appliances.

[0121] In addition, it should be noted that the step numbers described herein do not necessarily represent the order of steps, but are merely a reference numeral. The order may be changed depending on the specific circumstances, as long as the technical objective of this application can be achieved.

[0122] Figure 13 A block diagram of a system 2 for evaluating the freshness of an object according to an exemplary embodiment of this application is shown.

[0123] like Figure 13 As shown, the system 2 may include the following components:

[0124] - Semiconductor gas sensors, configured to acquire VOC detection data about the object to be identified, including, for example, a first semiconductor gas sensor 211, a second semiconductor gas sensor 212, a third semiconductor gas sensor 213 and / or a fourth semiconductor gas sensor 214, etc.

[0125] - Category recognition model 22, which is configured to determine the category of the object to be identified based on VOC detection data about the object acquired through various semiconductor gas sensors; and

[0126] - A freshness recognition model, which evaluates the freshness status of an object based on a VOC detection dataset about the object to be identified, by means of a freshness recognition model belonging to the category of the determined object. This includes, for example, a first freshness recognition model 231 for determining the freshness status of strawberries and a second freshness recognition model 232 for determining the freshness status of lychees.

[0127] And / or a third freshness identification model for determining the freshness status of bayberries, etc.

[0128] Optionally, the system 2 may further include a data construction module 24, which is configured to merge VOC detection data acquired by a semiconductor gas sensor into VOC detection data of a dimension belonging to the semiconductor gas sensor, thereby constructing a multi-dimensional VOC detection dataset.

[0129] Optionally, the system 2 may also include a notification module 25, which is configured to output notification information about the freshness status of the evaluated object.

[0130] It should be understood that the terms “first,” “second,” “third,” etc., used in this document are for descriptive purposes only and should not be construed as indicating or implying relative importance, nor should they be construed as implicitly specifying the number of technical features indicated.

[0131] Although specific embodiments have been described above, these embodiments are not intended to limit the scope of this application, even when only a single embodiment is described with respect to a particular feature. The feature examples provided in this application are intended for illustrative purposes and not for limitation, unless otherwise stated. In practice, multiple features may be combined with each other as needed and where technically feasible. Various substitutions, modifications, and alterations are also conceived without departing from the spirit and scope of this application.

Claims

1. A method for training a model for evaluating freshness of an object, the method comprising: Step S1: labeling VOC detection data about objects of respective categories acquired by respective semiconductor gas sensors (111, 112, 113, 114) with labels respectively indicative of freshness states of the objects of the respective categories; and Step S2: training a category recognition model (22) for determining a category of an object and respective freshness recognition models (231, 232, 233) for determining freshness states of the objects of the respective categories based on at least the labeled VOC detection data sets. The step S1 comprises:

2. The method of claim 1, wherein, Step S101: cyclically temperature modulating the plurality of semiconductor gas sensors (111, 112, 113, 114) at a preset temperature interval, and sampling VOC detection data about objects of respective categories acquired by respective semiconductor gas sensors (111, 112, 113, 114) within the cyclic temperature modulation period, for example, by a sliding sampling window technique; and Step S102: labeling the VOC detection data acquired by respective semiconductor gas sensors (111, 112, 113, 114) within respective cyclic temperature modulation periods with labels respectively indicative of freshness states of the objects of the respective categories based on temperature characteristics of the respective semiconductor gas sensors (111, 112, 113, 114). The method further comprises:

3. The method of claim 2, wherein, Step S11: merging the VOC detection data acquired by one semiconductor gas sensor (111, 112, 113, 114) within respective cyclic temperature modulation periods into one dimension of VOC detection data attributed to the semiconductor gas sensor (111, 112, 113, 114), and assigning labels to the respective dimension of VOC detection data based on the labeled labels, thereby constructing a multi-dimensional and labeled VOC detection data set. The method further comprises:

4. The method of claim 3, wherein, Step S12: normalizing one dimension of VOC detection data based on a maximum VOC detection data value and a minimum VOC detection data value in the respective dimension of VOC detection data. The method further comprises:

5. The method of claim 3, wherein, Step S13: performing feature importance analysis on the respective dimensions of VOC detection data, and merging the respective dimensions of VOC detection data based on the results of the feature importance analysis to reduce the dimensionality of the VOC detection data set. Assigning weight factors attributed to the dimensions of VOC detection data with low feature importance to the dimensions of VOC detection data with low feature importance based on the results of the feature importance analysis, and merging the product of the dimensions of VOC detection data with low feature importance and the weight factors attributed to the dimensions of VOC detection data into the dimensions of VOC detection data with high feature importance, wherein, in particular, the weight factors of the dimensions of VOC detection data with feature importance lower than a preset threshold are set to zero.

6. The method of claim 5, wherein, ​ 7. The method of any one of claims 1 to 6, wherein, the plurality of semiconductor gas sensors (111, 112, 113, 114) are constructed from the same or different semiconductor materials, wherein semiconductor gas sensors (111, 112, 113, 114) constructed from different semiconductor materials have different temperature characteristics; and / or the category recognition model (22) and / or the freshness recognition models (231, 232, 233) are configured as deep learning models.

8. A system (1) for training a model for assessing freshness of an object, the system (1) being configured to perform the method according to any one of claims 1 to 7, wherein, The system (1) comprises the following components: a plurality of semiconductor gas sensors (111, 112, 113, 114) configured to acquire VOC detection data about objects of respective categories; a data labeling module (12) configured to label the VOC detection data about objects of respective categories acquired by the respective semiconductor gas sensors (111, 112, 113, 114) with labels representing freshness states of the corresponding categories of objects, respectively; and a model training module (13) configured to train a category recognition model (22) for determining categories of objects and respective freshness recognition models (231, 232, 233) for determining freshness states of objects of the corresponding categories based on at least the labeled VOC detection data sets.

9. The system (1) according to claim 8, wherein The system (1) further comprises a data construction module (14) configured to merge the VOC detection data acquired by one semiconductor gas sensor (111, 112, 113, 114) within respective cyclic temperature modulation periods into a dimension of VOC detection data assigned to this semiconductor gas sensor (111, 112, 113, 114), respectively, and to assign labels to the respective dimension of VOC detection data based on the labeled labels, respectively, thereby constructing a multi-dimensional and labeled VOC detection data set.

10. A method for assessing freshness of objects, the method comprising: a step S1’ of determining a category of an object to be identified based on VOC detection data about the object to be identified acquired by a semiconductor gas sensor (211, 212, 213, 214) by a category recognition model (22); and a step S2’ of assessing a freshness state of the object based on the VOC detection data set about the object to be identified by a freshness recognition model (231, 232, 233) assigned to the determined category of the object.

11. The method of claim 10, wherein, The method further comprises: a step S11’ of merging VOC detection data acquired by one semiconductor gas sensor (211, 212, 213, 114) of a plurality of semiconductor gas sensors (211, 212, 213, 214) into a dimension of VOC detection data assigned to this semiconductor gas sensor (211, 212, 213, 214), respectively, thereby constructing a multi-dimensional VOC detection data set.

12. The method of claim 11, wherein, The step S11’ comprises: Step S111': performing cyclic temperature modulation on the plurality of semiconductor gas sensors (211, 212, 213, 214) in a preset temperature interval, respectively, and sampling and acquiring VOC detection data of each semiconductor gas sensor (211, 212, 213, 214) about the object to be identified in the cyclic temperature modulation period, for example, by a sliding sampling window technique; and Step S112': merging the VOC detection data acquired by each semiconductor gas sensor (211, 212, 213, 214) in each cyclic temperature modulation period into one-dimensional VOC detection data belonging to the semiconductor gas sensor (211, 212, 213, 214), thereby constructing a multi-dimensional VOC detection data set.

13. The method of claim 11 or 12, wherein, The method further comprises: Step S12': normalizing one-dimensional VOC detection data based on the maximum VOC detection data value and the minimum VOC detection data value in the corresponding one-dimensional VOC detection data, respectively.

14. The method of claim 11 or 12, wherein, The method further comprises: Step S13': performing feature importance analysis on each one-dimensional VOC detection data, and merging each one-dimensional VOC detection data based on the analysis result of the feature importance, so as to reduce the dimension of the VOC detection data set.

15. The method of claim 14, wherein, Based on the analysis result of the feature importance, a weight factor belonging to the one-dimensional VOC detection data is assigned to the one-dimensional VOC detection data with low feature importance, and the product of the one-dimensional VOC detection data with low feature importance and the weight factor belonging to the one-dimensional VOC detection data is merged into the one-dimensional VOC detection data with high feature importance, wherein, in particular, the weight factor of the one-dimensional VOC detection data with feature importance lower than a preset threshold is set to zero.

16. The method of claim 11 or 12, wherein, The plurality of semiconductor gas sensors (211, 212, 213, 214) are constructed of the same or different semiconductor materials, wherein the semiconductor gas sensors (211, 212, 213, 214) constructed of different semiconductor materials have different temperature characteristics; and / or The category recognition model (22) and / or the freshness recognition model (231, 232, 233) are configured as deep learning models.

17. The method of any one of claims 10 to 12, wherein, The method further comprises: Step S3': outputting notification information about the freshness state of the evaluated object.

18. A system (2) for assessing freshness of an object, the system (2) being configured to perform the method according to any one of claims 10 to 17, wherein, The system (2) comprises the following components: Semiconductor gas sensors (211, 212, 213, 214) configured to acquire VOC detection data about the object to be identified, respectively; A category recognition model (22) configured to determine the category of the object to be identified based on the VOC detection data about the object to be identified acquired by each semiconductor gas sensor (211, 212, 213, 214); and A freshness recognition model (231, 232, 233) configured to evaluate the freshness state of the object based on the VOC detection data set about the object to be identified by the freshness recognition model (231, 232, 233) belonging to the determined category of the object.

19. The system (2) according to claim 18, wherein The system (2) further comprises: a data building module (24) configured to merge VOC detection data acquired by one semiconductor gas sensor (211, 212, 213, 214) into a dimension of VOC detection data pertaining to this semiconductor gas sensor (211, 212, 213, 214), respectively, thereby building a multi-dimensional set of VOC detection data; and / or a notification module (25) configured to output notification information about the freshness status of the evaluated object.

20. A computer program product, e.g. a computer readable program carrier, comprising or storing computer program instructions, which, when executed by a processor, assist in at least implementing the steps of the method according to any one of claims 1 to 7 and 10 to 17.

Images