Internet of Things equipment classification method and device and electronic equipment

Abstract

The invention provides an Internet of Things equipment classification method and device and electronic equipment, and relates to the technical field of Internet of Things. The method comprises the following steps: performing feature extraction on equipment information of to-be-classified equipment through a feature extraction network to obtain to-be-classified features; encoding the to-be-classified features through the encoding network to obtain to-be-matched features; obtaining a matching result of the to-be-matched feature and at least one device feature in a feature library, and obtaining a matching result of the to-be-matched feature; if the matching result of the to-be-matched features is that the to-be-matched features are matched with target equipment features in a feature library, taking target classification features associated with the target equipment features as classification features of the to-be-classified equipment; and decoding the classification features of the to-be-classified equipment through the decoding network to obtain the equipment classification of the to-be-classified equipment. Therefore, the target classification feature associated with the target equipment feature can be used as the classification feature of the to-be-classified equipment, so that equipment classification is realized, and the speed and accuracy of equipment classification are improved.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of Internet of Things, and in particular, to a method, apparatus and electronic device for classifying Internet of Things devices. Background Art

[0002] The Internet of Things connects physical devices to the Internet to achieve intelligent interaction between objects and between people. In the Internet of Things architecture, different types of devices need to be connected. For example, in the smart home environment, there are smart door locks, smart cameras, smart home appliances, smart lighting systems, etc.; in the field of industrial Internet of Things, there are industrial sensors, programmable logic controllers, industrial robots, etc.; in the field of intelligent transportation, there are on-board diagnostic systems, traffic light controllers, smart parking equipment, etc.

[0003] At present, IoT device access mainly adopts the method of pre-setting a rule list and manually classifying. First, classification rules are formulated based on some known characteristics of the device, such as device type, functional purpose, data transmission method, etc., and a rule list is formed. When a new IoT device needs to access the network, its category is determined by comparing the various attributes of the device with the content in the preset rule list, and then the device is successfully accessed according to the access process of the corresponding category.

[0004] As the number of devices accessing the network continues to increase and the functions and characteristics of IoT devices are complex and diverse, the existing IoT device access method based on manual classification of preset rule lists cannot intelligently classify the access devices. The classification efficiency is low when facing a large number of device accesses, and it is easy to make misjudgments or missed judgments, which reduces the accuracy of IoT device classification. Summary of the invention

[0005] To solve the above problems, the present application provides a method, device and electronic device for classifying IoT devices, which can solve the technical problems of low efficiency and poor accuracy of device classification methods.

[0006] To achieve the above-mentioned purpose, in a first aspect, the present application provides a method for classifying IoT devices, which comprises: performing feature extraction on device information of a device to be classified through a feature extraction network to obtain features to be classified; performing encoding on the features to be classified through an encoding network to obtain features to be matched; obtaining a matching result between the features to be matched and at least one device feature in a feature library to obtain a matching result of the features to be matched, wherein the matching result of the features to be matched includes that the features to be matched match the target device features in the feature library, or that the features to be matched do not match at least one device feature in the feature library; wherein each device feature in the feature library is associated with a classification feature, and the target device feature is any device feature in the feature library; if the matching result of the features to be matched is that the features to be matched match the target device features in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; performing decoding on the classification features of the device to be classified through a decoding network to obtain a device classification of the device to be classified.

[0007] The IoT device classification method provided by the present application is classified according to the characteristics of device information. The to-be-matched features are obtained by feature extraction and encoding, and the to-be-matched features are matched with the device features in the feature library that already contains the classification relationship. If the target device features that are the same as the to-be-matched features are matched in the feature library, the target classification features associated with the target device features can be determined according to the classification relationship in the feature library, and the target classification features are used as the classification features of the to-be-classified devices, thereby realizing device classification and improving the speed and accuracy of device classification.

[0008] In an optional implementation, obtaining a matching result between a feature to be matched and at least one device feature in a feature library to obtain a matching result of the feature to be matched includes: if the matching result of the feature to be matched is that the feature to be matched does not match at least one device feature in the feature library, performing feature extraction on the feature to be matched to obtain a simplified feature; performing semantic translation on the simplified feature to obtain an approximate feature; obtaining a matching result between the approximate feature and at least one device feature in the feature library to obtain a matching result of the approximate feature, wherein the matching result of the approximate feature includes that the approximate feature matches the target device feature in the feature library, or that the approximate feature does not match at least one device feature in the feature library. In this way, when there is no device feature matching the feature to be matched in the feature library, it is also possible to find an approximate feature with a similar meaning to the feature to be matched by using a semantic translation method, and perform feature matching through the approximate feature to improve the comprehensiveness and accuracy of device classification.

[0009] In an optional implementation, after obtaining the matching result of the approximate feature with at least one device feature in the feature library and obtaining the matching result of the approximate feature, the IoT device classification method further includes: if the matching result of the approximate feature is that the approximate feature matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; the classification feature of the device to be classified is decoded through a decoding network to obtain the device classification of the device to be classified; the feature to be matched is marked as a device feature associated with the target classification feature, and the feature to be matched is stored in the feature library. In this way, the device classification is achieved by searching for the target device feature that matches the approximate feature in the feature library and determining the target classification feature associated with the feature to be matched.

[0010] In an optional implementation, after obtaining the matching result of the approximate feature with at least one device feature in the feature library, the IoT device classification method further includes: if the matching result of the approximate feature is that the approximate feature does not match at least one device feature in the feature library, the feature to be matched is stored in an abnormal data set, and the abnormal data set is a data set containing the feature to be matched that is not matched to the target device feature in the feature library. In this way, when there is neither a device feature matching the feature to be matched nor a device feature matching the approximate feature in the feature library, the feature to be matched can also be stored in the abnormal data set for subsequent search for a device feature matching the feature to be matched, thereby improving the comprehensiveness of device classification.

[0011] In an optional implementation, after the step of storing the feature to be matched in the abnormal data set, the IoT device classification method further includes: obtaining abnormal features, the abnormal features are the features to be matched in the abnormal data set; optimizing the feature library to obtain an updated feature library; obtaining the matching result of the abnormal features and at least one device feature in the updated feature library to obtain the matching result of the abnormal features, the matching result of the abnormal features includes that the abnormal features match the target device features in the feature library, or that the abnormal features do not match at least one device feature in the feature library; if the matching result of the abnormal features is that the abnormal features match the target device features in the feature library, the target classification features associated with the target device features are used as the classification features of the device to be classified; decoding the classification features of the device to be classified through a decoding network to obtain the device classification of the device to be classified. In this way, feature matching can be re-executed for the abnormal features that are not matched to the target device, thereby improving the comprehensiveness and accuracy of device classification.

[0012] In an optional implementation, the feature library is optimized to obtain an updated feature library, including: obtaining newly added device features; storing the newly added device features in the feature library to obtain an updated feature library; wherein each newly added device feature in the updated feature library is associated with a classification feature. In this way, by adding device features to the feature library to optimize the feature library, the abnormal features can be matched again with the device features in the updated feature library, reducing the probability of device classification failure due to insufficient device features in the feature library.

[0013] In an optional implementation, feature extraction is performed on device information through a feature extraction network to obtain features to be classified, including: performing data filtering on the device information to obtain valid device information; performing text analysis on the valid device information to obtain features to be classified. In this way, by extracting features from device information, the amount of data is reduced and the speed of device classification is increased.

[0014] In an optional implementation, obtaining the matching result of the feature to be matched with at least one device feature in the feature library, and obtaining the matching result of the feature to be matched, includes: splitting the feature to be matched to obtain a single feature; traversing the device features in the feature library to find the device feature that is the same as the single feature, or the feature to be matched; if there is a device feature that is the same as the single feature in the feature library, or there is a device feature that is the same as the feature to be matched in the feature library, the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library; if there is no device feature that is the same as the single feature in the feature library, and there is no device feature that is the same as the feature to be matched in the feature library, the matching result of the feature to be matched is that the feature to be matched does not match at least one device feature in the feature library. In this way, splitting the feature to be matched and searching for device features that match the single feature and the feature to be matched in the feature library respectively can improve the feature matching degree, thereby improving the probability of matching the target device feature and improving the accuracy of device classification.

[0015] In an optional implementation, semantic translation is performed on simplified features to obtain approximate features, including: decoding the simplified features to obtain features to be translated; performing semantic analysis on the features to be translated to obtain concepts or attributes of the features to be translated; acquiring semantic data, and traversing the semantic data to obtain approximate words that are similar to the concepts or attributes of the features to be translated; and encoding the approximate words to obtain approximate features. In this way, approximate features that are similar to the features to be matched can be found in a large amount of semantic data, thereby improving the comprehensiveness of device classification.

[0016] In a second aspect, the present application also provides an IoT device classification device, comprising: a feature extraction module, configured to: perform feature extraction on device information of a device to be classified through a feature extraction network to obtain a feature to be classified; an encoding module, configured to: perform encoding on the feature to be classified through an encoding network to obtain a feature to be matched; a device feature matching module, configured to: obtain a matching result between the feature to be matched and at least one device feature in a feature library to obtain a matching result of the feature to be matched, wherein the matching result of the feature to be matched includes that the feature to be matched matches the target device feature in the feature library, or that the feature to be matched does not match at least one device feature in the feature library; wherein each device feature in the feature library is associated with a classification feature, and the target device feature is any device feature in the feature library; a classification feature determination module, configured to: if the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; a decoding module, configured to: perform decoding on the classification feature of the device to be classified through a decoding network to obtain a device classification of the device to be classified.

[0017] In a third aspect, the present application also provides an electronic device, comprising: one or more processors; a memory configured to store one or more programs; wherein, when the one or more programs are executed by one or more processors, the one or more processors implement the Internet of Things device classification method in the first aspect and any optional implementation method thereof.

[0018] It can be understood that the beneficial effects that can be achieved by the technical solutions of the second and third aspects provided above can refer to the beneficial effects of the first aspect and any optional implementation method thereof, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0020] Figure 1 A schematic diagram of the classification of Internet of Things devices provided in an embodiment of the present application; Figure 2 A schematic diagram of a manual classification method for IoT devices provided in an embodiment of the present application; Figure 3 A schematic diagram of a method for classifying IoT devices provided in an embodiment of the present application; Figure 4 A schematic diagram of a method for obtaining a matching result between a feature to be matched and at least one device feature in a feature library provided in an embodiment of the present application; Figure 5 A schematic diagram of an approximate feature matching method provided in an embodiment of the present application; Figure 6 A schematic diagram of a method for performing feature matching on features to be matched in an abnormal data set provided in an embodiment of the present application; Figure 7 A schematic diagram of the structure of an IoT device classification device provided in an embodiment of the present application; Figure 8 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0021] The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following embodiments do not represent all embodiments consistent with the present application.

[0022] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their ordinary and common meanings.

[0023] The terms "first", "second", "third", etc. in the specification and the above drawings of this application are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise specified. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

[0024] To facilitate understanding of the solution, the following explains the relevant terms: Internet of Things: A technology that connects and controls multiple physical devices through the Internet, allowing the devices to operate and interact automatically and intelligently.

[0025] Equipment Classification Transformer (ECT) model: It is an improved model of the neural network model. It can extract features from equipment information, convert the extracted features into vector form, and perform classification prediction and semantic translation on the features to obtain equipment classification.

[0026] Feature extraction: It is a data processing technology that selects or constructs representative and valuable features (or attributes) from raw data. These features can better describe the intrinsic structure and pattern of the data for subsequent data analysis, pattern recognition, machine learning and other tasks. Through feature extraction, high-dimensional, complex, and redundant data can be converted into low-dimensional and more informative data representations.

[0027] Semantic translation: is a translation method that focuses on the accurate communication of the semantic content of the source language in the target language. It aims to preserve the semantic information of the original text as much as possible, including lexical meaning, meaning implied by grammatical structure, contextual meaning and cultural connotation, and at the same time find the closest and most natural expression in the target language.

[0028] The Internet of Things connects physical devices to the Internet to achieve intelligent interaction between objects and between people and objects. In order to clearly classify the functions of connected devices, it is necessary to connect devices to the Internet according to different categories.

[0029] Figure 1 A schematic diagram of the classification of Internet of Things devices provided in an embodiment of the present application.

[0030] like Figure 1 As shown, the Internet of Things can include categories such as whole-house intelligence, environmental monitoring, equipment and facilities, and security monitoring. Among them, smart door locks, smart home appliances, smart lighting, smart curtains, etc. belong to the whole-house intelligence category; thermometers, hygrometers, pressure testers, etc. belong to the environmental monitoring category; water meters, electric meters, etc. belong to the equipment and facilities category; cameras, face recognition equipment, etc. belong to the security monitoring category.

[0031] In order to classify the devices connected to the Internet of Things, a device classification method can be used. Figure 2 A schematic diagram of a manual classification method for IoT devices provided in an embodiment of the present application. Figure 2 As shown, first execute step S201: formulate classification rules. According to some known characteristics of the device, such as device type, functional purpose, data transmission method, etc., formulate classification rules and form a rule list. Secondly, execute step S202: obtain device attributes. When a new IoT device is connected to the network, obtain the various attributes of the device. Then execute step S203: compare the rule list and device attributes. Manually compare the various attributes of the device with the contents in the rule list. Finally, execute step S204: determine the device classification. After determining the category to which the device belongs, implement device access according to the access process of the corresponding category.

[0032] However, when the number of devices connected to the network is large and their functions and characteristics are complex and diverse, the above-mentioned manual classification method of IoT devices based on rule list classification cannot intelligently classify the connected devices, resulting in low classification efficiency and prone to misjudgment or omission, reducing the accuracy of IoT device classification.

[0033] To solve the above problems, the embodiments of the present application provide an IoT device classification method, apparatus and electronic device. The IoT device classification method obtains device features of access devices through an ECT model, and automatically matches the device features with features stored in a feature library to determine the target classification of the device, thereby realizing intelligent classification of access devices and improving the speed and accuracy of device classification.

[0034] Next, a specific implementation of the IoT device classification method is introduced. The execution subject of the method can be an electronic device.

[0035] Figure 3 A schematic diagram of a method for classifying IoT devices provided in an embodiment of the present application. Figure 3 As shown, the IoT device classification method includes the following steps: S100: Perform feature extraction on device information of the device to be classified through a feature extraction network to obtain features to be classified.

[0036] In some embodiments, when the device to be classified is connected to the Internet of Things, the device information of the device to be classified is first obtained, including at least one of the device's hardware information, communication protocol, data transmission mode, energy consumption mode, and operation data. For example, the hardware information includes name, brand, function, etc.; the communication protocol includes WiFi, Bluetooth, etc.; the energy consumption mode includes battery, solar energy, etc.; the operation data includes temperature range, update time, etc.

[0037] In some embodiments, in order to obtain information suitable for classification from the device information of the device to be classified and improve the accuracy of device classification, it is necessary to perform feature extraction on the device information to obtain the features to be classified for device classification. In order to improve the effectiveness of the features to be classified and ensure the accuracy of device classification, when extracting features from the device information, it is necessary to perform data filtering to obtain valid device information; then perform text analysis on the valid device information to obtain the features to be classified. Specifically, the device information is input into the ECT model, and data filtering and text analysis are performed on the device information through the feature extraction network in the ECT model.

[0038] Exemplarily, the feature extraction network in the ECT model is a Transformer architecture, which consists of a multi-head attention mechanism and a feed-forward neural network, and is used to extract key text information. For example, if the device to be classified is a "smart desk lamp", when the "smart desk lamp" is connected to the Internet of Things, the initial device information obtained is "eye protection desk lamp", "battery", etc. After data filtering, the valid device information obtained is "eye protection desk lamp"; after performing text analysis on the "eye protection desk lamp", the final extracted feature to be classified is "desk lamp".

[0039] It should be noted that other architectures may also be used as feature extraction networks, such as recurrent neural networks, etc., which are not specifically limited in the embodiments of the present application.

[0040] In some embodiments, in order to improve the accuracy of feature extraction performed by the ECT model, the gradient descent method can be used to train the ECT model. The gradient descent method determines the direction of parameter update by calculating the gradient of the loss function with respect to the parameter θ. The calculation formula is as follows: (1) in, Represents the learning rate, which is a small value, such as 0.001, which determines the compensation for updating parameters; Represents the loss function on the entire data set. Compute the first-order derivative.

[0041] Specifically, a labeled training data set is used for model training, and the labeled training data set includes device information and feature labels. First, the device information in the training data set is input as input data to the ECT model to be trained to obtain the prediction results of the ECT model output; secondly, the training loss is calculated based on the prediction results and classification labels, and the training loss is compared with the preset loss threshold. If the training loss is greater than or equal to the preset loss threshold, the model parameters of the ECT model are adjusted according to the training loss; if the training loss is less than the preset loss threshold, the feature extraction stage training of the ECT model is completed.

[0042] S200: Encoding the features to be classified through an encoding network to obtain features to be matched.

[0043] In some embodiments, in order to facilitate the ECT model to process the features to be classified, the features to be classified need to be vectorized. The features to be classified are encoded by the encoding network to obtain the features to be matched. For example, the feature to be classified "desk lamp" is converted into a feature to be matched in the form of a sequence that can be recognized by the computer.

[0044] S300: Obtain a matching result between the feature to be matched and at least one device feature in the feature library to obtain a matching result of the feature to be matched.

[0045] In some embodiments, in order to improve the speed of device classification, feature matching can be automatically performed through the ECT model. To achieve feature matching, it is necessary to establish a feature library in the ECT model, which includes classification features and device features, and each device feature in the feature library is associated with a classification feature. As shown in Table 1, a classification feature can be associated with multiple device features, including name, function, model, manufacturer, etc.

[0046] Table 1 .

[0047] It should be noted that the device feature may also be other information related to the device to be classified, as long as it can reflect the characteristics of the device to be classified, and is not specifically limited in the embodiments of the present application.

[0048] In some embodiments, after obtaining the feature to be matched and determining the feature library in the ECT model, the matching result of the feature to be matched and at least one device feature in the feature library can be obtained through the ECT model to obtain the matching result of the feature to be matched. The matching result of the feature to be matched includes that the feature to be matched matches the target device feature in the feature library, or that the feature to be matched does not match at least one device feature in the feature library, and the target device feature is any device feature in the feature library.

[0049] In some embodiments, in order to improve the accuracy of obtaining matching results, the features to be matched may be split and then feature matching may be performed. Figure 4 A schematic diagram of a method for obtaining a matching result between a feature to be matched and at least one device feature in a feature library provided in an embodiment of the present application. Figure 4 As shown, when obtaining the matching result between the feature to be matched and at least one device feature in the feature library, step S401 is first performed: splitting the feature to be matched to obtain a single feature.

[0050] Exemplarily, after performing feature extraction and encoding on the classified device "smart desk lamp", a sequence corresponding to the feature to be matched "desk lamp" is obtained, and the sequence corresponding to the feature to be matched "desk lamp" is split to obtain two sequences corresponding to the single features "desk" and "lamp" respectively.

[0051] Then execute step S402: traverse the device features in the feature library to find a device feature that is the same as the single feature or the feature to be matched.

[0052] Specifically, multiple fully connected layers can be added to the ECT model to extract high-level semantic information between a single feature and the device features in the feature library to determine whether there is a device feature in the feature library that matches the feature to be matched. For example, the device features that match the sequences corresponding to "table", "lamp", and "desk lamp" are searched in the feature library.

[0053] Finally, step S403 is executed: the matching result is output.

[0054] In some embodiments, the matching result includes that the feature to be matched matches the target device feature in the feature library or that the feature to be matched does not match the target device feature in the feature library. If there is a device feature identical to the single feature in the feature library, or there is a device feature identical to the feature to be matched in the feature library, the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library.

[0055] For example, if the feature to be matched is the sequence corresponding to "desk lamp", the single feature obtained by splitting is the sequence corresponding to "desk" and "lamp", and there is a sequence corresponding to the device feature "lamp" in the feature library, which is the same as the sequence corresponding to the single feature "lamp", indicating that the sequence corresponding to "lamp" is the target device feature, then the matching result is that the feature to be matched matches the target device feature in the feature library.

[0056] In some embodiments, if there is no device feature identical to the single feature in the feature library, and there is no device feature identical to the feature to be matched in the feature library, the matching result of the feature to be matched is that the feature to be matched does not match at least one device feature in the feature library.

[0057] For example, if the feature to be matched is the sequence corresponding to "desk lamp", the single feature obtained by splitting is the sequence corresponding to "desk" and "lamp", but the sequence corresponding to "desk lamp", "desk", and "lamp" does not exist in the feature library, indicating that the target device feature does not exist in the feature library, then the matching result is that the feature to be matched does not match the target device feature in the feature library.

[0058] It should be noted that in the process of obtaining the device information of the device to be classified, multiple device information can be obtained, so multiple features to be matched can be generated in step S200. For example, the device to be classified is a "smart camera", and the features to be matched obtained include sequences corresponding to "face recognition", "monitoring", "Wi-Fi", etc. Then, in step S300, the matching results of multiple features to be matched and at least one device feature in the feature library can be obtained respectively, so as to improve the comprehensiveness of feature matching and the accuracy of device classification.

[0059] In some embodiments, since the feature library cannot fully cover the device features of all connected devices, it will lead to the problem that the device features that are the same as the feature to be matched or the single feature cannot be matched because the device features in the feature library are not comprehensive. Therefore, if the matching result of the feature to be matched is that the feature to be matched does not match at least one device feature in the feature library, the ECT model can further perform semantic translation on the feature to be matched, obtain approximate features and perform feature matching.

[0060] Figure 5 A schematic diagram of an approximate feature matching method provided in an embodiment of the present application. Figure 5As shown, when performing approximate feature matching, step S501 is first performed: feature extraction is performed on the features to be matched to obtain simplified features.

[0061] Exemplarily, if the feature to be matched is a sequence corresponding to "lighting in front of the door", feature extraction is first performed to obtain a sequence corresponding to the simplified feature "lighting".

[0062] Next, step S502 is performed: semantic translation is performed on the simplified feature to obtain an approximate feature. In some embodiments, by performing semantic translation, an approximate feature having similar semantics to the simplified feature can be obtained, thereby expanding the range of feature matching and improving the accuracy of feature matching.

[0063] Specifically, firstly, the simplified features are decoded to obtain the features to be translated; secondly, semantic analysis is performed on the features to be translated to obtain the concepts or attributes of the features to be translated; then, semantic data is obtained and traversed to obtain approximate words that are similar to the concepts or attributes of the features to be translated; finally, the approximate words are encoded to obtain approximate features. For example, if the simplified feature is the sequence corresponding to "lighting", the sequence corresponding to the approximate feature "light" can be obtained through semantic translation.

[0064] Then execute step S503: obtain the matching result of the approximate feature and at least one device feature in the feature library to obtain the matching result of the approximate feature.

[0065] For example, if the approximate feature is the sequence corresponding to "light", and there is a sequence corresponding to the device feature "light" in the feature library, it means that the sequence corresponding to "light" is the target device feature, that is, the approximate feature matches the target device feature in the feature library; conversely, if there is no sequence corresponding to the device feature "light" in the feature library, it means that the approximate feature does not match any of the device features in the feature library.

[0066] Finally, step S504 is executed: determining whether the approximate feature matches the target device feature in the feature library. If the approximate feature matches the target device feature in the feature library, step S5051 is executed: using the target classification feature associated with the target device feature as the classification feature of the device to be classified, and decoding the classification feature of the device to be classified through the decoding network to obtain the device classification of the device to be classified.

[0067] For example, if the sequence corresponding to the approximate feature "light" matches the sequence corresponding to the target device feature "light" in the feature library, the sequence corresponding to the target classification feature "whole house smart" associated with the sequence corresponding to the target device feature "light" in the feature library is obtained as the classification feature of the device to be classified. Then, the sequence corresponding to "whole house smart" is decoded to obtain the device classification of the device to be classified as "whole house smart".

[0068] Then execute step S5052: mark the feature to be matched as a device feature associated with the target classification feature, and store the feature to be matched in a feature library.

[0069] Before semantic translation, the feature library does not include the sequence corresponding to "lighting", and the sequence corresponding to the simplified feature "lighting" obtained through semantic translation is similar in meaning to the sequence corresponding to "light", and it can be known that the sequence corresponding to the feature to be matched "front door lighting" is also similar in meaning to the sequence corresponding to "light". Therefore, the sequences corresponding to the simplified feature "lighting" and the feature to be matched "front door lighting" can also be stored in the feature library as device features to improve the comprehensiveness of the feature library and the accuracy of the IoT device classification method.

[0070] In some embodiments, the purpose of the IoT device classification method is to classify all access devices. When the device features of a certain access device cannot find matching target device features in the feature library, it means that the ECT model needs to be further optimized to achieve the purpose of accurate device classification. Therefore, if the matching result of the approximate feature is that the approximate feature does not match at least one device feature in the feature library, execute step S5061: store the to-be-matched feature in the abnormal data set. It is used to perform feature matching again on the to-be-matched features that have not matched the target device features in the feature library after optimizing the ECT model.

[0071] Exemplarily, if the feature to be matched is the sequence corresponding to "living room air conditioner", the simplified feature is the sequence corresponding to "air conditioner", and the approximate feature is the sequence corresponding to "temperature control", and the matching result after the ECT model executes feature matching is that the matching result of the approximate feature is that the approximate feature does not match at least one device feature in the feature library, then the sequence corresponding to the feature to be matched "living room air conditioner" is stored in the abnormal data set, waiting for feature matching to be executed again.

[0072] In some embodiments, in order to improve the accuracy of the device classification method, the ECT model may be optimized, and then the optimized ECT model may be used to perform feature matching on the features to be matched in the abnormal data set.

[0073] Figure 6 A schematic diagram of a method for performing feature matching on features to be matched in an abnormal data set provided in an embodiment of the present application. Figure 6 As shown, when performing feature matching on the to-be-matched features in the abnormal data set, step S601 is first performed: obtaining abnormal features, wherein the abnormal features are the to-be-matched features in the abnormal data set.

[0074] Then execute step S602: optimize the feature library to obtain an updated feature library. In some embodiments, the method of optimizing the feature library to obtain an updated feature library includes: obtaining a newly added device feature; storing the newly added device feature in the feature library to obtain an updated feature library. It should be noted that in order to ensure that each device feature in the feature library can find a corresponding classification feature, the newly added device feature should be a device feature associated with a classification characteristic in the feature library, or, when adding the newly added device feature to the feature library, a newly added classification feature associated with the newly added device feature is added.

[0075] Exemplarily, a sequence corresponding to a newly added device feature "air conditioner" is added to the feature library, and the classification feature associated therewith is determined to be a sequence corresponding to "whole house intelligence".

[0076] As another example, a sequence corresponding to a newly added device feature “navigation” is added to the feature library, and a sequence corresponding to a newly added classification feature “intelligent transportation” associated therewith is added to the feature library.

[0077] In some embodiments, in addition to the method of optimizing the feature library, other methods may be used to optimize the ECT model to improve the speed and accuracy of device classification.

[0078] Exemplarily, the ECT model is further trained by an incremental learning method. Specifically, a new training data set is obtained and gradually added to the ECT model for incremental training. During the training process, the incremental ratio of the training data can be dynamically adjusted according to the similarity analysis results of the new and old data.

[0079] It should be understood that the optimization method for the ECT model may also include other conventional neural network model optimization methods, such as transfer learning, etc., which are not specifically limited in the embodiments of the present application.

[0080] Finally, step S603 is executed: obtaining the matching result of the abnormal feature and at least one device feature in the updated feature library to obtain the matching result of the abnormal feature. The matching result of the abnormal feature includes that the abnormal feature matches the target device feature in the feature library, or that the abnormal feature does not match at least one device feature in the feature library. After optimizing the ECT model, the optimized ECT model can be used to perform feature matching on the abnormal feature again. If the matching result of the abnormal feature is that the abnormal feature matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified, and the classification feature of the device to be classified is decoded through the decoding network to obtain the device classification of the device to be classified, thereby realizing device classification.

[0081] For example, a new device feature is added to the feature library, and the associated classification feature is the sequence corresponding to "whole house intelligence". If the abnormal feature is the sequence corresponding to "living room air conditioner", the simplified feature is the sequence corresponding to "air conditioner". At this time, feature matching is performed, and the simplified feature is the same as the target device feature in the feature library. Therefore, the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library.

[0082] S400: If the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified.

[0083] In some embodiments, the device features and classification features of the connected IoT devices are associated. For example, the device features associated with the sequence corresponding to the classification feature "whole house intelligence" can be the sequences corresponding to "curtains", "lights", "door locks", "air conditioners", etc. Therefore, after obtaining the target device features, the ECT model can quickly find the target classification features associated with the target device features, and then obtain the device classification, thereby improving the speed of device classification.

[0084] For example, if the device to be classified is a "vertical air conditioner", after being connected to the Internet of Things, the sequence corresponding to the feature to be matched "living room air conditioner" is obtained through ECT model processing, and the simplified feature is the sequence corresponding to "air conditioner". After performing feature matching, the simplified feature is obtained to be the same as the sequence corresponding to the device feature "air conditioner" in the feature library. Therefore, the sequence corresponding to "air conditioner" is the target device feature, and the sequence corresponding to its associated classification feature "whole house intelligence" in the feature library is the target classification feature.

[0085] S500: Decoding the classification features of the device to be classified through a decoding network to obtain a device classification of the device to be classified.

[0086] In some embodiments, after the target classification features are acquired, the decoding network in the ECT model is used to decode the target classification features to obtain a device classification that is easy for the user to understand, and finally the classification of the device to be classified is achieved.

[0087] For example, after obtaining the target classification feature of the device to be classified "vertical air conditioner" as the sequence corresponding to "whole house intelligence", it is necessary to convert it into an expression form that is easy for users to understand. Therefore, the sequence corresponding to "whole house intelligence" is decoded, and the device classification of "vertical air conditioner" is obtained as "whole house intelligence".

[0088] Some embodiments of the present application also provide an IoT device classification device, Figure 7 This is a schematic diagram of the structure of an IoT device classification device provided in an embodiment of the present application. Figure 7As shown, the IoT device classification device includes: The feature extraction module 701 is configured to: perform feature extraction on the device information of the device to be classified through a feature extraction network to obtain features to be classified.

[0089] The encoding module 702 is configured to: perform encoding on the features to be classified through an encoding network to obtain features to be matched.

[0090] The device feature matching module 703 is configured to: obtain the matching result between the feature to be matched and at least one device feature in the feature library, and obtain the matching result of the feature to be matched, wherein the matching result of the feature to be matched includes that the feature to be matched matches the target device feature in the feature library, or that the feature to be matched does not match at least one device feature in the feature library; wherein each device feature in the feature library is associated with a classification feature, and the target device feature is any device feature in the feature library.

[0091] The classification feature determination module 704 is configured to: if the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library, use the target classification feature associated with the target device feature as the classification feature of the device to be classified.

[0092] The decoding module 705 is configured to: perform decoding on the classification features of the device to be classified through a decoding network to obtain the device classification of the device to be classified.

[0093] Some embodiments of the present application also provide an electronic device, comprising: a memory for storing a computer program; A processor is used to execute a computer program to implement the steps of any method provided in the first aspect.

[0094] For example, Figure 8 A schematic diagram of an electronic device provided in an embodiment of the present application. Figure 8 As shown, the electronic device includes a processor 801, at least one communication bus 802, a user interface 803, at least one external communication interface 804 and a memory 805. The communication bus 802 is configured to realize the connection and communication between these components. The user interface 803 may include a display screen, and the external communication interface 804 may include a standard wired interface and a wireless interface. The memory 805 stores a computer program. The processor 801 is used to execute the computer program stored in the memory 805.

[0095] Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims

1. A method for classifying IoT devices, characterized in that: The method comprises: Perform feature extraction on the device information of the device to be classified through a feature extraction network to obtain features to be classified; Encoding the features to be classified through an encoding network to obtain features to be matched; Obtaining a matching result between the feature to be matched and at least one device feature in the feature library, obtaining a matching result of the feature to be matched, wherein the matching result of the feature to be matched includes that the feature to be matched matches a target device feature in the feature library, or that the feature to be matched does not match any of the at least one device feature in the feature library; wherein each device feature in the feature library is associated with a classification feature, and the target device feature is any device feature in the feature library; If the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; The classification features of the device to be classified are decoded by a decoding network to obtain the device classification of the device to be classified.

2. The method for classifying IoT devices according to claim 1, characterized in that: The obtaining a matching result between the feature to be matched and at least one device feature in the feature library to obtain the matching result of the feature to be matched includes: If the matching result of the feature to be matched is that the feature to be matched does not match the at least one device feature in the feature library, performing feature extraction on the feature to be matched to obtain a simplified feature; Performing semantic translation on the simplified features to obtain approximate features; Obtain a matching result between the approximate feature and at least one device feature in the feature library to obtain a matching result of the approximate feature, wherein the matching result of the approximate feature includes that the approximate feature matches the target device feature in the feature library, or that the approximate feature does not match any of the at least one device feature in the feature library.

3. The method for classifying IoT devices according to claim 2, characterized in that: After obtaining the matching result between the approximate feature and at least one device feature in the feature library, the method further includes: If the matching result of the approximate feature is that the approximate feature matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; Decoding the classification features of the device to be classified by the decoding network to obtain the device classification of the device to be classified; The feature to be matched is marked as the device feature associated with the target classification feature, and the feature to be matched is stored in the feature library.

4. The method for classifying IoT devices according to claim 2, characterized in that: After the step of obtaining the matching result between the approximate feature and at least one device feature in the feature library to obtain the matching result of the approximate feature, the method further includes: If the matching result of the approximate feature is that the approximate feature does not match the at least one device feature in the feature library, the feature to be matched is stored in an abnormal data set, which is a data set containing the feature to be matched that is not matched to the target device feature in the feature library.

5. The method for classifying IoT devices according to claim 4, characterized in that: After the step of storing the to-be-matched features in the abnormal data set, the method further comprises: Acquire an abnormal feature, where the abnormal feature is the feature to be matched in the abnormal data set; Optimizing the feature library to obtain an updated feature library; Obtaining a matching result between the abnormal feature and at least one device feature in the updated feature library, and obtaining a matching result of the abnormal feature, wherein the matching result of the abnormal feature includes that the abnormal feature matches the target device feature in the feature library, or that the abnormal feature does not match any of the at least one device feature in the feature library; If the matching result of the abnormal feature is that the abnormal feature matches the target device feature in the feature library, the target classification feature associated with the target device feature is used as the classification feature of the device to be classified; The classification features of the device to be classified are decoded by a decoding network to obtain the device classification of the device to be classified.

6. The method for classifying IoT devices according to claim 5, characterized in that: The step of optimizing the feature library to obtain an updated feature library includes: Get new device features; The newly added device feature is stored in the feature library to obtain the updated feature library; wherein each of the newly added device features in the updated feature library is associated with one of the classification features.

7. The method for classifying IoT devices according to claim 1, characterized in that: The performing feature extraction on the device information through a feature extraction network to obtain features to be classified includes: Performing data filtering on the device information to obtain valid device information; Perform text analysis on the valid device information to obtain the features to be classified.

8. The method for classifying IoT devices according to claim 1, characterized in that: The obtaining a matching result between the feature to be matched and at least one device feature in the feature library to obtain the matching result of the feature to be matched includes: Splitting the feature to be matched to obtain a single feature; Traversing the device features in the feature library to find the device features that are the same as the single feature or the feature to be matched; If the device feature identical to the single feature exists in the feature library, or the device feature identical to the feature to be matched exists in the feature library, the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library; If the device feature identical to the single feature does not exist in the feature library, and the device feature identical to the feature to be matched does not exist in the feature library, the matching result of the feature to be matched is that the feature to be matched does not match the at least one device feature in the feature library.

9. The method for classifying IoT devices according to claim 2, characterized in that: The performing semantic translation on the simplified features to obtain approximate features includes: Decoding the simplified features to obtain features to be translated; Performing semantic analysis on the feature to be translated to obtain a concept or attribute of the feature to be translated; Acquiring semantic data, and traversing the semantic data to obtain similar words that are close to the concept or attribute of the feature to be translated; Encoding is performed on the approximate words to obtain the approximate features.

10. An IoT device classification device, characterized in that: The device comprises: The feature extraction module is configured to: perform feature extraction on the device information of the device to be classified through a feature extraction network to obtain features to be classified; The encoding module is configured to: encode the features to be classified through an encoding network to obtain features to be matched; The device feature matching module is configured to: obtain a matching result between the feature to be matched and at least one device feature in the feature library, and obtain a matching result of the feature to be matched, wherein the matching result of the feature to be matched includes that the feature to be matched matches a target device feature in the feature library, or that the feature to be matched does not match any of the at least one device feature in the feature library; wherein each device feature in the feature library is associated with a classification feature, and the target device feature is any device feature in the feature library; The classification feature determination module is configured to: if the matching result of the feature to be matched is that the feature to be matched matches the target device feature in the feature library, use the target classification feature associated with the target device feature as the classification feature of the device to be classified; The decoding module is configured to: decode the classification features of the device to be classified through a decoding network to obtain the device classification of the device to be classified.

11. An electronic device, characterized in that: include: one or more processors; a memory configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the Internet of Things device classification method according to any one of claims 1-9.

Images