Data classification method, vehicle and storage medium

Abstract

The embodiment of the invention provides a data classification method, a vehicle and a storage medium, and the method comprises the steps: carrying out the recognition of original data transmitted in the Internet of Vehicles, and obtaining a plurality of original elements contained in the original data; based on the element features of the multiple original elements, first data is selected from the original data, and the contribution degree of the multiple first elements contained in the first data to data classification of the original data is larger than the contribution degree of the multiple second elements contained in the second data to data classification of the original data; the second data is used for representing data except the first data in the original data; inputting the first data into a feature extraction model, and performing feature extraction on the first data to obtain data features of the first data; and inputting the data features into a data classification model for data classification to obtain a target classification result of the original data. According to the method and the device, the technical problem of relatively low accuracy of classifying the data transmitted in the Internet of Vehicles in related technologies is solved.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing, and specifically, to a data classification method, a vehicle, and a storage medium. Background Art

[0002] As a key platform connecting vehicles, road infrastructure, pedestrians, and cloud services, the Internet of Vehicles (IoV) enables real-time information exchange between vehicles through advanced communication technologies, ensuring safe and stable operation. However, the widespread adoption of IoV also presents significant cybersecurity challenges, particularly malicious intrusions targeting both internal and external vehicle networks, such as denial-of-service attacks, replay attacks, and spoofing attacks, which severely impact driving safety and data privacy. While current cybersecurity defense mechanisms, such as firewalls and intrusion detection systems, can detect and prevent cyberattacks to a certain extent, these defense mechanisms primarily rely on simple statistical analysis, resulting in low accuracy in classifying data transmitted within IoV systems in practice.

[0003] There is currently no good solution to the above problems. Summary of the Invention

[0004] The embodiments of the present application provide a data classification method, a vehicle, and a storage medium to at least solve the technical problem in the related art of low accuracy in classifying data transmitted in the Internet of Vehicles.

[0005] According to one aspect of an embodiment of the present application, a data classification method is provided, including: identifying original data transmitted in a vehicle network to obtain multiple original elements contained in the original data; selecting first data from the original data based on element features of the multiple original elements, wherein the contribution of the multiple first elements contained in the first data to the data classification of the original data is greater than the contribution of the multiple second elements contained in the second data to the data classification of the original data, and the second data is used to characterize data other than the first data in the original data; inputting the first data into a feature extraction model to extract features of the first data to obtain data features of the first data; inputting the data features into a data classification model to perform data classification to obtain a target classification result of the original data, wherein the target classification result is used to characterize whether there is a security risk in the original data and the risk type of the security risk.

[0006] Furthermore, based on the element features of multiple original elements, first data is selected from the original data, including: obtaining at least one data center corresponding to the original data, and data distribution corresponding to at least one data center, wherein different data centers correspond to different data categories, and the data categories include: there is no security risk in the original data, there is a security risk in the original data, and when there is a security risk in the original data, the risk type of the security risk, and the data distribution is used to characterize the distribution of data points in at least one data center; based on at least one data center and the data distribution, a first distance between any element feature and multiple first element features, and a second distance between any element feature and multiple second element features are obtained, wherein the first element feature is the same as the data center corresponding to any element feature, and the second element feature is different from the data center corresponding to any element feature; based on the first distance, the second distance, the prior probability of the data center corresponding to any element feature and the data distribution, the feature weight of any element feature is determined; based on the feature weight, a first element is selected from multiple original elements, and the data corresponding to the first element in the original data is determined as the first data.

[0007] Furthermore, the data classification model includes: a first classification model; inputting data features into the data classification model for data classification to obtain a target classification result of the original data, including: inputting data features into the first classification model for data classification to obtain a first classification result; evaluating the first classification result based on a risk assessment strategy to obtain a classification evaluation result, wherein the classification evaluation result is used to characterize the accuracy of the first classification result; in response to the classification evaluation result being that the accuracy of the first classification result meets a preset accuracy condition, determining the first classification result as the target classification result.

[0008] Furthermore, the data classification model also includes: at least one second classification model, and the model types of any one of the first classification model and the at least one second classification model are different; the above method also includes: in response to the classification evaluation result that the accuracy of the first classification result does not meet the preset accuracy condition, inputting the data features into at least one second classification model for data classification to obtain at least one second classification result; based on the first classification result and at least one second classification result, determining the target classification result.

[0009] Furthermore, based on the first classification result and at least one second classification result, a target classification result is determined, including: grouping at least one second classification result and the first classification result to obtain at least one classification group, wherein any classification results contained in the same classification group are the same; performing feature extraction on any classification group to obtain classification group features of any classification group; based on the classification group features, selecting a target classification group from at least one classification group; and determining the classification results contained in the target classification group as the target classification result.

[0010] Furthermore, the first classification model includes: a self-attention module, a feature matching module and a data classification module; the data features are input into the first classification model for data classification to obtain a first classification result, including: based on the self-attention module, performing self-attention processing on the data features to obtain self-attention features; based on the feature matching module, matching the self-attention features with at least one data category to obtain a feature matching result, wherein the feature matching result is used to characterize the degree of matching between the first data and any data category; based on the data classification module, determining the first classification result according to the feature matching result.

[0011] Furthermore, the first classification result is evaluated based on the risk assessment strategy to obtain a classification evaluation result, including: obtaining a feature matching result corresponding to the first classification result; evaluating the first classification result based on the feature matching result to obtain an evaluation index of the first classification result; in response to the evaluation index satisfying a preset evaluation condition, determining that the classification evaluation result is that the accuracy of the first classification result satisfies the preset accuracy condition; in response to the evaluation index not satisfying the preset evaluation condition, determining that the classification evaluation result is that the accuracy of the first classification result does not satisfy the preset accuracy condition.

[0012] Furthermore, the feature extraction model includes: a convolution feature extraction module, a convolution feature conversion module, a forward time series feature extraction module, a backward time series feature extraction module and a time series feature fusion module; the first data is input into the feature extraction model to perform feature extraction on the first data to obtain data features of the first data, including: using the convolution feature extraction module to perform feature extraction on the first data to obtain the convolution features of the first data; using the convolution feature conversion module to convert the convolution features to obtain time series data corresponding to the convolution features; using the forward time series feature extraction module to perform feature extraction on the time series data to obtain forward time series features, and using the backward time series feature extraction module to perform feature extraction on the time series data to obtain backward time series features; using the time series feature fusion module to fuse the forward time series features and the backward time series features to obtain data features.

[0013] According to another aspect of an embodiment of the present application, a data classification device is also provided, including: a data identification module, used to identify the original data transmitted in the Internet of Vehicles, and obtain multiple original elements contained in the original data; a data selection module, used to select first data from the original data based on the element features of the multiple original elements, wherein the contribution of the multiple first elements contained in the first data to the data classification of the original data is greater than the contribution of the multiple second elements contained in the second data to the data classification of the original data, and the second data is used to characterize data other than the first data in the original data; a feature extraction module, used to input the first data into a feature extraction model to extract features of the first data and obtain data features of the first data; a data classification module, used to input the data features into a data classification model to perform data classification and obtain a target classification result of the original data, wherein the target classification result is used to characterize whether there is a security risk in the original data and the risk type of the security risk.

[0014] Furthermore, the data selection module is also used to: obtain at least one data center corresponding to the original data, and the data distribution corresponding to at least one data center, wherein different data centers correspond to different data categories, and the data categories include: there is no security risk in the original data, there is a security risk in the original data, and when there is a security risk in the original data, the risk type of the security risk, and the data distribution is used to characterize the distribution of data points in at least one data center; based on at least one data center and the data distribution, obtain a first distance between any element feature and multiple first element features, and a second distance between any element feature and multiple second element features, wherein the first element feature is the same as the data center corresponding to any element feature, and the second element feature is different from the data center corresponding to any element feature; determine the feature weight of any element feature based on the first distance, the second distance, the prior probability of the data center corresponding to any element feature and the data distribution; based on the feature weight, select a first element from multiple original elements, and determine the data corresponding to the first element in the original data as the first data.

[0015] Furthermore, the data classification model includes: a first classification model; the data classification module is also used to: input data features into the first classification model for data classification to obtain a first classification result; evaluate the first classification result based on a risk assessment strategy to obtain a classification evaluation result, wherein the classification evaluation result is used to characterize the accuracy of the first classification result; in response to the classification evaluation result being that the accuracy of the first classification result meets a preset accuracy condition, determine that the first classification result is a target classification result.

[0016] Furthermore, the data classification model also includes: at least one second classification model, and the model types of any one of the first classification model and the at least one second classification model are different; the above-mentioned device also includes: a reclassification module, which is used to input data features into at least one second classification model for data classification in response to the classification evaluation result that the accuracy of the first classification result does not meet the preset accuracy conditions, and obtain at least one second classification result; a result determination module, which is used to determine the target classification result based on the first classification result and at least one second classification result.

[0017] Furthermore, the result determination module is also used to: group at least one second classification result and the first classification result to obtain at least one classification group, wherein any classification results contained in the same classification group are the same; perform feature extraction on any classification group to obtain classification group features of any classification group; based on the classification group features, select a target classification group from at least one classification group; and determine the classification results contained in the target classification group as the target classification results.

[0018] Furthermore, the first classification model includes: a self-attention module, a feature matching module and a data classification module; the data classification module is also used to: based on the self-attention module, perform self-attention processing on the data features to obtain self-attention features; based on the feature matching module, match the self-attention features with at least one data category to obtain a feature matching result, wherein the feature matching result is used to characterize the degree of matching between the first data and any data category; based on the data classification module, determine the first classification result according to the feature matching result.

[0019] Furthermore, the data classification module is also used to: obtain a feature matching result corresponding to the first classification result; evaluate the first classification result based on the feature matching result to obtain an evaluation index of the first classification result; in response to the evaluation index satisfying a preset evaluation condition, determine that the classification evaluation result is that the accuracy of the first classification result satisfies the preset accuracy condition; in response to the evaluation index not satisfying the preset evaluation condition, determine that the classification evaluation result is that the accuracy of the first classification result does not satisfy the preset accuracy condition.

[0020] Furthermore, the feature extraction model includes: a convolution feature extraction module, a convolution feature conversion module, a forward time series feature extraction module, a backward time series feature extraction module and a time series feature fusion module; the feature extraction module is also used to: use the convolution feature extraction module to extract features of the first data to obtain the convolution features of the first data; use the convolution feature conversion module to convert the convolution features to obtain time series data corresponding to the convolution features; use the forward time series feature extraction module to extract features of the time series data to obtain forward time series features, and use the backward time series feature extraction module to extract features of the time series data to obtain backward time series features; use the time series feature fusion module to fuse the forward time series features and the backward time series features to obtain data features.

[0021] According to another aspect of an embodiment of the present application, a vehicle is further provided, comprising: a memory storing an executable program; and a processor for running the program, wherein the method of each embodiment of the present application is executed when the program is running.

[0022] According to another aspect of an embodiment of the present application, a computer-readable storage medium is also provided, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present application.

[0023] According to another aspect of the embodiments of the present application, a computer program product is further provided, including a computer program, which implements the methods in various embodiments of the present application when executed by a processor.

[0024] According to another aspect of an embodiment of the present application, a computer program product is also provided, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present application is implemented.

[0025] According to another aspect of the embodiments of the present application, a computer program is further provided, which implements the methods in various embodiments of the present application when executed by a processor.

[0026] In an embodiment of the present application, the original data transmitted in the Internet of Vehicles is identified to obtain multiple original elements contained in the original data; based on the element features of the multiple original elements, first data is selected from the original data; the first data is input into a feature extraction model to extract features of the first data to obtain data features of the first data; the data features are input into a data classification model to perform data classification to obtain target classification results of the original data. By screening out the first data that contributes more to the data classification process from the original data, and using the data classification model to classify the original data based on the first data, the influence of the second data with lower contribution on the data classification process can be reduced, thereby ensuring the accuracy of the target classification results, and thus solving the technical problem of low accuracy in classifying data transmitted in the Internet of Vehicles. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0028] Figure 1 is a flow chart of a data classification method according to an embodiment of the present application;

[0029] Figure 2 is a schematic diagram illustrating a feature extraction process according to an embodiment of the present application;

[0030] Figure 3 is a flow chart of a data classification method according to an embodiment of the present application;

[0031] Figure 4 This is a structural block diagram of a data classification device according to an embodiment of the present application. DETAILED DESCRIPTION

[0032] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0033] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in a sequence other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0034] According to an embodiment of the present application, an embodiment of a method for data classification is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0035] In this embodiment, a data classification method is provided. Figure 1 is a flow chart of a data classification method according to an embodiment of the present application. Figure 1 As shown, the process includes the following steps:

[0036] Step S102 : identifying the original data transmitted in the Internet of Vehicles to obtain a plurality of original elements contained in the original data.

[0037] The above-mentioned original elements may refer to elements that can assist the classification system in classifying the original data to determine whether there is an intrusion attack in the original data, and may include but are not limited to: source IP address (Internet Protocol Address) and destination IP address, source port number and destination port number, transmission protocol type, data packet frequency and size, timestamp and time interval, etc.

[0038] In an alternative embodiment of this embodiment, to accurately classify raw data transmitted within the IoV network and improve its efficiency, the classification system may first identify the raw data transmitted within the IoV network to determine multiple raw elements within the raw data for classifying the raw data. Based on these multiple identified raw elements, the system then accurately detects potential threats within the vehicle network, ensuring vehicle safety while preventing irrelevant elements within the raw data from affecting the classification process. This improves the classification system's efficiency in identifying the raw data's category.

[0039] Step S104: selecting first data from the original data based on the element features of the plurality of original elements.

[0040] The contribution of the multiple first elements contained in the first data to the data classification of the original data is greater than the contribution of the multiple second elements contained in the second data to the data classification of the original data, and the second data is used to represent the data in the original data other than the first data.

[0041] The above-mentioned contribution degree may refer to the importance of different original elements to the classification process of the original data. The greater the contribution degree of the original element, the greater the importance of the original element to the classification process, and the corresponding accuracy of classifying the original data based on the first data containing the original element is higher.

[0042] In an optional solution of this embodiment, considering that there are certain limitations when using the above-mentioned multiple original elements to classify raw data, for example, the original elements may contain a large amount of noise, causing the classification system to over-focus on features unrelated to the data category during the data classification process, thereby affecting the accuracy and efficiency of the classification system in data classification. Therefore, in the actual process of classifying the raw data, the classification system can filter out first data with a high correlation with the data category of the original data from the raw data, and then use the first data to determine the data category of the raw data, thereby improving the accuracy of classifying the raw data. Considering the correlation between the first data and the data category of the original data, it can be reflected in the discriminative power of the multiple first elements contained in the first data and their contribution to the classification process. Therefore, in order to improve the accuracy of the selected first data, the classification system can first extract features from the multiple original elements after identifying them, such as converting timestamps into time intervals, converting communication volume into location-based access frequency, identifying key fields in the data packet header, etc., to obtain element features of the original elements. Then, the original element is evaluated based on the extracted element features to determine the contribution of the original element to data classification of the original data. Finally, the corresponding first data is selected from the original data based on the determined contribution of different original elements.

[0043] Step S106: Input the first data into a feature extraction model to perform feature extraction on the first data to obtain data features of the first data.

[0044] In an optional solution of this embodiment, in order to more deeply understand the attributes of the first data, thereby improving the accuracy of the classification system in classifying the original data based on the first data, after selecting the first data, the classification system can input the first data into a feature extraction model to extract features of the first data using the feature extraction model, thereby obtaining more comprehensive and more accurate data features of the first data, such as dynamic changes in time series, hidden patterns of data packet loads, statistical characteristics of communication patterns, temporal and spatial correlations of geographic location data, and other features, thereby improving the accuracy of the classification system in classifying the original data based on the data features.

[0045] Step S108: input the data features into the data classification model to perform data classification, and obtain the target classification result of the original data.

[0046] The target classification result is used to characterize whether there is a security risk in the original data and the risk type of the security risk.

[0047] In an optional solution of this embodiment, after extracting the data features of the first data, the classification system can input the data features into a data classification model to classify the original data using the data classification model to obtain a target classification result of the original data, thereby using the target classification result to reflect whether there are security risks such as unauthorized access and data tampering in the original data, as well as the risk type of the existing security risks.

[0048] In an embodiment of the present application, the original data transmitted in the Internet of Vehicles is identified to obtain multiple original elements contained in the original data; based on the element features of the multiple original elements, first data is selected from the original data; the first data is input into a feature extraction model to extract features of the first data to obtain data features of the first data; the data features are input into a data classification model to perform data classification to obtain target classification results of the original data. By screening out the first data that contributes more to the data classification process from the original data, and using the data classification model to classify the original data based on the first data, the influence of the second data with lower contribution on the data classification process can be reduced, thereby ensuring the accuracy of the target classification results, and thus solving the technical problem of low accuracy in classifying data transmitted in the Internet of Vehicles.

[0049] Furthermore, based on the element features of multiple original elements, first data is selected from the original data, including: obtaining at least one data center corresponding to the original data, and data distribution corresponding to at least one data center, wherein different data centers correspond to different data categories, and the data categories include: there is no security risk in the original data, there is a security risk in the original data, and when there is a security risk in the original data, the risk type of the security risk, and the data distribution is used to characterize the distribution of data points in at least one data center; based on at least one data center and the data distribution, a first distance between any element feature and multiple first element features, and a second distance between any element feature and multiple second element features are obtained, wherein the first element feature is the same as the data center corresponding to any element feature, and the second element feature is different from the data center corresponding to any element feature; based on the first distance, the second distance, the prior probability of the data center corresponding to any element feature and the data distribution, the feature weight of any element feature is determined; based on the feature weight, a first element is selected from multiple original elements, and the data corresponding to the first element in the original data is determined as the first data.

[0050] In an optional scheme of this embodiment, in order to improve the rationality and accuracy of the filtered first data, the classification system can first cluster the original data according to the data category to which the original data may belong, such as the data category that the original data does not have security risks, the original data has security risks, and the risk type of the security risk when the original data has security risks, to obtain at least one data set, and determine the data centers and data distributions corresponding to different data sets, so as to use the data centers to distinguish different data categories, and use the corresponding data distributions to reflect the distribution of data points in different data centers.

[0051] After clustering different data categories to obtain different data sets, in order to select representative data from the different data sets, that is, to select data from the different data sets that has a high correlation with the data category corresponding to the data set, as the first data, the classification system can determine the first distance between any element feature and multiple first element features, and the second distance between any element feature and multiple second element features based on the data center and data distribution of any data set. The determined first distance and second distance are then used to filter out the above-mentioned first data from the original data, wherein the data center corresponding to the above-mentioned first element feature can be the same as the data center corresponding to any element feature currently requiring distance calculation, and the data center corresponding to the above-mentioned second element feature can be different from the data center corresponding to any element feature currently requiring distance calculation.

[0052] For ease of understanding, assume that there are currently two data sets A and B. When performing distance calculation, the classification system can first select any data point A1 from data set A. The element feature A1 of the original element contained in any data point A1 can be regarded as any of the above-mentioned element features. While determining the element feature A1, the classification system can select other data points from data set A, such as data point A2. The element feature A2 of the original element contained in the data point A2 can be regarded as the above-mentioned first element feature. At this time, by calculating the distance between the element feature A1 and the element feature A2, the first distance corresponding to the two can be obtained. Similarly, the classification system can select a data point B1 from data set B. The element feature B1 of the original element contained in the data point B1 can be regarded as the above-mentioned second element feature. At this time, by calculating the distance between the element feature A1 and the element feature B1, the second distance corresponding to the two can be obtained.

[0053] In an optional solution of this embodiment, after calculating the first distance and the second distance corresponding to any element feature, in order to accurately determine the relevance of any element feature to the data category to which it belongs, the classification system can further introduce the prior probability and data distribution of the data center corresponding to the any element feature, so as to evaluate the element feature by using the first distance and the second distance corresponding to any element feature, as well as the prior probability and data distribution of the corresponding data center, and obtain the feature weight of the element feature, thereby using the feature weight to reflect the importance of the element feature. Finally, after determining the feature weights corresponding to all element features, the classification system can use the feature weights to select a suitable first element from multiple original elements, and determine the data corresponding to the first element in the original data as the above-mentioned first data.

[0054] For ease of understanding, the formula for determining the feature weight of any element feature can be:

[0055]

[0056] Among them, ω(S j ) represents the sample S j The feature weight of ω(S j0 ) represents the initial weight, which can be 0; H i and M i (C) represents sample S j The i-th nearest neighbor of the same type and different types of samples; diff(S j ,R,H i ) represents samples N and H i In feature S jThe difference between the two; m represents the number of iterations of the algorithm; k represents the number of neighboring samples; P(C) represents the probability of labeling C; class(R) represents the sample S j The label of p(class(R)) represents the sample S j The probability of belonging to the label; D c Represents the number of subclusters of category C in the entire dataset; Indicates the weight of the heterogeneous C in the current cluster distribution; diff(S j ,R,M i (C)) represents samples R and M i (C) In feature S j The difference on M i (C) indicates class The i-th nearest neighbor sample in .

[0057] Furthermore, the data classification model includes: a first classification model; inputting data features into the data classification model for data classification to obtain a target classification result of the original data, including: inputting data features into the first classification model for data classification to obtain a first classification result; evaluating the first classification result based on a risk assessment strategy to obtain a classification evaluation result, wherein the classification evaluation result is used to characterize the accuracy of the first classification result; in response to the classification evaluation result being that the accuracy of the first classification result meets a preset accuracy condition, determining the first classification result as the target classification result.

[0058] The above-mentioned first classification model may refer to a pre-trained classification model, which may include but is not limited to: support vector machine, decision tree, convolutional neural network, recurrent neural network or long short-term memory network models. In order to improve classification efficiency, a softmax classifier may be selected as the above-mentioned first classification model.

[0059] In an optional solution of this embodiment, the data classification model may include at least a first classification model. When the data classification model is used to process the data features of the first data, the classification system may directly input the data features into the first classification model to classify the original data based on the data features of the first data using the first classification model to obtain the first classification result. Considering that the final result output by any model may have errors, in order to avoid data classification errors on the original data, after obtaining the first classification result, the classification system may also evaluate the first classification result according to a pre-set risk assessment strategy, such as using a risk assessment model, or output the first classification result in a user interface, and the user actively evaluates the first classification result to obtain a corresponding classification evaluation result, so as to reflect the accuracy of the first classification result using the classification evaluation result. If the classification evaluation result shows that the accuracy of the first classification result meets the preset accuracy condition, such as the accuracy of the first classification result is greater than the preset accuracy threshold, or the user believes that the first classification result is relatively accurate based on the current accuracy, then the accuracy of the first classification result obtained can be considered to be high, and the corresponding classification system can directly determine the first classification result as the target classification result.

[0060] It should be noted that when the softmax classifier is used as the first classification model, in order to ensure the accuracy of classification, the cross entropy loss function can be used as the loss function of the softmax classifier to evaluate the classification error. The cross entropy loss function can be expressed as:

[0061]

[0062] Among them, N represents the number of samples, M is the category; i is the i-th sample; y ic is the label of the i-th sample category C; p ic is the probability that the i-th sample belongs to category c.

[0063] Furthermore, the data classification model also includes: at least one second classification model, and the model types of any one of the first classification model and the at least one second classification model are different; the above method also includes: in response to the classification evaluation result that the accuracy of the first classification result does not meet the preset accuracy condition, inputting the data features into at least one second classification model for data classification to obtain at least one second classification result; based on the first classification result and at least one second classification result, determining the target classification result.

[0064] The above-mentioned second classification model may also refer to a pre-trained classification model. In order to avoid the same processing trajectories between classification models of the same type, which may lead to the same classification errors in the process of classifying the original data, the model types of the first classification model set above and any one of the at least one second classification models are different. For example, the KNN (K-Nearest Neighbors, K-nearest neighbor algorithm) model can be selected as the above-mentioned second classification model.

[0065] In an optional solution of this embodiment, if the classification evaluation result obtained by the evaluation shows that the accuracy of the first classification result does not meet the preset accuracy condition, it can be considered that the accuracy of the current first classification result is low and cannot be directly used as the target classification result. At this time, the classification system can input the extracted data features into at least one second classification model for data classification to obtain at least one second result, and then use the first classification result and at least one second classification result to determine the above-mentioned target classification result. For example, the classification system can also evaluate the obtained second classification result to determine the accuracy of the second classification result, and then determine the classification result with the highest accuracy between the first classification result and the second classification result as the above-mentioned target classification result, thereby ensuring the accuracy of the target classification result.

[0066] Furthermore, based on the first classification result and at least one second classification result, a target classification result is determined, including: grouping at least one second classification result and the first classification result to obtain at least one classification group, wherein any classification results contained in the same classification group are the same; performing feature extraction on any classification group to obtain classification group features of any classification group; based on the classification group features, selecting a target classification group from at least one classification group; and determining the classification results contained in the target classification group as the target classification result.

[0067] In an optional scheme of this embodiment, in order to improve the rationality of determining the target classification result from the first classification result and at least one second classification result, the classification system can first group the above-mentioned at least one second classification result and the first classification result, that is, divide the same classification results into the same classification group to obtain at least one classification group, and then perform feature extraction on any classification group to obtain the classification group features of any classification group, such as the number of classification results contained in any classification group, the average value of accuracy and other features, and finally select the target classification group from at least one classification group based on the extracted classification group features, and determine the classification results contained in the target classification group as the target classification result. For example, the classification system can determine the classification group with the largest number of classification results as the above-mentioned target classification group, or determine the classification group with the largest average value of accuracy as the above-mentioned target classification group, thereby ensuring the rationality of the determined target classification result.

[0068] For ease of understanding, assume there are five classification results: Result 1: No Risk, Result 2: Risk A, Result 3: Risk A, Result 4: Risk A, and Result 5: Risk B. The classification system can then classify Result 1 into the first classification group, Result 234 into the second classification group, and Result 5 into the third classification group. When performing feature extraction on the grouped results, the number of classification results contained in each group can be extracted. The classification group with the largest number, i.e., the second classification group, is selected as the target classification group. The classification result containing Risk A in the corresponding target group can be used as the target classification result.

[0069] Furthermore, the first classification model includes: a self-attention module, a feature matching module and a data classification module; the data features are input into the first classification model for data classification to obtain a first classification result, including: based on the self-attention module, performing self-attention processing on the data features to obtain self-attention features; based on the feature matching module, matching the self-attention features with at least one data category to obtain a feature matching result, wherein the feature matching result is used to characterize the degree of matching between the first data and any data category; based on the data classification module, determining the first classification result according to the feature matching result.

[0070] In an optional scheme of this embodiment, in order to improve the rationality of classifying the original data, the first classification model may include at least: a self-attention module, a feature matching module and a data classification module. Correspondingly, when using the first classification model to classify the original data, the classification system can first use the self-attention module to perform self-attention processing on the data features to obtain the corresponding self-attention features, and then use the feature matching module to match the self-attention features and at least one data category to determine the degree of matching between the first data and any data category, thereby obtaining the corresponding feature matching results, and finally use the data classification module to determine the corresponding first classification result according to the matched feature matching results. For example, the classification system can use the feature matching result with the highest matching degree as the above-mentioned first classification result, or can display the feature matching results to the user in the user interface, and then the user determines the above-mentioned first classification result.

[0071] Furthermore, the first classification result is evaluated based on the risk assessment strategy to obtain a classification evaluation result, including: obtaining a feature matching result corresponding to the first classification result; evaluating the first classification result based on the feature matching result to obtain an evaluation index of the first classification result; in response to the evaluation index satisfying a preset evaluation condition, determining that the classification evaluation result is that the accuracy of the first classification result satisfies the preset accuracy condition; in response to the evaluation index not satisfying the preset evaluation condition, determining that the classification evaluation result is that the accuracy of the first classification result does not satisfy the preset accuracy condition.

[0072] In an optional scheme of this embodiment, in order to ensure the accuracy of the evaluation of the first classification result, the classification system can first obtain the feature matching result corresponding to the first classification result, and then use the feature matching result to evaluate the first classification result to obtain the corresponding evaluation index. For example, the classification system can convert the matching degree corresponding to the feature matching result into an evaluation score of the first classification result. The higher the matching degree, the higher the evaluation score, which means that the accuracy of the first classification result is higher. Alternatively, a probability model can be used to determine the probability that the original data belongs to the corresponding data category based on the feature matching result. The higher the matching degree, the higher the probability that the original data belongs to the corresponding quantity category, and the higher the accuracy of the first classification result. Correspondingly, if the obtained evaluation index can meet the preset evaluation conditions, the classification system can determine that the classification evaluation result is that the accuracy of the first classification result meets the preset accuracy conditions; if the obtained evaluation index does not meet the preset evaluation conditions, the classification system can classify the evaluation result as that the accuracy of the first classification result does not meet the preset accuracy conditions. For example, if the converted evaluation score is greater than or equal to the preset score threshold, it can be considered that the evaluation index can meet the preset evaluation conditions; if the converted evaluation score is less than the preset score threshold, it can be considered that the evaluation index cannot meet the preset evaluation conditions.

[0073] Furthermore, the feature extraction model includes: a convolution feature extraction module, a convolution feature conversion module, a forward time series feature extraction module, a backward time series feature extraction module and a time series feature fusion module; the first data is input into the feature extraction model to perform feature extraction on the first data to obtain data features of the first data, including: using the convolution feature extraction module to perform feature extraction on the first data to obtain the convolution features of the first data; using the convolution feature conversion module to convert the convolution features to obtain time series data corresponding to the convolution features; using the forward time series feature extraction module to perform feature extraction on the time series data to obtain forward time series features, and using the backward time series feature extraction module to perform feature extraction on the time series data to obtain backward time series features; using the time series feature fusion module to fuse the forward time series features and the backward time series features to obtain data features.

[0074] In an optional solution of this embodiment, in order to ensure the accuracy of the data features obtained by extracting features from the first data, the feature extraction model may include at least the above-mentioned convolution feature extraction module, convolution feature conversion module, forward time series feature extraction module, backward time series feature extraction module and time series feature fusion module. Correspondingly, when extracting features from the first data, the classification system may first use the convolution feature extraction module to extract features from the first data to obtain the convolution features of the first data, and then use the convolution feature conversion module to convert the convolution features to obtain the time series data corresponding to the convolution features. Then, the forward time series feature extraction module is used to extract features from the time series data to obtain forward time series features, and the backward time series feature extraction module is used to extract features from the time series data to obtain backward time series features. Finally, the forward time series features and the backward time series features are fused using the time series feature fusion module to obtain data features with higher accuracy.

[0075] For ease of understanding, Figure 2 is a schematic diagram of a feature extraction process according to an embodiment of the present application, such as Figure 2 The feature extraction model for extracting features from the first data may be a CNN-BiLSTM model (Convolutional Neural Network-Bidirectional Long Short-Term Memory Network, a composite model of a convolutional neural network and a bidirectional long short-term memory network), which may be divided into a convolutional neural network CNN part and a bidirectional long short-term memory network BiLSTM part.

[0076] When extracting spatial features in the convolutional neural network part, the classification system can first take the first data as input in the form of 11*11, and perform 11*11*32 convolution processing, 5*5*32 pooling processing, 2*2*64 convolution processing and other operations on the first data in sequence to convolve the first data and obtain a 1*1*64 output Ouput, that is, the spatial feature of the first data, and then normalize the convolution feature to obtain the time series data corresponding to the convolution feature as the input of the bidirectional long short-term memory network BiLSTM.

[0077] When extracting time series features in a bidirectional long short-term memory network, an input layer, a first hidden layer, a second hidden layer, and an output layer connected in sequence can be configured in the bidirectional long short-term memory network, wherein the input layer can receive the time series data output by the convolutional neural network part, the first hidden layer and the second hidden layer can extract time series features from the time series data, the output layer can output the extracted time series features, and finally the time series features output by the output layer are fused to obtain a feature result with higher accuracy, i.e., the data features of the first data. In any hidden layer, the classification system can perform forward time series feature extraction and backward time series feature extraction operations on the input of any hidden layer, respectively, to obtain forward time series features and backward time series features with higher accuracy. In the output layer,

[0078] Specifically, suppose the input data sequence is X=[x1,x2,··,x n ], where x i is the embedding representation of the i-th in the sequence. In the process of feature extraction of convolutional neural network, the classification system can first transform the data sequence through the embedding layer to obtain the corresponding feature vector E = [e1, e2, ..., e n ], and then convolve the feature vector E through d convolution kernels of different sizes to obtain the new feature F = [f1, f2, ..., f d ], where f j It is obtained by convolution of the feature vector E with the jth convolution kernel. After obtaining the new feature F, the classification system can use the Relu function to perform nonlinear mapping on the new feature F to obtain an output feature sequence A = [a1, a2, ..., a d After all convolution kernels have completed feature extraction, the extracted features are integrated to obtain a new feature sequence V. In order to further extract global features and reduce the number of model parameters, the classification system can also adopt an average pooling strategy to divide the feature sequence V into several sub-matrices of the same size and perform an average pooling operation on each sub-matrix to obtain D average values ​​P = [p1, p2, ..., p D ].

[0079] In the process of performing feature extraction of the bidirectional long short-term memory network, the classification system can use the forward LSTM layer and the reverse LSTM layer to process the output of the convolutional neural network, and then merge the processing results of the two to form a comprehensive hidden state y t .

[0080]

[0081]

[0082]

[0083] In the bidirectional long short-term memory network, the classification system can also use the Sigmoid function and the tanh function as activation functions. The corresponding Sigmoid function formula can be:

[0084]

[0085] The formula for the tanh function can be:

[0086]

[0087] The sigmoid function strictly limits output values ​​to between [0, 1], increasing the model's nonlinear expressiveness and helping the network learn complex data patterns. The tanh function can be used to generate candidate memory cell states, better representing the dynamic changes in data.

[0088] To facilitate understanding of the above classification process, Figure 3 is a flow chart of a data classification method according to an embodiment of the present application. Figure 3As shown, in the process of classifying the original data, the classification system can first preprocess the original data, such as performing digitization, normalization, and standardization operations on the original data in sequence to obtain preprocessed data, and screening the preprocessed data to obtain first data, and then input the first data into the feature extraction model to use the feature extraction model to extract the data features of the first data, and finally use the first classification model to perform self-attention processing, feature matching, data classification and other processing operations on the first data in sequence to obtain the corresponding first classification result, so as to preliminarily determine the data category to which the original data belongs. After obtaining the first classification result, the classification system can also use a three-branch decision classifier to evaluate the first classification result to determine whether the first classification result belongs to the boundary domain, that is, whether the accuracy of the first classification result can meet the preset accuracy condition. If the accuracy meets the preset accuracy condition, it can be considered that the first classification result does not belong to the boundary domain. The classification system can directly determine the first classification result as the target classification result to determine the data category to which the original data belongs; if the accuracy does not meet the preset accuracy condition, it can be considered that the first classification result belongs to the boundary domain. At this time, the classification system can perform boundary domain classification on the original data, that is, use at least one second classification model to reclassify the original data to obtain at least one second classification result, and then determine the target classification result based on the first classification result and at least one second classification result to accurately determine the data category to which the original data belongs.

[0089] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0090] According to an embodiment of the present application, a device embodiment of a data classification method is provided. It should be noted that the device can be used to execute the above-mentioned data classification method.

[0091] Figure 4 is a structural block diagram of a data classification device according to an embodiment of the present application. Figure 4 As shown, the device includes: a data identification module 402, a data selection module 404, a feature extraction module 406 and a data classification module 408.

[0092] Among them, the data identification module 402 is used to identify the original data transmitted in the Internet of Vehicles and obtain multiple original elements contained in the original data; the data selection module 404 is used to select the first data from the original data based on the element features of the multiple original elements, wherein the contribution of the multiple first elements contained in the first data to the data classification of the original data is greater than the contribution of the multiple second elements contained in the second data to the data classification of the original data, and the second data is used to characterize the data other than the first data in the original data; the feature extraction module 406 is used to input the first data into the feature extraction model to extract the features of the first data and obtain the data features of the first data; the data classification module 408 is used to input the data features into the data classification model to perform data classification and obtain the target classification results of the original data, wherein the target classification results are used to characterize whether there are security risks in the original data and the risk type of the security risks.

[0093] Furthermore, the data selection module is also used to: obtain at least one data center corresponding to the original data, and the data distribution corresponding to at least one data center, wherein different data centers correspond to different data categories, and the data categories include: there is no security risk in the original data, there is a security risk in the original data, and when there is a security risk in the original data, the risk type of the security risk, and the data distribution is used to characterize the distribution of data points in at least one data center; based on at least one data center and the data distribution, obtain a first distance between any element feature and multiple first element features, and a second distance between any element feature and multiple second element features, wherein the first element feature is the same as the data center corresponding to any element feature, and the second element feature is different from the data center corresponding to any element feature; determine the feature weight of any element feature based on the first distance, the second distance, the prior probability of the data center corresponding to any element feature and the data distribution; based on the feature weight, select a first element from multiple original elements, and determine the data corresponding to the first element in the original data as the first data.

[0094] Furthermore, the data classification model includes: a first classification model; the data classification module is also used to: input data features into the first classification model for data classification to obtain a first classification result; evaluate the first classification result based on a risk assessment strategy to obtain a classification evaluation result, wherein the classification evaluation result is used to characterize the accuracy of the first classification result; in response to the classification evaluation result being that the accuracy of the first classification result meets a preset accuracy condition, determine that the first classification result is a target classification result.

[0095] Furthermore, the data classification model also includes: at least one second classification model, and the model types of any one of the first classification model and the at least one second classification model are different; the above-mentioned device also includes: a reclassification module, which is used to input data features into at least one second classification model for data classification in response to the classification evaluation result that the accuracy of the first classification result does not meet the preset accuracy conditions, and obtain at least one second classification result; a result determination module, which is used to determine the target classification result based on the first classification result and at least one second classification result.

[0096] Furthermore, the result determination module is also used to: group at least one second classification result and the first classification result to obtain at least one classification group, wherein any classification results contained in the same classification group are the same; perform feature extraction on any classification group to obtain classification group features of any classification group; based on the classification group features, select a target classification group from at least one classification group; and determine the classification results contained in the target classification group as the target classification results.

[0097] Furthermore, the first classification model includes: a self-attention module, a feature matching module and a data classification module; the data classification module is also used to: based on the self-attention module, perform self-attention processing on the data features to obtain self-attention features; based on the feature matching module, match the self-attention features with at least one data category to obtain a feature matching result, wherein the feature matching result is used to characterize the degree of matching between the first data and any data category; based on the data classification module, determine the first classification result according to the feature matching result.

[0098] Furthermore, the data classification module is also used to: obtain a feature matching result corresponding to the first classification result; evaluate the first classification result based on the feature matching result to obtain an evaluation index of the first classification result; in response to the evaluation index satisfying a preset evaluation condition, determine that the classification evaluation result is that the accuracy of the first classification result satisfies the preset accuracy condition; in response to the evaluation index not satisfying the preset evaluation condition, determine that the classification evaluation result is that the accuracy of the first classification result does not satisfy the preset accuracy condition.

[0099] Furthermore, the feature extraction model includes: a convolution feature extraction module, a convolution feature conversion module, a forward time series feature extraction module, a backward time series feature extraction module and a time series feature fusion module; the feature extraction module is also used to: use the convolution feature extraction module to extract features of the first data to obtain the convolution features of the first data; use the convolution feature conversion module to convert the convolution features to obtain time series data corresponding to the convolution features; use the forward time series feature extraction module to extract features of the time series data to obtain forward time series features, and use the backward time series feature extraction module to extract features of the time series data to obtain backward time series features; use the time series feature fusion module to fuse the forward time series features and the backward time series features to obtain data features.

[0100] An embodiment of the present application further provides a vehicle, comprising: a memory storing an executable program; and a processor for running the program, wherein the method of each embodiment of the present application is executed when the program is running.

[0101] An embodiment of the present application further provides a computer-readable storage medium, which includes a stored executable program, wherein when the executable program is running, the device where the computer-readable storage medium is located is controlled to execute the methods in various embodiments of the present application.

[0102] An embodiment of the present application further provides a computer program product, including a computer program, which implements the methods in various embodiments of the present application when executed by a processor.

[0103] An embodiment of the present application further provides a computer program product, including a non-volatile computer-readable storage medium, wherein the non-volatile computer-readable storage medium is used to store a computer program, and when the computer program is executed by a processor, the method in each embodiment of the present application is implemented.

[0104] The embodiments of the present application further provide a computer program, which, when executed by a processor, implements the methods in the above-mentioned embodiments of the present application.

[0105] In the above embodiments of the present application, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, please refer to the relevant description of other embodiments.

[0106] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only exemplary. For example, the division of units can be a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0107] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple units. Some or all of the units may be selected to achieve the purpose of the present embodiment according to actual needs.

[0108] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0109] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program code.

[0110] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A data classification method, characterized in that: include: Identifying original data transmitted in the Internet of Vehicles to obtain multiple original elements contained in the original data; selecting first data from the original data based on element features of the multiple original elements, wherein contribution of the multiple first elements included in the first data to data classification of the original data is greater than contribution of the multiple second elements included in the second data to data classification of the original data, and the second data is used to represent data in the original data other than the first data; Inputting the first data into a feature extraction model to extract features of the first data to obtain data features of the first data; The data features are input into a data classification model for data classification to obtain a target classification result of the original data, wherein the target classification result is used to characterize whether there is a security risk in the original data and the risk type of the security risk.

2. The method according to claim 1, characterized in that Selecting first data from the original data based on element features of the plurality of original elements includes: Obtaining at least one data center corresponding to the original data and a data distribution corresponding to the at least one data center, wherein different data centers correspond to different data categories, and the data categories include: the original data does not have a security risk, the original data does have a security risk, and, if the original data does have a security risk, the risk type of the security risk; the data distribution is used to characterize the distribution of data points in the at least one data center; Based on the at least one data center and the data distribution, obtaining a first distance between any element feature and a plurality of first element features, and a second distance between the any element feature and a plurality of second element features, wherein the first element feature and the data center corresponding to the any element feature are the same, and the second element feature and the data center corresponding to the any element feature are different; Determining a feature weight of the any element feature based on the first distance, the second distance, a priori probability of a data center corresponding to the any element feature, and data distribution; Based on the feature weight, a first element is selected from the multiple original elements, and data corresponding to the first element in the original data is determined as the first data.

3. The method according to claim 1, characterized in that The data classification model includes: a first classification model; inputting the data features into the data classification model to perform data classification, and obtaining the target classification result of the original data, including: Inputting the data features into a first classification model to perform data classification to obtain a first classification result; Evaluating the first classification result based on a risk assessment strategy to obtain a classification evaluation result, wherein the classification evaluation result is used to characterize the accuracy of the first classification result; In response to the classification evaluation result indicating that the accuracy of the first classification result meets a preset accuracy condition, the first classification result is determined to be the target classification result.

4. The method according to claim 3, characterized in that The data classification model further includes: at least one second classification model, wherein the model type of the first classification model and any one of the at least one second classification model is different; the method further includes: In response to the classification evaluation result indicating that the accuracy of the first classification result does not meet the preset accuracy condition, inputting the data features into at least one second classification model for data classification to obtain at least one second classification result; A target classification result is determined based on the first classification result and the at least one second classification result.

5. The method according to claim 4, characterized in that Determining a target classification result based on the first classification result and the at least one second classification result includes: Grouping the at least one second classification result and the first classification result to obtain at least one classification group, wherein any classification results included in the same classification group are the same; Performing feature extraction on any classification group to obtain a classification group feature of the any classification group; selecting a target classification group from the at least one classification group based on the classification group characteristics; The classification results included in the target classification group are determined as the target classification results.

6. The method according to claim 3, characterized in that The first classification model includes: a self-attention module, a feature matching module and a data classification module; the data features are input into the first classification model for data classification to obtain a first classification result, including: Based on the self-attention module, performing self-attention processing on the data feature to obtain a self-attention feature; Based on the feature matching module, matching the self-attention feature with at least one data category to obtain a feature matching result, wherein the feature matching result is used to represent the matching degree between the first data and any data category; Based on the data classification module, the first classification result is determined according to the feature matching result.

7. The method according to claim 3, characterized in that The first classification result is evaluated based on the risk assessment strategy to obtain a classification assessment result, including: Obtaining a feature matching result corresponding to the first classification result; Evaluating the first classification result based on the feature matching result to obtain an evaluation index of the first classification result; In response to the evaluation indicator satisfying a preset evaluation condition, determining that the classification evaluation result is that the accuracy of the first classification result satisfies the preset accuracy condition; In response to the evaluation indicator not satisfying a preset evaluation condition, determining the classification evaluation result as an accuracy of the first classification result not satisfying the preset accuracy condition.

8. The method according to claim 1, characterized in that The feature extraction model includes: a convolution feature extraction module, a convolution feature conversion module, a forward time series feature extraction module, a backward time series feature extraction module and a time series feature fusion module; the first data is input into the feature extraction model to extract the first data to obtain data features of the first data, including: Using the convolution feature extraction module to perform feature extraction on the first data to obtain convolution features of the first data; The convolution feature conversion module is used to convert the convolution feature to obtain time series data corresponding to the convolution feature; Using the forward time series feature extraction module to extract features from the time series data to obtain forward time series features, and using the backward time series feature extraction module to extract features from the time series data to obtain backward time series features; The forward time series features and the backward time series features are fused using the time series feature fusion module to obtain the data features.

9. A vehicle, characterized in that: include: a memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 8 when running.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored executable program, wherein when the executable program is run, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 8.

Images