Flow feature extraction method and device, medium and equipment
By using the attention mechanism and kernel function complementarity to perform feature extraction in the network management control system, the problem of inaccurate feature extraction caused by limited computing power is solved, and accurate and fast feature data extraction is achieved.
Patent Information
- Application Number
- CN202510742010.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-05
- Publication Date
- 2025-09-26
AI Technical Summary
The feature extraction method based on neural network has the problem of inaccurate feature extraction due to limited computing power in network management control systems.
The attention mechanism is used for feature screening, combined with the predetermined target linear kernel function and target nonlinear kernel function, and feature extraction is performed through the complementary method of different kernel functions.
It achieves accurate and rapid extraction of feature data from target traffic data, improves the accuracy of feature extraction, and lays the foundation for subsequent abnormal traffic detection.
Smart Images
Figure CN120710892A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a flow feature extraction method, device, medium and equipment. Background Art
[0002] Traffic feature extraction for optical transmission equipment network management involves extracting relevant features from the management process of network elements (NEs). This is done to better describe and analyze the behavior and performance of NEs during control and monitoring. Common methods for extracting features from NMS data include neural network-based feature extraction.
[0003] However, when it comes to network management and control system scenarios, neural network-based feature extraction methods cannot fully meet the computing power requirements of such methods due to the limited computing power of network management-related equipment. At the same time, there is also the problem of inaccurate feature extraction. Summary of the Invention
[0004] In view of this, the present invention provides a flow feature extraction method, device, medium and equipment, the main purpose of which is to solve the current problem of inaccurate flow feature extraction.
[0005] To solve the above problems, this application provides a traffic feature extraction method, including:
[0006] For the target traffic data to be extracted, feature screening is performed using the attention mechanism to obtain initial feature data corresponding to the target traffic data;
[0007] For the initial feature data, using a predetermined target linear kernel function, extracting first feature data corresponding to the target flow data;
[0008] For the initial feature data, using a predetermined target nonlinear kernel function, extracting second feature data corresponding to the target flow data;
[0009] Target characteristic data corresponding to the target flow data is determined based on the first characteristic data and the second characteristic data.
[0010] Optionally, the feature screening of the target traffic data to be extracted using an attention mechanism to obtain initial feature data corresponding to the target traffic data specifically includes:
[0011] Creating an initial matrix corresponding to the target flow data for the target flow data, wherein the initial matrix includes a plurality of eigenvectors and a plurality of eigenvalues;
[0012] By using the attention weights in the predetermined target attention function, each eigenvalue and each eigenvector in the initial matrix are weighted to obtain an initial eigenvalue matrix and an initial eigenvector matrix to obtain the initial feature data.
[0013] Optionally, extracting first feature data corresponding to the target flow data using a predetermined target linear kernel function for the initial feature data specifically includes:
[0014] Based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, the target linear kernel function is used to extract the first eigenvalue matrix corresponding to the initial eigenvalue matrix and the first eigenvector matrix corresponding to the initial eigenvector matrix to obtain the first feature data.
[0015] Optionally, for the initial feature data, using a predetermined target nonlinear kernel function to extract and obtain second feature data corresponding to the target flow data, specifically includes:
[0016] Based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, the target nonlinear kernel function is used to extract the second eigenvalue matrix corresponding to the initial eigenvalue matrix and the second eigenvector matrix corresponding to the initial eigenvector matrix to obtain the second feature data.
[0017] Optionally, determining target characteristic data corresponding to the target flow data based on the first characteristic data and the second characteristic data specifically includes:
[0018] Based on a first weight value corresponding to a target linear kernel function and a second weight value corresponding to a target nonlinear kernel function, the first feature data and the second feature data are merged to obtain the target feature data.
[0019] Optionally, the method further includes predetermining the target attention function, the target linear kernel function, and the target nonlinear kernel function, specifically including:
[0020] Collecting a number of sample flow data and label feature data corresponding to each sample flow data;
[0021] Perform feature screening on each sample traffic data based on the initial attention function to obtain initial sample feature data corresponding to each sample traffic data;
[0022] For each of the initial sample characteristic data, using an initial linear kernel function, extracting first sample characteristic data corresponding to each of the sample flow data;
[0023] For each of the initial sample characteristic data, using an initial nonlinear kernel function, extracting second sample characteristic data corresponding to each of the sample flow data;
[0024] Determining target sample characteristic data corresponding to each of the sample flow data based on each of the first sample characteristic data and each of the second sample characteristic data;
[0025] Based on the target sample feature data corresponding to each sample flow data and the label feature data corresponding to each sample flow data, the parameters in the initial attention function, the initial linear kernel function and the initial target nonlinear kernel function are adjusted until the target attention function, the target linear kernel function and the target nonlinear kernel function are obtained when it is determined based on the target sample feature data that the predetermined conditions are met.
[0026] Optionally, after obtaining the target characteristic data, the method further includes: reviewing the target characteristic data to obtain reviewed target characteristic data.
[0027] To solve the above problems, the present application provides a flow feature extraction device, comprising:
[0028] A screening module is used to perform feature screening on the target traffic data to be extracted using an attention mechanism to obtain initial feature data corresponding to the target traffic data;
[0029] A first extraction module is configured to extract first feature data corresponding to the target flow data using a predetermined target linear kernel function for the initial feature data;
[0030] A second extraction module is configured to extract, from the initial feature data, second feature data corresponding to the target flow data using a predetermined target nonlinear kernel function;
[0031] A determination module is used to determine target characteristic data corresponding to the target flow data based on the first characteristic data and the second characteristic data.
[0032] To solve the above problems, the present application provides a storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the steps of any of the above-mentioned traffic feature extraction methods.
[0033] To solve the above problems, the present application provides an electronic device, which includes at least a memory and a processor, wherein a computer program is stored on the memory, and the processor implements the steps of any of the above-mentioned traffic feature extraction methods when executing the computer program on the memory.
[0034] The traffic feature extraction method, device, medium and equipment in this application first obtain initial feature data by performing preliminary feature screening on the traffic data based on the attention mechanism, and then further extract features from the initial feature data in combination with a predetermined target linear kernel function and a target nonlinear kernel function, that is, feature analysis and extraction are performed through the complementary manner of different kernel functions, which can accurately and quickly extract target feature data from the target traffic data.
[0035] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are specifically listed below. BRIEF DESCRIPTION OF THE DRAWINGS
[0036] Various other advantages and benefits will become apparent to those skilled in the art upon reading the detailed description of the preferred embodiment below. The accompanying drawings are for illustration purposes only and are not to be considered as limiting the present invention. The same reference symbols are used throughout the drawings to represent the same components. In the drawings:
[0037] Figure 1 This is a flow chart of a method for extracting traffic characteristics according to an embodiment of the present application;
[0038] Figure 2 This is a flow chart of a method for extracting traffic characteristics according to another embodiment of the present application;
[0039] Figure 3 This is a structural block diagram of a flow feature extraction device according to another embodiment of the present application;
[0040] Figure 4 This is a structural block diagram of an electronic device according to another embodiment of the present application. DETAILED DESCRIPTION
[0041] Various aspects and features of the present application are described herein with reference to the accompanying drawings.
[0042] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but merely as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.
[0043] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.
[0044] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.
[0045] It should also be understood that although the present application has been described with reference to certain specific examples, those skilled in the art will readily be able to implement many other equivalent forms of the present application.
[0046] The above and other aspects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings.
[0047] Specific embodiments of the present application will be described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments described are merely examples of the present application and may be implemented in a variety of ways. Familiar and / or repetitive functions and structures are not described in detail to avoid obscuring the present application with unnecessary or redundant details. Therefore, the specific structural and functional details described herein are not intended to be limiting, but rather serve merely as a basis and representative basis for the claims to teach those skilled in the art to variously utilize the present application with substantially any suitable detailed structure.
[0048] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," which may all refer to one or more of the same or different embodiments according to the present application.
[0049] The present application embodiment provides a method for extracting traffic characteristics, such as Figure 1 As shown, including:
[0050] Step S101: For the target traffic data to be extracted, feature screening is performed using an attention mechanism to obtain initial feature data corresponding to the target traffic data;
[0051] In this step, the target traffic data can specifically be traffic data / service data of the optical transmission equipment. In a specific implementation, an initial matrix can be constructed based on the target traffic data. This matrix contains a number of eigenvectors and a number of eigenvalues. Then, using the attention weights in a predetermined attention function, each eigenvalue and each eigenvector in the initial matrix can be weighted to obtain an initial eigenvalue matrix and an initial eigenvector matrix, thereby obtaining the initial feature data.
[0052] Step S102: extracting first feature data corresponding to the target flow data using a predetermined target linear kernel function for the initial feature data;
[0053] During the specific implementation of this step, the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data can be respectively input into the target linear kernel function, and the target linear kernel function can be used to output the first eigenvalue matrix corresponding to the initial eigenvalue matrix and the first eigenvector matrix corresponding to the initial eigenvector matrix to obtain the first feature data.
[0054] Step S103, extracting second feature data corresponding to the target flow data using a predetermined target nonlinear kernel function for the initial feature data;
[0055] In this step, the target nonlinear kernel function may specifically be a radial basis kernel function. In a specific implementation process, the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data may be inputted into the target nonlinear kernel function / radial basis kernel function, respectively. The target nonlinear kernel function / radial basis kernel function may be used to output a second eigenvalue matrix corresponding to the initial eigenvalue matrix and a second eigenvector matrix corresponding to the initial eigenvector matrix, respectively, to obtain the second feature data.
[0056] Step S104: determining target characteristic data corresponding to the target flow data based on the first characteristic data and the second characteristic data.
[0057] During the specific implementation of this step, the first feature data and the second feature data may be merged according to the weight values corresponding to the target linear kernel function and the target nonlinear kernel function, thereby obtaining the target feature data.
[0058] The traffic feature extraction method, device, medium and equipment in this application first obtain initial feature data by performing preliminary feature screening on the traffic data based on the attention mechanism, and then further extract features from the initial feature data in combination with a predetermined target linear kernel function and a target nonlinear kernel function, that is, feature analysis and extraction are performed through the complementary manner of different kernel functions, which can accurately and quickly extract target feature data from the target traffic data.
[0059] Based on the above embodiment, another embodiment of the present application provides a flow feature extraction method, including the following steps:
[0060] Step S201: creating an initial matrix corresponding to the target flow data, wherein the initial matrix includes a plurality of eigenvectors and a plurality of eigenvalues;
[0061] In this step, an initial matrix M can be pre-constructed for the target traffic data. Specifically, since the traffic data contains several data elements, such as protocol header information, protocol type identifier, timestamp, source address, destination address, and so on, these data elements can be vectorized and used as matrix elements to construct an initial matrix M containing several eigenvectors. Simultaneously, the eigenvalues corresponding to the eigenvectors in this initial matrix can be determined based on the non-zero eigenvectors. Therefore, the initial matrix contains several eigenvectors and several eigenvalues.
[0062] Step S202, using the attention weights in the predetermined target attention function, weighting each eigenvalue and each eigenvector in the initial matrix to obtain an initial eigenvalue matrix and an initial eigenvector matrix to obtain the initial feature data;
[0063] In this step, the attention weight can be obtained in advance through training.
[0064] Step S203, based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, respectively use the target linear kernel function to extract the first eigenvalue matrix corresponding to the initial eigenvalue matrix and the first eigenvector matrix corresponding to the initial eigenvector matrix to obtain the first feature data.
[0065] In the specific implementation of this step, the target linear kernel function formula can be shown as the following formula (1):
[0066]
[0067] Among them, x i 、y j are two n-dimensional vectors in the input space; x ik and y jk Represents x i and y j The kth vector of ; c1 is a constant term obtained through training and used to control the complexity of the model.
[0068] In this step, by inputting the initial eigenvector matrix into the above formula (1), the first eigenvector matrix corresponding to the initial eigenvector matrix can be output. Similarly, by inputting the initial eigenvalue matrix into the above formula (1), the first eigenvalue matrix corresponding to the initial eigenvalue matrix can be output.
[0069] Step S203, based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, respectively use the target nonlinear kernel function to extract the second eigenvalue matrix corresponding to the initial eigenvalue matrix and the second eigenvector matrix corresponding to the initial eigenvector matrix to obtain the second feature data.
[0070] In the specific implementation of this step, the target nonlinear kernel function formula can be shown as the following formula (2):
[0071] K2(x i ,y j )=(γ×x i ×y j +c2) d (2)
[0072] Among them, x i 、y j are two n-dimensional vectors in the input space; γ is an adjustable hyperparameter, obtained in advance through training, used to scale the inner product value of the vector (γ = 1 can be taken in this embodiment); c2 is a constant term, obtained in advance through training, used to control the complexity of the model to avoid the model being too complex (overfitting) in this article; d is the order of the kernel function, obtained in advance through training, which determines the degree of nonlinearity of the model (d = 2 can be taken in this embodiment).
[0073] In the specific implementation process of this step, the initial eigenvector matrix can be input into the above formula (2) to output the second eigenvector matrix corresponding to the initial eigenvector matrix. Similarly, the initial eigenvalue matrix can be input into the above formula (2) to output the second eigenvalue matrix corresponding to the initial eigenvalue matrix.
[0074] Step S204: Based on the first weight value corresponding to the target linear kernel function and the second weight value corresponding to the target nonlinear kernel function, the first feature data and the second feature data are merged to obtain the target feature data;
[0075] In this step, the first weight value δ and the second weight value ω can be pre-set and adjusted according to actual needs, and then the first feature data and the second feature data can be merged using the following formula (3).
[0076] K f =δK1+ωK2(3)
[0077] Among them, K fis the merged target feature data; K1 is the first feature data; K2 is the second feature data; δ is the first weight value corresponding to the target linear kernel function; ω is the second weight value corresponding to the target nonlinear kernel function. In this embodiment, combined with the actual structure of the optical transmission application and the properties of the data packet transmission, δ = 0.6 and ω = 0.4 can be taken.
[0078] During the specific implementation of this embodiment, the first eigenvector matrix in the first feature data and the second eigenvector matrix in the second feature data can be merged based on the first weight value δ and the second weight value ω, respectively, to obtain a target eigenvector matrix. Similarly, the first eigenvalue matrix in the first feature data and the second eigenvalue matrix in the second feature data can be merged based on the first weight value δ and the second weight value ω, respectively, to obtain a target eigenvalue matrix. Finally, based on the target eigenvector matrix and the target eigenvalue matrix, the target feature data is obtained. In this embodiment, by adopting a multi-kernel function complementary method for feature extraction, the loss of eigenvalues can be avoided, the accuracy of feature extraction can be improved, and the foundation for subsequent abnormal traffic identification based on the extracted feature data is laid.
[0079] Step S205: review the target feature data to obtain reviewed target feature data.
[0080] During the specific implementation of this step, after obtaining the target feature data, the target feature data may be further filtered and screened based on predetermined rules or manually to obtain the target feature data.
[0081] In the specific implementation process of this embodiment, before executing step S201, the initial attention function, the initial linear kernel function and the initial nonlinear kernel function can also be trained to obtain the target attention function, the target linear kernel function and the target nonlinear kernel function. The specific process can be combined with Figure 2 As shown, the following steps are included:
[0082] Step 1: Collect a number of sample traffic data and label feature data corresponding to each sample traffic data;
[0083] Step 2: Perform feature screening on each sample traffic data based on the initial attention function to obtain initial sample feature data corresponding to each sample traffic data;
[0084] During the specific implementation of this step, an initial sample matrix corresponding to each sample flow data / data packet may be constructed, wherein the initial sample matrix includes a plurality of eigenvectors and a plurality of eigenvalues;
[0085] Then, the initial attention weights in the initial attention function are used to perform weighted processing on each eigenvalue and each eigenvector in the initial sample matrix to obtain an initial sample eigenvalue matrix and an initial sample eigenvector matrix to obtain the initial sample feature data.
[0086] In this step, each initial sample matrix M contains n query vectors Query with a feature dimension of d, and also contains a key value matrix K and a eigenvalue matrix V with the same dimension.
[0087] Mapping can be performed by using the attention function. The specific process can be expressed as:
[0088]
[0089] Among them, QK is used to measure the similarity between each query vector Query and the key information vector Key. is the scaling factor, and the attention score is normalized by the Softmax function to obtain the weight coefficient. The larger the weight coefficient, the greater the proportion of the corresponding eigenvalue vector in the output result.
[0090] That is, the similarity or correlation between the initialized query vector Query and the key information vector Key is calculated to obtain the original score; then the original score calculated in the previous step is normalized; and then weighted according to the weight coefficient set by the initialization is performed to obtain the initial sample feature data.
[0091] Step 3: for each of the initial sample characteristic data, using an initial linear kernel function, extract and obtain first sample characteristic data corresponding to each of the sample flow data;
[0092] During the specific implementation of this step, based on the initial sample eigenvalue matrix and the initial sample eigenvector matrix in the initial sample feature data, the initial linear kernel function is used to extract the first sample eigenvalue matrix corresponding to the initial sample eigenvalue matrix and the first sample eigenvector matrix corresponding to the initial sample eigenvector matrix to obtain the first sample feature data.
[0093] Step 4: for each of the initial sample characteristic data, using an initial nonlinear kernel function, extract and obtain second sample characteristic data corresponding to each of the sample flow data;
[0094] During the specific implementation of this step, based on the initial sample eigenvalue matrix and the initial sample eigenvector matrix in the initial sample feature data, the initial nonlinear kernel function is used to extract the second sample eigenvalue matrix corresponding to the initial sample eigenvalue matrix and the second sample eigenvector matrix corresponding to the initial sample eigenvector matrix to obtain the second sample feature data.
[0095] Step 5: determining target sample characteristic data corresponding to each sample flow data based on each of the first sample characteristic data and each of the second sample characteristic data;
[0096] During the specific implementation of this step, the first sample eigenvector matrix and the second sample eigenvector matrix can be merged based on the initial first weight value δ corresponding to the target linear kernel function and the initial second weight value ω corresponding to the target nonlinear kernel function to obtain the target sample eigenvector matrix. Similarly, the first sample eigenvalue matrix and the second sample eigenvalue matrix can be merged based on the initial first weight value δ corresponding to the target linear kernel function and the initial second weight value ω corresponding to the target nonlinear kernel function to obtain the target sample eigenvalue matrix. Finally, the characteristic data of each target sample can be determined based on each target sample eigenvector matrix and each target sample eigenvalue matrix.
[0097] Step six: Based on the target sample feature data corresponding to each sample flow data and the label feature data corresponding to each sample flow data, the parameters in the initial attention function, the initial linear kernel function, and the initial target nonlinear kernel function are adjusted until the target attention function, the target linear kernel function, and the target nonlinear kernel function are obtained when it is determined based on the target sample feature data that the predetermined conditions are met.
[0098] During the specific implementation of this step, the error value can be determined based on the target sample feature data and label feature data corresponding to the same sample flow data, thereby determining the error value corresponding to each sample flow data, and then adjusting the parameters in the initial attention function, the initial linear kernel function, and the initial target nonlinear kernel function according to each error value, that is, adjusting the attention weight in the initial attention function, the constant term c1 in the initial linear kernel function, the hyperparameter γ in the initial nonlinear kernel function, the constant term c2 in the initial nonlinear kernel function, the order d of the kernel function in the initial nonlinear kernel function, the initial first weight value δ, and the initial second weight value ω. When each error value is less than a predetermined threshold, it is determined that the predetermined condition is met. Thus, the adjusted attention weight, the constant term c1 in the initial linear kernel function, the hyperparameter γ in the initial nonlinear kernel function, the constant term c2 in the initial nonlinear kernel function, the order d of the kernel function in the initial nonlinear kernel function, the first weight value δ, and the initial second weight value ω can be used as the final parameters to obtain the target attention function, the target linear kernel function, the target nonlinear kernel function, the first weight value corresponding to the target linear kernel function, and the second weight value corresponding to the target nonlinear kernel function. Subsequently, feature extraction can be performed directly based on the target attention function, the target linear kernel function and the target nonlinear kernel function.
[0099] In this embodiment, the traffic data in the optical transmission equipment network management system has the characteristics of high traffic volume, fast data transmission speed, and high data noise. Directly using all the features in the original network management data stream to train and predict the anomaly detection model will result in large dispersion of the prediction model, huge model training sample data, low model accuracy, and low efficiency. In this application, an attention mechanism is first used to extract features from the traffic data. This mechanism can focus on relevant parts when processing input data, thereby improving the performance and generalization ability of the model. That is, the attention mechanism can filter out key data that needs attention, and then perform feature analysis by using different kernel functions in a complementary manner. The final feature extraction result is achieved after expert intervention and confirmation, thereby being able to better adapt to the application scenarios of optical transmission equipment network management data, improve the accuracy of feature extraction, and lay the foundation for subsequent accurate and rapid abnormal traffic detection based on the extracted feature data.
[0100] Another embodiment of the present application provides a flow feature extraction device, such as Figure 3 As shown, including:
[0101] A screening module 11 is configured to perform feature screening on the target traffic data to be extracted using an attention mechanism to obtain initial feature data corresponding to the target traffic data;
[0102] A first extraction module 12 is configured to extract first feature data corresponding to the target flow data using a predetermined target linear kernel function for the initial feature data;
[0103] A second extraction module 13 is configured to extract second feature data corresponding to the target flow data using a predetermined target nonlinear kernel function for the initial feature data;
[0104] The determining module 14 is configured to determine target characteristic data corresponding to the target flow data based on the first characteristic data and the second characteristic data.
[0105] During the specific implementation of this embodiment, the screening module is specifically used to: create an initial matrix corresponding to the target traffic data for the target traffic data, the initial matrix including a number of eigenvectors and a number of eigenvalues; use the attention weights in the predetermined target attention function to perform weighted processing on each eigenvalue and each eigenvector in the initial matrix, obtain an initial eigenvalue matrix and an initial eigenvector matrix, so as to obtain the initial feature data.
[0106] During the specific implementation of this embodiment, the first extraction module is specifically used to: based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, respectively use the target linear kernel function to extract the first eigenvalue matrix corresponding to the initial eigenvalue matrix and the first eigenvector matrix corresponding to the initial eigenvector matrix to obtain the first feature data.
[0107] During the specific implementation of this embodiment, the second extraction module is specifically used to: based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, respectively use the target nonlinear kernel function to extract the second eigenvalue matrix corresponding to the initial eigenvalue matrix and the second eigenvector matrix corresponding to the initial eigenvector matrix to obtain the second feature data.
[0108] In the specific implementation process of this embodiment, the determination module is specifically used to: merge the first feature data and the second feature data based on the first weight value corresponding to the target linear kernel function and the second weight value corresponding to the target nonlinear kernel function to obtain the target feature data.
[0109] During the specific implementation of this embodiment, the flow feature extraction device also includes a parameter adjustment module; the parameter adjustment module is used to: collect a number of sample flow data and label feature data corresponding to each sample flow data; perform feature screening on each sample flow data based on the initial attention function to obtain initial sample feature data corresponding to each sample flow data; for each of the initial sample feature data, use the initial linear kernel function to extract the first sample feature data corresponding to each of the sample flow data; for each of the initial sample feature data, use the initial nonlinear kernel function to extract the second sample feature data corresponding to each of the sample flow data; based on each of the first sample feature data and each of the second sample feature data, determine the target sample feature data corresponding to each of the sample flow data; based on the target sample feature data corresponding to each sample flow data and the label feature data corresponding to each sample flow data, adjust the parameters in the initial attention function, the initial linear kernel function and the initial target nonlinear kernel function until the target attention function, the target linear kernel function and the target nonlinear kernel function are obtained when it is determined that the predetermined conditions are met based on the target sample feature data.
[0110] During the specific implementation of this embodiment, the traffic feature extraction device further includes an audit module, and the audit module is used to audit the target feature data after obtaining the target feature data to obtain the audited target feature data.
[0111] The traffic feature extraction device in this embodiment obtains initial feature data by first performing preliminary feature screening on the traffic data based on the attention mechanism, and then further extracts features from the initial feature data in combination with a predetermined target linear kernel function and a target nonlinear kernel function. That is, feature analysis and extraction are performed by using different kernel functions in a complementary manner, so that target feature data can be accurately and quickly extracted from the target traffic data.
[0112] Another embodiment of the present application provides a storage medium storing a computer program. When the computer program is executed by a processor, the following method steps are implemented:
[0113] Step 1: For the target traffic data to be extracted, use the attention mechanism to perform feature screening to obtain initial feature data corresponding to the target traffic data;
[0114] Step 2: for the initial feature data, using a predetermined target linear kernel function, extract first feature data corresponding to the target flow data;
[0115] Step 3: for the initial feature data, using a predetermined target nonlinear kernel function, extract and obtain second feature data corresponding to the target flow data;
[0116] Step 4: Determine target feature data corresponding to the target flow data based on the first feature data and the second feature data.
[0117] The specific implementation process of the above method steps can be found in the embodiments of any of the above traffic feature extraction methods, and this embodiment will not be repeated here.
[0118] The storage medium in this application obtains initial feature data by first performing preliminary feature screening on the traffic data based on the attention mechanism, and then further extracts features from the initial feature data in combination with a predetermined target linear kernel function and a target nonlinear kernel function, that is, feature analysis and extraction are performed by complementing different kernel functions, so that target feature data can be accurately and quickly extracted from the target traffic data.
[0119] Another embodiment of the present application provides an electronic device, such as Figure 4 As shown, it at least includes a memory 1 and a processor 2. The memory 1 stores a computer program. When the processor 2 executes the computer program on the memory 1, it implements the following method steps:
[0120] Step 1: For the target traffic data to be extracted, use the attention mechanism to perform feature screening to obtain initial feature data corresponding to the target traffic data;
[0121] Step 2: for the initial feature data, using a predetermined target linear kernel function, extract first feature data corresponding to the target flow data;
[0122] Step 3: for the initial feature data, using a predetermined target nonlinear kernel function, extract and obtain second feature data corresponding to the target flow data;
[0123] Step 4: Determine target feature data corresponding to the target flow data based on the first feature data and the second feature data.
[0124] The specific implementation process of the above method steps can be found in the embodiments of any of the above traffic feature extraction methods, and this embodiment will not be repeated here.
[0125] The electronic device in this application obtains initial feature data by first performing preliminary feature screening on the traffic data based on the attention mechanism, and then further extracts features from the initial feature data in combination with a predetermined target linear kernel function and a target nonlinear kernel function, that is, by performing feature analysis and extraction in a complementary manner of different kernel functions, it is possible to accurately and quickly extract target feature data from the target traffic data.
[0126] The specific implementation process of the above method steps can be found in any of the above embodiments of the CT image processing method, and will not be repeated in this embodiment.
[0127] The above embodiments are merely exemplary embodiments of the present application and are not intended to limit the scope of the present application. The scope of protection of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and scope of protection of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the scope of protection of the present application.
Claims
1. A flow feature extraction method, characterized in that: include: For the target traffic data to be extracted, feature screening is performed using the attention mechanism to obtain initial feature data corresponding to the target traffic data; For the initial feature data, using a predetermined target linear kernel function, extracting first feature data corresponding to the target flow data; For the initial feature data, using a predetermined target nonlinear kernel function, extracting second feature data corresponding to the target flow data; Target characteristic data corresponding to the target flow data is determined based on the first characteristic data and the second characteristic data.
2. The method according to claim 1, wherein The method of using the attention mechanism to perform feature screening on the target traffic data to be extracted to obtain initial feature data corresponding to the target traffic data specifically includes: Creating an initial matrix corresponding to the target flow data for the target flow data, wherein the initial matrix includes a plurality of eigenvectors and a plurality of eigenvalues; By using the attention weights in the predetermined target attention function, each eigenvalue and each eigenvector in the initial matrix are weighted to obtain an initial eigenvalue matrix and an initial eigenvector matrix to obtain the initial feature data.
3. The method according to claim 1, wherein The extracting, from the initial feature data, first feature data corresponding to the target flow data using a predetermined target linear kernel function specifically includes: Based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, the target linear kernel function is used to extract the first eigenvalue matrix corresponding to the initial eigenvalue matrix and the first eigenvector matrix corresponding to the initial eigenvector matrix to obtain the first feature data.
4. The method according to claim 1, wherein For the initial feature data, a predetermined target nonlinear kernel function is used to extract and obtain second feature data corresponding to the target flow data, specifically including: Based on the initial eigenvalue matrix and the initial eigenvector matrix in the initial feature data, the target nonlinear kernel function is used to extract the second eigenvalue matrix corresponding to the initial eigenvalue matrix and the second eigenvector matrix corresponding to the initial eigenvector matrix to obtain the second feature data.
5. The method according to claim 1, wherein The determining, based on the first characteristic data and the second characteristic data, target characteristic data corresponding to the target flow data specifically includes: Based on a first weight value corresponding to a target linear kernel function and a second weight value corresponding to a target nonlinear kernel function, the first feature data and the second feature data are merged to obtain the target feature data.
6. The method according to claim 2, wherein The method further includes predetermining the target attention function, the target linear kernel function, and the target nonlinear kernel function, specifically including: Collecting a number of sample flow data and label feature data corresponding to each sample flow data; Perform feature screening on each sample traffic data based on the initial attention function to obtain initial sample feature data corresponding to each sample traffic data; For each of the initial sample characteristic data, using an initial linear kernel function, extracting first sample characteristic data corresponding to each of the sample flow data; For each of the initial sample characteristic data, using an initial nonlinear kernel function, extracting second sample characteristic data corresponding to each of the sample flow data; Determining target sample characteristic data corresponding to each of the sample flow data based on each of the first sample characteristic data and each of the second sample characteristic data; Based on the target sample feature data corresponding to each sample flow data and the label feature data corresponding to each sample flow data, the parameters in the initial attention function, the initial linear kernel function and the initial target nonlinear kernel function are adjusted until the target attention function, the target linear kernel function and the target nonlinear kernel function are obtained when it is determined based on the target sample feature data that the predetermined conditions are met.
7. The method according to claim 1, wherein After obtaining the target feature data, the method further includes: The target characteristic data is reviewed to obtain reviewed target characteristic data.
8. A flow feature extraction device, characterized in that: include: A screening module is used to perform feature screening on the target traffic data to be extracted using an attention mechanism to obtain initial feature data corresponding to the target traffic data; A first extraction module is configured to extract first feature data corresponding to the target flow data using a predetermined target linear kernel function for the initial feature data; A second extraction module is configured to extract, from the initial feature data, second feature data corresponding to the target flow data using a predetermined target nonlinear kernel function; A determination module is used to determine target characteristic data corresponding to the target flow data based on the first characteristic data and the second characteristic data.
9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the flow feature extraction method according to any one of claims 1 to 7 are implemented.
10. An electronic device, characterized in that: The method comprises at least a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program in the memory, the steps of the flow feature extraction method according to any one of claims 1 to 7 are implemented.