Distributed radar multi-interference source identification method, electronic equipment and storage medium

Abstract

The invention provides a distributed radar multi-interference source identification method, electronic equipment and a storage medium, and the method comprises the steps: constructing a graph structure through signals received by a distributed radar, enabling nodes to represent the signals, and enabling edges to represent whether the nodes are interfered by the same type or not; then, a real part and an imaginary part of the signal are processed through a two-channel convolutional neural network, each channel comprises two convolutional layers, a pooling layer and a full connection layer, and signal features are extracted; then, the features are input into a graph self-attention network for interference type classification. And finally, the signal features with classification information enter a decoupling graph convolutional network, node features are decomposed into a plurality of independent subspaces, nodes influenced by the same interference source are identified, and a prediction adjacency matrix is constructed to distinguish different interference sources. According to the invention, the identification precision of the interference type in the complex electromagnetic environment can be improved, and the efficiency and accuracy of the model in large-scale data processing can be enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of radar interference source identification, and in particular to a distributed radar multi-interference source identification method, electronic equipment and storage medium. Background Art

[0002] Traditional radar interference source identification methods rely on manually extracted features, which often fail to fully reflect the essential characteristics of interference signals, resulting in low identification accuracy. Secondly, with the development of technology, radar systems are facing more complex interference environments, especially when facing multiple jammers working together and new interference methods based on digital technology. The collaborative work of multiple jammers increases the probability of interference overlap in different domains, making classification and identification more difficult, and existing methods are difficult to handle complex interference scenarios. In addition, the distinction between radar interference sources usually requires parameter estimation of the interference, while existing interference parameter estimation methods are often limited to specific interference types, making it difficult to accurately estimate the interference parameters of each component in the complex interference. Summary of the invention

[0003] In a first aspect of the present disclosure, a method for identifying multiple interference sources of a distributed radar is provided, comprising the following steps: The signals received by the distributed radar are used as nodes to construct a graph structure, where the edges connecting the nodes indicate that the corresponding connected nodes are subject to the same type of interference; Decomposing the real part and the imaginary part of the signal, extracting the signal feature of the real part signal through a first channel, and extracting the signal feature of the imaginary part signal through a second channel, wherein the first channel and the second channel both include a convolutional neural network structure, which sequentially includes a first convolutional layer, a second convolutional layer, a pooling layer, and a fully connected layer; The extracted real part signal features and imaginary part signal features are input as nodes into the graph self-attention network, and the interference type in the signal is classified according to the signal features, each signal feature corresponds to one or more feature vectors, and the feature vectors represent the attributes of the corresponding nodes; The signal with interference type classification information is input as a node into the decoupled graph convolutional network. By decomposing the node features into multiple independent subspaces, it is identified whether the nodes are affected by the same interference source, and then a prediction adjacency matrix is ​​constructed to distinguish different interference sources.

[0004] In combination with the first aspect, the graph structure is ,in yes A collection of radar nodes, is the set of edges, is a node feature matrix, where each row corresponds to a node’s feature vector.

[0005] In combination with the first aspect, the real signal includes an in-phase baseband signal of the distributed radar, and the imaginary signal includes an orthogonal baseband signal of the distributed radar.

[0006] In combination with the first aspect, the first convolution layer and the second convolution layer output multiple channels, and the output channels of the first convolution layer are less than the output channels of the second convolution layer. All convolution layers adopt the SAME boundary mode. The second convolution layer outputs to the pooling layer to compress the feature map and average the values ​​of each channel.

[0007] In combination with the first aspect, the feature representation of the output of the fully connected layer is: , in, is the feature representation of all nodes, For the The feature vector of a node, Indicates the total number of nodes, is the dimension of each feature vector, Represents the dimension of the original features.

[0008] In combination with the first aspect, the classifying the interference type in the signal by the signal feature includes: For each pair of connected nodes, calculate an attention coefficient; Normalize the nodes for calculating attention coefficients; According to the normalized attention coefficient, the feature representation of each node is updated.

[0009] In combination with the first aspect, the attention coefficient is calculated using the following formula: , in, Representation Node For Node The degree of attention, is the activation function, is the parameter vector, is the weight matrix used to linearly transform the node feature vector, and Respectively represent nodes and nodes The original feature vector of .

[0010] In combination with the first aspect, the normalization processing of the node for calculating the attention coefficient uses the following formula: , in, Is a node For Node The normalized attention coefficient of Represents the set of all neighbor nodes of a node, It is a subspace.

[0011] In combination with the first aspect, the updating of the feature representation of each node utilizes the following formula: , in, is the updated feature vector, is a non-linear activation function.

[0012] In combination with the first aspect, the method further includes using multi-head attention to concatenate the updated output features after updating the feature representation of each node: .

[0013] In combination with the first aspect, the decomposing node features into multiple independent subspaces, identifying whether the nodes are affected by the same interference source, and then constructing a prediction adjacency matrix to distinguish different interference sources includes: The feature vector of each node Project to In different subspaces, initialize Embedding vector for channels: , in and It is The parameters of the subspace, For the A low-dimensional projection of a subspace; For each node, its neighborhood information is used to update its decoupled representation. The calculation formula is: , , , , in, is the number of iterations, Representation Node and Between The connection probability caused by the factor is Indicated in Aggregate nodes in subspaces Relative to the node The importance of satisfying , and for all ,have and , and is a trainable trade-off coefficient that will be adjusted during the training process, and its adjustment range is within; The interference type of each node is judged. If it is judged that a certain type of interference does not exist at a certain node, the corresponding subspace output will be set to 0, otherwise it will remain unchanged; Using the formula , construct the predicted adjacency matrix, yes The matrix of is the number of nodes, is the feature dimension of each node, yes The transposed matrix of is the activation function.

[0014] According to a second aspect of the present disclosure, there is provided an electronic device, comprising: one or more processors; A storage unit is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors can implement the distributed radar multi-interference source identification method.

[0015] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the computer program is executed by a processor, the distributed radar multiple interference source identification method can be implemented.

[0016] Beneficial effects: The present invention provides a distributed radar multi-interference source identification method, which constructs a distributed radar interference perception network model and conducts a comprehensive analysis of data from multiple radar nodes, thereby achieving accurate identification of different types of interference sources, which is conducive to the effective acquisition of parameter information in subsequent tasks. This method utilizes the collaboration between various radar sites, enhances the system's anti-interference capability in complex electromagnetic environments and the accuracy of interference type identification, thereby greatly improving the accuracy of multi-interference source identification and the performance of subsequent radar collaborative anti-interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a process flow of a distributed radar multi-interference source identification method according to an embodiment of the present disclosure; Figure 2 An electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0018] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the embodiments of the present disclosure.

[0019] The terms used in the disclosed embodiments are only for the purpose of describing specific embodiments and are not intended to limit the disclosed embodiments. The singular forms of "a", "said" and "the" used in the disclosed embodiments and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0020] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the disclosed embodiments, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the disclosed embodiments, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0021] like Figure 1 FIG. 1 is a flow chart of a distributed radar multi-interference source identification method according to an embodiment of the present disclosure, including: S101: Using signals received by the distributed radar as nodes to construct a graph structure, wherein the edges connecting the nodes indicate that the corresponding connected nodes are subject to the same type of interference.

[0022] Specifically, in a distributed radar system, the signals received by each radar node can be regarded as nodes in a graph structure. This graph structure is an abstract data structure used to represent objects (nodes or vertices) and the relationships (edges) between these objects. When we consider interference identification, building such a graph model can help us better understand the propagation pattern and impact range of interference in radar networks.

[0023] Node: Each signal received by the radar node is considered a node. This means that each node represents the information received by the radar at a specific location.

[0024] Edge: If two nodes are affected by the same type of interference, there will be an edge connecting the two nodes. This edge indicates that there is some kind of correlation between the two nodes, that is, they may be affected by the same interference source.

[0025] In such a graph structure, the connections between nodes not only indicate physical proximity or direct communication, but more importantly, reflect the interaction in the interference environment. For example, if there are multiple radar nodes in a certain area, and these nodes are affected by the same interference source due to geographical proximity or other reasons, then these nodes will form a closely connected subgraph in the graph, which helps to identify the interference pattern in the area.

[0026] Beneficial effects: Taking the signals received by distributed radars as nodes and establishing connections based on common interference types can not only help us intuitively depict the existence and impact of interference in the entire radar network, but also provide strong support for further interference suppression strategies. This method integrates the ideas of graph theory and opens up a new way to understand and solve complex electromagnetic interference problems.

[0027] Furthermore, the graph structure is ,in yes A collection of radar nodes, is the set of edges, is a node feature matrix, where each row corresponds to a node’s feature vector.

[0028] Specific, collection Represents A collection of radar nodes. Each node corresponds to a radar receiving station or sensor, which is responsible for collecting electromagnetic wave information at a specific geographical location. These nodes form the basic framework of the entire graph.

[0029] The set of edges contains all the connections between nodes. An edge indicates that there is some form of relationship between nodes. Specifically in the interference identification scenario, it means that the two nodes are affected by the same type of interference. The existence of an edge and its weight (if any) reflect the strength of the correlation of the interference between nodes.

[0030] Each row in the node feature matrix corresponds to a node feature vector. It can contain various information, such as signal strength, spectrum characteristics, time series data, etc., which are used to describe the important parameters of the radar site status represented by the node. For example, in the interference identification task, the feature vector may include information such as the frequency component, power level, and arrival angle of the interference signal.

[0031] S102: Decomposing the real part and the imaginary part of the signal, extracting the signal features of the real part signal through the first channel, and extracting the signal features of the imaginary part signal through the second channel, wherein the first channel and the second channel both include a convolutional neural network structure, which sequentially includes a first convolutional layer, a second convolutional layer, a pooling layer, and a fully connected layer.

[0032] Specifically, the signal received by the distributed radar is decomposed into a real signal and an imaginary signal, wherein the real signal includes an in-phase baseband signal of the distributed radar, and the imaginary signal includes an orthogonal baseband signal of the distributed radar.

[0033] The in-phase baseband signal refers to a signal component that is in phase with the carrier wave, is used to carry information, and can be modulated or demodulated independently of the orthogonal signal.

[0034] Orthogonal baseband signals refer to signal components with a phase difference of 90 degrees (π / 2 radians) relative to the carrier, which can carry information independently without interfering with the information transmission of the in-phase component.

[0035] Beneficial effects: It not only helps to analyze the received information more accurately, but also provides convenient conditions for subsequent signal processing. In addition, through such a representation method, digital signal processing algorithms can be more easily implemented.

[0036] Furthermore, in order to remove redundant information and improve the efficiency and accuracy of subsequent identification, it is necessary to effectively extract the key features of the radar signal of each node and realize dimensionality reduction processing of the signal.

[0037] Specifically, the in-phase baseband signal is input into the first channel, and the orthogonal baseband signal is input into the second channel, wherein the first channel and the second channel both include a convolutional neural network structure, which sequentially includes a first convolutional layer, a second convolutional layer, a pooling layer, and a fully connected layer, wherein the first convolutional layer outputs 32 channels with a convolution kernel size of 5×5, and the second layer outputs 128 channels. A dilated convolution is used with a size of 3 and an expansion rate of 1, which helps to expand the receptive field without increasing the number of parameters and capture a wider range of spatial relationships.

[0038] Optionally, each convolutional layer implements the SAME boundary mode to keep the signal dimension size consistent before and after convolution. After passing through two convolutional blocks, the global average pooling layer averages the values ​​of each channel of the feature map and reduces the feature dimension.

[0039] Then it passes through a fully connected layer to prepare for subsequent graph structure processing. The feature representation of the final output of the fully connected layer is: , in, is the feature representation of all nodes, For the The feature vector of a node, Indicates the total number of nodes, is the dimension of each feature vector, Represents the dimension of the original features.

[0040] S103: The extracted real part signal features and imaginary part signal features are input as nodes into the graph self-attention network, and the interference type in the signal is classified according to the signal features. Each signal feature corresponds to one or more feature vectors, and the feature vectors represent the attributes of the corresponding nodes.

[0041] Specifically, the Graph Self-Attention Network is a neural network model designed specifically for processing graph-structured data. It enhances the learning ability of node features by introducing a self-attention mechanism. This model can apply an attention mechanism to each node in the graph, allowing each node to dynamically adjust the way it aggregates information based on the importance of its neighboring nodes, thereby capturing more complex patterns and dependencies.

[0042] In the present disclosure, the input of the graph self-attention network is a set of different radar signals, and each node has its own feature vector, which are the signal features extracted previously.

[0043] Further, classifying the interference type in the signal according to the signal feature includes: For each pair of connected nodes, calculate an attention coefficient; Normalize the nodes for calculating attention coefficients; According to the normalized attention coefficient, the feature representation of each node is updated.

[0044] Specifically, for each pair of connected nodes (i.e., edges) in the graph, an attention coefficient is calculated, which reflects the importance of the target node relative to the source node.

[0045] The attention coefficient is calculated using the following formula: , in, Representation Node For Node The degree of attention, is the activation function, is the parameter vector, is the weight matrix used to linearly transform the node feature vector, and Respectively represent nodes and nodes The original feature vector of .

[0046] Next, in order to ensure that these attention coefficients can be used as weights, they are usually softmax normalized so that the sum of the attention coefficients corresponding to each node’s neighbor nodes is 1.

[0047] The nodes for calculating the attention coefficient are normalized using the following formula: , in, Is a node For Node The normalized attention coefficient of Represents the set of all neighbor nodes of a node, It is a subspace.

[0048] Finally, the feature representation of each node is updated by weighted averaging the features of neighboring nodes using the normalized attention coefficients. This step allows each node to learn its new feature representation based on the relevance or importance of its neighboring nodes.

[0049] Update the feature representation of each node using the following formula: , in, is the updated feature vector, is a non-linear activation function.

[0050] Optionally, in order to further improve the fitting ability of the network, the present invention also adopts a multi-head attention mechanism. Separate attention mechanisms are used and their results are concatenated: .

[0051] Since in the defined scenario, with the node The relevant first-order neighboring nodes already contain all the information for identifying the same type of interference, so only one layer of graph self-attention network is needed. That is: , in, Is a node The output feature vector obtained after the graph self-attention layer is Is a node The new feature vector obtained after applying the graph self-attention mechanism. In a single-layer GAT, this means that the node not only considers its own features, but also considers the information from its direct neighbor nodes, which are weighted summed according to the attention coefficient.

[0052] The output result will be used as a mask for the representation learned by the decoupled graph convolutional network to enhance the network's ability to identify interference sources. In addition, the loss function used for classification is binary cross entropy: .

[0053] in, Represents the total number of samples or nodes, which is the number of all radar nodes that need to be interfered with. For nodes , is a true binary label (0 or 1) indicating whether a specific type of interference exists. Is the model for the node The predicted probability of whether there is a specific type of interference, that is, the probability that the model believes that the node belongs to the positive class. When the loss depends mainly on , encouraging the model to predict the probability of the positive class as accurately as possible; and when When , which encourages the model to correctly predict the probability of the negative class.

[0054] S104: Input the signal with interference type classification information as a node into the decoupled graph convolutional network, decompose the node features into multiple independent subspaces, identify whether the nodes are affected by the same interference source, and then construct a prediction adjacency matrix to distinguish different interference sources.

[0055] Specifically, in the original input graph, the edge is defined as two nodes receiving the same type of interference, and the relationship includes two nodes receiving interference from the same interference source and two nodes receiving interference from different interference sources under the same interference type.

[0056] In order to further distinguish these two potential relationships, the present invention adopts a decoupled graph convolutional network to predict new connection relationships between nodes, that is, to determine whether two nodes are affected by the same interference source, thereby achieving the separation of interference sources and laying a solid foundation for subsequent collaborative anti-interference.

[0057] Assume that different types of interference sources are independent of each other and do not affect each other. For the node input of the decoupled graph convolutional network, the key features extracted by the radar node are at least independent components, which indicates that at least decoupling is required potential factors. For a node , its potential representation Expressed as ,in Used to describe nodes No. aspects.

[0058] The network first transforms the feature vector Project to In different subspaces, we initialize Embedding vector for channels:

[0059] in and It is The parameters of the subspace, For the In order to fully understand the node No. In each case, information needs to be extracted from its neighborhood. This involves using and To build .

[0060] Therefore, three assumptions are made about the relationship between nodes: Hypothesis 1: The generation of links between nodes originates from the interaction of various hidden factors.

[0061] Assumption 2: If the node Its adjacent nodes In the subspaces have similarities, which indicates that the link between these two nodes may be attributed to the factor .

[0062] Assumption 3: If the adjacent node is show similarities in all aspects, forming a The clustering of subspaces, then the factor Most likely a connecting node The reason for this particular group of neighboring nodes.

[0063] Assumptions 1 and 2 are consistent with the definition of the radar node graph structure in the present invention. The generation of links in the distributed radar node graph originates from the reception of similar signals, i.e., nodes Its adjacent nodes At the same time, receiving type of interference.

[0064] When decoupled If the subspaces can be well matched to the interference types, then the link between the two nodes can be attributed to the factor .

[0065] Assumption 3 explains the principle of using decoupled graph convolutional networks for link prediction. The subspaces can be clustered by calculating the similarity distance, and these clusters are likely to correspond to different interference sources under the same interference type.

[0066] According to the above three assumptions, the decoupled representation of nodes can be learned by utilizing the decoupling layer based on the dynamic allocation mechanism: , , , , in, is the number of iterations. Representation Node and Between The connection probability caused by the factor is Indicated in Aggregate nodes in subspaces Relative to the node They satisfy , and for all ,have and . and is a trainable trade-off coefficient that will be adjusted during the training process, and its adjustment range is Within.

[0067] The dynamic allocation mechanism will iteratively infer and , and build At the same time, this mechanism allows the coexistence of multiple relationships between nodes. It should be noted that there are a total of decoupling layers, and at each layer middle, The value of is finally assigned to .

[0068] Decoupling layer output nodes The decoupled embedding is all The connection of components learned in channels, i.e. , Since the main goal of the disentangled graph convolutional network is to enable the graph encoder to generate disentangled representations for each latent factor . Extracted through dynamic routing mechanism Different factors are designed to focus on different reasons for connections between nodes. Channels capture mutually exclusive information associated with each latent factor.

[0069] In order to improve the decoupling effect and obtain a more interpretable explanation, the present invention imposes constraints on these subspaces when executing the dynamic allocation mechanism to ensure that the focus perspectives of the K subspaces are concentrated on different interference types. The specific method is to use the judgment result of the classifier on the interference type as a mask. As shown in the following formula: , When the classifier determines When the interference type does not exist, the corresponding The output of the subspace Set to 0, otherwise no operation is performed. The final representation learned after all nodes are masked is: , The final predicted adjacency matrix is ​​given by The inner product of is constructed, that is Since the upper triangular part of the adjacency matrix contains all the edges, the upper triangular part of the link prediction adjacency matrix is ​​compared with the label and the binary cross entropy between them is calculated: , The loss function of the entire network framework is: , in is a hyperparameter used to control the balance between the decoupled graph convolutional network and the classifier.

[0070] Beneficial effects: The present invention provides a distributed radar multi-interference source identification method, which constructs a distributed radar interference perception network model and conducts a comprehensive analysis of data from multiple radar nodes, thereby achieving accurate identification and detection of different types of interference sources, which is beneficial to the effective acquisition of parameter information in subsequent tasks. This method utilizes the collaboration between various radar sites, enhances the system's anti-interference capability in complex electromagnetic environments and the accuracy of interference type identification, thereby greatly improving the accuracy of multi-interference source identification and the performance of subsequent radar collaborative anti-interference.

[0071] The electronic device 200 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 200 may include, but is not limited to, a processor 201 and a memory 202. Those skilled in the art will appreciate that Figure 2 It is only an example of the electronic device 200 and does not constitute a limitation of the electronic device 200. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0072] The processor 201 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0073] The memory 202 may be an internal storage unit of the electronic device 200, for example, a hard disk or memory of the electronic device 200. The memory 202 may also be an external storage device of the electronic device 200, for example, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 200. Further, the memory 202 may also include both an internal storage unit of the electronic device 200 and an external storage device. The memory 202 is used to store the computer program 203 and other programs and data required by the electronic device. The memory 202 may also be used to temporarily store data that has been output or is to be output.

[0074] In the embodiments provided in the present disclosure, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0075] If the integrated module / 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 present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0076] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. A distributed radar multi-interference source identification method, characterized in that: The following steps are involved: The signals received by the distributed radar are used as nodes to construct a graph structure, where the edges connecting the nodes indicate that the corresponding connected nodes are subject to the same type of interference; Decomposing the real part and the imaginary part of the signal, extracting the signal feature of the real part signal through a first channel, and extracting the signal feature of the imaginary part signal through a second channel, wherein the first channel and the second channel both include a convolutional neural network structure, which sequentially includes a first convolutional layer, a second convolutional layer, a pooling layer, and a fully connected layer; The extracted real part signal features and imaginary part signal features are input as nodes into the graph self-attention network, and the interference type in the signal is classified according to the signal features, each signal feature corresponds to one or more feature vectors, and the feature vectors represent the attributes of the corresponding nodes; The signal with interference type classification information is input as a node into the decoupled graph convolutional network. By decomposing the node features into multiple independent subspaces, it is identified whether the nodes are affected by the same interference source, and then a prediction adjacency matrix is ​​constructed to distinguish different interference sources.

2. The method according to claim 1, characterized in that The graph structure is ,in yes A collection of radar nodes, is the set of edges, is a node feature matrix, where each row corresponds to a node’s feature vector.

3. The method according to claim 1, characterized in that The real signal includes an in-phase baseband signal of the distributed radar, and the imaginary signal includes an orthogonal baseband signal of the distributed radar.

4. The method according to claim 1, characterized in that The first convolution layer and the second convolution layer output multiple channels, and the output channels of the first convolution layer are less than the output channels of the second convolution layer. All convolution layers adopt the SAME boundary mode. The second convolution layer outputs to the pooling layer to compress the feature map and average the value of each channel.

5. The method according to claim 1, characterized in that The feature representation of the output of the fully connected layer is: ， in, is the feature representation of all nodes, For the The feature vector of a node, Indicates the total number of nodes, is the dimension of each feature vector, Represents the dimension of the original features.

6. The method according to claim 1, characterized in that The classifying the interference type in the signal by the signal feature comprises: For each pair of connected nodes, calculate an attention coefficient; Normalize the nodes for calculating attention coefficients; According to the normalized attention coefficient, the feature representation of each node is updated.

7. The method according to claim 6, characterized in that The calculation of the attention coefficient uses the following formula: ， in, Representation Node For Node The degree of attention, is the activation function, is the parameter vector, is the weight matrix used to linearly transform the node feature vector, and Respectively represent nodes and nodes The original feature vector of .

8. The method according to claim 7, characterized in that The normalization process for the nodes for calculating the attention coefficient uses the following formula: ， in, Is a node For Node The normalized attention coefficient of Represents the set of all neighbor nodes of a node, It is a subspace.

9. The method according to claim 8, characterized in that The updating of the feature representation of each node uses the following formula: ， in, is the updated feature vector, is a non-linear activation function.

10. The method according to claim 9, characterized in that The method further includes concatenating the updated output features using multi-head attention after updating the feature representation of each node: 。 11. The method according to claim 1, characterized in that: Decomposing node features into multiple independent subspaces, identifying whether nodes are affected by the same interference source, and then constructing a prediction adjacency matrix to distinguish different interference sources includes: The feature vector of each node Project to In different subspaces, initialize Embedding vector for channels: ， in and It is The parameters of the subspace, For the A low-dimensional projection of a subspace; For each node, its neighborhood information is used to update its decoupled representation. The calculation formula is: ， ， ， ， in, is the number of iterations, Representation Node and Between The connection probability caused by the factor is Indicated in Aggregate nodes in subspaces Relative to the node The importance of satisfying , and for all ,have and , and is a trainable trade-off coefficient that will be adjusted during the training process, and its adjustment range is within; The interference type of each node is judged. If it is judged that a certain type of interference does not exist at a certain node, the corresponding subspace output will be set to 0, otherwise it will remain unchanged; Using the formula , construct the predicted adjacency matrix, yes The matrix of is the number of nodes, is the feature dimension of each node, yes The transposed matrix of is the activation function.

12. An electronic device, characterized in that: include: one or more processors; A storage unit, used to store one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the distributed radar multi-interference source identification method according to any one of claims 1 to 11.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, it can implement the distributed radar multiple interference source identification method according to any one of claims 1 to 11.

Images