Graph kernel method-based graph transfer learning radar working mode identification method

Abstract

The invention discloses a graph transfer learning radar working mode identification method based on a graph kernel method, and belongs to the technical field of radar signal classification, and the method comprises the following steps: S1, carrying out the feature extraction of radar working mode pulse data in radar working mode identification tasks of a source domain and a target domain through employing the graph kernel method, forming a source domain feature and a target domain feature; s2, based on the source domain features and the target domain features, calculating a graph kernel similarity index of the source domain and the target domain; and S3, selecting a direct migration recognition algorithm or an adaptive multi-domain migration recognition algorithm based on the size of the graph kernel similarity index to carry out migration learning on the graph neural network model, and carrying out radar working mode recognition on a target domain radar working mode recognition task by using the trained graph neural network. By adopting the method provided by the invention, the working mode of the unknown radar can be quickly and accurately identified.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar signal classification, and in particular relates to a radar operating mode recognition method based on graph transfer learning using a graph kernel method. Background Art

[0002] Radar mode recognition technology is crucial. However, when faced with unknown radar systems, traditional radar mode recognition methods often rely on large amounts of labeled data and rich domain knowledge, which is difficult to meet in practical applications, especially in complex and ever-changing environments. Furthermore, with the advancement of radar technology, new radar systems are constantly emerging, and their operating modes are becoming more diverse and covert, which poses greater challenges to recognition.

[0003] When having a small amount of prior knowledge about unknown radars, that is, a small amount of sample data with target domain labels, how to effectively use the network trained with a large amount of known source domain radar working mode pulse data to perform unknown target domain radar working mode recognition is an urgent problem to be solved.

[0004] Graph neural networks and graph kernel methods are widely applicable in the field of two-dimensional data images, but there are still the following difficulties in adapting graph neural networks to one-dimensional data radar working mode recognition: First, pixels in two-dimensional images naturally have spatial adjacency relationships, and constructing subgraphs is relatively intuitive; and how to effectively map one-dimensional data into a graph structure to reflect pattern characteristics is a difficulty that needs to be overcome; secondly, the topological structure of the image grid graph is very regular, while the structure of the one-dimensional signal graph is usually highly irregular and data-driven, so how to use the constructed graph structure to effectively discriminate information on the one-dimensional signal is also a difficulty; finally, there is a challenge in adapting the graph neural network structure. Different radar working modes have different statistical characteristics and frequency domain characteristics, which requires the constructed graph neural network to have strong dynamic adaptability. Summary of the Invention

[0005] The purpose of the present invention is to provide a radar working mode recognition method based on graph neural network transfer learning of a graph kernel method, so as to solve the problem that the existing radar working mode recognition network cannot quickly and accurately identify the working mode of unknown radars with little prior knowledge in the field.

[0006] In order to achieve the above object, the technical solution of the present invention is as follows: The present invention relates to a radar operating mode recognition method based on graph transfer learning using a graph kernel method, which comprises the following steps: S1. Use the graph kernel method to extract features from radar working mode pulse data in the source domain and target domain radar working mode recognition tasks to form source domain features and target domain features; S2. Calculate the kernel similarity index of the source and target domains based on the source and target domain features. S3. Based on the size of the graph kernel similarity index, a direct transfer recognition algorithm or an adaptive multi-domain transfer recognition algorithm is selected to perform transfer learning on the graph neural network model, and the trained graph neural network is used to perform radar working mode recognition on the target domain radar working mode recognition task.

[0007] Preferably, the step of using the graph kernel method to extract features from radar working mode pulse data in the source domain and target domain radar working mode recognition tasks in S1 is: S1.1. Divide the radar working mode pulse sequence into the following categories according to the time sequence: n units; S1.2. Use the graph kernel method to find the source domain feature vector and the target domain feature vector. The source domain feature vector is expressed as: S i , i =1,2,…, P , i is the number of the source domain feature vector, P is the number of source domain feature vectors; the target domain feature vector is expressed as: T j , j =1,2,…, Q , j is the number of the source domain feature vector, Q is the number of source domain feature vectors; S1.3. Integrate the source domain eigenvector and the target domain eigenvector respectively to obtain the source domain eigenvector matrix and the target domain eigenvector matrix.

[0008] Preferably, the graph kernel method used in S1.2 adopts the PMK pyramid matching kernel method. The specific steps of using this method to find the source domain feature vector and the target domain feature vector are as follows: S1.2.1. Listed in chronological order n Each unit corresponds to each node in the graph neural network, forming a subgraph. Based on the connection or proximity of the nodes, the edges are constructed and the weights of the edges are calculated. The calculation formula is: , in, ω ab Representation node a and nodes b The weight of the edge between n a , n b Nodes representing the input graph neural network a and nodesb The specific characteristic value of is the scale parameter; S1.2.2. Construct for each subgraph K Layer pyramid, for the k The feature space of the layer pyramid is divided into A grid, k For which layer, d is the feature dimension of the subgraph; S1.2.3. Count the number of nodes in the subgraph in the grid and generate a histogram vector , the histogram vector is the corresponding source domain feature vector or target domain feature vector, The calculation formula of the histogram vector is as follows: , in, c Number the subgraph corresponding to the source domain or target domain; Indicates the k Layer c The eigenvalues ​​in the first grid of the subgraph are A grid.

[0009] Preferably, the calculation formula of the S2 graph kernel similarity index is: , in, k △ represents the graph kernel similarity index, G S is the source domain radar working mode subgraph, G T is the target domain radar working mode sub-graph, and Respectively represent the radar working mode sub-graphs in the source domain and target domain Histogram of the layer, I is the histogram intersection function, K is the number of layers of the sub-graph pyramid kernel method for the radar working mode in the source domain and the target domain; the expression of the histogram intersection function is: , in, n is the total number of grids in the graph neural network, i Number the mesh nodes.

[0010] Preferably, the method of selecting direct migration recognition algorithm or adaptive multi-domain migration recognition algorithm to transfer learning of graph neural network model based on the size of graph core similarity index is: setting similarity critical value αWhen the graph core similarity index is not less than the similarity critical value, the direct transfer recognition algorithm is selected to perform transfer learning on the graph neural network model; when the graph core similarity index is less than the similarity critical value, the adaptive multi-domain transfer recognition algorithm is selected to perform transfer learning on the graph neural network model.

[0011] Preferably, when the direct transfer recognition algorithm is selected in S3 to perform transfer learning on the graph neural network model, the specific method is: fix the model structure of the graph neural network model of the selected source domain radar working mode recognition task and the network parameters from the input layer to the feature extraction layer, use the labeled data in the target domain radar working mode recognition task to adaptively train the graph neural network model, and construct a network suitable for the target domain radar working mode recognition task.

[0012] Preferably, when the adaptive multi-domain transfer recognition algorithm is selected in S3 to perform transfer learning on the graph neural network model, the specific steps are as follows: S3.1. Select at least three source-domain radar operating mode recognition tasks in descending order of image kernel similarity metrics. S3.2. Select the source domain radar working mode recognition task with the largest graph kernel similarity index to build a graph neural network model. M layer; S3.3. Select an unselected source domain radar working mode recognition task in descending order of kernel similarity index; S3.4. The front of the graph neural network model l The parameters of the layer are fixed. l +1 to M Initialize the parameters of the layer, train the graph neural network model using the source domain radar working mode recognition task selected in S3.3, and update the graph neural network model l +1 to M Layer parameters; S3.5. Bring the source domain radar working mode recognition task selected in S3.3 into the graph neural network model constructed in S3.2 and the graph neural network model updated in S3.4, and compare the recognition rates. If the recognition rate of the updated graph neural network model is better, mark the graph neural network model as the first l The layer is general. If the recognition rate of the constructed graph neural network model is better, the first layer of the labeled graph neural network model l Layers have specificity; S3.6. Repeat S3.4-S3.5 until the source domain radar working mode recognition task selected in S3.2 is transferred to the source domain radar working mode recognition task selected in S3.3. MAll layers are judged as general or special, the layers marked as special are frozen, the layers marked as general are adaptively trained, and the migrated graph neural network model is updated; S3.7. Determine whether all the source domain radar working mode recognition tasks selected in S3.1 have been selected. If not, return to S3.3. If so, output the final graph neural network model.

[0013] Compared with the prior art, the technical solution provided by the present invention has the following beneficial effects: The present invention relates to a radar working mode recognition method based on graph transfer learning of graph kernel method, which aims at the unique properties of unknown radar few-sample signal data and introduces the use of transfer learning containing graph neural network for radar working mode recognition task; the constructed graph neural network with attention mechanism allows the importance weights of adjacent nodes to be dynamically calculated according to feature content when aggregating adjacent node information; and automatically updates the optimized adjacency matrix based on the graph kernel similarity feature optimization model calculated by graph kernel method; decomposes long-segment radar working mode signals into short-segment semantic signals, and combines the pyramid matching kernel method to effectively extract radar working mode features under long sequences, saving computational complexity while realizing feature extraction of one-dimensional radar signals; calculates the radar working mode features of the identified radar in the source domain The kernel similarity between the radar signal sequence and the radar signal sequence to be identified in the target domain is used to establish the distance relationship between the source domain and the target domain; for the radar working mode recognition kernel similarity between the source domain and the target domain is large, the transfer learning method is used to use deep network adaptive training to achieve the source domain radar working mode recognition classification knowledge can be transferred to the target domain with few labels and less prior knowledge; for the source domain and the target domain kernel similarity is small, the source domain and target domain gradual migration network is built by using multiple source domains, and the knowledge that can be transferred from multiple source domains to the target domain is transferred step by step to avoid the negative transfer problem caused by direct migration, and construct a network model suitable for the target domain radar working mode recognition task, thereby quickly and accurately identifying the working mode of unknown radars. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 This is a principle block diagram of a radar working mode recognition method based on graph kernel method through graph transfer learning according to the present invention; Figure 2 This is a network structure diagram of the graph neural network based on source domain radar working mode recognition of the present invention; Figure 3 Schematic diagram of the direct migration identification algorithm; Figure 4 Schematic diagram of the adaptive multi-domain migration recognition algorithm. DETAILED DESCRIPTION

[0015] In order to further understand the content of the present invention, the present invention is described in detail with reference to the examples. The following examples are used to illustrate the present invention but are not used to limit the scope of the present invention.

[0016] Refer to the attached Figure 1 As shown, the present invention relates to a radar working mode recognition method based on graph transfer learning of a graph kernel method, which includes the following steps: S1. Use the graph kernel method to extract features from the radar working mode pulse data in the source domain and target domain radar working mode recognition tasks to form source domain features and target domain features. The source domain and target domain are both radar working mode pulse datasets. The source domain is a radar working mode dataset with classification labels, and the target domain is a radar working mode dataset with rare labels. Rare labels refer to each classification containing only a few labeled data. The specific steps are as follows: S1.1. Divide the radar working mode pulse sequence into the following categories according to the time sequence: n units; S1.2. Use the graph kernel method to find the source domain feature vector and the target domain feature vector. The graph kernel method uses the PMK pyramid matching kernel method. The specific steps are as follows: S1.2.1. Listed in chronological order n The units correspond to each node in the graph neural network, that is, the graph neural network has n Nodes are formed into a subgraph. Edges are constructed based on whether the nodes are connected or close in time. The weight of the edges is calculated using the following formula: , in, ω ab Representation node a and nodes b The weight of the edge between n a , n b Nodes representing the input graph neural network a and nodes b The specific characteristic value of is the scale parameter; S1.2.2. Construct for each subgraph K Layer pyramid, for the k The feature space of the layer pyramid is divided into A grid, k For which layer, d is the feature dimension of the subgraph; S1.2.3. Count the number of nodes in the subgraph in the grid and generate a histogram vector , the histogram vector is the corresponding source domain feature vector or target domain feature vector, The calculation formula of the histogram vector is as follows: , in, c Number the subgraph corresponding to the source domain or target domain; Indicates the k Layer c The eigenvalues ​​in the first grid of the subgraph are grids; The source domain feature vector is expressed as: S i , i =1,2,…, P , i is the number of the source domain feature vector, P is the number of source domain feature vectors; the target domain feature vector is expressed as: T j , j =1,2,…, Q , j is the number of the source domain feature vector, Q is the number of source domain feature vectors; S1.3. Integrate the source domain eigenvectors and the target domain eigenvectors to obtain the source domain eigenvector matrix and the target domain eigenvector matrix. The source domain eigenvector matrix S Expressed as: , The target domain eigenvector matrix T Expressed as: .

[0017] S2. Based on the source domain features and the target domain features, calculate the image core similarity index of the source domain and the target domain. The calculation formula of the image core similarity index is: , in, k △ represents the graph kernel similarity index, G S For the source domain radar working mode recognition task, G T For the target domain radar working mode recognition task, and Respectively represent the radar working mode sub-graphs in the source domain and target domain Histogram of the layer, I is the histogram intersection function, L The number of layers for the radar working mode recognition task in the source and target domains; The expression of the histogram intersection function is: , in, n is the total number of grids in the graph neural network, i Number the mesh nodes.

[0018] S3. Based on the size of the graph kernel similarity index, select the direct transfer recognition algorithm or the adaptive multi-domain transfer recognition algorithm to perform transfer learning on the graph neural network model. The graph neural network model adopts the following Figure 2 The radar working pattern recognition network shown combines GNN with attention mechanism.

[0019] Its input is the radar working mode data sliced ​​in chronological order as mentioned above. The pulse sequence in each time unit is input into a node of the graph neural network to form a subgraph. The subgraphs are connected in the form of edges. A graph attention mechanism is added to each feature extraction layer to optimize the weight calculation of the edges between each subgraph. After the feature extraction of the graph neural network through several layers of graph attention mechanism, two fully connected layers are added to take over the features extracted by the above graph neural network. Finally, the extracted features are classified and identified through a classifier containing Softmax, and its activation function is RELU. The above steps achieve the task of identifying and classifying the original radar pulse data into different radar working modes.

[0020] The method of selecting direct transfer recognition algorithm or adaptive multi-domain transfer recognition algorithm based on the size of the graph kernel similarity index to transfer learning of the graph neural network model is: setting the similarity threshold α , when the graph kernel similarity index is not less than the similarity critical value, that is , indicating that the similarity between the two domains has reached a certain standard. At this time, the radar working mode classification knowledge in the source domain can be directly transferred to the target domain. Therefore, the direct transfer recognition algorithm is selected to perform transfer learning on the graph neural network model. The specific method of direct transfer learning is as follows: Figure 3 As shown in the figure, the model structure of the graph neural network model for the source domain radar working mode recognition task and the network parameters from the input layer to the feature extraction layer are fixed, and the graph neural network model is adaptively trained using the rare labeled data in the target domain radar working mode recognition task, and further constructed for the target domain radar working mode recognition task. The network, where the node feature vectors of the 0th layer in the source and target domain structures are h v0 express; C v0 It represents the feature vector after the attention mechanism extracts the surrounding associated node information. y i and yj They represent the source domain node classification output and target domain node classification output respectively.

[0021] When the kernel similarity index is less than the similarity critical value, that is, , it can be considered that the source domain and the target domain have little correlation. At this time, the radar working mode classification knowledge of the source domain cannot be directly transferred to the target domain. The features extracted by the neural network can be divided into general features and specific features. General features have little correlation with the task and can be directly used for other tasks; specific features have strong task relevance and are only applicable to specific tasks. The features in the front layers of the network show stronger generality, while the features in the back layers have stronger specificity. In order to avoid the negative transfer problem caused by directly migrating the target domain radar from the source domain radar working mode recognition model, an adaptive multi-domain transfer recognition algorithm is selected to perform transfer learning on the graph neural network model. The specific steps are as follows: Figure 4 As shown, S3.1. Select at least three source domain radar working mode recognition tasks in descending order of the image kernel similarity index. In this embodiment, three source domain radar working mode recognition tasks are selected. S A 、 S B 、 S C For example; S3.2. Select the source domain radar working mode recognition task with the largest kernel similarity index S A Building a graph neural network model N ( S A ), the graph neural network model has a total of M layer; S3.3. Select an unselected source domain radar working mode recognition task in descending order of image kernel similarity index S B ; S3.4. Graph Neural Network Model N ( S A ) before l The parameters of the layer are fixed. l +1 to M The parameters of the layer are initialized, and the source domain radar working mode recognition task selected in S3.3 is used. S B Train the graph neural network model and update the graph neural network model l +1 to M Layer parameters; S3.5. Substitute the source domain radar working mode recognition task selected in S3.3 into the graph neural network model constructed in S3.2 and the graph neural network model updated in S3.4, and compare the recognition rates. If the recognition rate of the updated graph neural network model is better, it proves that the source domain radar working mode recognition task selected in S3.2 is better. S A Migrate to the source domain radar working mode recognition task selected in S3.3 S B In the process, the first l The layer is general. If the recognition rate of the constructed graph neural network model is better, it proves that the source domain radar working mode recognition task selected in S3.2 S A Migrate to the source domain radar working mode recognition task selected in S3.3 S B In the process, the first l Layers have specificity; S3.6. Repeat S3.4-S3.5 until the source domain radar working mode recognition task selected in S3.2 is transferred to the source domain radar working mode recognition task selected in S3.3. M Each layer is judged as general or special, and then the S A Migrate to S B In the process N ( S A ) The generality and specificity of each layer, freeze the layers marked as specific, adaptively train the layers marked as general, and update the migrated graph neural network model, denoted as N ( S A* S B ); S3.7. Determine whether all the source domain radar working mode recognition tasks selected in S3.1 have been selected. If not, return to S3.3. If so, output the final graph neural network model and finally build a model suitable for the target task. T D Network N ( S A* S B* S C* T D ).

[0022] Finally, the trained graph neural network is used to perform radar working mode recognition on the target domain radar working mode recognition task.

[0023] The present invention has been described in detail above with reference to the embodiments. However, the contents described are only preferred embodiments of the present invention and should not be considered as limiting the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A radar operating mode recognition method based on graph transfer learning and graph kernel method, characterized by: It includes the following steps: S1. Use the graph kernel method to extract features from radar working mode pulse data in the source domain and target domain radar working mode recognition tasks to form source domain features and target domain features; S2. Calculate the kernel similarity index of the source and target domains based on the source and target domain features. S3. Based on the size of the graph kernel similarity index, a direct transfer recognition algorithm or an adaptive multi-domain transfer recognition algorithm is selected to perform transfer learning on the graph neural network model, and the trained graph neural network is used to perform radar working mode recognition on the target domain radar working mode recognition task.

2. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 1 is characterized by: The steps of using the graph kernel method to extract features from radar working mode pulse data in the source domain and target domain radar working mode recognition tasks in S1 are: S1.

1. Divide the radar working mode pulse sequence into the following categories according to the time sequence: n units; S1.

2. Use the graph kernel method to find the source domain feature vector and the target domain feature vector. The source domain feature vector is expressed as: S i , i =1,2,…, P , i is the number of the source domain feature vector, P is the number of source domain feature vectors; the target domain feature vector is expressed as: T j , j =1,2,…, Q , j is the number of the source domain feature vector, Q is the number of source domain feature vectors; S1.

3. Integrate the source domain eigenvector and the target domain eigenvector respectively to obtain the source domain eigenvector matrix and the target domain eigenvector matrix.

3. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 1 is characterized by: The graph kernel method used in S1.2 adopts the PMK pyramid matching kernel method. The specific steps of using this method to find the source domain feature vector and the target domain feature vector are as follows: S1.2.

1. Listed in chronological order n Each unit corresponds to each node in the graph neural network, forming a subgraph. Based on the connection or proximity of the nodes, the edges are constructed and the weights of the edges are calculated. The calculation formula is: ， in, ω ab Representation node a and nodes b The weight of the edge between n a , n b Nodes representing the input graph neural network a and nodes b The specific characteristic value of is the scale parameter; S1.2.

2. Construct for each subgraph K Layer pyramid, for the k The feature space of the layer pyramid is divided into A grid, k For which layer, d is the feature dimension of the subgraph; S1.2.

3. Count the number of nodes in the subgraph in the grid and generate a histogram vector , the histogram vector is the corresponding source domain feature vector or target domain feature vector, The calculation formula of the histogram vector is as follows: ， in, c Number the subgraph corresponding to the source domain or target domain; Indicates the k Layer c The eigenvalues ​​in the first grid of the subgraph are A grid.

4. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 2 is characterized by: The calculation formula of the S2 graph kernel similarity index is: ， in, k △ represents the graph kernel similarity index, G S is the source domain radar working mode subgraph, G T is the target domain radar working mode sub-graph, and Respectively represent the radar working mode sub-graphs in the source domain and target domain Histogram of the layer, I is the histogram intersection function, K is the number of layers of the sub-graph pyramid kernel method for the radar working mode in the source domain and the target domain; the expression of the histogram intersection function is: ， in, n is the total number of grids in the graph neural network, i Number the mesh nodes.

5. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 1 is characterized by: The method of selecting direct transfer recognition algorithm or adaptive multi-domain transfer recognition algorithm to transfer learning of graph neural network model based on the size of graph core similarity index is as follows: setting similarity critical value α ,When the graph kernel similarity index is not less than the similarity critical value, the direct transfer recognition algorithm is selected to perform transfer learning on the graph neural network model; When the graph kernel similarity index is less than the similarity critical value, the adaptive multi-domain transfer recognition algorithm is selected to perform transfer learning on the graph neural network model.

6. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 5 is characterized by: When the direct transfer recognition algorithm is selected in S3 to perform transfer learning on the graph neural network model, the specific method is: fix the model structure of the graph neural network model of the selected source domain radar working mode recognition task and the network parameters from the input layer to the feature extraction layer, use the labeled data in the target domain radar working mode recognition task to adaptively train the graph neural network model, and construct a network suitable for the target domain radar working mode recognition task.

7. The radar operating mode recognition method based on graph transfer learning and graph kernel method according to claim 5 is characterized by: When the adaptive multi-domain transfer recognition algorithm is selected in S3 to perform transfer learning on the graph neural network model, the specific steps are as follows: S3.

1. Select at least three source-domain radar operating mode recognition tasks in descending order of image kernel similarity metrics. S3.

2. Select the source domain radar working mode recognition task with the largest graph kernel similarity index to build a graph neural network model. M layer; S3.

3. Select an unselected source domain radar working mode recognition task in descending order of kernel similarity index; S3.

4. The front of the graph neural network model l The parameters of the layer are fixed. l +1 to M Initialize the parameters of the layer, train the graph neural network model using the source domain radar working mode recognition task selected in S3.3, and update the graph neural network model l +1 to M Layer parameters; S3.

5. Bring the source domain radar working mode recognition task selected in S3.3 into the graph neural network model constructed in S3.2 and the graph neural network model updated in S3.4, and compare the recognition rates. If the recognition rate of the updated graph neural network model is better, mark the graph neural network model as the first l The layer is general. If the recognition rate of the constructed graph neural network model is better, the first layer of the labeled graph neural network model l Layers have specificity; S3.

6. Repeat S3.4-S3.5 until the source domain radar working mode recognition task selected in S3.2 is transferred to the source domain radar working mode recognition task selected in S3.

3. M All layers are judged as general or special, the layers marked as special are frozen, the layers marked as general are adaptively trained, and the migrated graph neural network model is updated; S3.

7. Determine whether all the source domain radar working mode recognition tasks selected in S3.1 have been selected. If not, return to S3.

3. If so, output the final graph neural network model.

Images