Radar signal de-interleaving method based on complex network and Laplacian atlas clustering, program, equipment and storage medium

Abstract

The invention belongs to the technical field of radar signal de-interleaving, and particularly relates to a radar signal de-interleaving method based on a complex network and Laplacian atlas clustering. According to the method, radar pulse sequences are constructed into a complex network to mine the pulse parameter space incidence relation of the same radar, the complex network is reduced through a K nearest neighbor algorithm, and the connection strength between pulses from the same radar is enhanced; the number of the clustering centers is determined by determining the position of mutation of the characteristic value spectrum of the Laplacian matrix, so that wrong prediction of the number of the clustering centers is effectively avoided; radar pulse clustering is realized by using Laplacian spectrum decomposition by using a graph segmentation thought. The method does not depend on radar number prior information, can be applied to a complex electromagnetic environment with a large number of overlapped pulses, improves the accuracy of radar signal de-interleaving in the complex electromagnetic environment, and has high robustness.

Description

Technical Field

[0001] The present invention belongs to the technical field of radar signal deinterleaving, and in particular relates to a radar signal deinterleaving method based on complex network and Laplacian graph clustering. Background Art

[0002] Radar signal deinterleaving refers to the process of separating the pulse trains from the same radar from the randomly overlapping pulse trains detected by the receiver. Deinterleaving is a key link in electronic intelligence and electronic measurement systems, and is also an important basis for subsequent work such as positioning and tracking. With the rapid update and iteration of electromagnetic equipment, the widespread application of new radar systems such as low probability of intercept radars and multi-function radars has brought greater challenges to the field of radar signal deinterleaving, making the problems of "increased batches" and "wrong batches" more serious.

[0003] In the current deinterleaving algorithm based on pulse-to-pulse radar signals, it can be divided into single-parameter deinterleaving and multi-parameter deinterleaving according to the number of parameters. In the field of single-parameter deinterleaving, researchers mostly use pulse arrival time (TOA) as the deinterleaving parameter. At present, this type of method is only applicable to the case where the pulse repetition interval (PRI) is fixed and the number of radiation sources is small, and it is difficult to achieve complex radar signal deinterleaving. In the field of multi-parameter deinterleaving, researchers focus on the application and improvement of machine learning methods. Common radar signal multi-parameter deinterleaving algorithms include prototype-based clustering, density-based clustering and deep neural networks. Among them, clustering algorithms are sensitive to the initial cluster center and are limited by the setting of threshold parameters, which makes this type of algorithm less robust when deinterleaving unknown radar signals. Methods based on deep neural networks require a large amount of training data, and the scalability of the model is poor in complex electromagnetic environments, making it difficult to apply in actual scenarios.

[0004] Therefore, the problems of "over-batch" and "wrong batch" faced by the de-interleaving of complex radar signals are the research focus in the field of radar signal de-interleaving. Summary of the invention

[0005] The purpose of the present invention is to provide a radar signal deinterleaving method based on complex network and Laplacian graph clustering to address the problem that the rapid update iteration of complex system radars leads to an increased probability of "increased batches" (signals of the same radar are classified into multiple ones) and "wrong batches" (pulses belonging to a certain radar are wrongly classified into other radars) when deinterleaving radar signals.

[0006] After intercepting the interleaved radar pulse sequence, the present invention constructs the radar pulse sequence into a complex network to mine the spatial correlation of pulse parameters of the same radar, borrows the idea of ​​graph segmentation, and uses Laplace spectral decomposition to realize radar pulse clustering, thereby solving the problem of radar signal deinterleaving in complex electromagnetic environments.

[0007] A radar signal deinterleaving method based on complex network and Laplacian graph clustering includes the following steps:

[0008] Step 1: Obtain the time series X of multiple pulses of the radar signal, calculate the similarity between the pulses of the radar signal using the spatial correlation relationship of the pulses, and construct an adjacency matrix A with diagonal elements being 0 and the remaining elements being similarities;

[0009] Step 2: Construct the Laplacian matrix L corresponding to the adjacency matrix A. According to the graph characteristics of the Laplacian matrix, use the "mutation" of the eigenvalue spectrum to solve the number of cluster centers k, that is, the number of radars;

[0010] Step 3: Take the eigenvectors corresponding to the smallest k eigenvalues ​​in the matrix L to form the feature matrix F; select the first k row vectors of the feature matrix F as the initial clustering centers, use the clustering algorithm to cluster the radar signal pulses in X, output the predicted labels of each radar signal pulse, and realize radar signal deinterleaving.

[0011] Furthermore, the time series X of multiple pulses of the radar signal obtained in step 1 is specifically:

[0012] X={x1,x2,...,x i ,...,x N}

[0013] x i = {PW i ,RF i ,DOA i}

[0014] Among them, x i PW represents the pulse description word vector of the i-th pulse of the radar signal. i Indicates pulse width, RF i Indicates carrier frequency, DOA i represents the arrival time; N is the total number of pulses;

[0015] The specific calculation of the similarity between the pulses of the radar signal is:

[0016]

[0017] Among them, s ij represents the link weight between the i-th radar pulse and the j-th radar pulse, xi ,x j ∈X; c is the cutoff distance in the Gaussian kernel function.

[0018] Furthermore, the adjacency matrix A in step 1 is specifically:

[0019]

[0020] Furthermore, after obtaining the adjacency matrix A in step 1, the adjacency matrix A is simplified, K non-zero elements in the adjacency matrix A are retained, and the remaining elements are all set to zero; the value range of K is between 1% and 2% of the total number of pulses N.

[0021] Furthermore, the Laplacian matrix L corresponding to the adjacency matrix A constructed in step 2 is specifically:

[0022] L=DA

[0023]

[0024] Among them, the Laplacian matrix is ​​a matrix that represents a weighted undirected graph, D is the degree matrix of the weighted undirected graph, and the element D in the matrix D pq Calculated from the degree of each vertex:

[0025]

[0026] Among them, p and q are both pulse vertices, deg(v p ) is the sum of the weights of all adjacent edges of the pulse vertex p.

[0027] Furthermore, the number k of cluster centers in step 2 is calculated as follows:

[0028]

[0029] φ i =log a (1+λ i 2 / λ i+1 λ i-1 )

[0030] where λ1,...,λ N is the eigenvalue spectrum of the matrix L arranged in increasing order, that is, λ1<λ2<…<λ N ; a is an arbitrary constant and a>1.

[0031] Furthermore, in step 3, the smallest k eigenvalues ​​λ1,...,λ of the matrix L are taken. k The corresponding eigenvectors f1,...,f kThe feature matrix F is composed of k ) and use the k-means algorithm to cluster the radar signal pulses in X.

[0032] A computer device / equipment / system includes a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the radar signal deinterleaving method based on complex network and Laplacian graph clustering.

[0033] A computer-readable storage medium stores a computer program / instruction, which, when executed by a processor, implements the steps of the radar signal deinterleaving method based on complex network and Laplacian graph clustering.

[0034] A computer program product includes a computer program / instruction, which, when executed by a processor, implements the steps of the radar signal deinterleaving method based on complex network and Laplacian graph clustering.

[0035] The beneficial effects of the present invention are:

[0036] The present invention constructs radar pulse sequences into complex networks to mine the spatial correlation of pulse parameters of the same radar, simplifies the complex network through the K nearest neighbor algorithm, and strengthens the connection strength between pulses from the same radar; determines the number of cluster centers by determining the position of the "mutation" of the eigenvalue spectrum of the Laplace matrix, effectively avoiding the wrong prediction of the number of cluster centers; borrows the idea of ​​graph segmentation and uses Laplace spectrum decomposition to realize radar pulse clustering. The present invention does not rely on prior information on the number of radars, and can be applied in complex electromagnetic environments with a large number of pulse overlaps, improving the accuracy of radar signal deinterleaving in complex electromagnetic environments, and has high robustness. The present invention can solve the problem of increased probability of "increased batches" and "wrong batches" when radar signals are deinterleaved due to the rapid update iteration of complex radar systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 It is a schematic diagram of the radar signal deinterleaving method based on complex network and Laplacian graph clustering in the present invention.

[0038] Figure 2 It is a complex network visualization diagram.

[0039] Figure 3 It is a graph of deinterleaving accuracy and running time under different numbers of neighbor nodes.

[0040] Figure 4 This is a result diagram of radar signal deinterleaving using the method provided by the present invention.

[0041] Figure 5 This is the radar signal deinterleaving result diagram of the CNet algorithm.

[0042] Figure 6 This is the radar signal deinterleaving result diagram of the SD-ADPC algorithm.

[0043] Figure 7 This is the radar signal deinterlacing result diagram of the k-means algorithm.

[0044] Figure 8 It is a parameter setting data table of each radar radiation source in the embodiment of the present invention.

[0045] Fig. 9 It is a data table of deinterleaving results of different algorithms in the embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is further described below in conjunction with the accompanying drawings.

[0047] Radar signal pre-deinterleaving mainly uses five common pulse descriptors of radar signals as cluster deinterleaving parameters. The present invention sets the pulse arrival time according to PRI, and uses (pulse width, PW) pulse width, (Radio Frequency, RF) carrier frequency, and (direction of arrival, DOA) arrival time as characteristic parameters of the deinterleaved pulse. Given a time sequence containing N pulses as X = {x1, x2, ..., x i ,...,x N}, where x i = {PW i ,RF i ,DOA i} is a 3-dimensional parameter vector. In most cases, the number of radars for signal deinterleaving is unknown, and it is necessary to use a clustering algorithm to directly obtain the cluster centers and their number, and to complete radar signal deinterleaving by allocating all data points according to the defined distance. The final clustering result can be statistically analyzed in the form of each pulse prediction label sequence (dimension N).

[0048] According to the spatial correlation of pulses from the same radar, the present invention proposes a radar signal deinterleaving method based on complex network and Laplacian spectrum clustering. The present invention does not require prior knowledge of the number of radars, utilizes the spatial correlation of pulses, obtains a complex network through nearest neighbor selection, realizes complex network segmentation based on the spectrum characteristics of the Laplacian matrix and the characteristic space spectrum decomposition, and completes the deinterleaving of radar signals in a complex electromagnetic environment. The present invention can effectively avoid the wrong prediction of the number of cluster centers, and has higher deinterleaving accuracy and robustness.

[0049] Figure 1 This is a principle block diagram of the method described in the present invention. The traditional clustering method mainly classifies based on the characteristic distance of the signal, and does not take into account the intrinsic correlation between the signals of the same radiation source. The present invention constructs a complex graph network by calculating the distance between pulse signals, that is, converting the pulse signal correlation into a topological structure composed of a point set V and an edge set E, and mining the spatial correlation relationship of radar pulses through the topological structure, and then completing the radar signal deinterleaving by segmenting the topological structure.

[0050] The present invention calculates the distance d between the pulse signals. ij Similarity s ij Construct an adjacency matrix to represent a complex network. From the adjacency matrix, we can find out which pulse points there are and whether there is an edge connection between two pulse points. The distance and similarity can be expressed as

[0051]

[0052] Among them, x i With x j is the pulse characteristic parameter vector in the received pulse sequence, c is the cutoff distance in the Gaussian kernel function, s ij represents the link weight, i.e., similarity, between the i-th radar pulse and the j-th radar pulse; when the distance between the two pulses is smaller, the similarity between the two pulses is higher, and the weight of the edge between the two points in the constructed complex network is greater. The construction method of the present invention is more interpretable, and the adjacency matrix A constructed according to the similarity can be expressed as:

[0053]

[0054] Among them, the diagonal elements are 0, which means that there is no connection between the radar pulse and itself.

[0055] This embodiment uses Figure 8 The parameters in are used as input, and the complex network visualization results are constructed as follows Figure 2 The degree of overlap of pulse parameters in the complex network is much smaller than the parameter domain of the pulse description word, indicating that the complex network proposed by the present invention can effectively mine the internal correlation relationship of the same radar pulse.

[0056] After obtaining the adjacency matrix, if it is directly used as the input of spectral clustering, the feature vector will have a lot of redundant information. In order to extract key feature information that is more suitable for clustering, the adjacency matrix is ​​simplified before spectral clustering. The empirical value of the number of neighbor connections usually selected is generally between 1% and 2% of the total number of received pulses, which can ensure that pulses with high global similarity are closely connected. If there are too few neighbor nodes, pulses with high similarity may not be completely connected together, resulting in incompleteness of complex networks. Conversely, pulses from multiple different radars may be merged together, causing confusion in the clustering results.

[0057] In this embodiment, K is defined as the maximum number of neighbor nodes in each row of the adjacency matrix, that is, the K connections with the largest weights between each node and other nodes are retained. The results of deinterleaving accuracy and running time under different numbers of neighbor nodes are shown in Figure 3 When K=42, the deinterleaving accuracy is the highest, which is 99.14%, but the time complexity and accuracy are considered comprehensively. Figure 3 It shows that when K is within 1% to 2% of the total number of pulses, the deinterleaving accuracy of the algorithm of the present invention is not sensitive to the selection of K value, and performs well in deinterleaving complex interleaved radar signals, and can effectively suppress the "increased batch" and "wrong batch" problems caused by deinterleaving.

[0058] The present invention is inspired by the segmentation on the graph and uses the idea of ​​graph segmentation to complete the clustering of radar signals. Traditional clustering algorithms generally randomly select cluster centers, and the algorithm performance depends to a large extent on the selection of the number of cluster centers k, and the convergence speed is slow. In order to improve the performance of the clustering algorithm, the initial number of cluster centers of the present invention can be solved based on the graph spectrum characteristics of the Laplacian matrix. The Laplacian matrix, also known as the Kirchhoff matrix, is a matrix representing a weighted undirected graph. The Laplacian matrix corresponding to the complex network of the present invention and its standard form are:

[0059] L=DA

[0060]

[0061] Where D is the degree matrix of the graph, which is calculated from the degree of each vertex and can be expressed as follows:

[0062]

[0063] Among them, p and q are both pulse vertices, deg(v p ) is the sum of the weights of all adjacent edges of the pulse vertex p.

[0064] When L is composed of several parts, its eigenvalue spectrum will have a "mutation" near 0, that is, the eigenvalue spectrum arranged in increasing order λ1<λ2<…<λ NThe eigenvalue near this "mutation" will increase significantly, and the solution to the number of cluster centers can be achieved by the following formula:

[0065]

[0066] Among them, φ i =log a (1+λ i 2 / λ i+1 λ i-1 )(a is an arbitrary constant and a>1).

[0067] From the perspective of graph theory, clustering is the problem of graph segmentation. Clustering is equivalent to finding an optimization algorithm to segment the graph G into several subgraphs (corresponding to clusters or classes in clustering), so that the feature weights or similarities between subgraphs are small, while the feature weights or similarities within a subgraph are large.

[0068] For any real column vector In the present invention, f represents an eigenvector corresponding to any eigenvalue of the Laplacian matrix corresponding to the adjacency matrix constructed by the radar pulse sequence, and f' is the transpose of f. According to the Laplacian matrix L=DA, the following formula holds:

[0069] f'Lf=f'Df-f'Af

[0070] The RatioCut algorithm is used to solve the graph segmentation problem by finding the minimum sum of the feature weights of the edges. n}, then the objective function of RatioCut is:

[0071]

[0072] Among them, G i is the i-th subgraph of G, G i The complement of pq is the weight between vertex p and vertex q.

[0073] At the same time, the pth sample f of vector f p It can be defined as:

[0074]

[0075] Substituting the adjacency matrix A and the degree matrix D into the equation, we get:

[0076]

[0077] Then f p Substituting into the above formula, we can get:

[0078]

[0079] Where |V| is a constant representing the number of vertices in the graph. As can be seen from the above formula, minimizing the objective function of RatioCut is equivalent to minimizing f'Lf. Let λ be the eigenvalue of L, and f be the eigenvector corresponding to the eigenvalue λ, then Lf=λf, and multiplying both sides of the equation by f' on the left yields:

[0080] f'Lf=λf'f

[0081] Where f is a column vector, then the value of f'f is a real number, so minimizing f'Lf is equivalent to minimizing λ. We only need to solve the minimum eigenvalue of L and its corresponding eigenvector.

[0082] In summary, subgraph segmentation of the pulse sequence complex network G can be transformed into a spectral decomposition problem for solving the Laplacian matrix. By solving the eigenvalues ​​of the Laplacian matrix and its corresponding eigenvector, the goal of Laplacian graph clustering can be obtained. That is, mutually related vertices are as close as possible in the eigenspace of the Laplacian matrix; mutually unrelated vertices are as far away as possible in space.

[0083] The feature matrix F can be composed of the eigenvectors corresponding to the smallest k eigenvalues ​​of the Laplacian matrix. Each row in F can be regarded as a sample in the k-dimensional feature space. Sort the eigenvalues ​​obtained by L by λ1<λ2<…<λ k <…<λ N ,λ k The corresponding eigenvector is f k , then the feature matrix F can be expressed as

[0084] F=(f1,f2,…,f k )

[0085] Furthermore, the present invention uses the k-means algorithm to cluster the samples. The specific operation of the k-means algorithm is: select the first k row vectors of F as the initial cluster center; calculate the distance from the remaining sample points to the cluster center and assign them to the nearest class; if the class to which the sample belongs changes, update the cluster center point of the class; finally return k classes, that is, the predicted label of each pulse, wherein the convergence test of the algorithm uses the square sum criterion function.

[0086] The present invention constructs a complex network by utilizing the global spatial correlation of samples, and simplifies the complex network by using the K nearest neighbor algorithm, thereby strengthening the connection strength between pulses from the same radar and reducing the correlation between different radar pulses; the present invention determines the number of cluster centers by determining the position of the "mutation" of the eigenvalue spectrum of the Laplace matrix, thereby effectively avoiding the erroneous prediction of the number of cluster centers; inspired by the segmentation on the graph, the present invention applies the Laplacian spectral clustering algorithm to the field of radar signal deinterleaving, and has higher deinterleaving accuracy and robustness than the current multi-parameter radar signal deinterleaving method.

[0087] Based on the above content, the present invention provides a radar signal deinterleaving method based on complex network and Laplacian graph clustering, comprising the following steps:

[0088] Step 1: Obtain the time series of multiple pulses of radar signal X = {x1, x2, ..., x i ,...,x N}, using the spatial correlation relationship of the pulses, calculate the similarity s between the pulses of the radar signal ij , construct an adjacency matrix A with diagonal elements of 0 and the remaining elements as similarities;

[0089]

[0090] Step 2: Simplify the adjacency matrix A, retain K non-zero elements in the adjacency matrix A, and set all other elements to zero; the value range of K is between 1% and 2% of the total number of pulses N;

[0091] Step 3: Construct the Laplacian matrix L corresponding to the adjacency matrix A. According to the graph characteristics of the Laplacian matrix, use the "mutation" of the eigenvalue spectrum to solve the number of cluster centers k, that is, the number of radars;

[0092]

[0093] φ i =log a (1+λ i 2 / λ i+1 λ i-1 )

[0094] Step 4: Take the smallest k eigenvalues ​​λ1,...,λ of the matrix L k The corresponding eigenvectors f1,...,f k The feature matrix F is composed of k); select the first k row vectors of the feature matrix F as the initial clustering centers, use the k-means algorithm to cluster the radar signal pulses in X, output the predicted labels of each radar signal pulse, and realize radar signal deinterleaving.

[0095] The subgraph segmentation of the pulse sequence complex network G by the present invention can be converted into a spectral decomposition problem for solving the Laplacian matrix. By solving the eigenvalues ​​of the Laplacian matrix and its corresponding eigenvectors, the goal of Laplacian graph clustering can be obtained. That is, mutually related vertices are as close as possible in the characteristic space of the Laplacian matrix; mutually unrelated vertices are as far away as possible in space. The characteristic matrix F can be composed of eigenvectors corresponding to the smallest k eigenvalues ​​of the Laplacian matrix. Each row in F can be regarded as a sample in the k-dimensional feature space. The present invention uses the k-means algorithm to cluster the samples.

[0096] Embodiment 1:

[0097] In order to further verify the effectiveness of the algorithm in the present invention in dealing with the "increased batches" and "wrong batches" problems faced in radar signal deinterleaving, this embodiment uses simulation experiments to illustrate the process and demonstrate the effects, and selects PW, RF, and DOA of four radars as experimental simulation parameters. Radar 2 and Radar 4 are multi-function radars, and each radar has three working modes. Different radars or different working modes of the same radar will have overlapping parameters, and are closer to the actual receiver receiving signals. PW and RF have multiple modulation modes, including agile, staggered, and sliding. For specific parameter settings, see Figure 8 .

[0098] Compare the latest methods CNet, SD-ADPC and traditional algorithm k-means in recent years. Figure 8 The data in is used as the input of four de-interlacing algorithms, and the de-interlacing results with accuracy and adjusted Rand index (ARI) as evaluation indicators are shown in Fig. 9 As shown in the figure, the best deinterlacing accuracy of the algorithm of the present invention reaches 99.88%, and the ARI reaches 99.63%. The overall accuracy is much higher than that of the comparison algorithm and is most consistent with the real label. In order to further intuitively display the deinterlacing results of different algorithms, the t-SNE algorithm is used to map high-dimensional data points to two-dimensional space, and the data distribution results after visualizing the deinterlacing of radar signals are shown in the figure. Figures 4 to 7 Please note that the dimensionality reduction visualization method makes Figures 4 to 7 There is no parameter overlap in the data in order to more intuitively display the error pulse point. Figure 8 From the data in , we can see that the received radar pulses contain a lot of parameter overlap. Figures 4 to 7It can be seen that the error rate of the algorithm of the present invention in the clustering results of radar 2 and radar 4 is much lower than that of the comparison algorithm, and the clustering results of the four radars are relatively "clean". Because firstly, the algorithm of the present invention constructs a complex network through spatial correlation relationships, and pays more attention to the correlation relationship of each pulse in the global context. Secondly, the algorithm of the present invention uses the spectral clustering method to segment the complex topological structure to complete the clustering, which is more conducive to extracting the intrinsic characteristics of the radar pulse signal, while the SD-ADPC algorithm will lose feature information in the process of dimensionality reduction, resulting in a decrease in the deinterleaving accuracy. Therefore, the algorithm of the present invention can have a better performance in the deinterleaving problem of complex interwoven radar signals.

[0099] This embodiment models the intercepted radar pulse sequence in a complex electromagnetic environment as a complex network, and then uses the nearest neighbor algorithm to simplify the complex network to reduce the network complexity; on this basis, according to the graph characteristics of the Laplacian matrix, the "mutation" of the eigenvalue spectrum is used to solve the number of cluster centers, and finally the Laplacian eigenspace spectrum is used to decompose the complex network to complete the deinterleaving of radar signals in a complex electromagnetic environment. The method applied can still effectively sort out pulses from different radars when facing radar signals with severely overlapping parameters in a complex electromagnetic environment, and can achieve a higher deinterleaving accuracy rate.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A radar signal deinterlacing method based on complex network and Laplacian graph clustering, characterized in that: The following steps are involved: Step 1: Obtain the time series X of multiple pulses of the radar signal, calculate the similarity between the pulses of the radar signal using the spatial correlation relationship of the pulses, and construct an adjacency matrix A with diagonal elements being 0 and the remaining elements being similarities; Step 2: Construct the Laplacian matrix L corresponding to the adjacency matrix A. According to the graph characteristics of the Laplacian matrix, use the "mutation" of the eigenvalue spectrum to solve the number of cluster centers k, that is, the number of radars; Step 3: Take the eigenvectors corresponding to the smallest k eigenvalues ​​in the matrix L to form the characteristic matrix F; The first k row vectors of the feature matrix F are selected as the initial clustering centers, and the radar signal pulses in X are clustered using the clustering algorithm. The predicted labels of each radar signal pulse are output to achieve radar signal deinterleaving.

2. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 1, characterized in that: The time series X of multiple pulses of the radar signal obtained in step 1 is specifically: X={x1,x2,…,x i ,…,x N } x i ={PW i ,RF i ,DOA i } Among them, x i PW represents the pulse description word vector of the i-th pulse of the radar signal. i Indicates pulse width, RF i Indicates carrier frequency, DOA i represents the arrival time; N is the total number of pulses; The specific calculation of the similarity between the pulses of the radar signal is: Among them, s ij represents the link weight, i.e., similarity, between the i-th radar pulse and the j-th radar pulse, x i ,x j ∈X; c is the cutoff distance in the Gaussian kernel function.

3. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 2, characterized in that: The adjacency matrix A in step 1 is specifically:

4. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 3 is characterized by: After the adjacency matrix A is obtained in step 1, the adjacency matrix A is simplified, K non-zero elements in the adjacency matrix A are retained, and the remaining elements are all set to zero; the value range of K is between 1% and 2% of the total number of pulses N.

5. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 1, characterized in that: The Laplacian matrix L corresponding to the adjacency matrix A constructed in step 2 is specifically: L=DA Among them, the Laplacian matrix is ​​a matrix that represents a weighted undirected graph, D is the degree matrix of the weighted undirected graph, and the element D in the matrix D pq Calculated from the degree of each vertex: Among them, p and q are both pulse vertices, deg(v p ) is the sum of the weights of all adjacent edges of the pulse vertex p.

6. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 1, characterized in that: The calculation method of the number k of cluster centers in step 2 is: f i =log a (1+λ i 2 / l i+1 l i-1 ) where λ1,...,λ N is the eigenvalue spectrum of the matrix L arranged in increasing order, that is, λ1<λ2<…<λ N ; a is an arbitrary constant and a>1.

7. The radar signal deinterleaving method based on complex network and Laplacian graph clustering according to claim 6, characterized in that: In step 3, the smallest k eigenvalues ​​λ1,...,λ of the matrix L are taken. k The corresponding eigenvectors f1,...,f k The feature matrix F is composed of k ) and use the k-means algorithm to cluster the radar signal pulses in X.

8. A computer device / equipment / system comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer program product comprising a computer program / instructions, characterized in that: When the computer program / instructions are executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images